Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Economic aspects of obesity
(USC Thesis Other)
Economic aspects of obesity
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
ECONOMIC ASPECTS OF OBESITY
by
Andrew Messali, PharmD
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(PHARMACEUTICAL ECONOMICS AND POLICY)
May 2015
ii
DEDICATION
To my parents, whose guidance and sacrifices have shaped most aspects of my life.
To my sister, for her encouragement and support.
To my fiancé, whose love and support continues to make the future look better each day.
iii
ACKNOWLEDGEMENTS
My sincere thanks to my advisor and dissertation chair, Dr. Jason Doctor, for his consistent
support and good advice. Our regular discussions have been a primary source of academic
inspiration. I would also like to thank my other committee members, Dr. Neeraj Sood, Dr. Sarah-
Jeanne Salvy, Dr. Geoffrey Joyce, and Dr. Michael Goran. Finally, my thanks to several other
individuals whose professional guidance has been invaluable, including Dr. Joel Hay, Dr. Jeff
McCombs, Dr. Sepideh Varon, Dr. Michael Nichol, and Dr. Aniket Kawatkar.
iv
TABLE OF CONTENTS
DEDICATION ii
ACKNOWLEDGEMENTS iii
LIST OF TABLES vii
LIST OF FIGURES viii
ABSTRACT 1
CHAPTER 1: INTRODUCTION 3
Obesity Trend and Causal Theories 3
Health and Healthcare Costs of Obesity 5
Decision Analytic Models Applied to Obesity 6
Childhood Obesity, Teacher Evaluation, and
Academic Performance 8
Objectives of the Dissertation 9
References 11
CHAPTER 2: SIMULATION OF ADULT MEDICAL
COSTS AFTER BEHAVIORAL INTERVENTION
TO REDUCE ADOLESCENT OBESITY 15
Introduction 16
Methods 18
Model Structure and Transition Probabilities 18
Healthcare Costs 19
Population, Treatment Effect, and Time Horizon 20
v
Scenarios After Treatment 21
Model Validation 21
Results 22
Discussion 24
References 28
CHAPTER 3: THE EFFECT OF CHILDHOOD OBESITY ON
TEACHER EVALUATIONS IN GRADES K-8 40
Introduction 41
Methods 43
Data Source 43
Measures of Weight Status and Academic
Performance 44
Model Specification 45
Results 47
Discussion 48
Conclusions 51
References 52
CHAPTER 4: DO STATINS REDUCE THE HEALTH AND
HEALTHCARE COSTS OF OBESITY? 59
Introduction 60
Methods 62
The Future Elderly Model 62
Simulations 66
vi
Isolating the Impact of Obesity 69
Results 70
Effects of Statins on Life Expectancy and
Functional Status 70
Effects of Statins on Healthcare Costs 71
Cost-Effectiveness of Statins 73
Effects of Statins on the Costs of Obesity 74
Sensitivity Analyses 75
Discussion 77
References 81
Appendix A: Assigning Statin Use in the Future Elderly
Model 92
Data 92
Estimation 93
Statin Assignment 93
Appendix B: Effects of Statin Use on Health Dynamics and
Obesity in the Future Elderly Model 94
Appendix C: Sensitivity Analyses 95
Running Simulation Without Reassigning BMI 96
Simulating Only the Secondary Prevention
Effect of Statins 97
CHAPTER 5: CONCLUSIONS 108
vii
LIST OF TABLES
Table 2.1: Predicted Healthcare Costs Among Men
(Ages 18-80) 32
Table 2.2: Predicted Healthcare Costs Among Women
(Ages 18-80) 33
Table 3.1: Descriptive Statistics for ECLS-K Sample 55
Table 3.2: Longitudinal Models of Standardized Math Teacher
Evaluations and Test Scores 57
Table 3.3: Longitudinal Models of Standardized English
Teacher Evaluations and Test Scores 58
Table 4.1: Health Impacts of Statins 84
Table 4.2: Impact of Statins on Lifetime Medical Costs after
Age 50, $2009 thousands 85
Table 4.3: Cost per Health Gain of Statins 86
Table 4.4: Health and Healthcare Costs of Obesity in Both
Scenarios, $2009 thousands 87
Table 4.A1: Factors Influencing the Probability of Purchasing
Statins in MEPS, 2009 to 2011: Probit Estimates 99
Table 4.A2: 95% Confidence Interval of the Impact of Statins on
Lifetime Medical Costs after Age 50, $2009
thousands 100
Table 4.A3: Sensitivity of Impact of Statins on Life Expectancy
viii
and Expected QALYs after Age 50 to Alternate
Simulation Approaches 101
Table 4.A4: Sensitivity of Health and Healthcare Costs of Obesity
In Both Scenarios for Class 1 Obesity, $2009
thousands 102
ix
LIST OF FIGURES
Figure 2.1: Conceptual Markov Model Design 34
Figure 2.2: BMI Distribution Among Simulated Untreated
Cohorts and NHANES Respondents 35
Figure 2.3: Survival Curves of Simulated Untreated Cohorts and
2008 CDC Life Table Cohorts 36
Figure 2.4: Predicted Mean Annual Healthcare Costs by BMI
Among Gender and Age Groups 37
Figure 2.5: Predicted Prevalence of Obesity (BMI ≥ 30) Among
Men (Ages 18-80) 38
Figure 2.6: Predicted Prevalence of Obesity (BMI ≥ 30) Among
Women (Ages 18-80) 39
Figure 3.1: Cross-Sectional Effect of Child BMI on Standardized
Math Teacher Evaluations and Test Scores 56
Figure 3.2: Cross-Sectional Effect of Child BMI on Standardized
English Teacher Evaluations and Test Scores 56
Figure 4.1: Chronic Conditions Transitions in the Future Elderly
Model 88
Figure 4.2: Empirical Strategy 89
Figure 4.3: Life Expectancy Cost of Obesity in Both Scenarios
After Age 50 90
Figure 4.4: Difference-in-difference in Life Expectancy Gains
x
After Age 50, by Functional Status 91
Figure 4.A1: American Adults Purchasing Statins 103
Figure 4.A2: Use of Statins by Age and Obesity Status 104
Figure 4.A3: Comparison of Statin Use at Ages 51-52 in FEM
Simulations and MEPS Data 105
Figure 4.A4: Pathways Between Obesity and Selected Health
Outcomes in the FEM 106
Figure 4.A5: Sensitivity of Cost of Obesity to Alternative Simulation
Approaches 107
1
ABSTRACT
Obesity is widely recognized as one of the greatest modern public health challenges by
the US Surgeon General, The World Health Organization, and The Organization for Economic
Cooperation and Development. In 2013, the American Medical Association officially recognized
obesity as a disease in its own right, a decision that has generated some controversy. Despite
more than a century of research, new health risks and adverse economic outcomes associated
with obesity are still being discovered. Meanwhile, interventions that produce lasting cost-
effective results remain elusive. This dissertation attempts to contribute to our understanding of
the cost-effectiveness of obesity interventions, the impact of obesity on human capital
development, and the role that medical innovation can play in mitigating these costs.
Chapter two describes the use of a novel decision analytic model to estimate potential
reductions in medical expenditures after overweight adolescents participate in behavioral
interventions designed to reduce their body mass index. The model is designed and validated
using data from multiple sources and allows for the simulation of lifetime weight trajectories and
associated healthcare costs. Chapter three focuses on obesity in earlier childhood. Previous
research has indicated that childhood obesity may be associated with worse academic outcomes.
One theory asserts that teachers may have negative opinions and reduced expectations of their
obese or overweight students. This hypothesis is tested using a longitudinal dataset containing
standardized test scores and subjective teacher evaluations of students in grades K-8. Finally,
chapter four looks towards the future of obesity among the elderly. Given the difficulties in
adherence to diet and exercise interventions, some may be inclined to hope that medical
innovation will eventually mitigate the marginal cost of obesity by effectively treating all or most
2
of its associated complications. Using a previously developed dynamic model, the impact of the
development of statins (an example of medical innovation) on the marginal cost of obesity is
examined.
3
CHAPTER 1:
INTRODUCTION
1.1 Obesity Trend and Causal Theories
Since 1960, the height of an average American adult has increased by approximately 1
inch, while their weight has increased by over 24 pounds.[1, 2] However, the most commonly
used measure of adiposity in medical literature is the body mass index (BMI), which is equal to
an individual’s mass in kilograms divided by the square of their height in meters. The World
Health Organization (WHO) considers a BMI ≤ 18.5, 18.5 – 25, 25 – 30, 30 – 35, 35 – 40, and ≥
40 to be underweight, normal weight, and obese category I, II, or III, respectively.[3] Since 1960,
the prevalence of obesity among American adults has more than doubled. Perhaps even more
concerning is that the prevalence of class III obesity has increased more than six-fold. As of
2010, 36.1% of American adults were obese and 6.6% belonged to the class III obesity group
(BMI ≥ 40).[1] This trend of increasing adiposity is occurring in nearly every demographic
subgroup of Americans including men and women, most racial groups, young children and the
elderly.[4]
The trend of increasing adiposity among children and adolescents has caused particular
concern. In the last three decades alone, the prevalence of obesity among children and
adolescents (aged 2-19 years) has more than tripled, from 5.5% in 1980 to 16.9% in 2010.[5, 6]
In children and adolescents, obesity classification is done by comparison of a particular BMI to a
reference population. The most commonly used standards in the United States (US) are the 2000
BMI-for-age growth charts for boys and girls produced by the US Centers for Disease Control
and Prevention (CDC).[7, 8] A child or adolescent ≤ 19 years old is classified as underweight,
4
normal weight, overweight, or obese if they have a BMI ≤ 5
th
percentile, 5
th
– 85
th
percentile, 85
th
– 95
th
percentile, or ≥ 95
th
percentile, respectively.
The proximate causes of these obesity trends surely include increased caloric intake,
decreased caloric expenditure, or both. However, several theories of the ultimate causative
factors have been proposed. Philipson and Posner discuss how economic development has
caused many individuals to leave more physically demanding labor and seek employment in less
physically demanding service jobs, effectively raising the price of caloric expenditure.[9]
Meanwhile, technological innovation in agriculture has contributed to consistent reductions in
the real price of food. In a related paper, Lakdawalla and Philipson estimate that reductions in
caloric expenditure and increasing caloric intake may have accounted for roughly 60% and 40%
of the observed BMI increase over the last few decades, respectively.[10]
However, Cutler, Glaeser, and Shapiro point out that, while occupational shifts into less
strenuous jobs have certainly contributed to weight gain over the past century, this process had
significantly slowed by 1980, when obesity prevalence began to increase at its fastest pace.[11]
They stress the importance of caloric intake, particularly the frequency of snack consumption, as
the most important factor driving more recent obesity trends. Specifically, they point to
technological innovations such as controlled atmosphere transportation, hydrogen-peroxide
sterilization, “flavor barrier” chemicals, polyethylene plastics for storage and transportation, and
the microwave, which all became widely available in the 1970s and 1980s and significantly
reduced the supplier’s cost of delivering food and the consumer’s time cost of preparing food.
Other theories with mixed empirical support include the proliferation of passive
entertainment and communications technologies[12], urban sprawl and personal
transportation[13, 14], and regional lack of access to “quality” food[15, 16]. When considering
5
these proposed theories it is important to remember that relative increase in obesity rates appear
to have been very similar across demographic subgroups.[4] Although at any given time point
the prevalence of obesity is highest among lower income, lower education, and minority groups,
the BMI growth that these groups have experienced in recent decades has been similar to other
groups. This, combined with the similar BMI growth seen in most other developed countries[17],
makes any theory that affects certain subgroups disproportionately an unlikely explanation for
broader obesity trends.
1.2 Health and Healthcare Costs of Obesity
Childhood and adolescent obesity are associated with an increased risk of type II
diabetes, hypertension, hypercholesterolemia, orthopedic disease, sleep apnea, and asthma.[18-
21] There are also important costs related to the social stigma of obesity. Obese children have
more conduct, hyperactivity, and peer relationship problems.[22, 23] Obese adolescents also
report lower self-esteem and more depressive symptoms than peers with a normal BMI.[24] It
has been estimated that obese children incur an additional $220 (2006 USD) per year in direct
medical expenditures than their normal BMI peers, on average.[25]
However, the greatest cost of childhood obesity may very well be the increased risk of
adult obesity, a phenomenon commonly referred to as tracking. An obese 18-year old has a 68-
77% chance of being obese at age 35, more than four times the risk faced by an 18-year old with
a normal BMI.[26] Obesity in adulthood is associated with an increased risk of type II diabetes,
hypertension, hypercholesterolemia, heart attack, stroke, non-alcoholic fatty liver disease,
osteoarthritis, and several types of cancer.[3] The marginal cost of obesity in adulthood has
received considerably more attention among researchers, compared to childhood obesity. On
6
average, obese adults incur an additional $1,429 (2006 USD) per year in direct medical
expenditures, a 41.5% increase over their normal-BMI peers.[27] In aggregate, this represented
about $146.6 billion, or 9.1% of all medical spending in the US in 2006.
While the relationship between adult obesity and mortality has at times been described as
controversial[28], the most recent meta-analysis of 97 studies and over 2.8 million individuals
found that increased mortality risk was only associated with BMI ≥ 35 (class II or III
obesity).[29] Earlier studies had found larger effects of more moderate obesity on mortality risk.
Some have pointed out that many of the earlier studies did not adequately adjust for confounding
factors, such as smoking status, which is negatively correlated with obesity and positively
correlated with mortality risk. It may also be the case that the true effect of obesity on mortality
risk has declined in recent years as medical innovation, particularly related to the treatment of
cardiovascular disease, has mitigated some of the health risks associated with obesity.
1.3 Decision Analytic Models Applied to Obesity
Markov models are decision analysis tools that are particularly useful when a decision
problem involves continuous risk over a defined time period and when the timing of events is
important.[30] The simplest decision analytic model, the decision tree, often represents these
situations poorly. The applicability of Markov models to medical applications was first described
in 1983.[31] Since then, Markov models have quickly become one of the most common tools for
analysis of medical decision making and cost-effectiveness analysis in medicine.
Markov models assume that disease processes can be represented using a series of
discrete states of health, the probabilities of transitioning between those states of health over a
given time period, and the consequences of existing in those states of health for a given time
7
period. The length of each discrete time period is usually referred to as the cycle length.
Additionally, it is assumed that the probability of transitioning into each health state at the
beginning of each new cycle depends on an individual’s current health state, but does not depend
on an individual’s health state in previous cycles. This feature, sometimes referred to as the
“Markov assumption,” does not prohibit the dependence of transition probabilities on other risk
factors.
The accurate representation of obesity with Markov models is complicated, primarily
because selection of the most appropriate health states is difficult. The exact relationship
between obesity and some diseases often lacks consensus. It is also likely that some obesity-
related health risks remain undiscovered. Even if all obesity-associated diseases were known,
they can’t be simply described as discrete health states because they are not mutually exclusive.
All possible combinations of these diseases would need to be included as separate health states
so that all states could be considered mutually exclusive. Chapter two describes an attempt to
simplify the application of Markov models to obesity by using BMI ranges, rather than obesity-
associated diseases, as discrete health states. This novel Markov model is used to predict the
lifetime healthcare expenditures of obese adolescents.
More advanced decision analytic models abandon the Markov assumption and allow
parameters to depend on the previous experience of the simulated individual. For example,
transition probabilities can depend not only on the current health state an individual occupies, but
on their previous states of health as well. These dynamic models are especially useful when the
duration spent within health states or the path each individual takes to a given health state is
important. Chapter four describes the use of a previously developed dynamic model of the
8
health, wealth and longevity of elderly Americans to study the impact of medical innovation on
the marginal cost of obesity.
1.4 Childhood Obesity, Teacher Evaluation, and Academic Performance
There exists a growing literature investigating the relationship between childhood obesity
and academic performance. Academic performance is generally recognized as an important
indicator of human capital development. Even small reductions in academic performance early in
life could conceivably have very large economic consequences later in life. Although several
studies have noted an association between obesity and various measures of academic
performance, such as standardized exam scores, the causal nature of this relationship continues
to be debated.[32-35] Obese children are more likely to come from households of lower
socioeconomic status (SES)[36], which is also associated with lower performance on many of
these tests.[37] Obesity may also be an indicator of micronutrient deficiencies or less physical
activity, which may affect energy and concentration during school and while studying.[38, 39]
Obesity may be associated with increased health-related school absenteeism.[40] Peer bullying
may contribute to depression and low self-esteem, which may be associated with lower academic
performance.[39] Finally, discrimination from teachers may result in lower expectations or
encouragement.[41]
The results of studies that have attempted to identify the causal effect of obesity on
academic performance have been mixed. A 2011 review of 29 studies founded that, while the
findings of individual studies often conflicted, the overall pattern of findings suggested a small
but statistically significant negative impact of obesity on educational performance.[42] However,
9
several studies did not control for readily observable confounding factors, such as SES, and
many more did not address the issue of unobservable confounding factors.
The potential for discrimination among teachers against obese students has received
considerably less attention in the literature. It has recently been shown that childhood obesity is
associated with lower teacher evaluations of math, science, and English abilities.[32] However,
given the previously discussed evidence that obese students may indeed perform worse in these
subjects, lower teacher evaluations alone cannot be assumed to indicate discrimination. For this
reason we attempt to compare the effect of childhood obesity on teacher evaluations to its effect
on standardized test scores. A larger effect on the former, compared to the later, would represent
stronger evidence of discrimination. After the decades of research psychologists have devoted to
the effects of teacher expectations on future academic performance[43], childhood obesity’s
effect on teacher evaluations deserves more consideration. Chapter three describes the use of a
rich longitudinal dataset containing both teacher evaluations and standardized test scores of
children in grades K-8 to better isolate the effect of child obesity on both.
1.5 Objectives of the Dissertation
The impacts of the obesity epidemic on individual and public health cannot be overstated.
This dissertation will contribute to our understanding of the cost-effectiveness of obesity
prevention, the effect of obesity on human capital development, and the role of medical
innovation in treating the complications of obesity. Chapter two describes the development a
novel Markov model to evaluate the cost-effectiveness of behavioral interventions to reduce
adolescent obesity. This model is used to predict healthcare expenditures throughout life under
several hypothetical scenarios where the effectiveness of such programs is varied. Chapter three
10
is devoted to the analysis of a longitudinal dataset containing several years of child standardized
exam scores, subjective teacher evaluations, and other pertinent variables. The effects of
childhood obesity on teachers’ evaluations of their students’ abilities and their standardized test
performance are investigated. Finally, chapter four considers the development of statin
medications as an example of medical innovation aimed at treating obesity-related illnesses and
uses a previously development dynamic model to estimate the effect of statin development on the
marginal cost of obesity.
11
References
1. Fryar CD CM, Ogden CL. Prevalence of obesity among adults: United States, trends
1963-1965 through 2009-2010: National Center for Health Statistics; 2012.
2. Flegal KM, Carroll MD, Kit BK, Ogden CL. Prevalence of obesity and trends in the
distribution of body mass index among US adults, 1999-2010. JAMA : the journal of the
American Medical Association. 2012 Feb 1;307(5):491-7.
3. James WPT J-LR, Mhurchu CN, Kalamara E, Shayeghi M, Rigby NJ, Nishida C,
Rodgers A. Chapter 8: Overweight and Obesity (High Body Mass Index). In: Ezzati M LA,
Rodgers A, Murray CJL, editor. Comparative Quantification of Health Risks. Geneva,
Switzerland: World Health Organization; 2004.
4. Sturm R, An R. Obesity and economic environments. CA: a cancer journal for clinicians.
2014 May 22.
5. Fryar CD CM, Ogden CL. Prevalence of obesity among children and adolescents: United
States, trends 1963-1965 through 2009-2010: National Center for Health Statistics; 2012.
6. Flegal KM, Ogden CL, Yanovski JA, Freedman DS, Shepherd JA, Graubard BI, et al.
High adiposity and high body mass index-for-age in US children and adolescents overall and by
race-ethnic group. The American journal of clinical nutrition. 2010 Apr;91(4):1020-6.
7. Grummer-Strawn LM RC, Krebs NF. Use of World Health Organization and CDC
growth charts for children aged 0-59 months in the United States. Atlanta, GA: Centers for
Disease Control and Prevention; 2010.
8. Flegal KM, Ogden CL. Childhood obesity: are we all speaking the same language?
Advances in nutrition. 2011 Mar;2(2):159S-66S.
9. Philipson T PR. The long-run growth in obesity as a functional of technological change.
In: Research NBoE, editor. NBER Working Paper Series Cambridge, MA1999.
10. Lakdawalla D PT. The growth of obesity and technological change. Economics and
Human Biology. 2009;7(3):283-93.
11. Cutler DM GE, Shapiro JM. Why have Americans become more obese? Journal of
Economic Perspectives. 2003;17(3):93-118.
12. Thorp AA, Owen N, Neuhaus M, Dunstan DW. Sedentary behaviors and subsequent
health outcomes in adults a systematic review of longitudinal studies, 1996-2011. American
journal of preventive medicine. 2011 Aug;41(2):207-15.
12
13. Feng J, Glass TA, Curriero FC, Stewart WF, Schwartz BS. The built environment and
obesity: a systematic review of the epidemiologic evidence. Health & place. 2010
Mar;16(2):175-90.
14. Durand CP, Andalib M, Dunton GF, Wolch J, Pentz MA. A systematic review of built
environment factors related to physical activity and obesity risk: implications for smart growth
urban planning. Obesity reviews : an official journal of the International Association for the
Study of Obesity. 2011 May;12(5):e173-82.
15. Caspi CE, Sorensen G, Subramanian SV, Kawachi I. The local food environment and
diet: a systematic review. Health & place. 2012 Sep;18(5):1172-87.
16. Cummins S, Flint E, Matthews SA. New neighborhood grocery store increased awareness
of food access but did not alter dietary habits or obesity. Health affairs. 2014 Feb;33(2):283-91.
17. Cecchini M SF. Obesity Update 2014: Organization for Economic Cooperation and
Development; 2014.
18. Freedman DS, Mei Z, Srinivasan SR, Berenson GS, Dietz WH. Cardiovascular risk
factors and excess adiposity among overweight children and adolescents: the Bogalusa Heart
Study. The Journal of pediatrics. 2007 Jan;150(1):12-7 e2.
19. Li C, Ford ES, Zhao G, Mokdad AH. Prevalence of pre-diabetes and its association with
clustering of cardiometabolic risk factors and hyperinsulinemia among U.S. adolescents:
National Health and Nutrition Examination Survey 2005-2006. Diabetes care. 2009
Feb;32(2):342-7.
20. Gilliland FD, Berhane K, Islam T, McConnell R, Gauderman WJ, Gilliland SS, et al.
Obesity and the risk of newly diagnosed asthma in school-age children. American journal of
epidemiology. 2003 Sep 1;158(5):406-15.
21. Taylor ED, Theim KR, Mirch MC, Ghorbani S, Tanofsky-Kraff M, Adler-Wailes DC, et
al. Orthopedic complications of overweight in children and adolescents. Pediatrics. 2006
Jun;117(6):2167-74.
22. Cornette R. The emotional impact of obesity on children. Worldviews on evidence-based
nursing / Sigma Theta Tau International, Honor Society of Nursing. 2008;5(3):136-41.
23. Schwartz MB, Puhl R. Childhood obesity: a societal problem to solve. Obesity reviews :
an official journal of the International Association for the Study of Obesity. 2003 Feb;4(1):57-
71.
13
24. Griffiths LJ DC, Hill A. Is obesity associated with emotional and behavioral problems in
children? Findings from the Millennium Cohort Study. International Journal of Pediatric
Obesity. 2011;6:e423-e32.
25. Finkelstein EA TJ. Public health interventions for addressing childhood overweight:
Analysis of the business case. Health Policy and Ethics. 2007;98(3):411-5.
26. Guo SS, Wu W, Chumlea WC, Roche AF. Predicting overweight and obesity in
adulthood from body mass index values in childhood and adolescence. The American journal of
clinical nutrition. 2002 Sep;76(3):653-8.
27. Finkelstein EA, Trogdon JG, Cohen JW, Dietz W. Annual medical spending attributable
to obesity: payer-and service-specific estimates. Health affairs. 2009 Sep-Oct;28(5):w822-31.
28. Adams KF, Schatzkin A, Harris TB, Kipnis V, Mouw T, Ballard-Barbash R, et al.
Overweight, obesity, and mortality in a large prospective cohort of persons 50 to 71 years old.
The New England journal of medicine. 2006 Aug 24;355(8):763-78.
29. Flegal KM, Kit BK, Orpana H, Graubard BI. Association of all-cause mortality with
overweight and obesity using standard body mass index categories: a systematic review and
meta-analysis. JAMA : the journal of the American Medical Association. 2013 Jan 2;309(1):71-
82.
30. Sonnenberg FA, Beck JR. Markov models in medical decision making: a practical guide.
Medical decision making : an international journal of the Society for Medical Decision Making.
1993 Oct-Dec;13(4):322-38.
31. Beck JR, Pauker SG. The Markov process in medical prognosis. Medical decision
making : an international journal of the Society for Medical Decision Making. 1983;3(4):419-58.
32. Zavodny M. Does weight affect children's test scores and teacher assessments
differently? Economics of education review. 2013 Jun;34:135-45.
33. Shore SM, Sachs ML, Lidicker JR, Brett SN, Wright AR, Libonati JR. Decreased
scholastic achievement in overweight middle school students. Obesity. 2008 Jul;16(7):1535-8.
34. Gable S, Krull JL, Chang Y. Boys' and girls' weight status and math performance from
kindergarten entry through fifth grade: a mediated analysis. Child development. 2012 Sep-
Oct;83(5):1822-39.
35. Clark D SJ, Viglietti GC. Children's Weight and Academic Performance in Elementary
School: Cause for Concern? Analyses of Social and Public Policy. 2009;9(1):185-204.
14
36. Wilkinson R MM. Copenhagen, Denmark: World Health Organization; 2003.
37. SR S. Socioeconomics status and academic achievement: A meta-analytic review of
research. Review of Educational Research. 2005;75(3):417-53.
38. Averett SL SD. Food for thought: The cognitive effects of childhood malnutrition in the
United States. 2007.
39. Wang F, Veugelers PJ. Self-esteem and cognitive development in the era of the
childhood obesity epidemic. Obesity reviews : an official journal of the International Association
for the Study of Obesity. 2008 Nov;9(6):615-23.
40. Taras H, Potts-Datema W. Obesity and student performance at school. The Journal of
school health. 2005 Oct;75(8):291-5.
41. JJ S. The effect of body weight on adolescent academic performance. Southern Economic
Journal. 2007;73(4):871-900.
42. Caird J KJ, Oliver K, Oliver S, O'Mara A, Stansfield C, Thomas J. Childhood obesity and
educational attainment. London, United Kingdom: University of London; 2011. Report No.:
1901.
43. R R. Interpersonal expectancy effects: A 30-year perspective. Current Directions in
Psychological Science. 1994;3:176-9.
15
CHAPTER 2:
SIMULATION OF ADULT MEDICAL COSTS AFTER BEHAVIORAL
INTERVENTION TO REDUCE ADOLESCENT OBESITY
Andrew J Messali
*1
Jason N Doctor
1
June 2014
Abstract
Since 1980, the prevalence of obesity among adolescents (aged 12-19 years) has quadrupled.
Given the strong tracking of obesity from adolescence into adulthood, the current prevalence of
obesity among adults and the associated healthcare costs are likely remain high without
significant intervention. Behavioral interventions have been the primary method of treating
adolescent obesity. Using recently published literature evaluating the effectiveness of these
programs, as well as data from the 2004 – 2010 rounds of the Medical Expenditure Panel Survey
(MEPS), we have constructed a simulation model to predict BMI trajectories and medical
expenditures from ages 18 to 80 after overweight adolescents participate in these behavioral
interventions. We model several potential BMI trajectories to represent adoption of varying
degrees of healthy behavior change that may result in varying levels of obesity risk reduction.
When no healthy behavior changes are included in the model, we find between $1,288 and
$1,783 (2013 US$) in reduced adult medical expenditures per child. After incorporating various
degrees of healthy behavior change, we find between $5,924 and $25,528 (2013 US$) of savings
per child. Savings were significantly greater for women compared to men. We show that
significant reductions in lifetime medical costs can make these programs cost-effective if they
lower lifetime obesity risk by as little as 25%.
Keywords: obesity, weight, body mass index, behavior, habits, healthcare costs
* Corresponding author. Email: messali@usc.edu
1 Leonard D. Schaeffer Center for Health Policy and Economics, University of Southern
California, Los Angeles, USA
16
2.1 Introduction
The past three decades have witnessed a dramatic rise in the prevalence of both child and
adolescent obesity in the US. In 1980, the prevalence of obesity among children aged 6-11 years
and adolescents aged 12-19 years were 7% and 5%, respectively. By 2012, obesity among
children had doubled (18%) and obesity among adolescents had quadrupled (21%).[1-3] Given
the strong tracking of childhood obesity in adulthood, the prevalence of obesity among US adults
(currently 35%) seems likely to remain high without significant intervention.[1, 4]
Among children and adolescents, overweight is defined as a body mass index (BMI) ≥
the 85
th
percentile of sex-specific BMI-for-age charts developed by the Centers for Disease
Control and Prevention (CDC).[1, 5] Childhood and adolescent obesity is defined as BMI ≥ the
95
th
percentile.[1, 5] Among adults, overweight is defined as a BMI ≥ 25.5 and obesity is
generally categorized as class I (BMI 30-34.9), class II (BMI 35-39.9), or class III (BMI ≥
40).[1]
During childhood and adolescence, obesity is associated with increased risk of type II
diabetes, hypertension, hypercholesterolemia, orthopedic disease, and asthma.[6-9] Obese
adolescents also report low self-esteem and other depressive symptoms more than their non-
obese peers.[10, 11] After an obese child becomes an obese adult, even more significant medical
costs are incurred. Obesity in adulthood is known to increase the risk of type II diabetes,
hypertension, dyslipidemia, heart attack, stroke, non-alcoholic fatty liver disease, osteoarthritis,
and several types of cancer.[12] In one commonly cited study, after adjustment for demographic
factors, economic factors, and smoking status, the obese incurred and average of $1,400 (42%)
more medical costs than the non-obese in 2006.[13] In aggregate, this represented $147 billion,
or 9.1% of all US healthcare spending in 2006.
17
The majority of efforts to treat childhood or adolescent obesity rely on behavioral
interventions. Behavioral interventions are primarily aimed at modifying diet, increasing
physical activity, and reducing sedentary behaviors. They can be administered in a variety of
settings by a variety of healthcare professionals. These programs are commonly administered in
a clinic setting. Recent published reviews of the effectiveness of behavioral interventions for
childhood/adolescent obesity treatment have concluded that these programs are effective means
of reducing child/adolescent BMI over the short term.[14, 15] Long-run outcomes are less clear.
There have been published reports of successful childhood weight loss maintained up to 10 years
after conclusion of the intervention.[16] However, most of the clinical and economic costs of
obesity do not manifest until well beyond that point.[17]
In the absence of long-term data, simulation models are often used to predict long-term
outcomes and/or evaluate the cost-effectiveness of healthcare interventions and policies. Because
of their simplicity, Markov models are perhaps the most common models used to conduct these
analyses. Because Markov models require a set of mutually exclusive health states, typically the
stages of the disease being modeled or a limited set of comorbid conditions, it is difficult to
model obesity as a Markov process. The complications of obesity are numerous, not fully
understood, and not mutually exclusive. To avoid these problems, we have developed a Markov
model of BMI states, rather than disease states. This model projects individual BMI trajectories
from age 18 to 80 and associated healthcare costs for a cohort of 18 year olds with a given BMI
distribution. We use this model to predict the healthcare costs of adolescents that participate in
behavioral interventions to treat or prevent adolescent obesity and those that do not. From these
predictions we find the potential economic benefit of such interventions.
18
2.2 Methods
2.2.1 Model Structure and Transition Probabilities
The model defines 27 mutually exclusive BMI states. Death serves as a collecting state.
The BMI states range from <15 to ≥ 40, by increments of 1 kg/m
2
(Figure 1). All possible
transitions between each BMI state are permitted, however transitions out of the death state are
not permitted.
Transition probabilities were calculated using data from the Medical Expenditure Panel
Survey (MEPS). MEPS is a two-year repeating panel survey representative of the non-
institutionalized US population conducted by the Agency for Healthcare Research and Quality
(AHRQ). Each panel follows a cohort for two years and asks each participant to self-report his or
her height and weight twice, one year apart. We used six MEPS panels (2004-2010) and
weighted each by its sample size. In total, all six panels included 95,262 respondents. Individuals
< 18 years of age and individuals that left the survey scope for any other reason besides death
were excluded. This left a pooled total of 65,787 individuals, with an average of 10,964
individuals per panel. Self-reported height and weight were adjusted using previously published
estimates of age- and sex-specific differences between self-reported and measured height and
weight prior to calculation of BMI.[18] An average of 805 (7.3%) individuals per panel required
the imputation of a missing BMI value using the average one-year change in BMI among the
respondents without missing data.
Transition probabilities were calculated by determining the proportion of individuals
within a given BMI state in year one that was found to be in each BMI state in year two.
Similarly, the probability of transitioning from any BMI state into the death state was calculated
using proportion of individuals in each BMI state in year one that had died before their second
19
BMI report. To allow the transition probabilities to be age- and sex-specific, six different
transition matrices were defined – males aged 18-40, males aged 41-60, males aged >60, females
aged 18-40, females aged 41-60, and females aged >60. Calculation of transition probabilities
was conducted in SAS using appropriate survey procedures to account for the MEPS sampling
strategy.[19]
2.2.2 Healthcare costs
Healthcare costs were first modeled using a two-stage generalized linear model, similar
to Finkelstein et al.[13] This approach utilizes a logistic regression model to predict the
likelihood that each individual has nonzero healthcare cost and a generalized linear model with a
gamma distribution and log link function to predict the value of healthcare costs. This
specification is intended to mitigate the zero-mass and skewness problems commonly found in
healthcare cost data.[20] Both models included dummies for each BMI state, age, gender, race,
census region, education, marital status, income, insurance coverage, and smoking status as
independent variables. BMI was entered into the regression models using binary dummy
variables, each representing a range of 1 kg/m
2
, to allow for a possible non-linear relationship
between BMI and healthcare costs.
Each BMI state was assigned a gamma distribution of healthcare costs. The parameters of
these gamma distributions were calculated using the predicted mean and standard deviation of
total healthcare expenditures from the regression model, stratified by BMI state. Similar to the
transition probabilities, weighted averages from the six MEPS panels were used and the cost
distributions are also age- and sex-specific. At the beginning of each iteration of the simulation
(each individual within the cohort), a new cost value was sampled for each BMI state.
20
Resampling healthcare costs associated with each BMI state on each iteration is intended to
reflect individual-level heterogeneity.
Consistent with budget projections from the Congressional Budget Office (CBO), the
model assumes excess growth in healthcare spending, beyond per-capita growth in gross
domestic product (GDP), of 1.5% per year.[21] All results are discounted 3% annually to 2013
US dollars. Analysis of healthcare expenditures was done in STATA, also using appropriate
survey procedures to account for the design of the MEPS.[22]
2.2.3 Population, Treatment Effect, and Time Horizon
The simulated population was a hypothetical cohort of 1000 individuals considered
overweight at age 18. The initial BMI distribution of this cohort was derived from the observed
BMI distribution of all 18 year olds in the 2004-2010 MEPS data with a BMI ≥ the 85
th
percentile of sex-specific BMI-for-age. This population is intended to be representative of
adolescents that have been referred to an obesity treatment behavioral intervention program,
likely administered in a clinic setting. The first treatment effect of behavioral intervention is
implemented as a shift in the BMI state distribution at the start of the simulation. Behavioral
intervention participants are given a new BMI reduced by 1.45 kg/m
2
at the start of the
simulation. This estimate of weight loss over the course of treatment was taken from a recently
published meta-analysis.[14] The models run for 62 cycles, each representing one year, and is
censored when the cohort is 80 years old. Execution of the simulation was conducted using
TreeAGE.[23]
21
2.2.4 Scenarios After Treatment
After a behavioral intervention concludes, several things can happen. Some participants
will quickly regain any weight they’ve lost. Among these individuals, any reductions in medical
costs during the period of time shortly after treatment can be safely ignored. We therefore
assume that, all else equal, the lifetime healthcare costs of these individuals are equal to those
individuals who do not participate in the behavioral intervention. Other participants may
maintain the weight lost over the course of the intervention, but fail to adopt any new healthy
habits that might alter their BMI trajectory throughout later life. Finally, some participants might
be able to maintain their initial weight loss and adopt healthy habits that will alter their future
BMI trajectory in a favorable way. In the model, this is operationalized by adjustment of the
BMI state transition probabilities. We model twelve possible degrees of healthy habit adoption:
25%, 50%, 75%, and 100% risk reduction of class III obesity (BMI ≥ 40), class II and III obesity
(BMI ≥ 35), and class I, II, and III obesity (BMI ≥ 30).
2.2.5 Model Validation
The model is validated in two ways. First, the simulation was run for general population
male and female non-participants and the BMI state distributions at age 30 and 50 were
compared to an external dataset, the National Health and Nutrition Examination Survey
(NHANES) (Figure 2). To ensure a reliable sample size, NHANES data from 2005-2010 were
pooled (weighted by panel sample size) and respondents 25-35 and 45-55 years old were used
for comparison. Data from the NHANES was not used at any point in development of the model,
but the target population is identical to the MEPS. Therefore, if the BMI state transition
probabilities were calculated correctly, the BMI state distributions at each age should be similar.
22
Second, the survival curves of general population male and female non-participants were
compared to survival curves obtained from the US CDC representative of the entire 2008 US
population (Figure 4).[24] The survival curves assess how accurately the transition probabilities
governing the risk of death were calculated.
2.3 Results
BMI state distributions among simulated non-participants and NHANES respondents
were well matched among both men and women at age 30 and 50 (Figure 2). Among females,
the largest difference was found among 50 year olds expected to have a BMI between 28.0 and
28.9. The simulation predicted that 7.9% of 50 year old women would fall into this category, but
only 5.1% of female NHANES respondents 45-55 years old had such a BMI. Among males, the
largest difference occurred among 30 year olds expected to have a BMI between 23.0 and 23.9.
The simulation predicted that 9.7% of 30 year old men would have such a BMI, but only 8.2% of
males NHANES respondents 25-35 years old did so. The survival curves of simulated non-
participants were very similar to survival curves generated from CDC life tables, until age ≥ 60,
at which point simulated men and women had a lower death hazard compared to the CDC life
tables (Figure 3). This discrepancy resulted in a roughly 10 percentage point increase in the
proportion of individuals alive at age 80, compared to the general US population.
Predicted annual healthcare costs generally increased with BMI among men and women
of all age groups (Figure 4). However, among individuals ≥ 60 years old, being underweight
(BMI ≥ 18.5) was also associated with significantly higher healthcare costs. Among both men
and women, predicted annual healthcare spending increased with age. At all ages, predicted
annual healthcare spending among women was greater than among men.
23
Figure 5 and Figure 6 display the predicted prevalence of obesity (BMI ≥ 30) among the
simulated cohort of men and women, respectively, as they age. All cohorts that participate in the
behavioral intervention start with a lower prevalence of obesity at the age of 18, compared to
non-participants, because they all received the initial 1.45 kg/m
2
reduction in BMI. However,
when no further alterations were made to transition probabilities (participants did not adopt any
new healthy habits that alter their future risk of obesity), the population prevalence of obesity
steadily and becomes indistinguishable from the population that was never treated by age 45.
When participants do adopt healthy habits reducing their risk of future obesity the population
prevalence is substantially reduced.
Predicted healthcare costs among men between the ages of 18 and 80 are presented in
Table 1. Men that do not take part in the behavioral interventions spend an average of $367,525
on healthcare throughout these 62 years of life. Men that take part in the intervention and are
able to maintain their 1.45 kg/m
2
BMI reduction spend an average of $1,288 less on healthcare as
a result of this weight loss in adolescence. Adoption of healthy behaviors that result in 25% -
100% risk reduction in BMI ≥ 40, BMI ≥ 35, and BMI ≥ 30 are associated with healthcare cost
savings of $5,924 - $20,909.
Predicted healthcare costs among women between the ages of 18 and 80 are presented in
Table 2. Women that do not take part in the behavioral interventions spend an average of
$459,337 on healthcare throughout these 62 years of life. Women that take part in the
intervention and are able to maintain their 1.45 kg/m
2
BMI reduction spend an average of $1,783
less on healthcare as a result of this weight loss in adolescence. Adoption of healthy behaviors
that result in 25% - 100% risk reduction in BMI ≥ 40, BMI ≥ 35, and BMI ≥ 30 are associated
with healthcare cost savings of $7,419 - $25,528.
24
2.4 Discussion
As one might suspect, behavioral interventions that help adolescents avoid obesity later
in life appear to produce significant reductions in healthcare spending. Weight loss in
adolescence without the adoption of healthy habits that would continue to reduce the risk of
obesity in later life was associated with much smaller reductions in healthcare costs that were not
statistically significant. Therefore, it seems that behavioral interventions administered in
adolescence should be primarily concerned with the teaching of healthy habits and the
reinforcement of these habits so that they might last as long as possible. Such efforts may include
nutrition education, physical education, or education aimed at reducing specific sedentary
behaviors, such as television viewing.
It is well known that individuals tend to form habits in these behaviors. Economists have
historically described this habitual behavior using rational addiction models.[25, 26] In these
models, greater consumption in the current time period increases consumption in later time
periods (reinforcement), while simultaneously reducing the marginal utility of consumption in
later time periods (tolerance). Consumers are assumed to weigh the benefits of current
consumption against the costs of future addiction. Empirical research has found evidence that
total calories[27], carbohydrates[28], and carbonated soft drinks[29] can be rationally addictive.
Psychologists have described habits as actions that are taken automatically in the
presence of contextual cues.[30] Like economists, psychologists emphasize the role that previous
experience can have on reinforcing future decisions. It is also widely agreed among
psychologists that many behaviors related to nutrition[31] and physical activity[32] are habitual.
However, the focus on external cues, which is generally ignored by economists, may prove to be
25
a useful component of effective behavioral interventions. After a behavior is repeated in the
same context it becomes associated with that context and subsequent repetition of the behavior
requires less conscious attention and motivation.[33, 34] A randomized and controlled trial
evaluating weight loss advice based on a psychological habit formation model is underway.[35]
In this study, patients are being advised to integrate recommended weight loss behaviors within
the context of daily routines.
Epstein et al. have published what we believe is the longest follow-up of participants in
randomized trials evaluating behavioral interventions for the treatment of pediatric obesity.[16]
In 2007, they reported that 44.4% of their intent-to-treat subjects had maintained ≥ 1 point
reduction in BMI Z-score 10 years after treatment. The behavioral interventions pioneered by
Epstein et al. were very similar to the behavioral interventions included in the Ho et al. meta-
analysis, which informed the treatment effect implemented in our model.[14] However, the
Epstein et al. participants were around 10 years of age.[16] It may be the case that younger
children are more likely to remember and continue practicing learned healthy habits compared to
adolescents. Although the work of Epstein et al. can’t be directly translated to our evaluation of
programs for adolescents, it demonstrates that long-term success is possible.
Because we did not explicitly model the cost of administering the behavioral intervention
programs, our estimates of reduced medical costs can be interpreted as break-even points for
cost-effectiveness. The cost of administering these programs is rarely published. However, one
recent report of the administrative costs of a four-month clinic-based behavioral intervention in
rural areas found that the total cost per child was $872.[36] Some of these costs, such as training
provided to post-doctoral psychologists that led groups during nutrition and exercise education,
were fixed. If more adolescents participated these fixed costs would not increase and the
26
resulting cost per child would be reduced. However, it may not be feasible to rely entirely on
post-doctoral graduate students (paid $20/hour in this study) to conduct larger programs. Despite
this, our models suggest that investments of this size can be easily recouped if the intervention
reduces lifetime obesity risk by as little as 25%.
Understanding the potential economic benefits of adolescent obesity treatment and
prevention is becoming increasingly important as their coverage by public and private payers
expands. The Early and Periodic Screening, Diagnostic, and Treatment (EPSDT) benefit covers
obesity screening and counseling for children under 21 enrolled in state Medicaid programs.[37]
Lack of awareness of this benefit among patients and providers has historically limited its
uptake.[38, 39] However, some states have developed additional programs specifically aimed at
screening and/or treating child or adolescent obesity among young Medicaid enrollees. The
United States Preventive Services Task Force has recently given a grade B recommendation to
“comprehensive, intensive, behavioral interventions to improve weight status in children 6 and
older.”[40] Under the Patient Protection and Affordable Care Act, such interventions must now
be covered without cost sharing.
This research is subject to several important limitations. First, BMI state transitions are
modeled as a Markov process. This mean that an individual that recently became obese samples
an annual healthcare spending value from the same distribution as an individual that has been
obese for many years. This is a consequence of the lack of memory in Markov processes.
However, it should be noted that the probability of making large gains/losses in BMI over short
periods of time is small. Consistent with reality, the majority of simulated individuals that
become obese do so slowly and gradually. Second, the model was censored after 62 cycles, when
the simulated cohorts were 80 years old. Because the MEPS is representative of the non-
27
institutionalized US population, the simulated cohorts become less representative of the broader
US population after age ≥ 60. By censoring the simulation before everyone had died, we are
implicitly assuming that healthcare costs beyond the age of 80 are not different among
behavioral intervention participants and non-participants. While this is probably not entirely true,
a relatively small proportion of the lifetime healthcare costs attributable to obesity occur after the
age of 65.[17] Finally, any omitted variables the regression of healthcare costs could potentially
bias our estimates of the relationship between healthcare costs and obesity. However, it has been
shown that a large portion of the differences in costs between obese and non-obese adults can be
directly linked to diseases that are known to be associated with obesity.[41]
28
References
1. Ogden CL, Carroll MD, Kit BK, Flegal KM. Prevalence of childhood and adult obesity in
the United States, 2011-2012. JAMA : the journal of the American Medical Association. 2014
Feb 26;311(8):806-14.
2. Health, United States, 2011: With Special Features on Socioeconomic Status and Health.
Hyattsville, MD: National Center for Health Statistics; 2012.
3. Flegal KM, Ogden CL, Yanovski JA, Freedman DS, Shepherd JA, Graubard BI, et al.
High adiposity and high body mass index-for-age in US children and adolescents overall and by
race-ethnic group. The American journal of clinical nutrition. 2010 Apr;91(4):1020-6.
4. Guo SS, Wu W, Chumlea WC, Roche AF. Predicting overweight and obesity in
adulthood from body mass index values in childhood and adolescence. The American journal of
clinical nutrition. 2002 Sep;76(3):653-8.
5. Flegal KM, Ogden CL. Childhood obesity: are we all speaking the same language?
Advances in nutrition. 2011 Mar;2(2):159S-66S.
6. Freedman DS, Mei Z, Srinivasan SR, Berenson GS, Dietz WH. Cardiovascular risk
factors and excess adiposity among overweight children and adolescents: the Bogalusa Heart
Study. The Journal of pediatrics. 2007 Jan;150(1):12-7 e2.
7. Li C, Ford ES, Zhao G, Mokdad AH. Prevalence of pre-diabetes and its association with
clustering of cardiometabolic risk factors and hyperinsulinemia among U.S. adolescents:
National Health and Nutrition Examination Survey 2005-2006. Diabetes care. 2009
Feb;32(2):342-7.
8. Gilliland FD, Berhane K, Islam T, McConnell R, Gauderman WJ, Gilliland SS, et al.
Obesity and the risk of newly diagnosed asthma in school-age children. American journal of
epidemiology. 2003 Sep 1;158(5):406-15.
9. Taylor ED, Theim KR, Mirch MC, Ghorbani S, Tanofsky-Kraff M, Adler-Wailes DC, et
al. Orthopedic complications of overweight in children and adolescents. Pediatrics. 2006
Jun;117(6):2167-74.
10. Cornette R. The emotional impact of obesity on children. Worldviews on evidence-based
nursing / Sigma Theta Tau International, Honor Society of Nursing. 2008;5(3):136-41.
29
11. Schwartz MB, Puhl R. Childhood obesity: a societal problem to solve. Obesity reviews :
an official journal of the International Association for the Study of Obesity. 2003 Feb;4(1):57-
71.
12. Macroeconomics and Health: Investing in Health for Economic Development. Geneva,
Switzerland: World Health Organization; 2001 2001.
13. Finkelstein EA, Trogdon JG, Cohen JW, Dietz W. Annual medical spending attributable
to obesity: payer-and service-specific estimates. Health affairs. 2009 Sep-Oct;28(5):w822-31.
14. Ho M, Garnett SP, Baur L, Burrows T, Stewart L, Neve M, et al. Effectiveness of
lifestyle interventions in child obesity: systematic review with meta-analysis. Pediatrics. 2012
Dec;130(6):e1647-71.
15. Wang Y WY, Wilson RF, Bleich S, Cheskin L, Weston C, Showell N, Fawole O, Lau B,
Segal J. Childhood Obesity Prevention Programs: Comparative Effectiveness Review and Meta-
Analysis. Comparative Effectiveness Review No. 115. Rockville, MD: Agency for Healthcare
Research and Quality; 2013.
16. Epstein LH, Paluch RA, Roemmich JN, Beecher MD. Family-based obesity treatment,
then and now: twenty-five years of pediatric obesity treatment. Health psychology : official
journal of the Division of Health Psychology, American Psychological Association. 2007
Jul;26(4):381-91.
17. Finkelstein EA, Trogdon JG, Brown DS, Allaire BT, Dellea PS, Kamal-Bahl SJ. The
lifetime medical cost burden of overweight and obesity: implications for obesity prevention.
Obesity. 2008 Aug;16(8):1843-8.
18. Merrill RM, Richardson JS. Validity of self-reported height, weight, and body mass
index: findings from the National Health and Nutrition Examination Survey, 2001-2006.
Preventing chronic disease. 2009 Oct;6(4):A121.
19. SAS Institute. The SAS System for Windows. Release 9.2. Cary, NC: The SAS Institute;
2011.
20. Manning WG, Basu A, Mullahy J. Generalized modeling approaches to risk adjustment
of skewed outcomes data. Journal of health economics. 2005 May;24(3):465-88.
21. The 2013 Long-Term Budget Outlook: The Congressional Budget Office; 2013.
22. StataCorp. Stata Statistical Software: Release 12. College Station, TX: Statacorp LP;
2011.
30
23. TreeAge Software. TreeAge Pro 2012. Williamstown, MA: TreeAge Software, Inc.;
2012.
24. E A. United States Life Tables, 2008. Hyattsville, MD: National Center for Health
Statistics; 2012.
25. Becker GS MK. A Theory of Rational Addiction. Journal of Political Economy.
1988;96(4):675-700.
26. Gruber J KB. Is Addiction "Rational"? Theory and Evidence. Quarterly Journal of
Economics. 2001;116(4):1261-303.
27. Cawley J. Rational Addiction, the Consumption of Calories and Body Weight: University
of Chicago; 1999.
28. Richards T PP, Tegene A. Obesity and Nutrient Consumption: A Rational Addiction?
Contemporary Economic Policy. 2007;25:309-24.
29. Liu X LR. Evidence of rational addiction to carbonated soft drinks? China Agricultural
Economic Review. 2012;4(3):300-17.
30. Neal DT WW, Labrecque JS, Lally P. How do habits guide behavior? Perceived and
actual triggers of habits in daily life. Journal of Experimental Social Psychology. 2012;48:492-8.
31. van Riet J SS, Dagevos H, de Brujin G-J. The importance of habits in eating behaviour.
An overview and recommendations for future research. Appetite. 2011;57:585-96.
32. Verplanken B MO. Predicting habit: The case of physical exercise. Psychology of Sport
and Exercise. 2008;9:15-26.
33. Lally P WJ, Gardner B. Experiences of habit formation: a qualitative study. Psychology
of Health and medicine. 2011;16(4):484-9.
34. Gardner B dBG, Lally P. A systematic review and meta-analysis of applications of the
Self-Report Habit Index to nutrition and physical activity behaviors. Annals of Behavioral
Medicine. 2011;42(2):174-87.
35. Beeken RJ CH, Morris S, Leurent B, Omar R, Nazareth I, Wardle J. Study protocol for
the 10 top tips (10TT) trial: Randomised controlled trial of habit-based advice for weight control
in general practice. BMC Public Health. 2012;12:667.
36. Janicke DM, Sallinen BJ, Perri MG, Lutes LD, Silverstein JH, Brumback B. Comparison
of program costs for parent-only and family-based interventions for pediatric obesity in
medically underserved rural settings. The Journal of rural health : official journal of the
31
American Rural Health Association and the National Rural Health Care Association. 2009
Summer;25(3):326-30.
37. Preventive and obesity-related services available to Medicaid enrollees: Department of
Health and Human Services; 2010.
38. Lee JS, Sheer JL, Lopez N, Rosenbaum S. Coverage of obesity treatment: a state-by-state
analysis of Medicaid and state insurance laws. Public health reports. 2010 Jul-Aug;125(4):596-
604.
39. Wilensky S WR, Rosenbaum S. Strategies for improving access to comprehensive
obesity prevention and treatment services for Medicaid-enrolled children: George Washington
University Department of Health Policy and the Robert Wood Johnson Foundation; 2006.
40. Force USPST, Barton M. Screening for obesity in children and adolescents: US
Preventive Services Task Force recommendation statement. Pediatrics. 2010 Feb;125(2):361-7.
41. How Does Obesity in Adults Affect Spending on Health Care?: The Congressional
Budget Office; 2010.
32
Table 2.1 – Predicted Healthcare Costs Among Men (Ages 18-80)
Mean 95% CI (Mean) Difference 95% CI (Difference)
Untreated 364,025 361,143 366,412 Reference Reference
No Lasting Effects 362,737 360,007 365,464 1,288 -2,711 4,556
Avoid BMI ≥ 40
25% Reduction 357,601 354,976 360,229 6,425 2,542 10,163
50% Reduction 356,578 353,536 359,313 7,448 3,934 11,518
75% Reduction 352,778 350,175 355,391 11,247 7,389 14,856
100% Reduction 351,524 348,911 354,165 12,501 8,742 16,299
Avoid BMI ≥ 35
25% Reduction 356,376 353,697 358,824 7,650 3,866 11,111
50% Reduction 356,382 353,780 359,118 7,643 3,711 11,184
75% Reduction 351,745 349,003 354,174 12,280 8,658 15,931
100% Reduction 350,960 348,000 353,330 13,065 9,547 16,792
Avoid BMI ≥ 30
25% Reduction 355,274 352,268 357,727 8,751 4,885 12,132
50% Reduction 352,997 350,524 355,341 11,028 7,593 14,737
75% Reduction 351,593 349,047 354,426 12,432 8,578 16,264
100% Reduction 346,616 344,217 348,789 17,409 15,100 19,906
SE – Standard Error. All costs are discounted 3% annually to 2013 US dollars. “Difference” represents reduced healthcare
costs relative to individuals that do not participate in the behavioral intervention.
33
Table 2.2 – Predicted Healthcare Costs Among Women
Mean 95% CI (Mean) Difference 95% CI (Difference)
Untreated 454,837 450,133 458,595 Reference Reference
No Lasting Effects 453,054 449,026 456,978 1,783 -4,117 5,803
Avoid BMI ≥ 40
25% Reduction 447,915 443,022 452,467 6,922 2,673 12,339
50% Reduction 444,928 440,783 449,012 9,908 4,096 15,645
75% Reduction 440,210 435,735 444,229 14,627 8,792 20,524
100% Reduction 441,142 437,538 444,745 13,694 7,924 18,904
Avoid BMI ≥ 35
25% Reduction 444,933 440,600 448,984 9,904 3,595 15,798
50% Reduction 441,578 436,842 445,507 13,258 6,856 19,135
75% Reduction 442,898 439,380 446,664 11,939 5,878 17,269
100% Reduction 434,353 430,317 437,916 20,484 15,431 26,562
Avoid BMI ≥ 30
25% Reduction 442,024 436,732 446,283 12,812 6,548 19,249
50% Reduction 441,731 437,425 445,235 13,105 7,547 18,883
75% Reduction 437,456 433,615 440,590 17,381 11,962 22,808
100% Reduction 433,809 430,359 437,028 21,028 16,797 25,184
SE – Standard Error. All costs are discounted 3% annually to 2013 US dollars. “Difference” represents reduced healthcare
costs relative to individuals that do not participate in the behavioral intervention.
34
Figure 2.1 – Conceptual Markov Model Design
BMI < 15 15 ≤ BMI < 16 16 ≤ BMI < 17 17 ≤ BMI < 18
BMI ≥ 40
Death
35
Figure 2.2 – BMI Distribution among Simulated Untreated Cohorts and NHANES Respondents
0
2
4
6
8
10
12
<15 16 18 20 22 24 26 28 30 32 34 36 38 ≥40
Percent (%)
BMI State
Men Age 30
0
2
4
6
8
10
12
<15 16 18 20 22 24 26 28 30 32 34 36 38 ≥40
Percent (%)
BMI State
Men Age 50
0
1
2
3
4
5
6
7
8
9
10
<15 16 18 20 22 24 26 28 30 32 34 36 38 ≥40
Percent (%)
BMI State
Women Age 30
0
1
2
3
4
5
6
7
8
9
10
<15 16 18 20 22 24 26 28 30 32 34 36 38 ≥40
Percent (%)
BMI State
Women Age 50
36
Figure 2.3 – Survival Curves of Simulated Untreated Cohorts and 2008 CDC Life Table Cohorts
0.5
0.6
0.7
0.8
0.9
1
15 25 35 45 55 65 75 85
Percent Survived (%)
Age (years)
Men
0.5
0.6
0.7
0.8
0.9
1
15 25 35 45 55 65 75 85
Percent Survived (%)
Age (years)
Women
37
Figure 2.4 – Predicted Mean Annual Healthcare Costs by BMI Among Gender and Age Groups
0
2000
4000
6000
8000
10000
12000
14000
16000
≤18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 ≥40
2013 US Dollars ($)
BMI State
Males 18-40 Males 41-60 Males 61+ Females 18-40 Females 41-60 Females 61+
38
Figure 2.5– Predicted Prevalence of Obesity (BMI ≥ 30) Among Men (Ages 18-80)
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
15 25 35 45 55 65 75 85
Untreated No Healthy Habits 25% RR Reduction
50% RR Reduction 75% RR Reduction 100% RR Reduction
39
Figure 2.6 – Predicted Prevalence of Obesity (BMI ≥ 30) Among Women (Ages 18-80)
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
50%
15 25 35 45 55 65 75 85
Untreated No Healthy Habits 25% RR Reduction
50% RR Reduction 75% RR Reduction 100% RR Reduction
40
CHAPTER 3:
THE EFFECT OF CHILDHOOD OBESITY ON TEACHER EVALUATIONS IN
GRADES K-8
Andrew J Messali
*1
Jason N Doctor
1
Neeraj Sood
1
September 2014
Abstract
Over the past four decades, the prevalence of obesity among children (aged 2-11) has tripled. In
addition to the health consequences of early obesity, growing research has indicated that it may
be associated with worse educational outcomes. It remains undetermined whether such an effect
is the result of physiological conditions (nutrient deficiencies, health-related absenteeism) or
social conditions (discrimination from peers or teachers). Given the documented effect of teacher
expectations on educational outcomes and the pervasiveness of negative biases against the obese,
it is important to understand the effect that childhood obesity may have on how teachers evaluate
their students’ abilities. We utilize data from a longitudinal survey of children, beginning in
kindergarten and ending in the 8
th
grade. Children repeatedly took standardized tests to measure
their English and math abilities and teachers were repeatedly asked to rate each child’s English
and math abilities. The dataset also contains height, weight, and other important child, family,
and school variables. We estimate several regression models to find the effect of childhood
obesity on objective test scores and subjective teacher evaluations. We also investigate
differences in these effects between schools with high vs. low average student BMI. We find that
childhood obesity has negative effects on English and math teacher evaluations, but not on test
scores. Furthermore, this obesity penalty is strongest in schools with lower average student BMI.
Keywords: obesity, weight, body mass index, child, education, discrimination
* Corresponding author. Email: messali@usc.edu
1 Leonard D. Schaeffer Center for Health Policy and Economics, University of Southern
California, Los Angeles, USA
41
3.1 Introduction
The United States and other developed nations are experiencing a growing epidemic of
obesity.[1] Among children and adolescents, overweight and obesity are defined as ≥ the 85
th
and ≥ the 95
th
percentile of sex-specific Body Mass Index (BMI)-for-age standards, developed by
the US Centers for Disease Control and Prevention (CDC), respectively.[2] The prevalence of
obesity among children aged 2-11 years old has more than tripled over the past four decades. As
of 2010, 12.1% and 18.0% of children aged 2-5 and 6-11 were obese, respectively.[3] The health
risks associated with obesity in childhood have been well documented and include type II
diabetes, hypertension, hypercholesterolemia, orthopedic disease, and asthma.[4-8] Tracking of
obesity from childhood into adulthood has also been well documented. An obese ten year old has
a 37% - 52% chance of being obese at age 35, depending on their gender and race, more than
triple to risk of a ten year old with a normal BMI.[9]
In addition to these negative health risks, a growing body of research has indicated that
childhood obesity may be associated with adverse educational outcomes. Some previous research
has described a negative relationship between child BMI or obese status and standard test
performance among children[10-12] an adolescents[13]. However, other literature has failed to
find this association.[14] Less attention has been paid to the possible effect of childhood obesity
on teacher evaluations of student abilities. This relationship may be of particular interest in
grades K-8, where grading is likely to be more subjective compared to high school.
Two decades ago, the National Education Association issued a report on weight
discrimination that concluded that schools were a venue of ongoing ostracism, stigmatization,
and discrimination against overweight and obese youth from nursery school through college.[15]
Although weight discrimination from parents, peers, educators, and administrators can all
42
theoretically affect educational outcomes, the present study focuses on educators. Educators have
demonstrated both implicit and explicit biases against the obese in several studies. In one
frequently quoted study, one-fifth of middle and high school teachers reported beliefs that obese
persons are untidy, less likely to succeed, more emotional, and more likely to have family
problems.[16] In the same study, forty-three percent of teachers strongly agreed that “most
people feel uncomfortable when they associate with obese people.” Another study found that
over 50% of elementary school principals cited lack of self-control and psychological problems
as major contributors to obesity.[17] Twenty-five percent of those principals believed teachers at
their school would not be supportive of implementing school-based treatment programs to help
obese children. These beliefs are not unique to educators, but are common among adults of
various professions.[18]
In 1948, sociologist Robert Merton coined the phrase “self-fulfilling prophecy.”[19]
Since then, several researchers have found evidence of such a phenomenon in education –
misguided expectations held by educators lead students to perform at levels consistent with those
expectations. One of the most famous demonstrations of this phenomenon in education was the
1968 Pygmalion experiment.[20] A non-verbal intelligence test was administered to all of the
children within an elementary school, but teachers were told that this was a new test being
developed to identify students likely to show dramatic academic improvement over the next
school year. In reality, likely “bloomers” were selected at random and made known to teachers.
After retesting two years later, the bloomers had higher IQ and were reported to be better
adjusted by their teachers, compared to the control students. These particular results are still
debated decades later, but a recent meta-analysis has concluded that teacher expectancy effects
have statistically significant, albeit modest effects on academic performance.[21] The existence
43
of expectancy effects implies that, if teachers were to view obese and non-obese students
differently, those opinions could theoretically translate into different outcomes.
Therefore, the present study seeks to determine whether elementary school teachers
discriminate against obese students. We will investigate the relationship between child BMI and
both math and English teacher evaluations and standardized test scores using three different
modeling strategies.
3.2 Methods
3.2.1 Data Source
The National Center for Education Statistics (NCES) maintains a dataset called the Early
Childhood Longitudinal Study (ECLS). We utilize the ECLS-K data, which come from a survey
of kindergarten students, their families, their teachers, and their school administrators in 1998.
The same cohort of students was followed until the 8
th
grade, in 2006. The sample was refreshed
in the 1
st
grade, resulting in a cohort nationally representative of 1
st
grade students in 1999. We
make use of data from the spring kindergarten, 1
st
grade, 3
rd
grade, 5
th
grade, and 8
th
grade survey
waves. Our analytic sample includes all students who were eligible to participate in each survey
wave (not necessarily every survey wave). Because the NCES only attempted to contact a
random sample of the children that moved after each survey, the number of children that were
eligible to participate changed over the course of follow-up. Due to modest nonresponse rates
among eligible children and their respective parents, teachers, and school administrators, various
regression models will include different sample sizes.
44
3.2.2 Measures of Weight Status and Academic Performance
The ECLS-K includes each child’s height and weight at each survey wave. These
measures are not reported by parents, but are instead measured by trained staff. This is an
advantage over many public-access datasets that rely on self-reported or parent-reported height
and weight, which are consistently biased.[22] From height and weight we calculate body mass
index and classify each student at each time point as either underweight, normal weight,
overweight, or obese according to the previously mentioned CDC criteria.
Also included in the ECLS-K are measures of academic performance via standardized
tests and teacher assessments. English and math tests were administered by trained staff at each
survey wave. The content validity of all tests has been previously validated. Consistent with
modern test theory, each student received a subset of all potential test questions. A statistical
model was used to estimate the difficulty of each question and the number of questions each
student would have answered correctly, had they answered every possible question. We use the
item response theory (IRT) scores, which represent the total number of correct responses
predicted, rather than the raw number of questions answered correctly on a given test.
Interpretation of changes in raw scores over time is complicated when the number or difficulty of
test items may change over time. Because they measure the underlying latent trait (ability), IRT
scores avoid these complications.
At each survey wave teachers were asked to rate each child, using a discrete score
ranging from one to five, on their skills and knowledge of specific areas within English and
math. Teachers were specifically instructed to consider the child’s current skills and knowledge,
not their potential or effort. The discrete ratings were converted to a continuous scale, also
ranging from one to five. We make use of these continuous scores throughout this study.
45
Both the test scores and teacher evaluations were standardized to a scale with a mean of
zero and standard deviation of one. This standardization allows for the comparison of regression
coefficients from models of the test scores and teacher evaluations.
3.2.3 Model Specifications
We begin with cross-sectional analysis at each survey wave. We estimate ordinary least
squares (OLS) linear regression models with the following specification:
Outcome
i
= α + Weight Variables
i
+ Child Controls
i
+ Family Controls
i
+ School Controls
i
+ ε
i
We then utilize the longitudinal nature of the data by including all observations from
eligible participants at each survey wave. Pooled OLS linear regression models (without fixed
effects) are estimated separately for each outcome of interest, each with survey wave fixed
effects:
Outcome
i,t
= α + Weight Variables
i,t
+ Round FE
t
+ Child Controls
i,t
+ Family Controls
i,t
+
School Controls
i,t
+ ε
i,t
The pooled OLS model is then re-estimated using the method of seemingly unrelated regression
(SUR).[23] SUR allows for the estimation of the correlation between the residuals of teacher
evaluation and test score regressions for each subject, as well as test the restriction of equality
between the simultaneously estimated parameters. Finally, we estimate separate regression
46
models using the longitudinal data while including child fixed effects to control for unobserved,
time-invariant differences between children:
Outcome
i,t
= α + Weight Variables
i,t
+ Child FE
i
+ Round FE
t
+ Child Controls
i,t
+ Family
Controls
i,t
+ School Controls
i,t
+ ε
i,t
In each specifications i indicates the student and t indicates the survey wave. Outcomes
will include English test scores, English teacher evaluations, math test scores, and math teacher
evaluations. Weight variables include child BMI and interactions between child BMI and
indicators of schools found to be in the ≤ 10
th
percentile or ≥ 90
th
percentile of average student
BMI. Child BMI is used instead of child BMI Z-score because it is more sensitive to change,
particularly among overweight and obese children. Child controls include gender, an indicator of
Caucasian race, and self-reported internalizing behavior problems on a 0-4 scale. Internalizing
behavior problems are negative or problematic behaviors that are directed inward and typically
include social withdrawal and feelings of loneliness or guilt.[24] Because they are time-invariant,
child gender and race are not included in models that include child fixed effects. Family controls
include mother’s age, number of siblings, and socioeconomic status (SES) quintile. School
controls include teacher’s age, census region, an indicator of suburban, urban, or rural
environment, the percent minority students, and indicators of schools found to be in the ≤ 10
th
percentile or ≥ 90
th
percentile of average student BMI. Child fixed effects control for all child-
specific omitted variables that are constant over time. Round fixed effects control for time-
specific effects that affect all children. All regressions utilize robust standard errors clustered at
the child level and survey sampling weights to ensure the representativeness of results.
47
3.3 Results
Table 1 presents descriptive statistics for eligible individuals at each survey wave. The
eligible sample size at each survey wave declined from 21,192 in the spring of kindergarten to
9,725 in the 8
th
grade. The prevalence of obesity increased from 7.33% in kindergarten to
13.09% in 8
th
grade. Similarly, the prevalence of overweight increased from 13.48% in
kindergarten to 19.43% in the 8
th
grade. The mean teacher evaluation scores declined slightly in
each successive survey wave except for the 5
th
grade.
Figures 1 and 2 plot the estimated BMI coefficients from cross-sectional OLS regressions
on standardized test scores and teacher assessments at each survey wave. These coefficients
represent the marginal effects of BMI on standardized test scores and evaluations. For example, a
one kg/m
2
increase in body mass index in the eighth grade is predicted to decrease teacher
evaluation of math abilities by 0.035 standard deviations. Increasing child BMI is associated
with significantly lower teacher evaluations in the third, fifth, and eight grades. However, BMI
does not appear to have any significant impact on objective test scores at any time.
Tables 2 and 3 present the longitudinal models of standardized math and English scores,
respectively. In both tables, the first two panels contain estimates from test and teacher scores
estimated by pooled OLS, the next two panels contain estimates using seemingly unrelated
regression, and the final two panels contain OLS estimates with child fixed effects. All models
contain survey wave fixed effects. In no model is child BMI significantly associated with math
or English test scores. Child BMI is significantly associated with worse math and English teacher
evaluations when the first two estimation strategies are used. However, when child fixed effects
are included this effect remains consistently negative but longer statistically significant at an
acceptable type I error rate of 5%. A one kg/m2 increase in child BMI is associated with a
48
reduction of 0.016 – 0.026 standard deviations in math teacher evaluations and a reduction of
0.011 – 0.019 standard deviations in English teacher evaluations.
The effect of child BMI on math and English teacher evaluations also appears to be
significantly worse among students attending schools with an average student BMI in the bottom
10
th
percentile of the sample. Students attending these low-BMI schools suffered 0.011 – 0.017
standard deviation reductions in their teacher evaluations for each one kg/m2 BMI increase, in
addition to the effects of BMI already mentioned. This means that the average total effect of
childhood obesity in low-BMI schools can be a reduced teacher evaluation of up to 0.043
standard deviations.
3.4 Discussion
We have shown that increased BMI among children in grades K-8 is associated with
lower teacher assessments in both math and English. Cross-sectional models reveled that BMI
did not appear to be significantly associated with lower teacher assessments until after the third
grade. However, after this point the effect appears to consistently grow in magnitude. One simple
explanation for this finding might be that BMI is truly associated with lower math and English
abilities and teachers are simply aware of this fact. Obese children are more likely to suffer from
asthma, sleep apnea, and other chronic conditions, which may affect absenteeism or attention in
class. Some previous research has found evidence of negative affect of child BMI on
standardized test scores[10-13], but other literature has failed to find such an association[14]. We
failed to find any evidence that child BMI was associated with significantly lower performance
on math and English test scores, therefore we do not believe that this can explain why teachers
would rate the abilities of high-BMI students lower.
49
Interestingly, the negative effect of BMI on teacher assessments was greater in schools
that had an average student BMI in the bottom 10
th
percentile. This result cannot be explained by
the biological effects of obesity. Rather, it suggests that teacher assessments are, at least partly,
influenced by the social environment of their school. If it were the case that teachers rated the
math and English abilities of obese students lower because they held unfavorable opinions of
obesity, such a result would be expected.
It should be noted that, although this effect is statistically significant, the magnitude is
modest. Each one kg/m2 increase in BMI translated into an average decrease of 0.011 – 0.026
standard deviations in teacher score. The difference in mean BMI between children considered
obese according the CDC growth charts and those considered normal weight ranged from 7 – 11
kg/m2 across the survey waves. Therefore, becoming obese should theoretically lower a child’s
teacher evaluation scores by roughly 0.2 standard deviations at most. This estimation is
consistent with the results of regressions using categorical weight definitions rather than BMI
which are not reported here but available upon request.
Our estimates were derived using three different model specifications. The first, pooled
OLS, assumed independence of the errors between the teacher and test score equations and
independence of all observations within a given child. Neither of these assumptions is likely to
be correct. The error terms from the teacher and test score equations likely contain similar
unobservable variables, some of which may be child-specific and time-invariant and other may
not be. Therefore, we included two additional model specifications. We estimated the pooled
OLS models simultaneously using the method of seemingly unrelated regression to account for
the cross-equation correlation in errors and we estimated the child fixed effects models to
account for child-specific time-invariant factors. Our estimates were reasonably stables across
50
the three models. However, the inclusion of child fixed effects did appear to reduce the statistical
significance of BMI.
Another recent examination of the effect of childhood obesity on teacher evaluations also
found a negative effect of BMI on math and English assessments.[25] The author of this study
concluded that, “BMI was more negatively related to teacher assessments of academic
performance than test scores.” Because we have found no significant relationship between BMI
and test scores, this assertion seems reasonable. However, simultaneous estimation of the teacher
and test score equations via seemingly unrelated regression allowed us to explicitly test this
hypothesis. Not surprisingly, we found that the BMI parameter estimates from the test and
teacher equations were significantly different from each other, for both math and English scores.
It should be noted that the teacher evaluations elicited as part of the ECLS-K survey
represent the opinions of teachers, not the grades that the child has received. There has been
some conflicting evidence regarding the relationship between childhood obesity and grade point
average (GPA).[26, 27] However, given the established effect of teacher expectations[28], the
present study indicates that teachers’ biases against obese student may plausibly impact earned
grades and further educational achievement.
This analysis is subject to a few important limitations. There may not be perfect overlap
between the specific subjects included in the subjective evaluations and objective tests. English
teacher assessments were more focused on writing abilities, whereas the standardized English
tests focused on reading abilities. The domains of the math tests and math teacher assessments
appeared to be better matched. However, in both cases the specific teacher questions used to
form the assessment score were consistently focused on how a child attempted to complete tasks,
rather than whether they found the correct answer. For example, 8
th
grade math teachers were
51
asked whether a child “uses representations to model mathematical ideas” or “talks about
reasoning in solving a problem.” While proficiency in these skills is likely correlated with high
test scores, they are not measuring exactly the same skill.
3.5 Conclusion
BMI is associated with lower teacher assessments of math and English abilities among
students enrolled in kindergarten through the 8
th
grade. This effect was greatest in schools with
low average student BMI. However, BMI was not significantly associated with math and English
test scores. The lower teacher evaluations given to obese students are therefore plausibly due to
implicit or explicit biases.
52
References
1. Cecchini M SF. Obesity Update 2014: Organization for Economic Cooperation and
Development; 2014.
2. Flegal KM, Ogden CL. Childhood obesity: are we all speaking the same language?
Advances in nutrition. 2011 Mar;2(2):159S-66S.
3. Flegal KM, Ogden CL, Yanovski JA, Freedman DS, Shepherd JA, Graubard BI, et al.
High adiposity and high body mass index-for-age in US children and adolescents overall and by
race-ethnic group. The American journal of clinical nutrition. 2010 Apr;91(4):1020-6.
4. Cornette R. The emotional impact of obesity on children. Worldviews on evidence-based
nursing / Sigma Theta Tau International, Honor Society of Nursing. 2008;5(3):136-41.
5. Freedman DS, Mei Z, Srinivasan SR, Berenson GS, Dietz WH. Cardiovascular risk
factors and excess adiposity among overweight children and adolescents: the Bogalusa Heart
Study. The Journal of pediatrics. 2007 Jan;150(1):12-7 e2.
6. Gilliland FD, Berhane K, Islam T, McConnell R, Gauderman WJ, Gilliland SS, et al.
Obesity and the risk of newly diagnosed asthma in school-age children. American journal of
epidemiology. 2003 Sep 1;158(5):406-15.
7. Li C, Ford ES, Zhao G, Mokdad AH. Prevalence of pre-diabetes and its association with
clustering of cardiometabolic risk factors and hyperinsulinemia among U.S. adolescents:
National Health and Nutrition Examination Survey 2005-2006. Diabetes care. 2009
Feb;32(2):342-7.
8. Taylor ED, Theim KR, Mirch MC, Ghorbani S, Tanofsky-Kraff M, Adler-Wailes DC, et
al. Orthopedic complications of overweight in children and adolescents. Pediatrics. 2006
Jun;117(6):2167-74.
9. Guo SS, Wu W, Chumlea WC, Roche AF. Predicting overweight and obesity in
adulthood from body mass index values in childhood and adolescence. The American journal of
clinical nutrition. 2002 Sep;76(3):653-8.
10. Castelli DM, Hillman CH, Buck SM, Erwin HE. Physical fitness and academic
achievement in third- and fifth-grade students. Journal of sport & exercise psychology. 2007
Apr;29(2):239-52.
11. Datar A, Sturm R. Childhood overweight and elementary school outcomes. International
journal of obesity. 2006 Sep;30(9):1449-60.
53
12. Florence MD, Asbridge M, Veugelers PJ. Diet quality and academic performance. The
Journal of school health. 2008 Apr;78(4):209-15; quiz 39-41.
13. JT S. The Effect of Body Weight on Adolescent Academic Performance. Southern
Economic Journal. 2007;73(4):871-900.
14. Kaestner R GM. Effects of Weight on Children's Educational Achievement. Cambridge,
MA: National Bureau of Economic Research; 2008.
15. Report on Discrimination Due to Physicial Size.: National Education Association; 1994.
16. Neumark-Sztainer D SM, Harris T. Beliefs and attitudes about obesity among teacher and
school health care providers working with adolescents. Journal of Nutrition Education.
1999;31:3-9.
17. Price JH, Desmond SM, Stelzer CM. Elementary school principals' perceptions of
childhood obesity. The Journal of school health. 1987 Nov;57(9):367-70.
18. Puhl RM, Andreyeva T, Brownell KD. Perceptions of weight discrimination: prevalence
and comparison to race and gender discrimination in America. International journal of obesity.
2008 Jun;32(6):992-1000.
19. RK M. The Self-Fulfilling Prophecy. The Antioch Review. 1948;8(2):193-210.
20. Rosenthal R JL. Pygmalion in the Classroom: Teacher Expectations and Student
Intellectual Development. New York: Holt, Rinehart & Winston; 1968.
21. R R. Interpersonal expectancy effects: A 30-year perspective. Current Directions in
Psychological Science. 1994;3:176-9.
22. Merrill RM, Richardson JS. Validity of self-reported height, weight, and body mass
index: findings from the National Health and Nutrition Examination Survey, 2001-2006.
Preventing chronic disease. 2009 Oct;6(4):A121.
23. A Z. An effecient method of estimating seemingly unrelated regressions and tests for
aggregation bias. Journal of the American Statistical Association. 1962;57(298):348-68.
24. Rapport MD DC, Chung K-M, Hustace K. Internalizing behavior problems and scholastic
acheievement in children: cognitive and behavioral pathways as mediators of otucome. Journal
of Clinical Child & Adolescent Psychoogy. 2001;30(4):536-51.
25. Zavodny M. Does weight affect children's test scores and teacher assessments
differently? Economics of education review. 2013 Jun;34:135-45.
54
26. MacCann C, Roberts RD. Just as smart but not as successful: obese students obtain lower
school grades but equivalent test scores to nonobese students. International journal of obesity.
2013 Jan;37(1):40-6.
27. Huang TT, Goran MI, Spruijt-Metz D. Associations of adiposity with measured and self-
reported academic performance in early adolescence. Obesity. 2006 Oct;14(10):1839-45.
28. Jussim L HK. Teacher expectations and self-fulfiling prophecies: Knowns and unknowns,
resolved and unresolved controversies. Personality and social psychology review. 2005;9(2):131-
55.
55
Table 3.1. Descriptive Statistics for ECLS-K Sample
Child Variables
Test Scores (mean/SD)
36.27 9.1 61.26 18.09 98.72 24.72 123.69 24.79 142.22 22.01
46.46 14.04 77.36 23.87 126 28.04 150 26.39 171 27.59
Teach Evaluations (mean/SD)
3.54 0.85 3.47 0.89 3.11 0.74 3.42 0.71 3.09 0.94
3.37 0.8 3.44 0.92 3.32 0.88 3.43 0.84 3.06 1.04
Body Mass Index (mean/SD)
16.4 2.31 16.9 2.87 18.66 3.89 20.58 4.76 22.9 6.29
Weight Status (N/%)
1451 7.33 1398 8.99 1793 12.93 1604 14.50 1273 14.43
2668 13.48 2264 14.56 2743 19.79 2567 23.20 1890 21.42
14368 72.61 10845 69.74 8626 62.22 6309 57.02 5233 59.31
1300 6.57 1043 6.71 702 5.06 585 5.29 427 4.84
Male (N/%)
10828 51.09 8943 51.16 7807 51.01 5987 50.65 4929 50.68
Internalizing Behavior Problems (mean/SD)
1.57 0.52 1.59 0.52 1.63 0.54 1.63 0.54 2.04 0.54
Race (N/%)
11706 55.24 9856 56.38 8664 56.61 6733 56.96 5922 60.89
3193 15.07 2482 14.2 1995 13.03 1342 11.35 1001 10.29
3745 17.67 3036 17.37 2771 18.11 2248 19.02 1701 17.49
2499 11.29 2068 11.83 1855 12.12 1480 12.52 1091 11.22
Household Variables
Mother's Age (years) (mean/SD)
33.35 6.58 34.66 6.58 37.9 6.51 39.81 6.67 42.18 6.6
Siblings (N/%)
1.73 1.1 1.79 1.06 1.81 1.04 1.82 1.06 1.77 1.04
Socioeconomic Status (N/%)
3754 17.71 2719 15.55 2166 14.15 1799 15.22 1414 14.54
3893 18.37 2989 17.1 2422 15.82 2020 17.09 1691 17.39
3968 18.72 3166 18.11 2632 17.2 2094 17.72 1759 18.09
4130 19.49 3340 19.11 2895 18.92 2438 20.63 1805 18.56
4346 20.51 3795 21.71 3374 22.05 2645 22.38 2140 22.01
School Variables
Region (N/%)
4961 23.41 3783 21.64 3350 21.89 2616 22.13 1978 20.34
3904 18.42 3121 17.85 2724 17.8 2109 17.84 1738 17.87
5255 24.8 4299 24.59 3756 24.54 2979 25.2 2633 27.07
7072 33.37 5678 32.48 4809 31.42 3693 31.24 3085 31.72
Urbanization (N/%)
8174 38.57 6575 37.61 5673 37.07 4263 36.07 3418 35.15
8737 41.23 6621 37.88 5470 35.74 4053 34.29 2819 28.99
4281 20.2 3582 20.49 3168 20.7 2615 22.12 2245 23.08
Minorities (%) (N/%)
10030 47.33 8183 46.81 6946 45.38 5411 45.78 4524 46.52
3364 15.87 2667 15.26 2359 15.41 1969 16.66 1811 18.62
2283 10.77 1761 10.07 1622 10.6 1117 9.45 1112 11.43
4985 23.52 3969 22.7 3441 22.48 2831 23.95 1893 19.47
Underweight
Normal
Overweight
Obese
English
Math
English
Math
2nd Quintile
1st Quintile
Other
Hispanic
Black
White
5th Quintile
4th Quintile
3rd Quintile
> 75
50-75
25-50
Eligible Sample Size 17481 15305 11820 9725 21192
South
Midwest
Northeast
West
< 25
Rural
City
Suburb
1st Grade 3rd Grade 5th Grade 8th Grade Kindergarten
56
Figure 3.1. Cross-Sectional Effect of Child BMI on Standardized Math Teacher and Test
Scores
BMI coefficient represents change in standardized score associated with one kg/m2 increase in
BMI at each time point.
Figure 3.2. Cross-Sectional Effect of Child BMI on Standardized English Teacher and Test
Scores
BMI coefficient represents change in standardized score associated with one kg/m2 increase in
BMI at each time point.
-0.05
-0.04
-0.03
-0.02
-0.01
0
0.01
Kindergarten 1st Grade 3rd Grade 5th Grade 8th Grade
BMI Coefficient
Math Teacher Math Test
-0.03
-0.025
-0.02
-0.015
-0.01
-0.005
0
0.005
0.01
Kindergarten 1st Grade 3rd Grade 5th Grade 8th Grade
BMI Coefficient
English Teacher English Test
57
Table 3.2. Longitudinal Models of Standardized Math Teacher and Test Scores
Coeff. SE Coeff. SE Coeff. SE Coeff. SE Coeff. SE Coeff. SE
BMI -0.026 0.003 *** -0.003 0.003 -0.018 0.003 ** -0.002 0.004 -0.016 0.003 ** -0.001 0.003
Top10*BMI 0.001 0.003 0.004 0.003 0.004 0.005 0.005 0.005 0.005 0.004 0.006 0.022
Bottom10*BMI -0.014 0.003 ** 0.002 0.003 -0.011 0.003 * -0.002 0.003 -0.013 0.005 * -0.003 0.002
Top10 -0.017 0.049 0.011 0.043 -0.010 0.041 -0.021 0.041 -0.014 0.049 -0.025 0.032
Bottom10 -0.007 0.037 0.014 0.032 0.006 0.032 0.015 0.021 -0.007 0.037 -0.019 0.024
Male -0.096 0.027 *** 0.080 0.030 *** -0.097 0.027 *** 0.078 0.030 ***
Caucasian 0.118 0.037 *** 0.239 0.039 *** 0.116 0.037 *** 0.237 0.039 ***
Internal Behavior Probs -0.039 0.010 *** -0.028 0.006 *** -0.040 0.010 *** -0.028 0.006 *** -0.025 0.012 ** < 0.001 0.004
Teacher's Age -0.007 0.001 *** -0.001 0.001 -0.007 0.001 *** -0.001 0.001 -0.005 0.001 *** < 0.001 0.001
Mother's Age 0.007 0.002 *** 0.011 0.002 *** 0.007 0.002 *** 0.011 0.002 *** 0.006 0.009 -0.001 0.004
# Siblings -0.037 0.013 *** -0.039 0.013 *** -0.036 0.013 *** -0.038 0.014 *** -0.027 0.029 0.023 0.017
SES Quintile (Ref = 1)
2 0.312 0.045 *** 0.359 0.042 *** 0.312 0.045 *** 0.359 0.042 *** 0.128 0.068 0.016 0.034
3 0.402 0.047 *** 0.538 0.046 *** 0.403 0.047 *** 0.541 0.046 *** 0.092 0.073 0.021 0.040
4 0.614 0.049 *** 0.757 0.050 *** 0.614 0.049 *** 0.759 0.050 *** 0.172 0.085 ** 0.066 0.047
5 0.784 0.053 *** 1.033 0.052 *** 0.784 0.053 *** 1.035 0.052 *** 0.166 0.098 * 0.115 0.057
Region (Ref = West)
Northeast -0.080 0.047 -0.135 0.050 *** -0.078 0.047 -0.133 0.050 *** 0.044 0.447 0.074 0.193
Midwest -0.048 0.044 -0.047 0.047 -0.047 0.044 -0.044 0.047 0.397 0.232 0.083 0.138
South -0.005 0.041 -0.042 0.042 -0.003 0.041 -0.037 0.042 -0.170 0.242 0.240 0.155
Community (Ref = Sub)
City 0.022 0.033 -0.004 0.034 0.021 0.033 -0.005 0.034 -0.025 0.087 0.023 0.049
Rural -0.088 0.036 ** -0.102 0.039 *** -0.090 0.036 * -0.105 0.039 *** -0.152 0.159 0.007 0.075
% Minorities (Ref = <25)
25-50 0.045 0.038 -0.010 0.040 0.045 0.038 -0.012 0.040 -0.070 0.057 -0.037 0.031
50-75 -0.064 0.054 -0.056 0.047 -0.064 0.054 -0.060 0.047 -0.093 0.080 0.024 0.043
> 75 -0.005 0.046 -0.209 0.047 *** -0.008 0.046 -0.212 0.047 *** -0.010 0.087 -0.034 0.059
Round FE (Ref = K)
1st Grade -0.078 0.028 *** 0.005 0.021 -0.078 0.028 *** 0.007 0.021 -0.067 0.028 ** 0.009 0.017
3rd Grade -0.108 0.030 *** -0.063 0.024 *** -0.110 0.031 *** -0.059 0.026 ** -0.085 0.045 * -0.015 0.024
5th Grade -0.133 0.027 *** -0.108 0.026 *** -0.138 0.029 *** -0.101 0.029 *** -0.091 0.062 -0.040 0.030
8th Grade -0.159 0.042 *** -0.158 0.032 *** -0.169 0.047 *** -0.148 0.038 *** -0.118 0.079 -0.101 0.039 ***
Observations
R-Squared
Pooled OLS Pooled OLS-SUR Child Fixed Effects
Test Teacher Test Teacher Test Teacher
26140 25504 26140 26542 26542 20504
0.155 0.1226 0.242 0.113 0.211 0.195
* p < 0.1 , ** p < 0.05 , *** p < 0.01
58
Table 3.3. Longitudinal Models of Standardized English Teacher and Test Scores
Coeff. SE Coeff. SE Coeff. SE Coeff. SE Coeff. SE Coeff. SE
BMI -0.019 0.001 *** -0.002 0.003 -0.016 0.001 *** -0.001 0.003 -0.011 0.001 * -0.003 0.003
Top10*BMI -0.002 0.003 0.004 0.003 -0.007 0.004 0.004 0.003 -0.006 0.005 0.006 0.006
Bottom10*BMI -0.017 0.003 *** -0.002 0.003 -0.014 0.003 ** 0.003 0.003 -0.011 0.004 * -0.002 0.002
Top10 0.015 0.042 0.002 0.034 -0.007 0.035 0.003 0.041 -0.038 0.044 0.001 0.034
Bottom10 0.054 0.034 -0.014 0.030 0.014 0.032 0.044 0.037 0.027 0.027 -0.004 0.030
Male -0.302 0.028 *** -0.186 0.030 *** -0.302 0.028 *** -0.186 0.030 *** - - - - - -
Caucasian 0.178 0.039 *** 0.146 0.039 *** 0.178 0.039 *** 0.146 0.039 *** - - - - - -
Internal Behavior Probs -0.038 0.007 *** -0.019 0.007 *** -0.038 0.007 *** -0.019 0.007 *** -0.010 0.006 -0.001 0.005
Teacher's Age -0.006 0.001 *** < 0.001 0.001 -0.006 0.001 *** < 0.001 0.001 -0.004 < 0.001 *** 0.000 0.001
Mother's Age 0.008 0.002 *** < 0.001 0.002 *** 0.008 0.002 *** < 0.001 0.002 *** 0.008 0.006 0.006 0.005
# Siblings -0.055 0.013 *** -0.075 0.013 *** -0.055 0.013 *** -0.075 0.013 *** 0.021 0.022 0.009 0.022
SES Quintile (Ref = 1)
2 0.356 0.045 *** 0.373 0.044 *** 0.356 0.045 *** 0.373 0.044 *** -0.006 0.067 -0.042 0.047
3 0.500 0.048 *** 0.528 0.047 *** 0.500 0.048 *** 0.528 0.047 *** -0.049 0.073 -0.079 0.055
4 0.724 0.050 *** 0.767 0.050 *** 0.724 0.050 *** 0.767 0.050 *** 0.034 0.077 -0.061 0.061
5 0.940 0.052 *** 1.067 0.057 *** 0.940 0.052 *** 1.067 0.057 *** 0.100 0.089 0.000 0.074
Region (Ref = West)
Northeast -0.143 0.046 *** -0.104 0.053 ** -0.143 0.046 *** -0.104 0.053 ** 0.158 0.376 -0.040 0.274
Midwest -0.124 0.047 *** -0.106 0.048 ** -0.124 0.047 *** -0.106 0.048 ** -0.154 0.234 0.187 0.149
South -0.059 0.042 -0.059 0.042 -0.059 0.042 -0.059 0.042 -0.095 0.324 0.030 0.187
Community (Ref = Sub)
City 0.044 0.033 -0.002 0.033 0.044 0.033 -0.002 0.033 0.019 0.071 -0.004 0.058
Rural -0.024 0.036 -0.060 0.042 -0.024 0.036 -0.060 0.042 0.027 0.106 0.023 0.092
% Minorities (Ref = <25)
25-50 0.076 0.039 * 0.017 0.037 0.076 0.039 * 0.017 0.037 -0.030 0.043 0.016 0.033
50-75 0.018 0.053 -0.037 0.054 0.018 0.053 -0.037 0.054 0.040 0.066 0.040 0.054
> 75 0.078 0.046 * -0.201 0.048 *** 0.078 0.046 * -0.201 0.048 *** 0.111 0.065 -0.023 0.066
Round FE (Ref = K)
1st Grade -0.025 0.026 0.056 0.023 ** -0.025 0.026 0.056 0.023 ** -0.030 0.025 0.045 0.020 **
3rd Grade -0.071 0.029 ** 0.030 0.029 -0.071 0.029 ** 0.030 0.029 -0.090 0.036 ** 0.043 0.030
5th Grade -0.083 0.029 *** -0.019 0.030 -0.083 0.029 *** -0.019 0.030 -0.075 0.044 * 0.012 0.037
8th Grade -0.163 0.034 *** -0.098 0.036 *** -0.163 0.034 *** -0.098 0.036 *** -0.186 0.057 *** -0.095 0.047 **
Observations
R-Squared
Pooled OLS
Teacher Test Test
25867
0.230 0.131
25864
0.166
Test Teacher Teacher
Pooled OLS-SUR Child Fixed Effects
26487
0.139
26487
0.196
25864
0.102
25817
* p < 0.1 , ** p < 0.05 , *** p < 0.01
59
CHAPTER 4:
DO STATINS REDUCE THE HEALTH AND HEALTHCARE COSTS OF OBESITY?
Etienne Gaudette
*1
, Dana P. Goldman
1
, Andrew Messali
1
, and Neeraj Sood
1
July 2014
Abstract
Obesity impacts both individual health and, given its high prevalence, total health care spending.
However, as medical technology evolves, health outcomes for a number of obesity-related
illnesses improve. This article examines whether medical innovation can mitigate the adverse
health and spending associated with obesity, using statins as a case study. Due to the relationship
between obesity and hypercholesterolemia, statins play an important role in the medical
management of obese individuals and the prevention of costly obesity-related sequelae. Using
well-recognized estimates of the health impact of statins and the Future Elderly Model (FEM) –
an established dynamic microsimulation model of health of Americans aged over 50 – we
estimate the changes in life expectancy, functional status and health care cost of obesity due to
the introduction and widespread use of statins. Life expectancy gains of statins are estimated to
be 5%-6% higher for obese than healthy-weight individuals, but most of this additional gain is
associated with some level of disability. Considering both medical spending and the value of
quality-adjusted life-years, statins do not significantly alter the costs of class 1 and 2 obesity
(BMI larger or equal to 30 and 35 kg/m
2
), and increase the costs of class 3 obesity (BMI larger
or equal to 40 kg/m
2
) by 1.2%. Although statins are very effective medications for lowering the
risk of obesity-associated illnesses, they do not significantly reduce the costs of obesity.
Keywords: obesity, weight, body mass index, behavior, habits, healthcare costs
* Corresponding author. Email: egaudett@healthpolicy.usc.edu
1 Leonard D. Schaeffer Center for Health Policy and Economics, University of Southern
California, Los Angeles, USA
60
4.1 Introduction
Between 1960 and 2002, the average height of American adults increased by approximately
1 inch, while average weight increased by over 24 pounds (11 kg).[1] As a result, during these
four decades the prevalence of obesity (a body mass index (BMI) ≥ 30 kg/m
2
) more than doubled
and the prevalence of extreme obesity (BMI ≥ 40 kg/m
2
) increased six-fold. As of 2010, 36.1%
of American adults were obese and 6.6% were extremely obese. A similar trend is apparent
among men and women, most racial subgroups, and the elderly.[1]
This increase in the prevalence of obesity is a serious concern from both a public health and
fiscal perspective. Obesity is associated with an increased risk of type II diabetes, ischemic heart
disease, ischemic stroke, hypertension, dyslipidemia, osteoarthritis, and several cancers.[2]
Published estimates of the healthcare costs attributable to obesity have consistently increased as
the epidemic has grown.[3-6] Finkelstein et al. find— after adjustment for demographic factors,
economic factors, and smoking status—that obese individuals incur 42% more direct medical
costs.[5] This translated into roughly 9.1% of all public and private healthcare spending, or about
$147 billion in 2006—with roughly one-quarter paid by Medicare.[5] Given the fact that most
obesity-associated diseases do not present themselves until later adulthood, this finding is not
surprising. However, it does underscore the importance of tackling the obesity epidemic in the
aging population of the United States.
The past decades have also seen numerous medical innovations, several of which have
become important components of the management of obesity-associated illnesses. However,
little is known about how these innovations affect the health and healthcare costs of obesity.
In this study we use the introduction of statins as a case study to shed light on this issue.
Statins offer an interesting example for several reasons. First, statins are perhaps one of the most
61
important innovations in recent medical history. Not only do they lower the risk of death from
obesity-associated illness, they also lower the risk of developing some of these diseases in the
first place.[7-12] Second, statin use has rapidly increased since 2000, with more than 40 million
adult Americans filling a prescription in 2011.
1
A recent study finds a large social value for this
widespread use, and suggests it may have prevented as many as 40,000 deaths and 100,000
hospitalizations for heart attacks and strokes in 2008 alone.[13] Third, statins are well studied,
allowing us to model the effects of statins on long term health outcomes and healthcare costs.
We note that other examples of medical innovations, such as bariatric surgery and
prescription weight loss drugs, directly reduce the weight of obese people. The former was
shown to be quite effective[14], but has recently plateaued around 120,000 procedures per
year.[15] Statin therapy differs from such innovations in that it is so widely used that it may
plausibly have had a noticeable impact on the costs of obesity. We also believe that statins are a
better case study for medical innovation as a whole, since they prevent diseases associated with
obesity rather than obesity itself.
Statins might affect obesity through multiple channels. On the one hand, statin therapy
lowers both the chance of contracting cardiovascular diseases and stroke and the health
consequences of these diseases.[7-12] Since obesity is an important risk factor for these diseases,
statin therapy likely reduces the health consequences of being obese. On the other hand, the
effects of statins on health care costs are a priori ambiguous because of competing factors. By
decreasing morbidity associated with obesity, statins plausibly diminish obesity-related costs at a
1
Authors’ calculations with Medical Expenditure Panel Survey (MEPS) data. A MEPS
respondent is defined as a statin user if he or she filled at least one prescription associated with
the “HMG-CoA reductase inhibitor” therapeutic subclass during a given year, or a prescription
of “Simcor”, “Advicor” or “Vytorin” (which combine statins with other active ingredients).
Therapeutic classes in MEPS correspond to Multum Lexicon variables from Cerner Multum, inc.
62
given age. However, by increasing the length of life of obese people, statins enable the
accumulation of healthcare costs over more years. To our knowledge, no study has thus far
attempted to quantify these effects of statins on obesity. In this study, we fill this gap in the
literature and ask the following question: How did the introduction and widespread use of statins
impact the health and health care costs of obesity?
To answer this question, we use the Future Elderly Model (FEM) – an established dynamic
microsimulation model of the health of Americans aged over 50.[16] Using the well-recognized
estimates of the health impacts of statins, we construct an FEM scenario in which statins have
not been discovered. By comparing life trajectories and medical costs of elderly Americans in
this scenario and the baseline version of the FEM, we estimate the change in the health and
health care cost of obesity due to the introduction and widespread use of statins. We focus our
analysis on the long-term impact of statins on life expectancy, disability, and health care costs
including total spending and spending by public programs.
The remainder of the paper is organized as follows. Section 2 provides an overview of the
FEM and details the cohort simulations conducted in our analysis. Section 3 presents the impact
of statins on obesity and how these differ by sex, race and ethnicity. Finally, we discuss the
implications of our results in Section 4.
4.2 Methods
4.2.1 The Future Elderly Model
We conduct simulations using the Future Elderly Model (FEM), a dynamic
microsimulation model that follows Americans aged 51 and older and projects their health and
63
medical spending over time. The FEM was initially developed by Goldman et al. (2004) to
forecast the implications of different medical technology scenarios on long-term health and
healthcare costs [16]. Its unique feature is to follow the evolution of individual-level health
trajectories, rather than the average or aggregate health characteristics of a cohort. It has proved
useful for a number of purposes in the recent past, including forecasting the future costs of
cancer [17] and obesity [18], assessing the benefits of risk prevention [19], and evaluating the
marginal cost of being obese [20] and the value of medical interventions for reducing its
prevalence [14].
The FEM simulates the lives of people from the age of 51 onward using the Health and
Retirement Survey (HRS), a biennial survey of the American population aged 51 and over that
has been ongoing since 1992. The HRS data is used to compute the health transition models at
the core of the FEM and the input population that goes into the simulations. It is supplemented
by the Medical Expenditure Panel Survey (MEPS), a set of large-scale surveys of the non-
institutionalized population of the United States, and the Medicare Current Beneficiary Survey
(MCBS), a nationally representative survey of Medicare beneficiaries, to project healthcare
spending and assess quality-of-life during the simulations. For each individual, the FEM takes
into account initial demographic characteristics and health conditions to project medical
spending, health conditions and behaviors, disability status, and quality of life. We describe the
model and methods briefly here; a complete technical appendix containing details on the
modeling is provided in the Electronic Supplementary Material.
The FEM’s core module calculates transition probabilities across various health states,
including mortality, functional status, and BMI based on the individual’s current characteristics.
These transition probabilities are modeled using first-order Markov processes that depend on a
64
battery of predictors: age, sex, education, race, ethnicity, BMI, smoking behavior, marital status,
employment, functional status and health conditions. We also control for baseline factors using a
series of initial health variables.
Health conditions are derived from HRS survey questions and include diabetes, high
blood pressure, heart disease, cancer (except skin cancer), stroke or transient ischemic attack,
and lung disease (either or both chronic bronchitis and emphysema). Transitions into illness and
death are synthesized in Figure 1, which is analyzed in further detail in Appendix B. The concept
of chronic conditions used in the simulations corresponds to having ever been diagnosed with a
condition. We thus treat them as absorbing: once individuals receive a diagnosis, they are
henceforth considered to have that condition.
2
The BMI variable is based on the self-reported
height and weight of HRS respondents and its evolution is projected with the estimates of a log-
linear model. We adopt a log-linear specification to account for the ‘thick’ right tail of the BMI
distribution in HRS. We adjust for height and weight self-reporting biases using the average BMI
underreporting by age group and sex in the National Health and Nutrition Examination Survey
(NHANES) data.[21] Functional status is measured by limitations in instrumental activities of
daily living (IADLs), activities of daily living (ADLs), and residence in a nursing home. The
IADL indicator is based on questions about difficulty using the phone, managing money, and
taking medications. The ADL indicator is based on respondents’ assessment of their ability to
conduct basic tasks such as dressing, eating, and bathing. For the purpose of this study, we
consider an individual as free of disability if he reported no IADL or ADL limitation and did not
live in a nursing home; as disabled if he reported at least one IADL or ADL limitation but did
not live in a nursing home; and we consider nursing home living as indicating the most severe
2
This interpretation is consistent with the HRS questionnaire, which asks respondents if they
were ever diagnosed with a condition.
65
functional status impairment. Unlike health conditions, we allow for transitions into and out of
functional states.
To evaluate quality-of-life, we predict quality-adjusted life years (QALYs) using the EQ-
5D, a commonly used quality-of-life index based on five health-related variables addressing
mobility, daily activities, self-care, anxiety, depression and pain. Using the MEPS data, we use
an ordinary least squares regression to fit derived EQ-5D quality adjustment scores as a function
of the chronic conditions and functional states included in the FEM simulations. This model is
then used to predict the quality of each person’s life-years in our simulations.
Based on two complementary medical spending data sources, a second FEM module
predicts an individual’s health spending with regards to health status (chronic conditions and
functional status), demographics (age, sex, race, ethnicity and education), nursing home status,
and mortality. Our definition of medical costs includes medical provider visits, hospital events,
inpatient stays, outpatient visits, emergency room visits, dental care, home health care,
optometry, other medical equipment and services, prescribed medicines, and nursing home stay.
Our estimates are based on spending data from the 2002-2004 MEPS for individuals under 65
years-old and the 2002-2004 MCBS for individuals aged 65 and over. We separately predict total
medical costs and medical costs paid for by the Medicare and Medicaid programs. The estimates
are based on pooled least squares regressions of each type of spending on risk factors, self-
reported conditions, and functional status, with spending inflated to current dollars using the
medical component of the consumer price index.
66
4.2.2 Simulations
We simulate the health trajectories of elderly Americans characterized by a range of
different BMIs and predict their healthcare costs with and without the possibility to use statins.
These simulations focus on a representative cohort of Americans aged 51 to 52 in 2010. For each
period, the spending module predicts medical expenditure on the person’s current state vector.
Then, the health module predicts survival, health transitions, BMI, functional status and QALY
for the next period,
3
using the FEM’s transition probabilities. The same process is repeated at
each time step until everyone in the cohort has died.
The baseline FEM provides projections consistent with the current medical technology,
and thus implicitly includes the health effects and costs of statins. In the remainder of this article,
we refer to the baseline FEM as the “With Statins” scenario. Our simulation strategy consists of
comparing the predictions of the “With Statins” scenario to a scenario in which statins were
never invented. We implement this “Without Statins” scenario in two steps. First, we augment
the FEM to identify individuals who currently use statins. Second, we remove the well-
recognized health impact of statins from individuals identified as statin users. The differences in
outcomes between the scenarios reveal the impact of statins.
Since there is no HRS question about statin use, we turn to MEPS Prescribed Medicine
File data, which contains detailed information about prescription drugs and associated costs.
Since statin use is a dichotomous variable, we use a probit model to estimate the probability that
an individual is using a statin within the FEM simulations. We define a MEPS respondent as a
statin user if he or she filled at least one prescription associated with the “HMG-CoA reductase
3
Since the HRS is biennial, we simulate health and costs over two-year periods.
67
inhibitor” therapeutic subclass
4
during a given year, or a prescription of “Simcor”, “Advicor” or
“Vytorin” (which combine statins with other active ingredients).
The MEPS data reveals the dramatic increase in statin use over the recent years. In 2000,
about 16 million Americans had filled at least one statin prescription in the previous year. By
2006, that number doubled. Then, by the end of the decade, the number of Americans purchasing
statins reached 40 million (Figure A1, in the Electronic Supplementary Material). Remarkably,
over half of the obese elderly population now uses statins (Figure A2, in the Electronic
Supplementary Material).
5
To take these factors into account, we restrict our analysis to 2009 to
2011 MEPS data. This ensures that our statins use imputation is consistent with 2010, the first
year of our simulations. More detail about this task and the results of our estimations are
presented in the Electronic Supplementary Material.
The benefits of statin therapy in primary and secondary prevention of cardiovascular disease
have been widely studied and well reported by many researchers. Table 1 (available in the
Electronic Supplementary Material) summarizes the most important findings from recent
literature.
As mentioned above, the “Without Statins” scenario alters the FEM’s health transition
module to realistically remove the health benefits of statins from individuals who are identified
as statin users by the simulations. For these individuals, we increase the probabilities of
contracting diseases and mortality, using the inverse of the health effects of statins documented
in Table 1 of the Electronic Supplementary Material:
4
Therapeutic classes in MEPS correspond to Multum Lexicon variables from Cerner Multum,
inc.
5
Authors’ calculations with Medical Expnditure Panel Survey (MEPS) data.
68
1. Heart disease: We increase the probability of contracting a heart disease by factors
with a mean of 1.30. The distribution used corresponds to the inverse of the risk ratios
of contracting a non-fatal
6
cardio-vascular disease among the primary prevention
population reported by the Cochrane collaboration.[12]
2. Stroke: We increase the probability of contracting a stroke by factors with a mean of
1.45, based on the Cochrane collaboration’s non-fatal stroke primary prevention result.
3. Mortality: The literature indicates that statins have differential mortality effects
depending on sex and whether they are used for primary prevention or secondary
prevention. We thus make the following adjustments:
a. For individuals who have never been diagnosed with a stroke or a heart disease,
we increase the mortality probability by factors with a mean of 1.16, based on
the primary prevention risk ratios reported by the Cochrane collaboration.
b. For men and women who have been diagnosed with a heart disease, we
increase the mortality probability by factors with means of 1.27 and 1.09,
respectively, based on Gutierrez et al.[10]
To account for uncertainty in statin effectiveness, we sampled new estimates of the
clinical effect of statins from the confidence intervals for relative risks reported by the literature.
Assuming parameters to be independent from each other, we drew 200 sets of risk ratio estimates
from a log-normal distribution and computed the factors described above. To obtain robust
simulation results for each set of estimates, we ran 100 repetitions of the cohort simulations for
each draw. We then computed and sorted the 200 means (one per draw) over these repetitions for
all our variables of interest. In the remainder of the article, the point estimates of our results
6
This is consistent with the FEM health transition models, which are estimated only for live
individuals.
69
correspond to the mean of each variable of interest across the 200 draws and their 100
repetitions. The bounds of the 95% confidence intervals correspond to the 5
th
lowest and highest
results for each variable of interest. These intervals can be interpreted as simulated 95%
confidence intervals with regards to the clinical uncertainty of the effectiveness of statins.
4.2.3 Isolating the Impact of Obesity
The key objective of this article is to isolate the difference in the impact of statins
between otherwise identical healthy-weight and obese individuals. A challenge in obtaining this
differential impact arises because of the obesity gradient in the distribution of chronic conditions.
For instance, the prevalence of diabetes and heart disease at age 51 is significantly higher among
the obese population than among the population with a healthy BMI.
To address this issue, we start by assigning a healthy BMI of 24.9 to each individual with
a BMI greater or equal to 25 in our 2010 cohort while keeping other states unchanged. We
conduct simulations of our “With Statins” and “Without Statins” scenarios for this healthy-
weight cohort. We find the impact of statins on lifetime medical costs, length of life, and quality
of life associated with this BMI. We then repeat this process for BMIs of 30 to 40 by units of 5.
The difference in the effect of statins obtained with one of these higher BMIs and the healthy
weight reflects the marginal effect of statins for individuals of that BMI, since the distribution of
health and economic statuses is identical otherwise. This strategy is outlined in Figure 2 and the
effects of obesity and statin use on health are expanded upon in greater detail in Appendix B, in
the Electronic Supplementary Material. Overall, we expect obese individuals to benefit more
from the existence of statins in the “With Statins” scenario and see their health deteriorate faster
in the “Without Statins” scenario.
70
4.3 Results
4.3.1 Effects of Statins on Life Expectancy and Functional Status
Our simulations show the impact of the widespread use of statins on life expectancy.
According to our results, a 51-year-old American with a healthy BMI gains on average 1.2 year
of life because of the existence of statins. This value corresponds to the difference in life
expectancy between the “With Statins” (33.2 years) and the “Without Statins” scenario (32.0
years) when imposing a BMI below 25 to the FEM’s 2010 cohort.
As expected, life expectancy gains from statins are greater for obese individuals. Figure 3
shows a measure of the health cost of obesity, the difference in life expectancy between
otherwise identical obese individuals and their healthy-weight counterparts. Each bar can be
interpreted as the estimated loss of life expectancy that can be expected if an average individual
traded a healthy weight for an obese weight. For a given BMI, the difference between the “With
Statins” and the “Without Statins” bars correspond to the differential gain in life expectancy of
statins for obese people (the difference-in-difference illustrated in Figure 2). When compared to
healthy-weight Americans, life expectancy gains are estimated to be 5%-6% (0.06 to 0.07 years)
higher for people with obese BMIs. While these differences are modest, they are statistically
significant
7
and support the notion that statins lower the health consequences of obesity. We
emphasize that these results do not mean that statins are not effective for obese people, but mean
that statins are almost as effective for people with a healthy BMI as they are for obese people.
As discussed in the Electronic Supplementary Material’s Appendix 2, as health evolves
with the simulations, obese individuals are more likely to be ill and benefit from the secondary
mortality prevention effect of statins (i.e. after the onset of heart disease). In contrast, healthy-
7
As detailed in section 2.2.1, statistical significance refers to the 95% confidence intervals with
regards to the clinical uncertainty of the effectiveness of statins.
71
weight individuals are more likely to have never had a stroke or a heart disease, and thus likely
benefit more from the primary prevention effects of statins. It therefore matters whether the life
expectancy gains of Figure 3 are in good health or disabled.
To examine the composition of these gains, we decompose the additional life expectancy
gains for obese individuals by functional status in Figure 4. The bars correspond to the
difference-in-difference in life expectancy with a functional status due to both the existence of
statins and to having an obese BMI. We find that the estimated additional gain in longevity for
obese people is predominantly composed of time living with a disability. Difference-in-
difference estimates of healthy life-years are either not significant (for obesity classes 1 and 2),
or significantly negative (for obesity class 3).
Taken together, these estimates indicate that statins somewhat reduce the health cost of
being obese as measured by life expectancy, but this gain mainly amounts to time in bad health.
4.3.2 Effects of Statins on Health Care Costs
Of course, by delaying the onset of chronic conditions and mortality, statins also impact
medical costs. To quantify this impact, we computed the present value of the expected lifetime
medical costs of our cohorts at age 51, discounted with a 3% interest rate. The detailed results
are shown in Table 2.
Since medical costs in MEPS and MCBS include prescription drugs, the direct cost of
statin therapy is included in both the “With Statins” and “Without Statins” projections. In 2010,
according to MEPS data, the average annual costs of statins prescriptions for statin users aged
over 51 was of $532. Of this amount, $125 and $26 were paid for by Medicare and Medicaid,
respectively. For the “With Statins” scenario, we subtracted these values from the FEM’s cost
72
projections for all individuals identified as statin users in the simulations and reassigned them to
the “Statins” row. For the “Without Statins” scenario, we removed these costs altogether from
the FEM’s cost projections.
8
Our estimates indicate that statins have a modest impact of about $25,000-$30,000 on
lifetime health care costs for individuals of all BMIs. Of this amount, 25% stems from the direct
costs of statins prescriptions, mainly paid for privately. The remainder corresponds to medical
costs linked to the health benefits of statins. These notably arise because of the secondary
prevention effect of statins, which enables individuals to live and require care longer after the
onset of cardiovascular diseases. Thus, 60% of the additional costs are incurred when individuals
are disabled or living in a nursing home. Nursing home living is the highest contributor to the
costs of statins and is largely paid for by the Medicaid program. Overall, Medicare and Medicaid
spending account for over half of the added costs.
While our estimates reveal a BMI gradient in the additional costs of statins, it is quite
small: $2,800 for class 1 obesity, $3,500 for class 2 obesity, and $4,000 for class 3 obesity. This
gradient is consistent with the life expectancy gains described in Section 4.3.1, and indicate that
statins increase the healthcare costs of being obese, albeit mildly.
8
By doing so, we assume that the direct costs of statin therapy for the elderly will be stable at
their 2010 levels in the future. This is consistent with the remainder of the FEM’s cost
projections, which surmise that costs of treatments remain proportional over time. The 2010
values however constitute a higher bound for the future cost of statin prescriptions, given
atorvastatin’s patent expiration in 2011 and rosuvastatin’s forthcoming patent expiration in 2016.
The 2010 prescription costs are converted in 2009 dollars, the base year for the FEM’s costs
projections, using the U.S. City Average CPI from the Bureau of Labor Statistics.
73
4.3.3 Cost-Effectiveness of Statins
In Table 3, we compare the additional health care costs of statins with their health benefits.
We consider total life expectancy gains (including time disabled or in a nursing home),
disability-free life-years and quality-adjusted life-years. Again, our results show a BMI gradient,
especially with regards to disability and quality of life. For instance, since individuals with class
3 obesity gain less healthy life-years and incur more costs, the estimated cost of statins per non-
disabled life-year is $10,900 higher for them than for healthy-weight individuals. Despite this
higher unit cost, we find that, for any reasonable value of life, statins are estimated to be quite
cost-effective for all health effectiveness measures and all BMIs. The expiration of atorvastatin’s
patent in 2011 has likely increased the use of generics and further improved the cost-
effectiveness of statins. We expect that rosuvastatin’s patent expiration in 2016 will continue this
trend.
According to MEPS data, 56.7 million Americans were aged 50 to 64 in 2010. Of these,
34% were obese and 26% filled a statin prescription during the previous year. Based on our
estimates, this cohort will have collectively lived 67.6 million life-years more than it would have
if statins did not exist. Of these, 33.2 million life-years will be lived free of disability. These
additional life-years gained by obese people will be associated with $1.5 trillion additional
lifetime healthcare costs, for an average of $22,200 per life-year gained. Earlier, we saw that our
estimates of life expectancy gains and costs of statins are modestly higher for obese individuals.
When aggregated to the level of 2010’s cohort aged 50 to 64, this gradient sums up to 1.3 million
additional life-years and $61 billion additional lifetime medical costs.
74
4.3.4 Effect of Statins on the Cost of Obesity
In the previous sections, we saw that the existence and widespread use of statins have large
longevity and health impacts for both healthy-weight and obese individuals, contributing about
one year to life expectancy at age 51. However, we noticed only a modest gradient of life
expectancy (and lower QALY gains) with BMI, so it is unclear at this stage whether statins have
a noticeable impact on the total costs of obesity. In other words, how do the additional longevity
gains and medical costs of statins for obese people compare to the health and medical care costs
of obesity?
In Table 4, we present estimates of the total costs of obesity in both scenarios, accounting
for both medical care and health costs. The medical care cost of being obese corresponds to the
additional lifetime medical costs at age 51 that can be expected when an individual moves from a
healthy BMI to an obese BMI.
9
In Panel A., our health cost concept refers to the loss in QALY
that can be expected when an individual moves from a healthy BMI to an obese BMI. We assign
a value in $2009 to this health cost by assuming a value of $150,000 per QALY [22]. In Panel B,
we conduct the same calculations with raw life expectancy losses, assuming a value of $100,000
per year.
In a world without statins, we estimate that 51-year-olds with obesity class 1, 2, and 3
would respectively incur $14,000, $21,000, and $25,800 higher lifetime healthcare costs than
average healthy-weight individuals, and could expect to live 1.9, 3.1, and 4.1 less QALYs. With
our assumption of the value of QALYs, the total costs of obesity class 1, 2, and 3 if statins did
not exist would be $303,500, $488,200, and $636,400.
9
These are computed with the present values presented in Table 2. For instance, the medical care
cost of class 2 obesity in the “With Statins” scenario ($24.5) corresponds to the difference
between the total medical costs with BMI = 35 and BMI < 25 in the “With Statins” column
(485.5 and 460.9, respectively).
75
In the “With Statins” scenario, the medical costs of obesity increases somewhat over the
“Without Statins” scenario, while the difference in QALYs between BMIs remain mostly
unchanged. This implies estimates of total costs of obesity class 1, 2, and 3 of $303,700,
$492,000, and $643,900, respectively.
The last rows show the difference in total costs due to the existence of statins. These values
correspond to the difference-in-difference in health and medical care cost when considering both
the existence of statins and obesity. Our estimates indicate that the availability of statins has no
significant impact on the healthcare costs of obesity class 1 and 2, and increase the costs of
obesity class 3 by $7,500 (1.2% of the cost of obesity if statins did not exist). This occurs
because statins are about as beneficial to the expected QALYs of all BMIs, and have a higher
impact on the lifetime medical costs of obese individuals. The added costs stem from the
additional time disabled gained by obese individuals, as illustrated by Figure 4.
In Panel B, we replicate the calculations of Panel A without accounting for quality-of-life. In
this panel, all length-of-life lost due to obesity are valued equally at $100,000 per year. With this
health cost concept, we find significantly positive effects of statins on the cost of obesity classes
1 and 2. These different findings reflect the reality that much of the difference in life expectancy
gains for obese individuals is spent with some level of disability. For both health concepts, we
find that the effect of statins is quite small relative to the cost of obesity.
4.3.5 Sensitivity Analyses
To further our understanding of the factors at play behind these results, we consider two
sensitivity analyses for the effect of statins on the costs of obesity.
76
First, a plausible explanation for the small impact of statins on the cost of being obese is our
very concept of the cost of obesity. In this paper, we reassigned the BMI of the FEM’s nationally
representative 2010 cohort to specific values and conducted separate simulations with these
synthetic cohorts (as described in Section 2.2.2). We thus interpreted the costs of obesity as the
health consequences and medical costs of trading a healthy BMI for an obese BMI at age 51,
when other variables (including health) remain unchanged. An alternative interpretation would
be that the costs of obesity should include comorbidities associated with obesity at the beginning
of the simulations (higher prevalence of diabetes, hypertension, heart disease, etc.).
Hypothetically, statins could have a larger impact under this interpretation. In Appendix C1, we
adopted this concept and conducted simulations without reassigning BMIs. While we found
larger effects of statins on the life expectancy of obese individuals, these effects remain small
relative to the costs of obesity.
Second, another factor that could be preventing statins from lowering the costs of obesity is
their effectiveness for the primary prevention of diseases and mortality. Since healthy-weight
individuals are more likely than obese individuals to avoid heart disease and stroke, they may
benefit from primary prevention effects of statins over more years, while obese individuals
benefit more from secondary prevention. To test this hypothesis, we conducted a second
sensitivity analysis, in which we only removed the secondary prevention effects of statins in the
“Without Statins” scenario. This analysis is presented in the Electronic Supplementary Material’s
Appendix C2. Our results do show an increase in the differential effectiveness of statins with
higher BMIs, which supports the notion that medical innovations aimed at preventing disease
may be more beneficial to non-obese people. To significantly reduce the costs of obesity,
innovations would thus need to treat, rather than prevent, disease. However, the differential
77
effects of statins on life expectancy and quality-of-life for obese people remain small in
comparison with the costs of obesity, and are associated with higher medical costs. When taking
both medical costs and health consequences into account, we do not find a significant net
reduction in the total costs of obesity.
Overall, both analyses confirm the robustness of the small effect of statins on the costs of
obesity.
4.5 Discussion
Obesity, given its high prevalence, impacts both individual health and the American fiscal
outlook. As medical technology evolves over time, health outcomes for a number of obesity-
related illnesses improve, changing the life outlook for obese people. In this article, we asked if
such innovation could effectively alleviate the health and health care costs of obesity. We used
statins, one of the most effective medical innovations of the recent decades, as a case study.
In many ways, statins constitute a best-case scenario of medical innovations capable of
lowering the burden of obesity. Statin therapy is inexpensive and has been shown to be quite
effective at both preventing the onset of diseases for which obesity is a risk factor and decreasing
their health impact.
Our main findings are fivefold. First, the current use of statins significantly extends life,
with estimated gains in life expectancy at age 51 of over one year for all BMIs. Second, life
expectancy gains are modestly higher for obese individuals than healthy-weight individuals.
Third, most of the additional gain in life expectancy of obese people is lived in poor health, with
at least some level of disability. Fourth, this additional time lived disabled is associated with
higher healthcare costs. Finally, when taking both quality of life and medical costs into account,
our simulations indicate that statins have little or no effect on the costs of obesity.
78
Our study has several limitations. Notably, we do not take into account the potential moral
hazard effect of statins in our “Without Statins” scenario: by lowering the probability of
contracting chronic diseases and reducing mortality, statins may reduce the incentive for
preventive behavior. In concept, the emergence of statins may explain some of the trend towards
higher obesity rates, by lowering the health cost of being obese. A recent study revealed that
caloric and fat intake of statins users have indeed grown from 1999 to 2010. However, the study
could not determine if the changes stemmed from an expanded use of statins among people who
eat more or from a change in the eating habits of statin users in response to the medical
efficiency of statins.[23] In another recent study, Kaestner and colleagues find that statin use is
associated with conflicting behavioral responses.[24] On the one hand, they find an increase in
BMI and the probability of being obese among both men and women, as well as an increase in
moderate or greater alcohol consumption. On the other hand, statin use leads to increased
exercise rates among men. Overall, their results appear to support a net substitution between the
medical effectiveness of statins and healthy behaviors. If such responses were introduced in our
model, they would reduce the health effectiveness of statins. Thus, our results can be viewed as
an upper bound of the impact of statins on the costs of obesity.
Also, this study is based on the premise that the health effects of statins reported by the
randomized clinical trials literature can be generalized to all statin users, regardless of the
number of filled prescriptions during the year or their characteristics. In practice, real world
effects of statins are bound to differ from those obtained in a clinical trial setting. However,
observational research has shown that the effectiveness of statins in community practice is
comparable to what has been observed in clinical trials [25]. Moreover, since the FEM does not
include biomarker data, cholesterol levels do not directly contribute to the probability of
79
mortality and contracting diseases in the simulation. Thus, even if the adjustments used in the
“Without Statins” scenario are correct, we may underestimate the probability of contracting
diseases and mortality for individuals identified as statin users. Since this is true for both
scenarios and other key risk factors, such as diabetes and BMI, are included in the transition
models, the bias in the difference between scenarios is likely to be small.
Finally, while we accounted for the clinical uncertainty of the effectiveness of statins, other
sources of uncertainty are more difficult to correctly account for and were ignored. For example,
there could also be changes over time in the mortality benefits of statin use, as other treatments
for clogged arteries and heart disease continue to improve; the medical care costs associated with
various conditions and states will change over time; and transition probabilities between health
states will also change over time with changes in medical technology. These are all reasons for
caution when predicting the changes in lifetime costs associated with a given treatment. Thus,
our results are indicative of the effect of the innovation of statins on lifetime outcomes when
assuming that medical technology remains constant over time, which is of course unrealistic.
The results presented here refute the notion that people do not need to watch their weight as
much as in the past because of the availability of statins and other medical innovations. This is
particularly true when taking factors such as quality of life and disability into account. Given
these results, policymakers and practitioners should direct greater attention towards obesity
prevention, as opposed to treatment of the consequences of obesity. This research should also
serve as an example to researchers studying the economics of obesity of the difficulty of
predicting the net effects on the costs of obesity from treatments that may, at first glance, seem to
have obviously significant effects.
80
While our estimates indicate that the innovation of statins increased life expectancy at age
51 by roughly 1.2 years, life expectancy gains have been only slightly greater for the obese
compared to individuals with a healthy BMI. The small marginal gain in life expectancy for the
obese compared to the non-obese (0.06 – 0.07 years) was mostly associated with some level of
disability. A similar story emerged with respect to lifetime healthcare costs. Statins increased
healthcare costs about roughly $25,000 - $30,000, but the marginal increase estimates for the
obese compared to the non-obese was relatively small ($2,800 - $4,000). Therefore, after valuing
the quantity and quality of differences in life expectancy, we find that the innovation of statins
has had little impact on simulated the cost of obesity (the sum of both health and healthcare
costs). Statins significantly reduce the risk of heart disease, stroke, and mortality, but because
they do so for both the obese and non-obese, they have not significantly reduced the cost of
obesity.
81
References
1. Fryar CD CM, Ogden CL, . Prevalence of Overweight, Obesity, and Extreme Obesity
Among Adults: United States, Trends 1960–1962 Through 2009–2010: National Center for
Health Statistics; 2012.
2. James WPT J-LR, Mhurchu CN, Kalamara E, Shayeghi M, Rigby NJ, Nishida C,
Rodgres A. Overweight and Obesity. Comparative Quantification of Health Risks: World Health
Organization; 2004.
3. Colditz GA. Economic costs of obesity. The American journal of clinical nutrition. 1992
Feb;55(2 Suppl):503S-7S.
4. Finkelstein EA, Fiebelkorn IC, Wang G. National medical spending attributable to
overweight and obesity: how much, and who's paying? Health affairs. 2003 Jan-Jun;Suppl Web
Exclusives:W3-219-26.
5. Finkelstein EA, Trogdon JG, Cohen JW, Dietz W. Annual medical spending attributable
to obesity: payer-and service-specific estimates. Health affairs. 2009 Sep-Oct;28(5):w822-31.
6. Cawley J, Meyerhoefer C. The medical care costs of obesity: an instrumental variables
approach. Journal of health economics. 2012 Jan;31(1):219-30.
7. Chen YH, Feng B, Chen ZW. Statins for primary prevention of cardiovascular and
cerebrovascular events in diabetic patients without established cardiovascular diseases: a meta-
analysis. Experimental and clinical endocrinology & diabetes : official journal, German Society
of Endocrinology [and] German Diabetes Association. 2012 Feb;120(2):116-20.
8. Cholesterol Treatment Trialists C, Mihaylova B, Emberson J, Blackwell L, Keech A,
Simes J, et al. The effects of lowering LDL cholesterol with statin therapy in people at low risk
of vascular disease: meta-analysis of individual data from 27 randomised trials. Lancet. 2012
Aug 11;380(9841):581-90.
9. de Vries FM, Denig P, Pouwels KB, Postma MJ, Hak E. Primary prevention of major
cardiovascular and cerebrovascular events with statins in diabetic patients: a meta-analysis.
Drugs. 2012 Dec 24;72(18):2365-73.
10. Gutierrez J, Ramirez G, Rundek T, Sacco RL. Statin therapy in the prevention of
recurrent cardiovascular events: a sex-based meta-analysis. Archives of internal medicine. 2012
Jun 25;172(12):909-19.
82
11. Mills EJ, Wu P, Chong G, Ghement I, Singh S, Akl EA, et al. Efficacy and safety of
statin treatment for cardiovascular disease: a network meta-analysis of 170,255 patients from 76
randomized trials. QJM : monthly journal of the Association of Physicians. 2011
Feb;104(2):109-24.
12. Taylor F, Huffman MD, Macedo AF, Moore TH, Burke M, Davey Smith G, et al. Statins
for the primary prevention of cardiovascular disease. The Cochrane database of systematic
reviews. 2013;1:CD004816.
13. Grabowski DC, Lakdawalla DN, Goldman DP, Eber M, Liu LZ, Abdelgawad T, et al.
The large social value resulting from use of statins warrants steps to improve adherence and
broaden treatment. Health affairs. 2012 Oct;31(10):2276-85.
14. Michaud PC, Goldman DP, Lakdawalla DN, Zheng Y, Gailey AH. The value of medical
and pharmaceutical interventions for reducing obesity. Journal of health economics. 2012
Jul;31(4):630-43.
15. Livingston EH. The incidence of bariatric surgery has plateaued in the U.S. American
journal of surgery. 2010 Sep;200(3):378-85.
16. Goldman DM, et al. Health status and medical treatment of the future elderly: Final
Report. Santa Monica, CA: RAND Corporation; 2004.
17. Bhattacharya J, Shang B, Su CK, Goldman DP. Technological advances in cancer and
future spending by the elderly. Health affairs. 2005;24 Suppl 2:W5R53-66.
18. Lakdawalla DN, Goldman DP, Shang B. The health and cost consequences of obesity
among the future elderly. Health affairs. 2005;24 Suppl 2:W5R30-41.
19. Goldman DP, Zheng Y, Girosi F, Michaud PC, Olshansky SJ, Cutler D, et al. The
benefits of risk factor prevention in Americans aged 51 years and older. American journal of
public health. 2009 Nov;99(11):2096-101.
20. Bhattacharya J, Sood N. Who pays for obesity? The journal of economic perspectives : a
journal of the American Economic Association. 2011 Winter;25(1):139-58.
21. Merrill RM, Richardson JS. Validity of self-reported height, weight, and body mass
index: findings from the National Health and Nutrition Examination Survey, 2001-2006.
Preventing chronic disease. 2009 Oct;6(4):A121.
22. Macroeconomics and Health: Investing in Health for Economic Development. Geneva,
Switzerland: World Health Organization; 2001 2001.
83
23. Sugiyama T, Tsugawa Y, Tseng CH, Kobayashi Y, Shapiro MF. Different Time Trends
of Caloric and Fat Intake Between Statin Users and Nonusers Among US Adults: Gluttony in the
Time of Statins? JAMA internal medicine. 2014 Apr 24.
24. Kaestner R, Darden M, Lakdawalla D. Are investments in disease prevention
complements? The case of statins and health behaviors. Journal of health economics. 2014
Jul;36:151-63.
25. Wei L, Ebrahim S, Bartlett C, Davey PD, Sullivan FM, MacDonald TM. Statin use in the
secondary prevention of coronary heart disease in primary care: cohort study and comparison of
inclusion and outcome with patients in randomised trials. Bmj. 2005 Apr 9;330(7495):821.
84
Table 4.1 Health Impact of Statins
Source Population Effects of statin therapy compared to placebo
Cochrane Collaboration
2013[12]
Primary prevention of
cardiovascular disease
Reduced all-cause mortality (OR 0.86, 95% CI 0.79-0.94)
Reduced fatal (RR 0.83, 95% CI 0.72-0.96) and non-fatal (RR 0.77, 95% CI 0.62-0.96) CVD
Reduced fatal (RR 0.82, 95% CI 0.7-0.96) and non-fatal (RR 0.67, 95% CI 0.59-0.76) CHD
Reduced non-fatal stroke (RR 0.69, 95% CI 0.58-0.83)
Chen 2012[7] Primary prevention of
cardiovascular disease
among diabetics
Reduced incidence of MACCE (RR 0.79, 95% CI 0.66-0.95)
Reduced the risk of stroke (RR 0.71, 95% CI 0.54-0.94)
No significant effect on all-cause mortality (RR 0.79, 95% CI 0.58-1.08).
Cholesterol Treatment
Trialists 2012[8]
Primary prevention of
cardiovascular disease
Reduced risk of major vascular events (RR 0.79, 95% CI 0.77-0.81, per 1.0 mmol/L reduction in LDL)
Reduced risk of vascular mortality (RR 0.85, 95% CI 0.77-0.95, per 1.0 mmol/L reduction in LDL)
Reduced all-cause mortality (RR 0.91, 95% CI 0.85-0.97 per 1.0 mmol/L reduction in LDL).
de Vries 2012[9] Primary prevention of
cardiovascular disease
among diabetics
Reduced major cardiovascular or cerebrovascular events (RR 0.75, 95% CI 0.67-0.85)
Reduced fatal/non-fatal stroke (RR 0.69, 95% CI 0.51-0.92)
Reduced fatal/non-fatal MI (RR 0.70, 95% CI 0.54-0.90)
No significant effect on all-cause mortality (RR 0.84, 95% CI 0.65-1.09)
Mills 2011[11] Primary and secondary
prevention of
cardiovascular disease
Reduced all-cause mortality (RR 0.90, 95% CI 0.86-0.94)
Reduced CVD mortality (RR 0.80, 95% CI 0.74-0.87)
Reduced fatal (RR 0.82, 95% CI 0.75-0.91) and non-fatal (RR 0.74, 95% CI 0.67-0.81) MI
Reduced revascularization (RR 0.76, 95% CI 0.70-0.81)
Reduced fatal/non-fatal strokes (RR 0.86, 95% CI 0.78-0.95)
Gutierrez 2012[10] Secondary prevention
of cardiovascular
disease
Reduced CVD events in women (RR 0.81, 95% CI 0.74-0.89) and men (RR 0.82, 95% CI 0.78-0.85)
Reduced all-cause mortality (RR 0.79, 95% CI 0.72-0.87) in men
Reduced stroke (RR 0.81, 95% CI 0.72-0.92) in men
No significant effect on all-cause mortality (RR 0.92, 95% CI 0.76-1.13) in women
No significant effect on stroke (RR 0.92, 95% CI 0.76-1.10) in women
Wei 2005[25] Secondary prevention
of cardiovascular
disease
Significant reduction in all-cause mortality among overall community population (RR 0.69, 95% CI 0.59-
0.80), women (RR 0.63, 95% CI 0.49-0.80), and age ≥ 65 (RR 0.72, 95% CI 0.61-0.84)
Significant reduction in fatal/non-fatal MI among overall community population (RR 0.82, 95% CI 0.71-
0.95), women (RR 0.69, 95% CI 0.54-0.88), and age ≥ 65 (RR 0.84, 95% CI 0.71-0.99)
RR: Relative risk; OR: Odds ratio; CVD: Cardiovascular disease; CHD: Coronary heart disease; MI: Myocardial infarction; MACCE: Major cardiovascular and
cerebrovascular events
85
Table 4.2 Impact of Statins on Lifetime Medical Costs after Age 50, $2009 thousands
With
Statins
Without
Statins Difference
Private
(%)
Medicare
(%)
Medicaid
(%)
Healthy weight (BMI <
25)
Total Medical Costs 460.9 435.5 25.4 40 35 25
Excluding Statins 456.1 435.5 20.6 32 38 30
Free of disability 288.9 283.6 5.2 48 48 4
Disabled 96.9 91.7 5.1 38 56 6
In a nursing home 70.4 60.2 10.2 21 24 55
Statins 4.8 0 4.8 72 24 5
Obese class 1 (BMI = 30)
Total Medical Costs 477.7 449.5 28.2 39 36 24
Excluding Statins 472.2 449.5 22.6 32 39 29
Free of disability 284.4 279.0 5.4 47 49 4
Disabled 116.5 110.2 6.3 38 57 6
In a nursing home 71.3 60.3 11.0 20 25 55
Statins 5.6 0 5.6 72 24 5
Obese class 2 (BMI = 35)
Total Medical Costs 485.5 456.6 28.9 39 37 24
Excluding Statins 479.9 456.6 23.4 31 40 29
Free of disability 277.9 272.5 5.4 46 50 4
Disabled 131.2 124.3 6.9 38 57 6
In a nursing home 70.8 59.8 11.0 20 25 55
Statins 5.5 0 5.5 72 24 5
Obese class 3 (BMI = 40)
Total Medical Costs 490.7 461.3 29.4 39 38 24
Excluding Statins 485.3 461.3 24.0 31 41 28
Free of disability 268.1 262.8 5.3 46 50 4
Disabled 147.3 139.6 7.7 37 57 6
In a nursing home 69.8 58.9 11.0 19 26 55
Statins 5.4 0 5.4 72 24 5
All amounts are in present value at age 51, computed with a 3% interest rate. BMI refers to “body mass
index”, defined as the ratio between mass of individuals, expressed in kilograms, and the square of height,
expressed in meters. The first two columns present lifetime medical costs at age 51 in the “With Statins” and
“Without Statins” scenarios. The third column presents the difference between the scenarios, and corresponds
to the additional medical costs due to the existence of statins. The last three columns decompose the
additional costs due to statins by spending source. In the rows, we decompose medical costs by type (total
medical costs excluding statins prescriptions and statin prescriptions) and by functional status. A “disabled”
functional status refers to reporting at least one instrumental activity of daily living (IADL) or activity of
daily living (ADL) limitation. The “free of disability” status refers to reporting no IADL or ADL limitation
and not living in a nursing home. “In a nursing home” indicates the most severe functional status impairment.
Confidence intervals for this table are presented in Table A2.
86
Table 4.3 Cost per Health Gain of Statins
Healthy weight
(BMI = 24.9)
Obese class 1
(BMI = 30)
Obese class 2
(BMI = 35)
Obese class 3
(BMI = 40)
Costs ($2009 thousands) 25.4 [12.1 - 38.0] 28.2 [13.3 - 41.9] 28.9
[13.5 -
42.9] 29.4 [13.2 - 44.0]
1. Life expectancy gain
Years 1.17 [0.6 - 1.6] 1.24 [0.7 - 1.7] 1.24 [0.7 - 1.7] 1.23 [0.7 - 1.7]
Cost/year 21.4 [15.9 - 24.9] 22.5 [16.1 - 26.3] 22.9
[16.3 -
26.9] 23.5 [16.8 - 27.5]
2. Disability-free life
expectancy gain
Years 0.59 [0.4 - 0.8] 0.59 [0.4 - 0.8] 0.57 [0.4 - 0.7] 0.54 [0.3 - 0.7]
Cost/year 42.7 [24.9 - 57.2] 47.2 [26.0 - 63.4] 49.9
[27.2 -
68.3] 53.6 [29.6 - 74.1]
3. Expected quality-
adjusted life-years (QALY)
gain
Years 0.88 [0.5 - 1.2] 0.90 [0.5 - 1.2] 0.88 [0.5 - 1.2] 0.86 [0.5 - 1.2]
Cost/year 28.5 [19.3 - 34.9] 31.0 [20.0 - 38.5] 32.4
[20.3 -
40.3] 33.8 [21.4 - 42.1]
Values in the table correspond to the difference in medical costs and life expectancy between the “With Statins” and the
“Without Statins” scenarios. Costs are in present value at age 51, computed with a 3% interest rate. BMI refers to “body mass
index”, defined as the ratio between mass of individuals, expressed in kilograms, and the square of height, expressed in
meters. Disability-free life expectancy refers to reporting instrumental activity of daily living (IADL) or activity of daily
living (ADL) limitations and not living in a nursing home. Quality-adjusted life-years (QALYs) adjust length of life for
quality based on a person’s chronic conditions and functional status. 95% confidence intervals with regards to the uncertainty
of the effectiveness of statins are presented in brackets.
87
Table 4.4 Health and Healthcare Costs of Obesity in Both Scenarios, $2009 thousands
A. Health costs measured by difference in expected
QALYs
B. Health costs measured by difference in unadjusted
life expectancy
Obese class 1
(BMI = 30)
Obese class 2
(BMI = 35)
Obese class 3
(BMI = 40)
Obese class 1
(BMI = 30)
Obese class 2
(BMI = 35)
Obese class 3
(BMI = 40)
Without Medical care cost ($) 14.0 21.0 25.8 14.0 21.0 25.8
Statins
[12.3 - 15.8] [18.6 - 23.5] [22.8 - 28.7]
[12.3 - 15.8] [18.6 - 23.5] [22.8 - 28.7]
Health cost
Years / QALYs lost 1.90 3.10 4.10
1.20 2.00 2.90
[1.9 - 2.0] [3.1 - 3.2] [4.0 - 4.1]
[1.1 - 1.2] [2.0 - 2.1] [2.8 - 2.9]
Value of years/QALYs lost
($) 289.5 467.2 610.6
116.3 201.1 287.8
[285.3 - 293.0] [461.6 - 472.5] [604.4 - 617.0]
[112.7 - 119.9] [195.6 - 205.9] [280.8 - 293.9]
Total cost of obesity ($) 303.5 488.2 636.4
130.3 222.2 313.6
[298.8 - 307.7] [481.8 - 494.3] [629.3 - 644.3]
[127.0 - 133.5] [217.7 - 226.4] [308.5 - 318.3]
With statins Medical care cost ($) 16.8 24.5 29.8
16.8 24.5 29.8
[15.9 - 17.6] [23.5 - 25.6] [28.3 - 30.9]
[15.9 - 17.6] [23.5 - 25.6] [28.3 - 30.9]
Health cost
Years / QALYs lost 1.90 3.10 4.10
1.10 1.90 2.80
[1.9 - 1.9] [3.1 - 3.1] [4.1 - 4.1]
[1.1 - 1.1] [1.9 - 2.0] [2.8 - 2.9]
Value of years/QALYs lost
($) 286.9 467.5 614.1
109.7 194.1 281.7
[283.3 - 289.8] [462.2 - 471.4] [608.2 - 619.1]
[106.6 - 112.1] [189.3 - 198.0] [276.5 - 286.4]
Total cost of obesity ($) 303.7 492.0 643.9
126.5 218.7 311.4
[300.5 - 306.4] [487.5 - 495.7] [638.2 - 648.4]
[123.8 - 128.6] [214.9 - 221.4] [307.0 - 315.3]
Difference
due to
statins
Total cost of obesity ($) -0.2 -3.8 -7.5
3.8 3.5 2.2
[-3.7 - 3.7] [-8.4 - 2.3] [-13.9 - -1.2]
[1.1 - 6.8] [0.4 - 6.8] [-1.2 - 5.4]
Percent (%) -0.1 -0.8 -1.2
2.9 1.6 0.7
[-1.2 - 1.2] [-1.7 - 0.5] [-2.2 - -0.2] [0.8 - 5.2] [0.2 - 3.0] [-0.4 - 1.7]
The medical care cost of obesity is the difference in the sum of expected medical costs at age 51 between identical individuals of the obese category and the
healthy-weight category (BMI<25) presented in Table 2. BMI refers to “body mass index”, defined as the ratio between mass of individuals, expressed in
kilograms, and the square of height, expressed in meters. The total cost of obesity is defined as the sum of the medical cost of obesity and the value of the
health cost of obesity, expressed in dollars. Medical costs are in present value, computed with a 3% interest rate. The value of health costs of obesity is
obtained by attributing a nominal value of $150,000 per Quality-adjusted life-year (QALY) lost due to obesity in panel A., and of $100,000 per life-year lost
in panel B. Quality-adjusted life-years (QALYs) adjust length of life for quality based on a person’s chronic conditions and functional status. 95% confidence
intervals with regards to the uncertainty of the effectiveness of statins are presented in brackets.
88
Figure 4.1 Chronic Conditions Transitions in the Future Elderly Model
Healthy Contract condition
Die
Stay healthy
1-q1-q2
q1
q2
1-p
Ill
Die
Stay ill
1-u
u
p
89
Figure 4.2 Empirical Strategy
Both the “With Statins” and the “Without Statins” scenarios are run four times: first, with all individuals given a
healthy BMI; then, with all individuals given an obese BMI (30, 35, or 40 in separate simulations). For each
scenario, the differences in outcomes between obese and healthy-weight simulations reveal the cost of obesity. The
difference in these differences (the difference-in-difference) reveals the impact of statins on the cost of obesity.
90
Figure 4.3 Life Expectancy Cost of Obesity in Both Scenarios after Age 50
BMI refers to “body mass index”, defined as the ratio between mass of individuals, expressed in kilograms, and the
square of height, expressed in meters. The “life expectancy cost of obesity” is the simulated difference in life
expectancy of otherwise identical 51-year-olds with obese BMIs (we consider values of 30, 35 and 40) compared to
individuals with a healthy BMI. 95% confidence intervals with regards to the uncertainty of the effectiveness of
statins are shown for each bar.
91
Figure 4 Difference-in-difference in Life Expectancy Gains after Age 50, by Functional Status
BMI refers to “body mass index”, defined as the ratio between mass of individuals, expressed in kilograms, and the
square of height, expressed in meters. A “disabled” functional status refers to reporting at least one instrumental
activity of daily living (IADL) or activity of daily living (ADL) limitation. The “free of disability” status refers to
reporting no IADL or ADL limitation and not living in a nursing home. “In a nursing home” indicates the most
severe functional status impairment. 95% confidence intervals with regards to the uncertainty of the effectiveness of
statins are shown for each bar.
92
APPENDIX A. Assigning Statin Use in the Future Elderly Model
A.1 Data
To estimate the probability that individuals are statin users within FEM simulations, we
utilize estimates based on Medical Expenditure Panel Survey (MEPS) data. The MEPS is a set of
large-scale surveys of families and individuals, their medical providers and employers across the
United States. The MEPS collects data on the specific health services that Americans use, how
frequently they use them, the cost of these services, and how they are paid for, as well as data on
the cost, scope, and breadth of health insurance held by and available to U.S. workers.
We use two components of the MEPS. The first is the Household Component (HC), which
collects data from a sample of families and individuals in selected communities across the United
States, drawn from a nationally representative sample of the civilian non-institutionalized
population of the United States. These constitute a subsample of households that participated in
the prior year's National Health Interview Survey (conducted by the National Center for Health
Statistics). During the household interviews, MEPS collects detailed information for each person
in the household, including demographic characteristics and health conditions. The second is the
Prescribed Medicines files, which provide detailed information on household-reported prescribed
medicines, which can be used to make estimates of prescribed medicine utilization and
expenditures. Each record on this event file represents a unique prescribed medicine event; that
is, a prescribed medicine reported as being purchased by the household respondent.
By merging the two components for years 2009-2011, we obtained an individual-level
database including individual characteristics and statin use. A MEPS respondent is defined as a
statin user if he or she filled at least one prescription associated with the “HMG-CoA reductase
inhibitor” therapeutic subclass during a given year, or a prescription of “Simcor”, “Advicor” or
“Vytorin” (which combine statins with other active ingredients). Therapeutic classes in MEPS
correspond to Multum Lexicon variables from Cerner Multum, inc. The years 2009-2011 are
chosen to surround 2010, the first year of our simulations. We also used MEPS 2008 data to find
the lag of statin use in 2009.
93
A.2 Estimation
Since we define statin use as a dichotomous variable, we use probit regressions on our
MEPS dataset to find the probability that individuals are using statins at each period in the FEM
simulations. We estimate these probabilities in two steps. The first regression finds the factors
influencing the probability that 50 to 53 year-old individuals use statins, based on information
common to the MEPS and the FEM: demographics, education, health conditions and BMI. The
regression estimates are used within FEM to assign the probability of statin use at ages 51-52, in
the first period of the simulations. The second regression uses the MEPS 2-year overlapping
panel design to find the probability of using statins for the population aged 53 and over. This
regression is limited to individuals in their second interview year, and adds age splines and the
use of statins in the previous year (by far the largest predictor of statin use) to the list of
regressors. After several specification checks, we chose nodes at age 70 and BMI 30 for the age
and log of BMI splines in order to maximize the fit. The estimates of this regression are used to
assign the transitions in and out of using statins in each period of the simulations. The results of
both regressions are presented in Table A1.
A.3 STATIN ASSIGNMENT
The use of these estimates in the assignment of statin use follows the same methodology
as the other non-absorbing binary outcomes in the FEM, which are detailed in the complete
technical appendix provided in the Electronic Supplementary Material. Since the FEM periods
are two-year long and the MEPS only follows respondents longitudinally for a second year, we
first assign statin use during the next year, which corresponds to the FEM’s mid-period. This
mid-period statin use then becomes the lag of statin use, based on which we assign statin use for
the next FEM period.
Thus, MEPS data is used to assign statin use to our FEM cohort, which is based on HRS
data. While the two surveys are representative of the American population they study, it is
initially unclear whether our assignment of statin use will be consistent with observed data. To
verify this, we use our methodology to assign a probability of statin use to each individual of the
FEM cohort of age 51-52 in 2010.
10
We then compare the average probability of using statins in
10
This cohort is the baseline for those used in the remainder of this article, before we reassign BMI.
94
this cohort to observed statin use in the MEPS for the same ages in years 2009 to 2011. We focus
on statin use by BMI and chronic condition diagnostic (Figure A3).
We find that our assigned probability of using statins mirrors quite well the observed
prevalence of statin use in the MEPS. We observe some divergence for individuals with very
high BMIs, for which statin use probability in FEM is about 10% lower than prevalence in the
MEPS. The MEPS however counted less than 35 respondents for these BMI values. It is of note
that we cannot compare statin use by age between the FEM simulations and the MEPS data
because of cohort effects, since we simulate the health of the cohort aged 51-52 in 2010. For
instance, our FEM cohort at age 71-72 (in 2030) is quite different from respondents of the same
age in the available MEPS data.
APPENDIX B. Effects of Statin Use and Obesity on Health Dynamics and
Mortality in the Future Elderly Model
As summarized in Figure A4, obesity has an intricate relationship with the health
conditions in the FEM. BMI is a significant predictor of key health conditions including diabetes,
stroke, heart diseases, and hypertension, which have important dynamic consequences for the life
trajectories of individuals. As shown in the figure, having one or several chronic conditions
increases the probability of contracting additional ones, which contribute to the probability of
mortality. To clarify the mechanisms at play in the simulations, we synthesize FEM transitions
into illness and death in Figure 1. This figure simultaneously applies to all chronic conditions in
the FEM.
An individual has probability p of already suffering from a chronic condition at a given
age, and probability 1-p of being free of that condition. Conditional on him being healthy, he has
a probability q
1
of contracting the disease, a probability q
2
of dying and a probability 1-q
1
-q
2
of
staying healthy in the next period. If the individual already has the disease, he has a probability u
of dying and a probability 1-u of surviving with the illness. Obesity increases the probability that
an individual is unhealthy in a given year (p) and the probability that a healthy individual
contracts a new condition (q
1
). The widespread use of statins has four impacts: it increases the
probability that an individual is healthy in a given year (1-p), decreases the probability of
95
contracting heart diseases and stroke (q
1
), and lowers the probability of dying if healthy (q
2
) or
afflicted with heart disease (u).
The net effect of the interaction of obesity and statins is complex. First, for otherwise
identical individuals free of heart disease and stroke, the reduction of q
1
due to statin use has
more impact on the obese, since their base probability of contracting the conditions is higher.
Second, since obesity does not directly enter the FEM’s mortality prediction and only
affects mortality through the onset of chronic conditions, the effect of statins on mortality (q
2
and
u) is identical for individuals differentiated only by their BMI.
Third, since obese people are more likely to be chronically ill in a given year (higher p),
obese statin users benefit relatively more from the secondary prevention of mortality (reduction
of u) than healthy-weight statin users. For the same reason, obese people gain relatively less
from the primary prevention effects of statins (q
1
and q
2
reductions). Thus, the worse health of
obese people as we simulate their health outcomes interacts in an a priori ambiguous manner
with the health effects of statins.
Finally, a factor that is absent from Figure 1 is the probability of using statins. As can be
seen in Table A1, statin use increases significantly with BMI until it reaches 30. This effect
unequivocally increases the health impact of statins for obese individuals.
APPENDIX C. Sensitivity Analyses
We consider two sensitivity analyses for the effect of statins on the costs of obesity: 1) without
reassigning the BMI of the cohort at age 51, and 2) simulating only the secondary prevention
effect of statins. The effects of these alternative approaches on life expectancy and quality-of-life
are detailed for all BMIs in Table A3. The baseline difference-in-difference results of Table 4 are
compared with these approaches in Table A4. For simplicity, we present only results for the
obesity class 1 category, which accounts for over 50% of the US obese population, in Table A4.
The difference-in-difference in the total cost of obesity is presented graphically for all obesity
classes in Figure A5.
96
C.1 Running Simulations Without Reassigning BMI
Throughout this article, we consider the cost of being obese as the consequences of an
average person of age 51 going from a healthy BMI to an obese BMI. Thus, we reassigned the
BMI of each individual in our cohort to specific BMIs, while keeping their other states
(comorbidities) unchanged (see Section 2.2.2). The difference in expected health outcomes and
medical spending between these synthetic cohorts revealed the cost of obesity. By construction,
this strategy removes the difference in other health variables linked to obesity. In other words,
our strategy considers that a person of average health faces the choice between obese and non-
obese BMIs, rather than a choice between the average health of obese and non-obese people.
To test whether statins have a larger effect when taking into account the lower health
status of obese people, we conduct an additional set of simulations. Instead of reassigning BMI
values, we remove from the simulations all individuals outside of the BMI category (BMI under
24.9 for “Healthy” and between 30 and 34.9 for “Obese class 1”, etc.).
As could be expected, the differential health of obese individuals has serious
consequences on longevity and quality of life. In Table A3, we find that the difference in life
expectancy and expected QALYs by BMI is much greater without reassigning BMI (the second
set of results in the table) than our baseline results (the first set of results). We also find that,
under this definition, the additional gain of obese individuals – the difference-in-difference - is
higher. According to our results, obese individuals can expect to gain 0.18-0.19 more years of
life expectancy than healthy-weight individuals, about three times more than in our baseline
simulations.
However, the costs of obesity are also much larger with this definition than they are
under our baseline approach. In Table A4, our baseline difference-in-difference results for obese
class 1 are shown in the first column,
11
and those obtained without reassigning BMI are
presented in the second column. We note that the larger confidence intervals for all statistics in
the table are caused by the smaller number of individuals going through the simulations for each
BMI category. While we find a net reduction of $5,000 in the costs of obesity with this
specification, it represents only 0.8% of the “Without Statins” costs of obesity and is not
significantly different from zero, as was the case in our baseline simulation.
11
This column reproduces the column “Obese class 1” of the Panel A in Table 4.
97
The Panel b of Figure A5 shows the effect of statins on the total costs of obesity for the
three obese categories visually. When compared to Panel a (our baseline results), the cost of
obesity are larger, while the differences between the “Without Statins” and the “With Statins”
remain small – and statistically insignificant – for all obese classes.
C.2 Simulating Only the Secondary Prevention Effect of Statins
Our second sensitivity analysis considers only the secondary prevention effects of statins.
The “Without Statins” scenario of this analysis removes only the secondary prevention effect of
statins, and thus corresponds to a scenario in which statins are used, but do not reduce the risk of
mortality after the onset of heart disease. This analysis serves two purposes. First, the small
impact of statin therapy on the cost of obesity can plausibly be due to its important primary
prevention effects, which may benefit healthy-weight individuals more than obese individuals.
The primary prevention effects can thus be too effective for statins to lower the costs of obesity.
Considering only the secondary prevention effects can thus better illustrate the potential of
medical innovation to reduce the costs of obesity when its benefits are targeted towards the
obese. Second, this exercise serves to decompose the overall effect of statins (the baseline
results) into its primary and secondary prevention components.
As we saw in Table 3, when considering both primary and secondary preventions,
individuals with class 1 obesity gain 6% more life expectancy than healthy-weight individuals
(1.24 vs. 1.17 years). However, most of the difference-in-difference is associated with some level
of disability, such that QALY gains are only 2% higher (0.90 vs 0.88 QALYs). When
considering only the secondary prevention effect (third set of results of Table A3), we find that
the gains of obese individuals expressed as a percent of the gains of healthy weight individual
increase. For instance, life expectancy and QALY gains of obese class 1 individuals are 14% and
8% higher than healthy-weight individuals, respectively. Thus, this analysis confirms the notion
that the secondary prevention effect of statins benefits obese individuals more, while the primary
prevention effect benefits healthy-weight individuals more.
However, the difference-in-difference in QALYs is quite small (0.02-0.03 QALYs)
relative to the costs of obesity, and is partially countered by higher medical costs. Thus, when
compared to the baseline results, the net reduction in the costs of obesity due to the secondary
prevention effect also does not significantly differ from zero (shown for class 1 obesity in Table
98
A5). When presented graphically (Panel c of Figure A5), the costs of obesity appear virtually
identical to our baseline results (Panel a).
99
Table 4.A1 Factors Influencing the Probability of Purchasing Statins in MEPS, 2009 to 2011:
Probit Estimates
1. Initialization Model 2. Transition Model
Pop. aged 50 – 53 Pop. aged 53+
Coefficient
Standard
error
Marginal
effect Coefficient
Standard
error
Marginal
effect
Demographics and
education
Black -0.19 ** (0.08) -0.039 -0.08
(0.06) -0.028
Hispanic -0.24 *** (0.09) -0.049 -0.10
(0.07) -0.033
Male 0.08
(0.10) 0.018 0.03
(0.06) 0.010
Less than high school 0.13 * (0.07) 0.030 -0.02
(0.06) -0.006
College education 0.26 *** (0.09) 0.058 0.08
(0.07) 0.027
Male and less than
high school -0.09
(0.13) -0.020 -0.16
(0.10) -0.054
Male and college 0.01
(0.10) 0.003 -0.04
(0.08) -0.013
Male and black -0.23 ** (0.12) -0.045 -0.06
(0.10) -0.021
Male and Hispanic -0.02
(0.12) -0.005 -0.02
(0.10) -0.007
Health Conditions
Cancer 0.22 ** (0.11) 0.055 -0.07
(0.06) -0.024
Diabetes 0.91 *** (0.06) 0.272 0.27 *** (0.05) 0.099
Heart diseases 0.40 *** (0.06) 0.103 0.21 *** (0.04) 0.076
High blood pressure 0.51 *** (0.05) 0.117 0.18 *** (0.04) 0.064
Lung diseases 0.05
(0.09) 0.011 0.05
(0.06) 0.017
Stroke 0.04
(0.11) 0.009 0.10 * (0.06) 0.037
Weight (loglinear
splines)
Log of BMI under 30 1.36 *** (0.22) 0.297 0.63 *** (0.15) 0.221
Log of BMI over 30 -0.12
(0.21) -0.026 -0.24
(0.19) -0.083
Age (linear splines)
Less than 70 years old
0.02 *** (0.00) 0.008
70 years and older
-0.01 *** (0.00) -0.004
Statin use in the
previous period
2.67 *** (0.04) 0.818
Constant -6.00 *** (0.72) -5.05 *** (0.55)
Observations 5,234
11,458
Pseudo R-squared 0.154 0.594
*** p<0.01, ** p<0.05, * p<0.1. BMI refers to “body mass index”, defined as the ratio between mass of individuals,
expressed in kilograms, and the square of height, expressed in meters. A MEPS respondent is defined as a statin user
if he or she filled at least one prescription associated with the “HMG-CoA reductase inhibitor” therapeutic subclass
during a given year, or a prescription of “Simcor”, “Advicor” or “Vytorin” (which combine statins with other active
ingredients). Therapeutic classes in MEPS correspond to Multum Lexicon variables from Cerner Multum, inc.
100
Table 4.A2 95% Confidence Interval of the Impact of Statins on Lifetime Medical Costs after
Age 50, $2009 thousands
With Statins Without Statins Difference
Healthy weight (BMI < 25)
Total Medical Costs [459.3 - 462.4] [422.9 - 448.7] [12.1 - 38.0]
Excluding Statins [454.5 - 457.5] [422.9 - 448.7] [7.2 - 33.2]
Free of disability [288.1 - 289.6] [280.9 - 286.7] [2.7 - 7.8]
Disabled [96.2 - 97.4] [88.2 - 95.3] [1.6 - 8.7]
In a nursing home [69.2 - 71.5] [52.9 - 68.2] [2.6 - 17]
Statins [4.8 - 4.8] [0 - 0] [4.8 - 4.8]
Obese class 1 (BMI = 30)
Total Medical Costs [476 - 479.3] [435.3 - 464.2] [13.3 - 41.9]
Excluding Statins [470.5 - 473.7] [435.3 - 464.2] [7.7 - 36.4]
Free of disability [283.5 - 285.0] [276.2 - 282.0] [2.6 - 8.1]
Disabled [115.8 - 117.1] [105.8 - 114.4] [2.1 - 10.6]
In a nursing home [70.2 - 72.3] [52.7 - 69.2] [2.5 - 18.2]
Statins [5.6 - 5.6] [0 - 0] [5.6 - 5.6]
Obese class 2 (BMI = 35)
Total Medical Costs [483.8 - 487.1] [441.6 - 471.5] [13.5 - 42.9]
Excluding Statins [478.3 - 481.5] [441.6 - 471.5] [8 - 37.3]
Free of disability [277.1 - 278.5] [269.7 - 275.5] [2.6 - 7.9]
Disabled [130.5 - 131.9] [119.3 - 128.8] [2.1 - 11.7]
In a nursing home [69.8 - 71.7] [51.8 - 68.9] [2.4 - 18.5]
Statins [5.5 - 5.6] [0 - 0] [5.5 - 5.6]
Obese class 3 (BMI = 40)
Total Medical Costs [489.0 - 492.1] [445.8 - 476.9] [13.2 - 44]
Excluding Statins [483.5 - 486.7] [445.8 - 476.9] [7.8 - 38.6]
Free of disability [267.2 - 268.7] [260.0 - 265.6] [2.5 - 7.7]
Disabled [146.5 - 148.1] [134.0 - 144.7] [2.5 - 13.1]
In a nursing home [68.7 - 70.8] [51.0 - 68.1] [2.1 - 18.8]
Statins [5.4 - 5.4] [0 - 0] [5.4 - 5.4]
This table shows 95% confidence intervals with regards to the uncertainty of the effectiveness of
statins for the first three columns of Table 2. All amounts are in present value at age 51, computed with
a 3% interest rate. BMI refers to “body mass index”, defined as the ratio between mass of individuals,
expressed in kilograms, and the square of height, expressed in meters. The first two columns present
lifetime medical costs at age 51 in the “With Statins” and “Without Statins” scenarios. The third
column presents the difference between the scenarios, and corresponds to the additional medical costs
due to the existence of statins. The last three columns decompose the additional costs due to statins by
spending source. In the rows, we decompose medical costs by type (total medical costs excluding
statins prescriptions and statin prescriptions) and by functional status. A “disabled” functional status
refers to reporting at least one instrumental activity of daily living (IADL) or activity of daily living
(ADL) limitation. The “free of disability” status refers to reporting .no IADL or ADL limitation and
not living in a nursing home. “In a nursing home” indicates the most severe functional status
impairment.
101
Table 4.A3 Sensitivity of Impact of Statins on Life Expectancy and Expected QALYs after Age
50 to Alternate Simulation Approaches
Life Expectancy (Years) Expected QALYs
Healthy Obese class
Healthy Obese class
weight 1 2 3 weight 1 2 3
Baseline
Results
With Statins 33.2 32.1 31.3 30.4 28.0 26.1 24.9 23.9
Without
Statins 32.0 30.9 30.0 29.2
27.1 25.2 24.0 23.0
Difference 1.17 1.24 1.24 1.23
0.88 0.90 0.88 0.86
Difference-in-
difference 0.07 0.07 0.06
0.02 0.00 -0.02
Percent
(%) 6% 6% 5%
2% 0% -3%
95% C.I.
[0.03
-
0.10]
[0.03
-
0.11]
[0.01
-
0.11]
[-0.01
-
0.04]
[-0.03
-
0.03]
[-0.06
-
0.01]
Without
Reassigning
BMI
With Statins 34.4 31.3 29.8 27.8
29.3 25.1 23.4 21.4
Without
Statins 33.3 30.1 28.5 26.5
28.5 24.2 22.5 20.6
Difference 1.09 1.27 1.27 1.28
0.83 0.91 0.88 0.87
Difference-in-
difference 0.18 0.18 0.19
0.08 0.05 0.04
Percent
(%) 16% 16% 17%
9% 7% 5%
95% C.I.
[0.08
-
0.27]
[0.06
-
0.29]
[0.02
-
0.34]
[0.00
-
0.14]
[-0.02
-
0.13]
[-0.05
-
0.14]
Reassigning
BMI -
Secondary
Prevention
Only
With Statins 33.2 32.1 31.3 30.4
28.0 26.1 24.9 23.9
Without
Statins 32.7 31.5 30.7 29.8
27.7 25.7 24.5 23.5
Difference 0.51 0.57 0.59 0.60
0.33 0.36 0.36 0.35
Difference-in-
difference 0.07 0.09 0.09
0.03 0.03 0.02
Percent
(%) 14% 17% 18%
8% 8% 6%
95% C.I.
[0.02
-
0.11]
[0.02
-
0.14]
[0.03
-
0.15]
[0.01
-
0.05]
[0.00
-
0.05]
[-0.00
-
0.04]
Obese classes 1, 2 and 3 refer to individuals with a BMI larger or equal to 30 and 35 kg/m
2
, respectively. BMI
refers to “body mass index”, defined as the ratio between mass of individuals, expressed in kilograms, and the
square of height, expressed in meters. The difference-in-difference corresponds to the difference between the life
expectancy / QALY gain of statins between obese and healthy-weight individuals. The percent line expresses the
difference-in-difference as a percentage of the life expectancy / QALY gained by healthy-weight individuals.
95% confidence intervals of the difference-in-difference with regards to the uncertainty of the effectiveness of
statins are presented in brackets. The “Baseline Results” column corresponds to our main simulation approach
with BMI reweighting described in Section 2.2.2. The “Without Reassigning” simulations remove all individuals
with a BMI outside of the studied category at ages 51-52 from the simulations. The “Reassigning BMI -
Secondary Prevention Only” uses the same individuals as our baseline simulations, but removes only the
secondary prevention effects of statins in the “Without Statins” scenario.
102
Table 4.A4 Sensitivity of Health and Healthcare Costs of Obesity in Both Scenarios for Class 1
Obesity, $2009 thousands
Baseline
Results
(Reassigning
BMI)
Without
Reassigning
BMI
Reassigning
BMI -
Secondary
Prevention
Only
Without
statins
Medical care cost ($) 14.0 22.0 13.3
[12.3 - 15.8] [17.8 - 26.9] [11.3 - 15.3]
Health cost
QALYs lost
1.93 4.32 1.94
[1.9 - 2.0] [4.2 - 4.5] [1.9 - 2.0]
Value of QALYs lost
($)
289.5 647.5 291.0
[285.3 - 293.0] [623.5 - 668.8] [286.8 - 295.1]
Total cost of obesity
($)
303.5 669.5 304.3
[298.8 - 307.7] [647.0 - 687.6] [301.2 - 307.1]
With statins Medical care cost ($)
16.8 28.4 16.8
[15.9 - 17.6] [24.0 - 32.2] [15.9 - 17.6]
Health cost
QALYs lost
1.91 4.24 1.91
[1.9 - 1.9] [4.1 - 4.4] [1.9 - 1.9]
Value of QALYs lost
($)
286.9 636.0 286.9
QALYs lost
[283.3 - 289.8] [611.7 - 659.0] [283.3 - 289.8]
Total cost of obesity
($)
303.7 664.4 303.7
[300.5 - 306.4] [643.7 - 684.7] [300.5 - 306.4]
Difference due
to statins
Total cost of obesity
($)
-0.2 5.1 0.6
[-3.7 - 3.7] [-5.2 - 13.2] [-0.8 - 2.0]
Percent (%)
-0.1 0.8 0.2
[-1.2 - 1.2] [-0.8 - 2.0] [-0.3 - 0.7]
This table shows the sensitivity of the results presented in the first column of Table 4 to alternative simulation
approaches. The medical care cost of obesity is the difference in the sum of expected medical costs at age 51
between identical individuals of the obese category and the healthy-weight category (BMI<25) presented in
Table 2. BMI refers to “body mass index”, defined as the ratio between mass of individuals, expressed in
kilograms, and the square of height, expressed in meters. The total cost of obesity is defined as the sum of the
medical cost of obesity and the value of the health cost of obesity, expressed in dollars. Medical costs are in
present value, computed with a 3% interest rate. The value of health costs of obesity is obtained by attributing a
nominal value of $150,000 per Quality-adjusted life-year (QALY) lost due to obesity. Quality-adjusted life-years
(QALYs) adjust length of life for quality based on a person’s chronic conditions and functional status. The
“Baseline Results” column corresponds to our main simulation approach with BMI reweighting described in
Section 2.2.2. The “Without Reassigning” simulations remove all individuals with a BMI outside of the studied
category at ages 51-52 from the simulations. The “Reassigning BMI - Secondary Prevention Only” uses the same
individuals as our baseline simulations, but removes only the secondary prevention effects of statins in the
“Without Statins” scenario. 95% confidence intervals with regards to the uncertainty of the effectiveness of
statins are presented in brackets.
103
Figure 4.A1 American Adults Purchasing Statins
Source: Medical Expenditure Panel Survey (MEPS). 95% confidence interval is shown in grey. A MEPS respondent
is defined as a statin user if he or she filled at least one prescription associated with the “HMG-CoA reductase
inhibitor” therapeutic subclass during a given year, or a prescription of “Simcor”, “Advicor” or “Vytorin” (which
combine statins with other active ingredients). Therapeutic classes in MEPS correspond to Multum Lexicon
variables from Cerner Multum, inc.
104
Figure 4.A2 Use of Statins by Age and Obesity Status
Non-obese (BMI < 30) Obese (BMI ≥ 30)
Source: Medical Expenditure Panel Survey (MEPS). BMI refers to “body mass index”, defined as the ratio between
mass of individuals, expressed in kilograms, and the square of height, expressed in meters. A MEPS respondent is
defined as a statin user if he or she filled at least one prescription associated with the “HMG-CoA reductase
inhibitor” therapeutic subclass during a given year, or a prescription of “Simcor”, “Advicor” or “Vytorin” (which
combine statins with other active ingredients). Therapeutic classes in MEPS correspond to Multum Lexicon
variables from Cerner Multum, inc.
105
Figure 4.A3 Comparison of Statin Use at Ages 51-52 in FEM Simulations and MEPS Data
a. Statin Use vs. BMI
BMI refers to “body mass index”, defined as the ratio between mass of individuals, expressed in kilograms, and the
square of height, expressed in meters. “MEPS” refers to the percent of individuals aged 51-52 years old using statins
in the 2009 to 2011 waves of the Medical Expenditure Panel Survey. “FEM” refers to our imputation of statin use in
the Future Elderly Model’s 2010 cohort of 51-52 year-olds. A MEPS respondent is defined as a statin user if he or
she filled at least one prescription associated with the “HMG-CoA reductase inhibitor” therapeutic subclass during a
given year, or a prescription of “Simcor”, “Advicor” or “Vytorin” (which combine statins with other active
ingredients). Therapeutic classes in MEPS correspond to Multum Lexicon variables from Cerner Multum, inc.
b. Statin Use vs. Chronic Diseases
MEPS FEM
“MEPS” refers to the percent of individuals aged 51-52 years old using statins in the 2009 to 2011 waves of the
Medical Expenditure Panel Survey. “FEM” refers to our imputation of statin use in the Future Elderly Model’s 2010
cohort of 51-52 year-olds. A MEPS respondent is defined as a statin user if he or she filled at least one prescription
associated with the “HMG-CoA reductase inhibitor” therapeutic subclass during a given year, or a prescription of
“Simcor”, “Advicor” or “Vytorin” (which combine statins with other active ingredients). Therapeutic classes in
MEPS correspond to Multum Lexicon variables from Cerner Multum, inc.
106
Figure 4.A4 Pathways Between Obesity and Selected Health Outcomes in the FEM
Obesity
Stroke
Heart Disease
Type II
Diabetes
Hypertension
Mortality
107
Figure 4.A5 Sensitivity of Costs of Obesity to Alternative Simulation Approaches
a. Baseline Results b. Without Reassigning BMI
c. Reassigning BMI – Secondary Prevention Only
This figure shows the sensitivity of the results presented in Panel A of Table 4 to alternative simulation
approaches. The medical care cost of obesity is the difference in the sum of expected medical costs at age 51
between identical individuals of the obese category and the healthy-weight category (BMI<25) presented in
Table 2. BMI refers to “body mass index”, defined as the ratio between mass of individuals, expressed in
kilograms, and the square of height, expressed in meters. The total cost of obesity is defined as the sum of the
medical cost of obesity and the value of the health cost of obesity, expressed in dollars. Medical costs are in
present value, computed with a 3% interest rate. The value of health costs of obesity is obtained by attributing a
nominal value of $150,000 per Quality-adjusted life-year (QALY) lost due to obesity. Quality-adjusted life-years
(QALYs) adjust length of life for quality based on a person’s chronic conditions and functional status. The
“Baseline Results” column corresponds to our main simulation approach with BMI reweighting described in
Section 2.2.2. The “Without Reassigning” simulations remove all individuals with a BMI outside of the studied
category at ages 51-52 from the simulations. The “Reassigning BMI - Secondary Prevention Only” uses the same
individuals as our baseline simulations, but removes only the secondary prevention effects of statins in the
“Without Statins” scenario. 95% confidence intervals with regards to the uncertainty of the effectiveness of
statins are shown for each bar.
108
CHAPTER 5:
CONCLUSIONS
The rising prevalence of obesity in the U.S. is now commonly described as an epidemic,
not only in popular media, but by US Centers for Disease Control and the World Health
Organization as well. This epidemic will likely have significant economic and social
implications, in addition to its more obvious affects on public health. This dissertation has
described some of these economic and social impacts of obesity in childhood, adulthood, and
later life. Chapter two described how a modeling framework commonly applied to more simple
disease states can be generalized and used to predict adolescent obesity’s effects on later
healthcare spending. Chapter three examined the relationship between body mass index,
standardized test scores, and teacher evaluations among children in grades K-8. Finally, chapter
four tested the theory that medical innovation has significantly reduced the costs of obesity
among the elderly, using statin medications as a case-study.
Treatment of obesity in adolescence is generally recommended to reduce complications
and healthcare costs in later adulthood. Chapter two demonstrated that, while this is a reasonable
goal, its success depends not on how much weight is lost at the time of treatment, but on the
treatments lasting effects. We considered the most common obesity intervention programs used
in adolescence, behavioral interventions that include diet and exercise counseling. We found that
a BMI reduction of 1.45 kg/m
2
at age 18, the mean BMI reduction reported in literature, was not
sufficient to produce statistically significant reduction in medical spending through age 80.
However, if behavioral intervention participants had a lifetime reduction in their risk of
becoming obese, reductions in medical spending could be quite large. For example, a 25%
reduction in the lifetime risk of only class III obesity (BMI ≥ 40) was associated with $6,425 and
109
$6,922 in reduced medical costs among men and women, respectively. Larger risk reductions in
class I or II obesity were associated with as much as $21,028 in savings through age 80.
The lifetime obesity risk reductions modeled in chapter two are hypothetical, but
intended to be representative of lasting behavior change. Such lasting success has been rarely
reported in the literature. Nonetheless, our work offers two primary contributions: First, an
example of how Markov models can be applied to a complex disease such as obesity. Markov
models describe disease processes as transitions between discreet states of health. As a result,
they have been difficult to apply to diseases that do not offer clearly discreet states. Obesity is
associated with a wide range of complications that are not mutually exclusive from each other.
However, as we have shown, the disease process can be modeled using BMI states instead of
disease states, so long as the relationship between BMI and the outcome of interest (medical
spending in our case) can be modeled. Second, estimates of reasonable investments in adolescent
BMI reduction. Because we did not model the cost of administering behavioral intervention
programs to overweight adolescence, our results can be interpreted as cut-off points for
investments that will return a positive net-benefit if the assumed obesity risk reductions are
achieved. As we have noted, our savings estimates are significantly larger than the few published
estimates of program administrative costs that we were able to find. Therefore, such behavioral
intervention might be considered cost-effective, but only if they can better demonstrate lasting
obesity risk reductions.
Researchers have voiced concerns about the impact of obesity among younger children
on their human capital development for some time. The Early Childhood Longitudinal Survey,
administered by the National Center for Education Statistics, offered an opportunity to study
these effects. In chapter three, we found that BMI among children in grades K-8 was
110
significantly associated with worse teacher evaluations of English and math abilities, although it
was not significantly associated with performance on English and math standardized tests. The
finding of an inverse association with teacher evaluations is consistent with the only other
publication to report such an effect. However, the literature on childhood obesity and
standardized test performance is mixed. We also reported that this negative effect of BMI on
teacher evaluations was strongest among schools with average student BMI in the bottom 10
th
percentile. Although this effect was weaker in models that included child fixed effects, it offers
some indication that the penalty of obesity on teacher evaluations is at least partly driven by
social conditions. This work also improved on previous research by including a control for child
behavioral problems, which has been associated with obesity, and modeling the error correlation
between English and math models using seemingly unrelated regression.
As the US population ages the issue of obesity among the elderly is becoming more
important. Roughly half of US adults over the age of 65 already use a statin medication. Because
statins reduce the risks of heart attacks and stroke, which are associated with obesity, it is
reasonable to assume that statins may reduce the costs of obesity. Chapter four tested this theory,
using the previously described Future Elderly Model. However, we concluded that the
development of statins did not significantly reduce the costs of obesity among Americans over
50 years old. Statins do significantly reduce the risk of heart attack and stroke, and they do
reduce mortality, but the gained life expectancy was almost entirely spent with some level of
disability. Statins also have similar clinical benefit among the obese and non-obese, making it
difficult for them to reduce costs attributable to obesity.
Future research is likely to expand on all three of the projects described in this
dissertation. More sophisticated modeling efforts to evaluate the cost-effectiveness of obesity
111
interventions, further analysis of the potential for obesity discrimination in schools and its effects
on human capital, and more careful consideration of medical innovation’s potential impact on the
cost of obesity will continue to improve our understanding of the economic aspects of obesity.
Abstract (if available)
Abstract
Obesity is widely recognized as one of the greatest modern public health challenges by the US Surgeon General, The World Health Organization, and The Organization for Economic Cooperation and Development. In 2013, the American Medical Association officially recognized obesity as a disease in its own right, a decision that has generated some controversy. Despite more than a century of research, new health risks and adverse economic outcomes associated with obesity are still being discovered. Meanwhile, interventions that produce lasting cost‐effective results remain elusive. This dissertation attempts to contribute to our understanding of the cost‐effectiveness of obesity interventions, the impact of obesity on human capital development, and the role that medical innovation can play in mitigating these costs. ❧ Chapter two describes the use of a novel decision analytic model to estimate potential reductions in medical expenditures after overweight adolescents participate in behavioral interventions designed to reduce their body mass index. The model is designed and validated using data from multiple sources and allows for the simulation of lifetime weight trajectories and associated healthcare costs. Chapter three focuses on obesity in earlier childhood. Previous research has indicated that childhood obesity may be associated with worse academic outcomes. One theory asserts that teachers may have negative opinions and reduced expectations of their obese or overweight students. This hypothesis is tested using a longitudinal dataset containing standardized test scores and subjective teacher evaluations of students in grades K-8. Finally, chapter four looks towards the future of obesity among the elderly. Given the difficulties in adherence to diet and exercise interventions, some may be inclined to hope that medical innovation will eventually mitigate the marginal cost of obesity by effectively treating all or most of its associated complications. Using a previously developed dynamic model, the impact of the development of statins (an example of medical innovation) on the marginal cost of obesity is examined.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
The causal-effect of childhood obesity on asthma in young and adolescent children
PDF
Advances and applications for economic evaluation methods in health technology assessment (HTA)
PDF
Essays on the equitable distribution of healthcare
PDF
Burden of illness in hemophilia A: taking the patient’s perspective
PDF
Essays on the economics of infectious diseases
PDF
Discriminating changes in health using patient-reported outcomes
PDF
New approaches using probabilistic graphical models in health economics and outcomes research
PDF
Evaluating approaches to reduce inappropriate antibiotic use in the United States
PDF
Value in health in the era of vertical integration
PDF
Impact of pharmacy-based transitional care on healthcare utilization and costs
PDF
Evaluating treatment options for metastatic, castration-resistant prostate cancer: a comprehensive value assessment
PDF
Essays in pharmaceutical and health economics
PDF
Three essays on emerging issues in hemophilia care
PDF
Essays on the use of microsimulation for health and economic policy analysis
PDF
Healthcare provider recommendations: a panacea to improving influenza vaccination rates?
PDF
The impact of Patient-Centered Medical Home on a managed Medicaid plan
PDF
Crohn’s disease: health outcomes and resource utilization in the biologic era
PDF
Hepatitis C in the post-interferon era: selected essays in health economics
PDF
An evaluation of risk predictors for disease progression in Veteran Affairs patients with chronic hepatitis C
PDF
Selected papers on the evaluation of healthcare costs of prematurity and necrotizing enterocolitis using large retrospective databases
Asset Metadata
Creator
Messali, Andrew James
(author)
Core Title
Economic aspects of obesity
School
School of Pharmacy
Degree
Doctor of Philosophy
Degree Program
Pharmaceutical Economics and Policy
Publication Date
04/17/2015
Defense Date
02/19/2015
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
database analysis,economic model,Health Economics,healthcare costs,OAI-PMH Harvest,obesity
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Doctor, Jason N. (
committee chair
), Salvy, Sarah-Jeanne (
committee member
), Sood, Neeraj (
committee member
)
Creator Email
drew.messali@me.com,messali@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c3-552150
Unique identifier
UC11297637
Identifier
etd-MessaliAnd-3329.pdf (filename),usctheses-c3-552150 (legacy record id)
Legacy Identifier
etd-MessaliAnd-3329.pdf
Dmrecord
552150
Document Type
Dissertation
Format
application/pdf (imt)
Rights
Messali, Andrew James
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Access Conditions
The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...
Repository Name
University of Southern California Digital Library
Repository Location
USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA
Tags
database analysis
economic model
healthcare costs
obesity