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
/
Factors influencing the decision and timing to undergo breast reconstruction after mastectomy for breast cancer in public hospital vs. private medical center from 2007 to 2013: a retrospective co...
(USC Thesis Other)
Factors influencing the decision and timing to undergo breast reconstruction after mastectomy for breast cancer in public hospital vs. private medical center from 2007 to 2013: a retrospective co...
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
FACTORS INFLUENCING THE DECISION AND TIMING TO UNDERGO
BREAST RECONSTRUCTION AFTER MASTECTOMY FOR BREAST
CANCER IN PUBLIC HOSPITAL VS. PRIVATE MEDICAL CENTER
FROM 2007 TO 2013: A RETROSPECTIVE COHORT STUDY
by
Chia-Wei Liu
A Thesis Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA In
Partial Fulfillment of the
Requirements for the Degree
MASTER OF SCIENCE
(BIOSTATISTICS)
May 2015
Copyright 2015 Chia-Wei Liu
1
Table of Contents
List of Tables ....................................................................................................................... 3
List of Figures ..................................................................................................................... 4
Acknowledgement .............................................................................................................. 5
Abstract ................................................................................................................................ 6
Introduction ......................................................................................................................... 7
Methods ................................................................................................................................ 9
Study Population ..................................................................................................................... 9
Data Collection ....................................................................................................................... 9
Statistical Analysis ............................................................................................................... 10
Results ................................................................................................................................. 13
Descriptive Analysis ............................................................................................................. 13
Patient Characteristics and Institution Factors ............................................... 13
Disease Characteristics .................................................................................. 13
Treatment and Surgical Factors ..................................................................... 14
Difference in the overall reconstruction rate or IBR rate between public vs. private
hospital .................................................................................................................................. 14
Overall Reconstruction Rate .......................................................................... 14
IBR Rate ........................................................................................................ 15
Association between overall reconstruction rate or IBR rate with predictor factors by
univariate analysis ................................................................................................................ 15
Overall Reconstruction Rate .......................................................................... 15
IBR Rate ........................................................................................................ 15
Multivariate model for predicting overall reconstruction rate or IBR rate .......................... 16
Overall Reconstruction Rate .......................................................................... 16
IBR Rate ........................................................................................................ 18
Timing of IBR rate or reconstruction rate ............................................................................ 19
Factors associated with patients who would undergo delayed breast reconstruction .......... 20
Multivariate non-mixture cure fraction model for estimating reconstruction rate and
fraction of patients who would undergo delayed breast reconstruction .............................. 20
Discussion ........................................................................................................................... 22
Reconstruction rate across institutions ................................................................................. 22
2
Predictors for reconstruction rates ........................................................................................ 23
Multivariate prediction Mmodels for reconstruction rates .................................................. 23
Analysis of Time-to-Event ................................................................................................... 25
Censoring Problem – A Demonstration ............................................................................... 27
Limitations ............................................................................................................................ 28
Reference ............................................................................................................................ 30
Tables .................................................................................................................................. 32
Figures ................................................................................................................................. 47
Supplementary Materials .............................................................................................. 49
3
List of Tables
Table 1. Characteristics of Patients by Institution ................................................. 32
Table 2. Association with IBR Rate or Overall Reconstruction Rate across
Institutions ..................................................................................................... 36
Table 3. Association between Overall Reconstruction Rate or IBR Rate with
Predictor Variables (Univariate Analysis and Institution-Adjusted) ............ 37
Table 4. Association between Overall Breast Reconstruction with Predictor
Variables (Multivariate Analysis) ................................................................. 39
Table 5. Association between Immediate Breast Reconstruction with Predictor
Variables (Multivariate Analysis) .................................................................. 41
Table 6. Hazard Ratio Estimates of Non-Mixture Cure Model for Patients with
Delayed Reconstruction or Patients Who Were Censored (Univariate
Analysis) ........................................................................................................ 43
Table 7. Hazard Ratio Estimates of Non-Mixture Cure Models for Patients with
Delayed Reconstruction or Patients Wo Were Censored (Multivariate
Analysis) ........................................................................................................ 45
Table 8. Reconstruction Probability Estimation when Controlling and Not
Controlling for Censored Observatiions ....................................................... 46
4
List of Figures
Figure 1. ROC Curves for Logistic Models. ......................................................... 47
Figure 2. Time to Reconstruction. ......................................................................... 48
5
Acknowledgement
I am most grateful to my advisor and mentor, Dr. Richard Sposto, for his invaluable
help, kind support and great advice on completing this thesis. I also thank the other
members of my thesis committee: Dr. Wendy Mack and Dr. Susan Groshen, for spending
time reviewing my thesis and giving comments. Great appreciation also goes to Dr. Lily
Tung and Dr. Julie E. Lang from Department of Surgery, USC/Norris Comprehensive
Cancer Center for their excellent research and priceless database.
This thesis is dedicated to all the patients involved in this study, and also my
beloved friends, family and my home country, Taiwan.
6
Abstract
Background/Objective: This study was performed to compare the rates of breast
reconstruction between a public safety-net hospital (LA County Medical Center (LAC)) vs.
a academic medical center (Keck Hospital of USC / Norris Cancer Center (Keck/Norris)),
with emphasis on immediate breast reconstruction (IBR). We hypothesized that both overall
reconstruction rate (the proportion of patients receiving reconstruction among the total
patients having mastectomy) and the IBR rate (the proportion of patients receiving IBR
among the total patients having reconstruction) were higher at Keck/Norris than LAC. We
also aimed to explore the factors that were associated with overall reconstruction rate or
IBR rate, and to estimate the time to reconstruction and the fraction of patients who would
not undergo delayed breast reconstruction (cure fraction).
Methods: We retrospectively collected all the data of patients with stage 0 – III breast
cancer who received mastectomy during Jan, 2006 to Dec, 2013 from Keck/Norris and
LAC. Univariate and multivariate logistic regressions were performed to compare the IBR
rate and overall reconstruction rate between the two institutions, as well as to evaluate the
correlation between IBR rate or overall reconstruction rate with various factors. Kaplan-
Meier method was used to scrutinize the time to reconstruction, and non-mixture cure
fraction model was performed for estimating the cure fraction.
Results: Of the 453 patients, 327 patients were from LAC and 126 were from Keck/Norris.
The odds ratio (OR) of reconstruction rate was 2.2 at Keck/Norris compared to LAC (p <
0.001) without adjustment for patient demographic factors, but was no longer significant
with adjustment. The OR of IBR rate was 21.2 compared from Keck/Norris to LAC (p <
0.001) without adjustment for patient demographic factors, and was 22.3 with adjustment (p
< 0.001). Based on the p < 0.25 entry criteria, factors that we kept in the multivariate overall
reconstruction rate model were insurance status (p < 0.001), race (p < 0.05), age (p < 0.001),
BMI (p < 0.05), smoking status (p = 0.08), occupation status (p = 0.18), personal history of
other cancer (p = 0.12), progesterone receptor (PR) status (p = 0.22) and radiotherapy (p <
0.001). For IBR rate we had institution (p < 0.001), BMI (p = 0.21), clinical stage (p < 0.01)
and radiotherapy (p < 0.01) in the model. The Kaplan Meier curves revealed that more
patients underwent IBR at Keck/Norris than at LAC. Excluding the IBR patients, the
average time to reconstruction at Keck/Norris was 20.3 ± 0.47 (SE) months, and was 37.4 ±
0.91 (SE) months at LAC. The estimated cure fractions were 0.71 and 0.99 at LAC and
Keck/Norris, respectively. The factors that we kept in the multivariate non-mixture cure
fraction model were institution (p < 0.01), insurance status (p <0.05), age = (p < 0.001),
smoking status (p = 0.10), occupation status (p = 0.062), clinical stage (p = 0.12), PR status
(p < 0.05).
Conclusions: The reconstruction rate did not differ between these two institutions, but the
utilization of IBR was higher at Keck/Norris than at LAC. The reconstruction rate was
significantly associated with insurance status, age, BMI, occupation status, and
radiotherapy, whereas the IBR rate was found to be significantly associated with institution,
clinical stage and radiotherapy. When excluding IBR patients, the average time to
reconstruction at Keck/Norris was 20.3 months, and was 37.4 months at LAC. Factors
including institution, insurance status and PR status were found to be significantly
associated with cure fraction.
7
Introduction
Breast cancer is the most common cancer among women in the U.S.
(Howlader N, 2012) Approximately 1 out of 8 women will develop breast cancer
during her life time (Howlader N, 2012). In 2015, an estimated 231,840 new cases of
invasive breast cancer are expected to be diagnosed in women, representing about
14.0% of all new cancer cases in the U.S. (American Cancer Society, 2015).
Mastectomy, which is the surgical removal of the entire breast to treat or prevent breast
cancer, is a standard treatment option for breast cancer. There are several types of
mastectomies including skin-sparing, nipple-sparing mastectomy and total mastectomy.
About 45% of 291,117 patients who were diagnosed with stages 0 – III breast cancer
during 1998 – 2011 underwent mastectomy in the U.S. (Kurian et al., 2014). Breast
reconstruction is usually provided as an option to women who undergo mastectomies,
either implants or autologous tissue, in order to rebuild the shape of the removed breast.
The timing of reconstruction surgery can be either immediate or delayed. Immediate
breast reconstruction (IBR) is done at the same time as mastectomy, while the delayed
type (DBR) is performed after the patient has healed from the mastectomy and after the
completion of adjuvant therapy. Generally speaking, IBR has less risk of being socially
or emotionally devastating, better cosmetic results and possibly fewer surgeries and
lower surgery cost. There is no marked elevation in the rate of cancer recurrence after
IBR(Zervoudis et al., 2014), but DBR is recommended to patients in need of post-
mastectomy radiotherapy to avoid irradiating the expander or implant(Alderman et al.,
2014).
The enactment of the Women’s Health and Cancer Rights Act (WHCRA) of
1998 made all breast reconstructions mandatory procedures to be covered within any
health plan offering coverage of mastectomies. The coverage includes both breast and
nipple reconstruction as well as contralateral breast operations for symmetry. A study
showed that the rate of patients having post-mastectomy reconstruction greatly
increased largely after the passing of the act, from 11.6 % in 1998 to 36.4% in
2011(Kummerow et al., 2015). Among all patients receiving reconstructions, another
study indicated that the rate of IBR increased from 20.8% in 1999 to 37.8% in 2008, an
average increase of 5% per year(Albornoz et al., 2013). The elevation in IBR rate may
correspond to the increase in implant use(Albornoz et al., 2013), but the actual reason
is still unclear. Despite the overall increase in overall reconstruction rate and IBR rate,
disparities in these rates among certain patient populations still exist. For example,
8
patient-level and treatment-level factors including age, race, insurance type, tumor
stage and grade, and post-mastectomy radiation have been shown to significantly
correlate with the rates of reconstruction.(Kruper et al., 2011; Lang et al., 2013; Yang
et al., 2013) The hospital setting in which patients receive care and the procedure
volume of the physician who performed the mastectomy also strongly predict the rate
of reconstruction.(Hershman et al., 2012)
Based on this background, a study controlling for patient and institution
factors and precise determination of an overall reconstruction patient population (IBR +
DBR) is necessary to clarify the causes for the secular rise of reconstruction rates,
including the overall reconstruction rate (the proportion of patients receiving
reconstruction among the total patients having mastectomy) and the IBR rate (the
proportion of patients receiving IBR among the total patients having reconstruction).
To address this need we undertook an analysis of all patients undergoing a mastectomy
at two institutions, LAC+USC Medical Center (LAC) and USC Keck Medical Center
and Norris Comprehensive Cancer Center (Keck/Norris). These two institutions are
within the same USC health care system, and they are managed by one multi-
disciplinary medical team so they have otherwise similar medical practice, but these
two institutions actually serve different populations due to the different hospital
settings (public safety-net hospital vs. academic medical center). Using these data, we
undertook an analysis which had three aims: 1) compare the rate of overall breast
reconstruction and the rate of IBR between these two institutions, adjusting for patient-
level factors; 2) explore the factors associated with overall reconstruction rate or IBR
rate by univariate and multivariate logistic regression model, and establish prediction
models for predicting these rates; and 3) establish a non-mixture cure fraction model
for investigating and addressing the censoring problem among patients who underwent
DBR.
9
Methods
Study Population
This study included all patients who were diagnosed with ductal carcinoma
in situ (DCIS), lobular carcinoma in situ (LCIS) or stage I, II, III breast cancer and
treated with mastectomy at LAC or Keck/Norris during January, 2006 to December,
2013. Any patient with inflammatory or stage IV breast cancer, or who underwent
breast conserving surgery were excluded. This study was approved by the
institutional review board at the University of Southern California Health Sciences
Campus.
Data Collection
Patients who met the study entry criteria were identified from electronic
medical record systems, which was Affinity at LAC and Cerner at Keck/Norris,
respectively. Patient data were then collected and managed using the REDCap
electronic data capture software hosted by the Southern California Clinical and
Translational Science Institute (SC-CTSI). Only authorized study personnel had the
access to the REDCap database. Datasets without personal identifiers were exported
from REDCap after data collection was completed. Only approved personnel could
access the personal identifiers data. Five categories of variables were collected in
this study:
1) Patient factors: race (Caucasian, African American, Hispanic, Asian or Other
race), age (< 50yrs and ≥ 50yrs), BMI (Underweight (BMI < 18.5), Healthy
weight (BMI: 18.5 - 25.0), Overweight (BMI: 25.0 - 30.0) or Obese (BMI > 30)),
insurance status (private or non-private (including patients without insurance or
with Medicare/Medicaid)), use of diabetes medication (yes/no), smoking status
(current smoker(yes) or not current smoker(no)), occupation status (employed
(including full-time and part-time jobs, or jobs without documenting working
hours) or unemployed), Charlson comorbidity score (categorized into 1 point, 2
points and ≥ 3 points), family history of breast cancer (yes/no), personal history
of other cancer (yes/no), type of prior cancer (breast, ovarian, uterine, colon,
thyroid, or other cancer), and prior surgical treatment of breast cancer
(lumpectomy or mastectomy).
2) Institution factors: was there a discussion between plastic surgeon and patient
regarding reconstruction (yes/no), had the subject ever declined reconstruction
10
during discussion (yes/no), time between referral to plastic surgery and plastic
surgery clinic appointment, and time between mastectomy and reconstruction.
3) Disease factors: clinical stage (information was obtained before mastectomy) (I,
II, III or in situ (DCIS/LCIS)), pathological/surgical stage (information was
obtained after mastectomy) (I, II, III, in situ (DCIS/LCIS) or complete pathologic
response (ypT0N0)), estrogen receptor (ER) status (positive, negative or not
documented), progesterone receptor (PR) status (positive, negative or not
documented), HER2 status (positive, negative or not documented), BRCA status
(positive, negative or not tested), type of BRCA (1 or 2) and recurrence of breast
cancer (yes/no).
4) Treatment factors: pre-mastectomy neoadjuvant therapy (yes/no), post-
mastectomy adjuvant chemotherapy (yes/no), post-mastectomy radiotherapy
(yes/no), and post-mastectomy endocrine therapy (yes/no).
5) Surgery factors: type of mastectomy (total, nipple-sparing or skin-sparing),
laterality of mastectomy, did the patient have any wound infection after
mastectomy (yes/no), did the patient receive reconstruction after mastectomy
(yes/no), timing of reconstruction (immediate, delayed or both), type of
reconstruction (autologous, implant or both), number of revisions after
reconstruction, and total number of surgeries.
6) Time factors: date of mastectomy and date of reconstruction
Statistical Analysis
The primary outcome of interest was the probability of undergoing
reconstruction following mastectomy, with an emphasis on IBR. Throughout this
thesis, we will use the word ”rate” interchangeably with the term “proportion of
patients”. All variables were summarized by descriptive statistics and comparisons
between LAC and Keck/Norris patients were determined by Chi-square test. For aim
1, univariate logistic regression was performed to compare the rate of IBR (the
proportion of patients receiving IBR among the total patients having reconstruction)
and the rate of overall breast reconstruction (the proportion of patients receiving
reconstruction among the total patients having mastectomy) at each institution, and
multivariate logistic regression was used to study these associations after adjustment
for insurance status, race, age, BMI, and comorbidity index.
For aim 2, univariate logistic regression was first performed to evaluate the
association between each variable with overall breast reconstruction rate or IBR rate.
11
Next, we selected the conventional variables including insurance status, age, BMI,
comorbidity index and all the other variables with p value smaller than 0.25 in the
univariate analysis for multivariate model building. Note that no surgery factors
were considered as candidate variables in this multivariate model, because we
wanted to focus on the covariates that reflected patient status before any surgical
procedures had been performed. The pathological/surgical stage was also excluded
from model building based on the same reason. After choosing the candidate
variables, we began with the model containing all the candidate variables, performed
manual backward elimination to rule out all the variables that had p values larger
than 0.25, and refitted the model with remaining variables. After multiple
verification and refitting steps, the final model contained variables with p values
smaller than 0.25. Automated selection methods including backward, forward and
stepwise eliminations were also used to verify the final model we obtained by
purposeful (manual) selection described above. The Hosmer and Lemeshow
test(Hosmer and Lemesbow, 1980) was then used to check the goodness of fit of the
final model. A one-step approximation to the parameter estimates was used to
approach the leave-one-out cross validation to estimate prediction errors (SAS
Institute Inc., 2009). Receiver operating characteristic (ROC) curve, the area
underneath the curve (AUC) and its confidence interval were plotted/estimated by
the SAS macro (%ROC) provided in (Gönen, 2007). Two criteria were used to
define an optimal threshold point from the ROC curve: 1) Close to the ideal: the
threshold that will make the resulting binary prediction closest to a perfect predictor,
which is the predictor that has 100% sensitivity and 100% specificity, and 2) Far
from random (Youden index): the threshold that will make the resulting binary
prediction most far from line of no-discrimination, which is the predictor based on
completely random guess (Gönen, 2007).
For aim 3, because some patients did not have any reconstruction up to the
date that we started the statistical analysis, we defined the date of lost to follow-up as
October, 01, 2014 for these patients. Those patients who had reconstruction were
coded as 1 (i.e., events) and those who did not have reconstruction up to date of lost
to follow-up were coded as 0 (i.e., censors). The distribution of time interval
between date of mastectomy and date of reconstruction or date of lost to follow-up
was studied by Kaplan-Meier plots. The mean of time to construction and its
standard error was estimated based on event time and did not include the censored
observation time (i.e., excluded those patients who did not undergo reconstruction up
12
to the date of lost to follow-up). Since all the Kaplan Meier curves appeared to
plateau quickly within 1 years after mastectomy, it was reasonable to assume that
certain proportion of patients did not have the possibility of undergoing
reconstruction anymore after certain time point (defined as “cure fraction”), while
the remaining proportion of patients still had the potential to undergo delayed
reconstruction (i.e., the “uncured”). Based on this assumption, we were able to study
factors that might lead to the plateau effect, and also the proportion of patients who
would/would not undergo reconstruction by using non-mixture cure fraction model.
Univariate non-mixture cure fraction models were used to evaluate potential factors
associated with the cure fraction (Lambert et al., 2007). All variables with p values
smaller than 0.25 were selected as candidates for a multivariate model. Likelihood
ratio test was used to compare multivariate models in order to find the best one.
Taking similar steps as described in aim 2, after multiple verification and refitting
steps, the final model contained variables with p values smaller than 0.25. The
multivariate non-mixture cure fraction model was used to model and estimate the
cure fraction and also the uncured fraction, which were those patients who would
eventually undergo DBR. Two different models were built: Model 1 was based on
the proportional excess hazards, which means the scale and shape parameters were
assumed not to vary by covariates. In the model 2 we also modeled the effects of the
covariates on the scale parameter but not on the shape parameter.
All the statistical analyses and reporting were performed using SAS
statistical software (version 9.4; SAS Institute Inc., Cary, NC) except aim 3. The
non-mixture cure fraction analysis in the aim 3 was performed using Stata 12
(StataCorp. 2011. Stata Statistical Software: Release 12. College Station, TX:
StataCorp LP).
13
Results
Descriptive Analysis
Patient Characteristics and Institution Factors
A total of 453 patients who underwent mastectomy from Jan., 2006 to
Dec., 2013 were identified through the electronic medical record systems. Among
these patients, 327 (72.2%) were from LAC and 126 (27.8%) were from
Keck/Norris. The predominant racial group was Hispanic (69.2%) at LAC, and
Caucasian (51.6%) at Keck/Norris (p < 0.001). More patients were over 50 years
old at Keck/Norris (68.3%) than at LAC (46.2%) (p < 0.001). At LAC, 79.4% of
patients were categorized as overweight or obese according to their BMI
compared to 45.3% of patients at Keck/Norris (p < 0.001). At LAC, 79.5% of
patients had Medicare/Medicaid, whereas 71.4% patients from Keck/Norris had
private insurance (p < 0.001). When offered at LAC, 26.0% of patients declined
reconstruction, whereas 9.5% of patients declined reconstruction at Keck/Norris (p
< 0.001). Note that institution was statistically significantly associated with
insurance status (p < 0.001 (Chi-Square Test, data not shown)), indicating the that
institution was a potential confounder in the relationship of insurance status and
the outcome (overall reconstruction rate or IBR rate), and vice versa. Other patient
characteristics and institution factors are summarized in Table 1.
Disease Characteristics
As classified by the American Joint Committee on Cancer staging
system, the majority of patients were in clinical stage II at both LAC and
Keck/Norris (49.2% and 44.4%, respectively). More patients had DCIS/LCIS at
Keck/Norris (24.6%) compared to LAC (19%) (p = 0.08). Based on the evaluation
of specimens obtained from mastectomy, the predominant patients at LAC were
still classified as stage II (38.3%), whereas most of the patients at Keck/Norris
were re-classified as stage I (36%) (p < 0.05). More patients had HER2-positive
breast cancer at LAC (24.5%) compared to Keck/Norris (15.9%) (p < 0.05). Note
that only limited patients obtained BRCA testing at both institutions (50.5% at
LAC; 26.2% at Keck/Norris), which may be due to the low proportion of patients
having breast cancer family history (30% at LAC; 54% at Keck/Norris). Other
disease factors are summarized in Table 1.
14
Treatment and Surgical Factors
Fewer patients at Keck/Norris received treatments in addition to surgery
than patients at LAC (Keck/Norris: 81.7%; LAC: 92.4%) (Supplementary 1).
Among these non-surgical cancer treatments, the predominant type was endocrine
therapy at both institutions(73.2% at LAC and 80.6% at Keck/Norris). At LAC,
89.6% of patients had total mastectomy, whereas only 77.8% at Keck/Norris (p <
0.001). More patients had a unilateral mastectomy at LAC (88.1%) compared to
Keck/Norris (71.4%) (p < 0.001). Fewer patients had wound infection after
mastectomy at Keck/Norris (12.7%) than at LAC (22.9%) (p < 0.05). No matter
whether the patients had ever declined reconstruction when it was offered, around
38.2% of patients from LAC received reconstruction after mastectomy (IBR or
DBR), and 57.9% patients from Keck/Norris had this surgery (p < 0.001). Among
those receiving reconstruction, 44.8% patients from LAC, and 94.5% patients
from Keck/Norris had IBR (p < 0.001). The major type of reconstruction was
autologous at LAC (52.8%), and implant at Keck/Norris (75.3%) (p <0.001).
Other treatment and surgery factors are summarized in Table 1.
Difference in the overall reconstruction rate or IBR rate between public
vs. private hospital
Overall Reconstruction Rate
Univariate analysis revealed that among all the patients who received
mastectomy, the odds of patients having reconstruction at Keck/Norris were 2.2
times the odds of patients having reconstruction at LAC (95% CI: 1.46 – 3.36, p <
0.001). Multivariate analysis with adjustment for insurance status, race, age, BMI,
and comorbidity index showed that the odds of having reconstruction at
Keck/Norris were only 1.3 times the odds of having reconstruction at LAC, and
this association was no longer statistically significant (95% CI: 0.61 – 2.59; p =
0.53). The multivariate analysis also showed the following results:
The odds of having reconstruction among patients with non-private insurance
plan was 0.3 times compared to patients with private plan (p < 0.01).
The odds of receiving reconstruction among patients less than 50 years old
was 3.6 times compared to patients over 50 years old (p < 0.01).
Comparing to the odds of having reconstruction among underweight/healthy
weight patients, the odds among obese patients was 0.7 times of the reference
odds, and the odds among overweight patients was 0.4 times of the reference
15
odds (p < 0.01). (Table 2).
IBR Rate
Univariate logistic analysis revealed that among all the patients who
received reconstruction, the odds of patients having IBR at Keck/Norris were 21.2
times the odds at LAC (95% CI: 7.3 – 61.39, p < 0.001). Multivariate analysis
with adjustment for patient-level factors including insurance status, race, age,
BMI, and comorbidity index showed that the odds of patients having IBR at
Keck/Norris remained 22.3 times the odds at LAC (95% CI: 3.75 - 132.36, p <
0.001) (Table 2).
Association between overall reconstruction rate or IBR rate with
predictor factors by univariate analysis
Overall Reconstruction Rate
Univariate logistic regression analysis showed that the following factors
were statistically significantly associated with reconstruction: institution,
insurance status, age, BMI, diabetes medication, occupation status, comorbidity
index, clinical stage, pathological/surgical stage, HER2 status, neoadjuvant
therapy, radiotherapy, laterality of mastectomy, type of mastectomy, type of
reconstruction, total number of revisions and total number of surgeries.
We were also interested in which variables would still be statistically
significant when controlling for institution, because institution was statistically
significantly associated with many predictor variables (data not shown), which
gave it potential to be a confounder in the relationship between the overall
reconstruction rate or IBR rate and other predictors. Multivariate analysis showed
that the following factors remained statistically significantly associated with
overall reconstruction rate after adjusting for institution: insurance status, age,
BMI, occupation status, BRCA status, comorbidity index, clinical stage,
pathological/surgical stage, HER2 status, neoadjuvant therapy, adjuvant
chemotherapy, radiotherapy, laterality of mastectomy, type of mastectomy, type of
reconstruction and total number of surgeries (Table 3).
IBR Rate
Univariate logistic regression analysis showed that the following factors
were statistically significantly associated with IBR rate: institution, insurance
16
status, race, age, BMI, comorbidity index, personal history of other cancer,
comorbidity index, clinical stage, pathological/surgical stage, HER2 status,
neoadjuvant therapy, adjuvant chemotherapy, radiotherapy, laterality of
mastectomy, type of mastectomy, type of reconstruction, and total number of
revisions.
We were also interested in which variables would still be statistically
significant when controlling for institution. Multivariate analysis showed that the
following predictor factors remained statistically significantly associated with IBR
rate after adjusting for institution: personal history of other cancer, clinical stage,
pathological/surgical stage, HER2 status, neoadjuvant therapy, radiotherapy,
laterality of mastectomy, type of mastectomy and type of reconstruction (Table 3).
Multivariate model for predicting overall reconstruction rate or IBR rate
Overall Reconstruction Rate
The following variables were selected as candidates for the multivariate
model based on their clinical or statistical importance (i.e., based on the univariate
analysis in Table 1 and the criterion of p < 0.25 in the univariate logistic
regression): institution, insurance status, race, age, BMI, diabetes medication use,
smoking status, occupation status, family history of breast cancer, personal history
of other cancer, BRCA status, comorbidity index, clinical stage, progesterone
receptor status, HER2 status, neoadjuvant therapy, radiotherapy and
chemotherapy. After multiple verification and refitting steps, the final model
contained the following variables: insurance status (p < 0.001), race (p < 0.05),
age (p < 0.001), BMI (p < 0.05), smoking status (p = 0.08), occupation status (p =
0.18), personal history of other cancer (p = 0.12), progesterone receptor (PR)
status (p = 0.22) and radiotherapy (p < 0.001). The parameter estimates were
interpreted as follows:
The odds of having reconstruction among patients with non-private
insurance was 0.32 times the odds among patients with private
insurance, with adjustment for other model variables (95% CI: 0.17 –
0.6).
Compared to Caucasians (reference odds), the odds of receiving
reconstruction among African Americans was 1.24 times (95% CI: 0.40
– 3.81), among Asians was 0.46 times (95% CI: 0.22 – 0.98), among
Hispanic was 1.23 times (95% CI: 0.64 – 2.37), and among all the other
17
races was 0.20 times (95% CI: 0.04 – 1.24), with adjustment for other
model variables.
The odds of having reconstruction among patients under 50 years old
was 2.91 times the odds among patients over 50 years old, with
adjustment for other model variables (95% CI: 1.78 – 4.74).
Compared to the patients who were underweight/healthy weight
(reference odds), the odds of receiving reconstruction among overweight
patients was 0.56 times (95% CI: 0.32 – 0.98), and among obese patients
was 0.39 times (95% CI: 0.21 – 0.73), with adjustment for other model
variables.
The odds of having reconstruction among patients who were not current
smokers was 0.49 times the odds among patients who were current
smokers, with adjustment for other model variables (95% CI: 0.22 –
1.09).
The odds of having reconstruction among patients who were employed
was 1.37 times the odds among patients who were unemployed, with
adjustment for other model variables (95% CI: 0.87 – 2.16).
The odds of having reconstruction among patients who did not have any
personal cancer history was 0.55 times the odds among patients had
personal cancer history, with adjustment for other model variables (95%
CI: 026 – 1.16).
The odds of having reconstruction among patients with negative PR
status was 0.75 time the odds among patients with positive PR status
(95% CI: 0.48 – 1.19), with adjustment for other model variables.
The odds of having reconstruction among patients not having any post-
surgery radiotherapy was 2.32 times the odds among patients having
post-surgery radiotherapy, with adjustment for other model variables
(95% CI: 1.44 – 3.73).
(See Table 4 for details)
The Hosmer and Lemeshow Goodness-of-Fit test indicated there was no
significant deviation from fit in this model (p = 0.86). The estimate of area under
ROC obtained by the resubstitution method is 0.780 (95% CI: 0.737 – 0.821),
whereas the approximation of the leave-one-out cross validation gave an estimate
of 0.729 (95% CI: 0.682 – 0.775). Thus, the probability which this model could
18
discriminate whether the patient would have post-mastectomy overall
reconstruction is around 72.9%. Both the methods of close to the ideal and far
from random (Youden index) defined the same optimal threshold for the model-
predicting probability as 0.437, which gives a true positive rate (sensitivity) of
64.6% and a false negative rate (1-specificity) of 31.1% (Figure 1).
IBR Rate
The following variables were selected as candidates for the multivariate
model based on their clinical or statistical importance (based on the univariate
analysis in Table 1 and the criterion of p < 0.25 in the univariate logistic
regression): institution, insurance status, race, age, BMI, family history of breast
cancer, personal history of other cancer, BRCA status, comorbidity index, clinical
stage, HER2 status, neoadjuvant therapy, radiotherapy and chemotherapy. After
multiple verification and refitting steps, the final model we arrived at contained
the following variables: institution (p < 0.001), BMI (p = 0.211), clinical stage (p
< 0.01) and radiotherapy (p < 0.01). The parameter estimates were interpreted as
follows:
The odds of having IBR at Keck/Norris was 30.5 times the odds at LAC,
with adjustment for other model variables (95% CI: 8.14 – 114.46).
Compared to those underweight/healthy weight patients (reference
odds), the odds of receiving IBR among overweight patients was 0.65
times (95% CI: 0.25 – 1.67), and the odds among obese patients was
0.38 times (95% CI: 0.13 – 1.12), with adjustment for other model
variables.
Compared to patients with in situ (DCIS/LCIS) breast cancer (reference
odds), the odds of receiving IBR among patient with stage I breast
cancer was 0.82 times (95% CI: 0.19 – 3.46), and the odds among all the
stage II patients was 0.20 times (95% CI: 0.07 – 0.55), and the odds
among all the stage III patients was 0.13 times (95% CI: 0.02 – 0.87),
with adjustment for other model variables.
The odds of having IBR among patents not having post-surgery
radiotherapy was 6.17 times the odds among patients having the
radiotherapy, with adjustment for other model variables (95% CI: 2.07 –
18.37).
(See Table 5 for details).
19
The Hosmer and Lemeshow Goodness-of-Fit test indicated there was no
significant deviation from fit in this model (p = 0.32) (data not shown). The
estimate of area under ROC obtained by the resubstitution method was 0.892
(95% CI: 0.847 – 0.937), whereas the approximation of the leave-one-out cross
validation gave an estimation of 0.858 (95% CI: 0.803 – 0.913). Thus, the
probability which this model can discriminate whether the patient would have
post-mastectomy IBR is around 85.8%. Both the methods of close to the ideal and
far from random (Youden index) defined the same optimal threshold for the
model-predicting probability as 0.50, which gives a true positive rate (sensitivity)
of 85.6% and a false negative rate (1-specificity) of 26.0% (Figure 1).
Timing of IBR rate or reconstruction rate
A total of 198 out of 452 study patients underwent reconstruction. The
majority of these 198 patients had their reconstruction within 12 months after
mastectomy (Fig. 2a). The average time to reconstruction and its standard error
were estimated based on event time and did not include the censored observation
time. The mean of time to reconstruction of all the study patients was 36.7 ± 1.37
(SE) months. Comparing the two institutions, over 50% of all the patients at
Keck/Norris had IBR (Table 1), and the average time to reconstruction was 9.2 ±
0.93 (SE) months. In contrast, less than 25% of patients at LAC had IBR and the
average time to reconstruction was 36.7 ± 1.37 (SE) months (Fig. 2b). A log-rank
test indicated that the distribution of time to breast reconstruction was statistically
significant different between two institutions (p < 0.001).
Next we excluded the patients with IBR and focused on patients with
DBR or did not have reconstruction up through the cutoff date (Oct., 01, 2014)
(censored). A total of 327 patients had DBR or were censored (Fig. 2c). Among
these patients, 73 had DBR, of which the majority had their reconstruction within
12 months after mastectomy. The average time to DBR was 38.6 ± 0.79 (SE)
months. Comparing between two institutions, less than 10% of patients underwent
DBR at Keck/Norris, and the average time to reconstruction was 20.3 ± 0.47 (SE)
months. In contrast, more than 25% of patients at LAC had DBR, and the average
time to reconstruction was 37.4 ± 0.91 (SE) months (Fig. 2d). A log-rank test
indicated that the distribution of time to reconstruction is statistically significant
different between two institutions (p < 0.01).
20
Factors associated with patients who would undergo delayed breast
reconstruction
Table 6 shows the estimated hazard ratios for patient with delayed
reconstruction or were censored based on the non-mixture cure fraction model,
with a logit link function for the fraction of patients not having breast
reconstruction (cure fraction). A proportional excess hazards was assumed in these
univariate models (i.e., not estimating the scale and shape parameters of Weibull
distribution). Excluding patients who had IBR, the hazard of not having
reconstruction among patients at Keck/Norris was 4.83 times the hazard at LAC
(95% CI: 1.67 – 13.91) (i.e., patients in LAC were more likely to receive
reconstruction at certain time point than patients in Keck/Norris, given both of
them had not yet received any reconstruction before that time point). The hazard
of not having reconstruction among patients aged over 50 was 3.93 times the
hazard among patients under 50, when excluding patients who had IBR (95% CI:
2.20 – 7.03). The hazard of not having reconstruction among employed patients
was 0.56 times the hazard among unemployed patients, when excluding patients
who had IBR (95% CI: 0.33 – 0.96). The hazard of not having reconstruction
among patients with a higher comorbidity index (≥ 3) was 0.32 times the hazard
among patients with lower index, when excluding patients who had IBR (95% CI:
0.18 – 0.57). Compared to the hazard of not having DBR among patients with
positive BRCA testing results (reference hazard), the hazard among patients with
negative testing result was 0.91 times (95% CI: 0.36 – 2.31), and the hazard
among patients who did not have BRCA testing was 2.22 times (95% CI: 0.88 –
5.64), when excluding patients who had IBR (Table 6).
Multivariate non-mixture cure fraction model for estimating
reconstruction rate and fraction of patients who would undergo delayed
breast reconstruction
The following variables were selected as candidates in the multivariate
non-mixture cure fraction model based on their clinical or statistical importance:
institution, insurance status, race, age, BMI, smoking, occupation status, BRCA
status, comorbidity index, clinical stage and progesterone receptor status. After
multiple verification and refitting steps, the final model contained the following
variables: institution (p < 0.01), insurance status (p < 0.05), age (p < 0.001),
21
smoking status (p = 0.101), occupation status (p = 0.062), clinical stage (p = 0.12)
and progesterone receptor status (p < 0.05) (Model 1 in Table 7). We also
modeled the effects of these candidate variables on the scale parameter (λ) (Model
2 in Table 7). The likelihood ratio test results showed that clinical stage had a
statistically significant effect on the scale parameter (p < 0.01). Based on the final
model, the estimated proportion of patients who would not receive delayed post-
mastectomy reconstruction at LAC was 0.711 (95% CI: 0.56 = 0.83), and the
estimated median of the time to censoring was 14.95 months (95% CI: 12.73 –
17.55). In Keck/Norris, the estimated proportion of censored patients was 0.99
(95% CI: 0.94 – 1.00) and the estimated median of the time to censored date was
9.51 months (95% CI: 6.71 – 13.48).
22
Discussion
Reconstruction rate across institutions
This current study of patients undergoing mastectomy enabled us to
compare the reconstruction rate and factors that are related to decision making and
timing of reconstructions across two institutions which are served by one medical
team. The similar distribution of treatment combinations in these two institutions
indicates the otherwise similar medical practices for treating breast cancer
(Supplementary 1), and almost equivalent rate of offering reconstruction to patients
at both institutions indicates the equivalent physician preference (Table 1: LAC vs.
Keck/Norris: 77% vs. 78%). When adjusting for patient-level factors, we found that
the odds of having IBR was significantly higher at Keck/Norris than LAC. Despite
the higher probability of having IBR at Keck/Norris, our results showed that the
odds of overall reconstruction did not differ between these two institutions,
suggesting that most of the patients at LAC selected DBR instead of IBR, which
agrees with the frequency results in Table 1.
Since the enactment of the Women’s Health and Cancer Rights Act
(WHCRA) in 1998, many studies have indicated a gradual increase of reconstruction
use over time. Moreover, some studies pointed out the disparity in the reconstruction
rate increase among different patient populations, especially patients with private vs.
non-private insurance plans. However, several previous studies were based on
nationwide registries, which were limited by evaluation of only the single admission
linked to each mastectomy performed(Hershman et al., 2012; Kruper et al., 2011;
Lang et al., 2013; Yang et al., 2013), it made these studies could therefore only focus
on the IBR patients and failed to account for the patients with DBR. In this hospital-
based study we were able to control for physician preference and to access both IBR
and DBR patient data. We found that the IBR rate did not differ between private and
non-private insurance after adjusting for patient-level factors, but the overall
reconstruction rate was notably higher among patients with private insurance plan.
This finding suggested that it might not be that the patients with non-private
insurance received less education regarding the reconstruction coverage under
WHCRA, as other studies indicated, but the patients with non-private insurance were
prone to choose DBR due to other reasons such as tumor stage or post-operative
radiation(Kuzmiak et al., 2008).
23
Predictors for reconstruction rates
Many factors affect reconstruction rates. Previous studies demonstrated that
several patient-level factors including age, race, tumor stage and grade, axillary
lymph node status, estrogen receptor status and post-mastectomy radiation were all
significantly associated with reconstruction rates(Albornoz et al., 2013; Hershman et
al., 2012; Kruper et al., 2011; Lang et al., 2013; Yang et al., 2013). Geographical
region was also shown to be an influential factor(Hershman et al., 2012; Polednak,
2000). In this current study, we found that the associations between IBR rate with
several patient-level factors including insurance status, race, age, comorbidity index
and BMI were no longer significant after adjusting for institution. On the other hand,
overall reconstruction rate stayed its statistically significant association with many
patient-level factors, after adjusting for institution. These factors include insurance
status, age, BMI, occupation status, BRCA status and comorbidity index. It was not
surprising to us that these patient-level factors would have different effects on IBR
rate and overall reconstruction rate after adjusting for institution, because the two
rates were basically different things. It might be just that some unknown institutional
elements have confounding effects on the relationship between these patient-level
predictors and IBR rate, but not on the relationship between these predictors and
overall reconstruction rate. In fact, many studies indicated the crucial roles of
institutional factors might play in reconstruction rates. A recent study revealed that
the hospital settings in which patients received surgery and care and the volume of
the surgeon who performed the mastectomy were strong predictors for IBR rate.
Women treated at larger hospitals and non-teaching hospitals were more likely to
undergo reconstruction, compared to patients treated at smaller hospitals and
teaching hospitals(Hershman et al., 2012). In our study, detailed information
regarding hospital characteristics including hospital volume and clinical operation
were not available. These information will be helpful for further clarifying whether
the hospital factors really interacted/confounded with these on the relationship
between IBR rate and patient-level factors.
Multivariate prediction Mmodels for reconstruction rates
By using the multivariate logistic regression analysis we developed
prediction models for both the IBR rate and the overall reconstruction rate. For the
IBR rate model, it is not surprising that radiotherapy was a strongly negative
24
predictor, because the guidelines of the American Society of Plastic Surgeons
(ASPS) indicated that irradiation to the expander or implant is associated with an
increased risk of post-surgery complications(Alderman et al., 2014) and thus,
patients might not consider IBR if they have already known that they will receive
radiotherapy in the future. Also, it is understandable that the Keck/Norris patients
had much higher probability of IBR compared to those in LAC, since we have
known that the majority of patients in Keck/Norris had private insurance and most
patients in LAC had non-private insurance, and many prior studies have highlighted
the higher propensity of having IBR among patients with private insurance plan
(Hershman et al., 2012; Iskandar et al., 2015; Yang et al., 2013). However, it is also
possible that the insurance status might just be a surrogate for socioeconomic status
of patients for predicting reconstruction rate. For example, those patients who had
non-private insurance might have lower income, and thus they might be less willing
to receive reconstruction because an extra procedure usually requires more time to
recovery, and those patients might not be able to afford their own living if they took
too many days off from work in order to fully recovered.
In addition, we found that compared to underweight/healthy weight patients,
both overweight (BMI: 25 – 30) and obese (BMI ≥ 30) patients had lower
probability to receive IBR, which is reasonable because manystudies have suggested
an association between higher BMI and the incidence of post-reconstruction
complications(Fischer et al., 2014a; Fischer et al., 2014b) The ASPS guidelines also
strongly recommended not to perform IBR for patients with BMI over 30(Alderman
et al., 2014). Finally, the result that later cancer stage associated with lower
propensity of having IBR is also consistent with the findings of other studies
(Iskandar et al., 2015; Lang et al., 2013).
This multivariate model gave an adjusted probability of 0.858, which is the
probability that this classifier (model) will rank a randomly chosen patients who is
having IBR (positive) higher than a randomly chosen one who is not having IBR
(negative). Based on the “conventional” definition for the best threshold, the optimal
threshold can be 0.50, which gives the true positive rate (sensitivity) of 85.6% and
the false negative rate (1-specificity) of 26.1%. By using this optimal threshold, this
model can help hospital staffs efficiently evaluate the potential of having IBR, and
thus give proper estimation of cost and the use of health care for each patients.
However, in the hospital context sometimes we might want to identify the patients
who are potentially less likely to receive IBR, and give them financial consultation
25
or education regarding the advantages of IBR, because IBR provides better cosmetic
benefits and lower surgery cost, requires less time to recovery as it basically
combines two surgeries at the same time, as well as compatible local cancer
recurrence and complication rates withDBR(Giacalone et al., 2010). Thus, we might
want to choose a cut point threshold which gives lower sensitivity (ability to detect
the patients who might have reconstruction) in trade of higher specificity (ability to
detect patients who night not have IBR), which means this threshold we choose for
the classifier should be larger than 0.5. The same logic is applicable to the overall
reconstruction rate model.
For reconstruction rate, we also built up a multivariate model to predict the
probability of having reconstruction. This model contains several patient-level
factors including insurance status race, age, BMI, disease stage and the use of
radiotherapy. Most of the associations and the trends we found in this model are
consistent with previous findings or medical practice guidelines. However, we
noticed that in this current study these non-current smokers had less propensity to
have breast reconstruction than current smokers, although this association was only
marginally significant (p = 0.09). Many previous studies have indicated that smoking
is associated with increased risk of complications(David W. Chang, 2000), and also
elevated probability of reconstruction failure(Eriksson et al., 2013). The ASPS
guidelines strongly discourage smoking patients undergoing breast reconstruction, or
at least a timely smoking cessation before reconstruction should be
conducted(Iskandar et al., 2015). In contrast, preliminary analysis of this study data
revealed that there was no significant difference in either number of revisions after
reconstruction, or in the number of infection after mastectomy (given the
reconstruction was performed), between the current smokers and non-current
smokers (data not shown).We do not know why our results were inconsistent with
the findings in the previous studies. Further investigation on the smoking cessation
education before reconstruction, or complication incidence rate after reconstruction
among smoking patients may be required to elucidate the reasons.
Analysis of Time-to-Event
According to the Kaplan-Meier plots, we noticed that there was a notably
high reconstruction rate close to the time of mastectomy, which corresponded to the
performance of IBR, and then the hazard rate was sharply decreased. This is not a
regular cumulative curve that we usually see, and the proportional hazard
26
assumption was not held at all and thus might affect the modelling of hazard ratios
(See Figure 2a and 2b). As a result, to avoid the problem of poor-fitting, we chose
not to include the IBR hazard, but only modeled the DBR hazard with survival
analysis techniques.
To model the DBR hazard we used the cure model instead of Cox
regression model. Due to the fact that a large proportion of patients not having
reconstruction (cure fraction) in this study, we were interested not only in the
reconstruction rate in the beginning of the follow-up time period, but also in
estimating the difference in the cure rate. Cox analysis is unable to separate the early
difference in reconstruction and the late difference in the ultimate cure rate, or give
the estimation for cure fraction. We would only know the reconstruction rate is
different across groups, but know nothing about the cure fraction if using Cox
regression. Thus, to analyze the cure fraction, we used the non-mixture cure fraction
model.
The univariate cure fraction model revealed that institution, age, occupation
status, BRCA status and comorbidity index were statistically significant predictors
for the cure fraction. All these significant predictors in the cure fraction model were
also significant in the univariate logistic regression we performed for the overall
breast reconstruction (See Table 3). Note that the comorbidity index was significant
but neither the clinical stage nor the pathological stage were significant predictors.
This results were puzzling, because the worse health condition usually corresponds
to later stage of cancer. A possible explanation for this correlation is that higher
score of the comorbidity index might be the result of overall worse health
conditions which was not absolutely related to breast cancer. Or there might be
collinearity among these predictors which can also result in this unexpected results,
especially when the sample size is relatively small to the total number of variables.
The final multivariate cure fraction model had statistically significant
predictors including institution, age, clinical stage and progesterone receptor status.
This model gave LAC an estimated cure fraction of 0.71 and Keck/Norris a fraction
of 0.99, which means the uncured (patients who still had chance to have DBR in the
end of follow-up) was 0.29 at LAC and only 0.01 at Keck/Norris. The selected
variable, clinical stage, also had statistically significant effect on the model scale
parameter. However, even though clinical stage has highly significant effect on the
scale parameter (p < 0.001), the scale and the significance of other parameter
estimates were not affected much after modelling scale parameter with clinical stage.
27
Based on the parsimonious principle for model building, we might want to ignore the
effect of clinical stage on scale parameter, and assume the excess hazard ratios are
proportional in each level of covariates in the future investigations.
Censoring Problem – A Demonstration
The censored portion, which is those patients who did not receive breast
reconstruction until the cutoff date, cannot be ignored in the analysis of this study.
However, in most of the analyses in this study we used logistic regression, which is a
methodology that is unable to accommodate censored observations or allow for
staggered entry of patients. In fact, the multivariate and univariate logistic regression
that we used to study the association between reconstruction rate and predictor
variables might underestimate the DBR rate, which caused the overestimation of the
IBR rate and underestimation of the overall reconstruction rate, and thus biased the
results. An obvious example of censoring problem is that from Table 1 we observed
the proportion of patients who had breast reconstruction was 43.7% among all the
patients, without considering censored problem. By the product-limit survival
method, the estimated rate of having breast reconstruction was 46.3%. The error rate
is -5.6%. Also, when stratifying by institution, the patient proportions of having
reconstruction were 38.2% and 57.9% at LAC and Keck/Norris, respectively,
without considering censored data. By product-limit method, we found they were
actually 42.0% and 58.1% at LAC and Keck/Norris, respectively. The error rates
were -9.0% and -0.3% at LAC and Keck/Norris, respectively, which were reasonable
because the patients at Keck/Norris are usually followed up better than patients at
LAC.
In Table 8, we used two important predictor variables of this study,
institution and insurance status, to demonstrate how the censored observations might
affect the model predictions. Model 1 was built for predicting the overall
reconstruction rate by logistic regression, without accounting for the censoring
effect. Model 2 was established by logistic regression for predicting the IBR rate
among all the study patients. This model was not affected by censored observations,
because neither numerator nor denominator was estimated based on number of DBR
patients. Model 3 was built using a non-cure fraction model, and controlled for
censored observations and was able to give the unbiased estimation of the DBR
probability. We used the following equation to estimate the overall reconstruction
rate based on the estimations from Model 2 and 3:
28
𝑃 𝑅 ′
= 𝑃 𝐼 𝐵𝑅 + ( 1 − 𝑃 𝐼 𝐵𝑅 ) × 𝑃 𝐷𝐵𝑅
We then used these models/equation to predict the overall reconstruction
rates for the four different groups of insurance status and hospital combinations. The
lower part of Table 8 shows the estimated overall reconstruction probabilities. By
Model 1, the estimated overall reconstruction rates were 0.373 ± 0.115, 0.632 ±
0.341, 0.395 ± 0.300 and 0.653 ± 0.211 (±SE) for LAC patients with non-private
insurance, LAC patients with private insurance, Keck/Norris patients with non-
private insurance and Keck/Norris patients with private insurance, respectively. By
the above equation based on Model 2 and 3, the estimated overall reconstruction
rates were 0.398 (LAC + Non-Private Insurance), 0.715 (LAC + Private Insurance),
0.414 (Keck/Norris + Non-Private Insurance) and 0.651 (Keck/Norris + Private
Insurance). The error rate (calculated by the difference between estimated rate of
Model 1 and estimated rate of Model 2 + 3) were -6.2% (LAC + Non-Private
Insurance), -11.6% (LAC + Private Insurance), -4.6% (Keck/Norris + Non-Private
Insurance) and 0.31% (Keck/Norris + Private Insurance). These results suggest a
mild but noticeable underestimation if we did not control for censored observations.
Note that although we might not be able to draw an unbiased conclusion by using
logistic regression in this context, logistic regression can still serve as a relatively
simple method to sift out the potential factors from the variable pool for further
analysis.
Limitations
Our study has some limitations. First, there were two noticeable problems in
the design of case report form: 1) Medicare and Medicaid were categorized together,
regardless the two insurance programs actually serve different populations; 2)
Although most of radiotherapy were supposed to be planned ahead before
undergoing mastectomy, the patient’s treatment plan may be changed to
incorporate/exclude the radiotherapy depending on the detailed observation of lesion
size, staging or margin sharpness during mastectomy, The case report form failed to
document how many patients changed their decision of receiving radiotherapy after
mastectomy. Second, the study results may have limitation in generalization as it is a
hospital-based study based on the data obtained from the USC health system. Many
studies have found that the reconstruction rate after mastectomy varies across
different geographical regions and thus, findings of this study may only be
applicable to Southern California (Lang et al., 2013). Third, the sample size is small
29
compared to other studies which used large nationwide data throughout the US. The
limited subjects might cause the unstable statistical inference. For example, there
were only 4 patients out of 57 in Keck/Norris with DBR for building the non-
mixture cure model. The cure fraction that we estimated based on this model might
not be reliable. Lastly, not only the limited sample size, but the problem of missing
values might also have biased the conclusion. More specifically, a larger proportion
of data was missing in the following 3 factors: occupation status (13.5% missing),
preoperative discussion regarding reconstruction (22% missing), and a whether the
patients had ever declined reconstruction when it was offered (22% missing). Care
should be taken when drawing any conclusions based on these data.
30
Reference
SAS Institute Inc. (2009). SAS/STAT® 9.2 user's guide. Cary, NC: SAS Institute Inc.
American Cancer Society (2015). Cancer Facts and Figures 2015. Atlanta, GA: American
Cancer Society.
Albornoz, C. R., Bach, P. B., Mehrara, B. J., et al. (2013). A paradigm shift in U.S. Breast
reconstruction: increasing implant rates. Plast Reconstr Surg 131, 15-23.
Alderman, A., Gutowski, K., Ahuja, A., Gray, D., and Postmastectomy Expander Implant
Breast Reconstruction Guideline Work, G. (2014). ASPS clinical practice guideline
summary on breast reconstruction with expanders and implants. Plast Reconstr Surg 134,
648e-655e.
David W. Chang, G. P. R., Baoguang Wang, Geoffrey L. Robb, Michael J. Miller, Gregory
R. D. Evans, Howard N. Langstein, Stephen S. Kroll (2000). Effect of smoking on
complications in patients undergoing free TRAM flap breast reconstruction. Plast Reconstr
Surg 105.
Eriksson, M., Anveden, L., Celebioglu, F., et al. (2013). Radiotherapy in implant-based
immediate breast reconstruction: risk factors, surgical outcomes, and patient-reported
outcome measures in a large Swedish multicenter cohort. Breast Cancer Res Treat 142, 591-
601.
Fischer, J. P., Tuggle, C. T., Au, A., and Kovach, S. J. (2014a). A 30-day risk assessment of
mastectomy alone compared to immediate breast reconstruction (IBR). J Plast Surg Hand
Surg 48, 209-215.
Fischer, J. P., Wes, A. M., Tuggle, C. T., and Wu, L. C. (2014b). Venous thromboembolism
risk in mastectomy and immediate breast reconstruction: analysis of the 2005 to 2011
American College of Surgeons National Surgical Quality Improvement Program data sets.
Plast Reconstr Surg 133, 263e-273e.
Gönen, M. (2007). Analyzing Receiver Operating Characteristic Curves with SAS® . Cary,
NC: SAS Institute Inc.
Giacalone, P. L., Rathat, G., Daures, J. P., Benos, P., Azria, D., and Rouleau, C. (2010). New
concept for immediate breast reconstruction for invasive cancers: feasibility, oncological
safety and esthetic outcome of post-neoadjuvant therapy immediate breast reconstruction
versus delayed breast reconstruction: a prospective pilot study. Breast Cancer Res Treat 122,
439-451.
Hershman, D. L., Richards, C. A., Kalinsky, K., et al. (2012). Influence of health insurance,
hospital factors and physician volume on receipt of immediate post-mastectomy
31
reconstruction in women with invasive and non-invasive breast cancer. Breast Cancer Res
Treat 136, 535-545.
Hosmer, D. W., and Lemesbow, S. (1980). Goodness of fit tests for the multiple logistic
regression model. Communications in Statistics - Theory and Methods 9, 1043-1069.
Howlader N, N. A., Krapcho M, Neyman N, Aminou R, Waldron W, Altekruse SF, Kosary
CL, Ruhl J, Tatalovich Z, Cho H, Mariotto A, Eisner MP, Lewis DR, Chen HS, Feuer EJ,
Cronin KA (eds) (2012). SEER Cancer Statistics Review, 1975-2009 (Vintage 2009
Populations). National Cancer Institute.
Iskandar, M. E., Dayan, E., Lucido, D., et al. (2015). Factors influencing incidence and type
of postmastectomy breast reconstruction in an urban multidisciplinary cancer center. Plast
Reconstr Surg 135, 270e-276e.
Kruper, L., Holt, A., Xu, X. X., et al. (2011). Disparities in reconstruction rates after
mastectomy: patterns of care and factors associated with the use of breast reconstruction in
Southern California. Ann Surg Oncol 18, 2158-2165.
Kummerow, K. L., Du, L., Penson, D. F., Shyr, Y ., and Hooks, M. A. (2015). Nationwide
trends in mastectomy for early-stage breast cancer. JAMA Surg 150, 9-16.
Kurian, A. W., Lichtensztajn, D. Y ., Keegan, T. H., Nelson, D. O., Clarke, C. A., and
Gomez, S. L. (2014). Use of and mortality after bilateral mastectomy compared with other
surgical treatments for breast cancer in California, 1998-2011. JAMA 312, 902-914.
Kuzmiak, C. M., Haberle, S., Padungchaichote, W., Zeng, D., Cole, E., and Pisano, E. D.
(2008). Insurance status and the severity of breast cancer at the time of diagnosis. Acad
Radiol 15, 1255-1258.
Lambert, P. C., Thompson, J. R., Weston, C. L., and Dickman, P. W. (2007). Estimating and
modeling the cure fraction in population-based cancer survival analysis. Biostatistics 8, 576-
594.
Lang, J. E., Summers, D. E., Cui, H., et al. (2013). Trends in post-mastectomy
reconstruction: a SEER database analysis. J Surg Oncol 108, 163-168.
Polednak, A. P. (2000). Geographic Variation in Postmastectomy Breast Reconstruction
Rates. Plast. Reconstr. Surg 106, 298.
Yang, R. L., Newman, A. S., Lin, I. C., et al. (2013). Trends in immediate breast
reconstruction across insurance groups after enactment of breast cancer legislation. Cancer
119, 2462-2468.
Zervoudis, S., Iatrakis, G., Tomara, E., Bothou, A., Papadopoulos, G., and Tsakiris, G.
(2014). Main controversies in breast cancer. World J Clin Oncol 5, 359-373.
32
Tables
Table 1. Characteristics of Patients by Institution
No. of Patients (%)
Characteristics
Total
(N = 453)
LA County Medical
Center
(N = 327 (72.2%))
Keck Hospital of USC /
Norris Cancer Center
(N = 126 (27.8%)) P-value
[1]
Patient and Institution Factors
Race
Caucasian 96 (21.2%) 31 (9.5%) 65 (51.6%) <0.001
African American 22 (4.9%) 12 (3.7%) 10 (7.9%)
Hispanic 244 (53.9%) 226 (69.1%) 18 (14.3%)
Asian 80 (17.7%) 52 (15.9%) 28 (22.2%)
Other 11 (2.4%) 6 (1.8%) 5 (4%)
Age
< 50yrs 216 (47.7%) 176 (53.8%) 40 (31.7%) <0.001
≥ 50yrs 237 (52.3%) 151 (46.2%) 86 (68.3%)
BMI
Underweight (BMI < 18.5) 8 (1.8%) 4 (1.2%) 4 (3.2%) <0.001
Healthy weight (BMI: 18.5 - 25.0) 144 (31.8%) 79 (24.2%) 65 (51.6%)
Overweight (BMI: 25.0 - 30.0) 163 (36%) 128 (39.1%) 35 (27.8%)
Obese (BMI > 30) 138 (30.5%) 116 (35.5%) 22 (17.5%)
Insurance Status
None 55 (12.1%) 54 (16.5%) 1 (0.8%) <0.001
Medicare/Medicaid 295 (65.1%) 260 (79.5%) 35 (27.8%)
Private 103 (22.7%) 13 (4%) 90 (71.4%)
Currently on diabetes medication
[2]
No 398 (87.9%) 279 (85.3%) 119 (94.4%) 0.008
Yes 55 (12.1%) 48 (14.7%) 7 (5.6%)
Smoking Status
No 410 (90.5%) 298 (91.1%) 112 (88.9%) 0.47
Yes 43 (9.5%) 29 (8.9%) 14 (11.1%)
Occupation Status
Unemployed 191 (42.2%) 136 (41.6%) 55 (43.7%) <0.001
Employed 201 (44.4%) 131 (40.1%) 70 (55.6%)
Missing 61 (13.5%) 60 (18.3%) 1 (0.8%)
Comorbidity Index
2 211 (46.6%) 170 (52%) 41 (32.5%) <0.001
≥3 242 (53.4%) 157 (48%) 85 (67.5%)
Family History of Breast Cancer
No 287 (63.4%) 229 (70%) 58 (46%) <0.001
Yes 166 (36.6%) 98 (30%) 68 (54%)
Personal History of Other Cancer
No 407 (89.8%) 297 (90.8%) 110 (87.3%) 0.27
Yes 46 (10.2%) 30 (9.2%) 16 (12.7%)
33
No. of Patients (%)
Characteristics
Total
(N = 453)
LA County Medical
Center
(N = 327 (72.2%))
Keck Hospital of USC /
Norris Cancer Center
(N = 126 (27.8%)) P-value
[1]
Type of prior cancer
Breast Cancer 20 (43.5%) 16 (53.3%) 4 (25%) 0.004
Ovarian Cancer 1 (2.2%) 1 (3.3%) 0 (0.0%)
Uterine Cancer 5 (10.9%) 5 (16.7%) 0 (0.0%)
Colon Cancer 2 (4.3%) 1 (3.3%) 1 (6.3%)
Thyroid Cancer 3 (6.5%) 3 (10%) 0 (0.0%)
Other Cancer 14 (30.4%) 3 (10%) 11 (68.8%)
Missing 1 (2.2%) 1 (3.3%) 0 (0.0%)
Prior Surgical Treatment of Breast Cancer
Lumpectomy 19 (95%) 15 (93.8%) 4 (100%) 0.61
Other 1 (5%) 1 (6.3%) 0 (0.0%)
Discussion with patient regarding reconstruction
Yes 352 (77.7%) 255 (78%) 97 (77%) 0.82
Not documented 101 (22.3%) 72 (22%) 29 (23%)
Patient declined reconstruction
No 200 (44.2%) 127 (38.8%) 73 (57.9%) <0.001
Yes 97 (21.4%) 85 (26%) 12 (9.5%)
Unknown 56 (12.4%) 44 (13.5%) 12 (9.5%)
Missing 100 (22.1%) 71 (21.7%) 29 (23%)
Disease Factors
Clinical Stage
I 76 (16.8%) 49 (15%) 27 (21.4%) 0.08
II 217 (47.9%) 161 (49.2%) 56 (44.4%)
III 65 (14.3%) 54 (16.5%) 11 (8.7%)
in situ (DCIS/LCIS) 93 (20.5%) 62 (19%) 31 (24.6%)
Missing 2 (0.4%) 1 (0.3%) 1 (0.8%)
Pathological/Surgical Stage
I 119 (26.3%) 74 (22.6%) 45 (35.7%) 0.006
II 157 (34.7%) 119 (36.4%) 38 (30.2%)
III 99 (21.9%) 78 (23.9%) 21 (16.7%)
in situ (DCIS/LCIS) 61 (13.5%) 40 (12.2%) 21 (16.7%)
complete pathologic response (ypT0N0) 17 (3.8%) 16 (4.9%) 1 (0.8%)
Estrogen Receptor Status
Positive 329 (72.6%) 230 (70.3%) 99 (78.6%) 0.08
Negative 124 (27.4%) 97 (29.7%) 27 (21.4%)
Progesterone Receptor Status
Positive 269 (59.4%) 189 (57.8%) 80 (63.5%) 0.27
Negative 184 (40.6%) 138 (42.2%) 46 (36.5%)
HER2 Status
Positive 100 (22.1%) 80 (24.5%) 20 (15.9%) 0.023
Negative 293 (64.7%) 199 (60.9%) 94 (74.6%)
DCIS/Not applicable 60 (13.2%) 48 (14.7%) 12 (9.5%)
34
No. of Patients (%)
Characteristics
Total
(N = 453)
LA County Medical
Center
(N = 327 (72.2%))
Keck Hospital of USC /
Norris Cancer Center
(N = 126 (27.8%)) P-value
[1]
BRCA Status
Positive 35 (7.7%) 31 (9.5%) 4 (3.2%) <0.001
Negative 163 (36%) 134 (41%) 29 (23%)
Not tested 255 (56.3%) 162 (49.5%) 93 (73.8%)
If BRCA
+
, BRCA 1 or 2
BRCA 1 18 (51.4%) 15 (48.4%) 3 (75%) 0.32
BRCA 2 17 (48.6%) 16 (51.6%) 1 (25%)
Recurrence of breast cancer
No 412 (90.9%) 295 (90.2%) 117 (92.9%) 0.38
Yes 41 (9.1%) 32 (9.8%) 9 (7.1%)
Local or distant recurrence
Local 3 (7.5%) 2 (6.5%) 1 (11.1%) 0.87
Distant 29 (72.5%) 23 (74.2%) 6 (66.7%)
Both 8 (20%) 6 (19.4%) 2 (22.2%)
Treatment Factors
Neoadjuvant Therapy
No 355 (78.4%) 246 (75.2%) 109 (86.5%) 0.009
Yes 98 (21.6%) 81 (24.8%) 17 (13.5%)
Adjuvant Chemotherapy
No 254 (56.1%) 171 (52.3%) 83 (65.9%) 0.009
Yes 199 (43.9%) 156 (47.7%) 43 (34.1%)
Radiotherapy
No 276 (60.9%) 182 (55.7%) 94 (74.6%) <0.001
Yes 177 (39.1%) 145 (44.3%) 32 (25.4%)
Endocrine Therapy
No 149 (32.9%) 106 (32.4%) 43 (34.1%) 0.73
Yes 304 (67.1%) 221 (67.6%) 83 (65.9%)
Surgery Factors
Type of Mastectomy
Total mastectomy 391 (86.3%) 293 (89.6%) 98 (77.8%) <0.001
Nipple-sparing mastectomy 17 (3.8%) 4 (1.2%) 13 (10.3%)
Skin-sparing mastectomy 45 (9.9%) 30 (9.2%) 15 (11.9%)
Laterality of mastectomy
Unilateral 378 (83.4%) 288 (88.1%) 90 (71.4%) <0.001
Bilateral 75 (16.6%) 39 (11.9%) 36 (28.6%)
Prophylactic Mastectomy
No 17 (22.7%) 13 (33.3%) 4 (11.1%) 0.021
Yes 58 (77.3%) 26 (66.7%) 32 (88.9%)
Wound Infection after Mastectomy
No 362 (79.9%) 252 (77.1%) 110 (87.3%) 0.015
Yes 91 (20.1%) 75 (22.9%) 16 (12.7%)
35
No. of Patients (%)
Characteristics
Total
(N = 453)
LA County Medical
Center
(N = 327 (72.2%))
Keck Hospital of USC /
Norris Cancer Center
(N = 126 (27.8%)) P-value
[1]
No. of infection
1 81 (89%) 65 (86.7%) 16 (100%) 0.30
2 9 (9.9%) 9 (12%) 0 (0.0%)
≥3 1 (1.1%) 1 (1.3%) 0 (0.0%)
Reconstruction
No 254 (56.1%) 201 (61.5%) 53 (42.1%) <0.001
Yes 198 (43.7%) 125 (38.2%) 73 (57.9%)
Missing 1 (0.2%) 1 (0.3%)
Timing of Reconstruction
Delayed 73 (36.9%) 69 (55.2%) 4 (5.5%) <0.001
Immediate 125 (63.1%) 56 (44.8%) 69 (94.5%)
Type of Reconstruction
Autologous 78 (39.4%) 66 (52.8%) 12 (16.4%) <0.001
Implant 86 (43.4%) 31 (24.8%) 55 (75.3%)
Both 34 (17.2%) 28 (22.4%) 6 (8.2%)
No. of Revisions
0 43 (21.7%) 34 (27.2%) 9 (12.3%) <0.001
1 68 (34.3%) 30 (24%) 38 (52.1%)
2 53 (26.8%) 42 (33.6%) 11 (15.1%)
≥3 34 (17.2%) 19 (15.2%) 15 (20.5%)
Total Number of Surgeries
[3]
1 249 (55%) 193 (59%) 56 (44.4%) <0.001
2 95 (21%) 53 (16.2%) 42 (33.3%)
≥3 109 (24.1%) 81 (24.8%) 28 (22.2%)
[1]
Significance was determined using Chi-square test.
[2]
Including oral meds or insulin.
[3]
including mastectomy
36
Table 2. Association with IBR Rate or Overall Reconstruction Rate across Institutions
Overall Reconstruction Rate
[1]
IBR Rate
[2]
Unadjusted
[3]
Adjusted
[4]
(Global P-value = <0.001) Unadjusted
[3]
Adjusted
[4]
(Global P-value = <0.001)
Variable
ORs
(95% CI) P-value
ORs
(95% CI) P-value
ORs
(95% CI) P-value
ORs
(95% CI) P-value
Institution
LAC (Ref.) <0.001 0.53 <0.001 <0.001
Keck/Norris
2.2
(1.46 - 3.36)
-
1.3
(0.61 - 2.59)
-
21.2
(7.3 - 61.79)
-
22.3
(3.75 - 132.36)
-
Insurance Status
Private (Ref.) <0.001 0.0009 <0.001 0.78
Non-Private
0.3
(0.2 - 0.51)
-
0.3
(0.15 - 0.61)
-
0.1
(0.06 - 0.31)
-
1.3
(0.25 - 6.29)
-
Race
Caucasian (Ref.) 0.23 0.23 0.004 0.99
African American
0.7
(0.29 - 1.86)
-
1.5
(0.54 - 4.21)
-
0.5
(0.11 - 2.31)
-
1.4
(0.18 - 10.83)
-
Hispanic
0.7
(0.41 - 1.06)
-
1.1
(0.59 - 2.06)
-
0.2
(0.1 - 0.5)
-
1.3
(0.43 - 4.19)
-
Asian
0.5
(0.27 - 0.92)
-
0.6
(0.28 - 1.17)
-
0.6
(0.19 - 1.67)
-
1.2
(0.28 - 5.06)
-
Other
0.5
(0.14 - 1.84)
-
0.6
(0.16 - 2.37)
-
0.2
(0.03 - 1.73)
-
1.5
(0.15 - 15.03)
-
Age
≥ 50yrs (Ref.) <0.001 0.009 0.007 0.51
< 50yrs
2.5
(1.69 - 3.63)
-
3.6
(1.38 - 9.14)
-
0.4
(0.23 - 0.79)
-
0.6
(0.13 - 2.78)
-
BMI
Underweight/Healthy weight (Ref.) <0.001 0.008 <0.001 0.14
Obese
0.6
(0.38 - 0.93)
-
0.7
(0.43 - 1.19)
-
0.4
(0.21 - 0.86)
-
0.8
(0.32 - 1.77)
-
Overweight
0.4
(0.25 - 0.64)
-
0.4
(0.24 - 0.73)
-
0.2
(0.1 - 0.47)
-
0.4
(0.14 - 1.02)
-
Comorbidity Score
≥3 (Ref.) <0.001 0.63 0.014 0.92
2
2.2
(1.47 - 3.14)
-
0.8
(0.32 - 2.01)
-
0.5
(0.26 - 0.86)
-
0.9
(0.2 - 4.21)
-
Abbreviation: Ref., reference group; OR, odds ratio; CI, confidence interval; LAC, LA County Medical Center; Keck/Norris, Keck Hospital of USC/Norris
Cancer Center.
[1]
The Overall reconstruction rate was the proportion of patients receiving reconstruction among the total patients having mastectomy.
[2]
The IBR rate was the proportion of patients receiving IBR among the total patients having reconstruction.
[3]
Unadjusted ORs were determined by logistic regression.
[4]
Adjusted ORs were calculated by logistic regression with institution, insurance status, race, age, BMI and comorbidity inside the model.
Note: All the P-values were estimated by Wald Chi-squared test.
37
Table 3. Association between Overall Reconstruction Rate or IBR Rate with Predictor Variables (Univariate Analysis
and Institution-Adjusted)
Overall Reconstruction IBR
P-value P-value
Variable
Reconstruction
Rate(%)
[1]
Unadjusted
[2]
Institution-
Adjusted
[3]
IBR Rate(%)
[4]
Unadjusted
[2]
Institution-
Adjusted
[3]
Institution
LAC 38.3 <0.001 - 44.8 <0.001 -
Keck/Norris 57.9 - - 94.5 - -
Insurance Status
Non-Private 37.5 <0.001 0.002 50.4 <0.001 0.57
Private 65 - - 88.1 - -
Race
Caucasian 53.1 0.23 0.35 82.4 0.004 0.99
African American 45.5 - - 70 - -
Hispanic 42.8 - - 51 - -
Asian 36.3 - - 72.4 - -
Other 36.4 - - 50 - -
Age
< 50yrs 55.3 <0.001 <0.001 55.5 0.007 0.18
≥ 50yrs 33.3 - - 74.7 - -
BMI
Underweight/Healthy weight 55.3 <0.001 0.012 77.4 <0.001 0.13
Overweight 42.3 - - 59.4 - -
Obese 32.8 - - 42.2 - -
Currently on diabetes medication
No 45.8 0.021 0.057 63.7 0.55 0.93
Yes 29.1 - - 56.3 - -
Smoking Status
No 42.8 0.18 0.22 64 0.49 0.22
Yes 53.5 - - 56.5 - -
Occupation Status
Unemployed 40.3 0.008 0.013 58.4 0.25 0.89
Employed 53.7 - - 66.7 - -
Family History of Breast Cancer
No 40.6 0.07 0.30 59.5 0.21 0.56
Yes 49.4 - - 68.3 - -
Personal History of Other Cancer
No 42.6 0.13 0.18 60.1 0.028 0.018
Yes 54.3 - - 84 - -
BRCA Status
Positive 48.6 0.08 0.007 52.9 0.07 0.60
Negative 50 - - 55.6 - -
Not tested 39.2 - - 71 - -
Comorbidity Index
2 53.8 <0.001 <0.001 55.8 0.014 0.21
≥3 35.1 - - 72.9 - -
Clinical Stage
I 44.7 0.004 0.011 85.3 <0.001 <0.001
II 41 - - 50.6 - -
III 29.7 - - 21.1 - -
in situ (DCIS/LCIS) 58.1 - - 85.2 - -
38
Overall Reconstruction IBR
P-value P-value
Variable
Reconstruction
Rate(%)
[1]
Unadjusted
[2]
Institution-
Adjusted
[3]
IBR Rate(%)
[4]
Unadjusted
[2]
Institution-
Adjusted
[3]
Pathological/Surgical Stage
I 47.9 0.002 0.005 75.4 <0.001 0.001
II 38.2 - - 56.7 - -
III 33.7 - - 36.4 - -
in situ (DCIS/LCIS) 63.9 - - 82.1 - -
complete pathologic response (ypT0N0) 52.9 - - 44.4 - -
Estrogen Receptor Status
Positive 43.5 0.81 0.58 65 0.37 0.43
Negative 44.7 - - 58.2 - -
Progesterone Receptor Status
Positive 46.8 0.12 0.16 61.9 0.64 0.91
Negative 39.3 - - 65.3 - -
HER2 Status
Positive 40.4 0.027 0.012 47.5 0.015 0.001
Negative 41.6 - - 63.1 - -
DCIS/Not applicable 60 - - 80.6 - -
Recurrence of breast cancer
No 44.5 0.33 0.41 63.4 0.79 0.42
Yes 36.6 - - 60 - -
Neoadjuvant Therapy
No 46.8 0.016 0.045 68.1 0.002 0.002
Yes 33 - - 37.5 - -
Adjuvant Chemotherapy
No 46.2 0.24 0.47 69.2 0.033 0.07
Yes 40.7 - - 54.3 - -
Radiotherapy
No 51.1 0.001 <0.001 75.2 <0.001 <0.001
Yes 32.4 - - 33.3 - -
Endocrine Therapy
No 45 0.73 0.77 65.7 0.60 0.70
Yes 43.2 - - 61.8 - -
Laterality of mastectomy
Unilateral 40.1 <0.001 0.004 55 0.001 0.002
Bilateral 62.7 - - 89.4 - -
Type of Mastectomy
Total mastectomy 35.1 <0.001 <0.001 48.2 <0.001 <0.001
Partial mastectomy 98.4 - - 96.7 - -
Type of Reconstruction
Autologous - - - 32.1 <0.001 <0.001
Implant - - - 94.2 - -
Both - - - 55.9 - -
Total Number of Revisions
0 - - - 44.2 0.005 0.28
1 - - - 75 - -
2 - - - 56.6 - -
≥3 - - - 73.5 - -
Total Number of Surgeries
[4]
1 6.5 <0.001 <0.001 100 0.12 0.92
2 77.9 - - 68.9 - -
≥3 99.1 - - 53.7 - -
[1]
The Overall reconstruction rate was the proportion of patients receiving reconstruction among the total patients having mastectomy.
[2]
Wald Chi-square test was performed to compare the odds of receiving IBR or Reconstruction among different stratifications of patients.
[3]
Wald Chi-square test was performed to compare the odds of receiving IBR or Reconstruction among different stratifications of patients, after controlling for
institutions.
[4]
The IBR rate was the proportion of patients receiving IBR among the total patients having reconstruction in each stratification.
39
Table 4. Association between Overall Breast Reconstruction with Predictor Variables (Multivariate Analysis)
Preliminary Model Final Model
Parameters ORs (95% CI) P-value
Global P-
value ORs (95% CI) P-value
Global P-
value
Institution
LAC (Ref.) - - 0.99 - -
Keck/Norris
0.994
(0.424 ,2.332)
0.99 -
Insurance Status
Private (Ref.) - - 0.010 - - <.001
None/Medicare/Medicaid
0.335
(0.145 ,0.772)
0.010
0.324
(0.174 ,0.6)
<.001
Race
Caucasian (Ref.) - - 0.026 - - 0.038
African American
1.126
(0.346 ,3.669)
0.84
1.24
(0.404 ,3.807)
0.71
Asian
0.434
(0.191 ,0.985)
0.046
0.459
(0.215 ,0.982)
0.045
Hispanic
1.264
(0.616 ,2.594)
0.52
1.231
(0.64 ,2.368)
0.53
Other
0.193
(0.033 ,1.129)
0.068
0.229
(0.042 ,1.24)
0.087
Age
≥ 50yrs (Ref.) - - 0.026 - - <.001
< 50yrs
3.699
(1.169 ,11.705)
0.026
2.906
(1.781 ,4.739)
<.001
BMI
Underweight/Healthy weight (Ref.) - - 0.012 - - 0.011
Obese
0.378
(0.196 ,0.729)
0.020
0.393
(0.212 ,0.727)
0.015
Overweight
0.529
(0.296 ,0.946)
0.54
0.562
(0.322 ,0.979)
0.65
Currently on diabetes medication
Yes (Ref.) - - 0.91 - -
No
1.049
(0.443 ,2.483)
0.91 -
Smoking Status
Yes (Ref.) - - 0.11 - - 0.080
No
0.5
(0.215 ,1.162)
0.11
0.485
(0.216 ,1.091)
0.080
Occupation Status
Unemployed (Ref.) - - 0.086 - - 0.18
Employed
1.529
(0.942 ,2.484)
0.086
1.37
(0.869 ,2.161)
0.18
Family History of Breast Cancer
Yes (Ref.) - - 0.35 - -
No
0.775
(0.454 ,1.324)
0.35 -
Personal History of Other Cancer
Yes (Ref.) - - 0.19 - - 0.12
No
0.593
(0.27 ,1.3)
0.19
0.55
(0.26 ,1.162)
0.12
BRCA Status
Positive (Ref.) - - 0.53 - -
Negative
0.631
(0.253 ,1.575)
0.55 -
Not tested
0.573
(0.214 ,1.531)
0.34 -
40
Preliminary Model Final Model
Parameters ORs (95% CI) P-value
Global P-
value ORs (95% CI) P-value
Global P-
value
Comorbidity Index
≥3 (Ref.) - - 0.60 - -
2
0.734
(0.23 ,2.348)
0.60 -
Clinical Stage
in situ (DCIS/LCIS) (Ref.) - - 0.72 - -
I
0.711
(0.263 ,1.926)
0.95 -
II
0.738
(0.291 ,1.873)
0.92 -
III
0.523
(0.165 ,1.658)
0.32 -
Progesterone Receptor Status
Positive (Ref.) - - 0.20 - - 0.22
Negative
0.718
(0.431 ,1.195)
0.20
0.753
(0.477 ,1.19)
0.22
HER2 Status
Positive (Ref.) - - 0.49 - -
DCIS/Not applicable
0.942
(0.308 ,2.882)
0.81 -
Negative
0.691
(0.368 ,1.301)
0.32 -
Neoadjuvant Therapy
Yes (Ref.) - - 0.45 - -
No
1.373
(0.602 ,3.133)
0.45 -
Chemotherapy
Yes (Ref.) - - 0.27 - -
No
1.413
(0.767 ,2.601)
0.27 -
Radiotherapy
Yes (Ref.) - - 0.064 - - <.001
No
1.746
(0.968 ,3.15)
0.064
2.322
(1.444 ,3.733)
<.001
Abbreviation: Ref., reference group; OR, odds ratio; CI, confidence interval; LA County, LA County Medical Center; Keck/Norris, Keck Hospital of USC/Norris
Cancer Center.
Note: 1. Overall reconstruction rate was the proportion of patients receiving reconstruction among the total patients having mastectomy.
2. Wald Chi-square test was performed to compare the odds of receiving reconstruction among different stratifications of patients.
41
Table 5. Association between Immediate Breast Reconstruction with Predictor Variables (Multivariate Analysis)
Preliminary Model Final Model
Parameters ORs (95% CI) P-value
Global P-
value ORs (95% CI) P-value
Global P-
value
Institution
LAC (Ref.) - - <.001 - - <.001
Keck/Norris
80.948
(6.884 ,951.895)
<.001
30.529
(8.143 ,114.464)
<.001
Insurance Status
Private (Ref.) - - 0.39 - -
Non-Private
2.613
(0.288 ,23.685)
0.39 -
Race
Caucasian (Ref.) - - 0.99 - -
African American
1.834
(0.097 ,34.597)
0.69 -
Asian
1.446
(0.214 ,9.753)
0.71 -
Hispanic
1.345
(0.336 ,5.377)
0.68 -
Other
1.206
(0.066 ,22.003)
0.90 -
Age
≥ 50yrs ( - - 0.60 - -
< 50yrs
0.603
(0.09 ,4.037)
0.60 -
BMI
Underweight/Healthy weight - - 0.33 - - 0.21
Obese
0.398
(0.117 ,1.345)
0.20
0.378
(0.128 ,1.115)
0.11
Overweight
0.624
(0.223 ,1.744)
0.98
0.65
(0.253 ,1.665)
0.89
Family History of Breast Cancer
Yes (Ref.) - - 0.77 - -
No
0.857
(0.302 ,2.43)
0.77 -
Personal History of Other Cancer
Yes (Ref.) - - 0.37 - -
No
0.522
(0.126 ,2.16)
0.37 -
BRCA Status
Positive (Ref.) - - 0.92 - -
Negative
0.71
(0.117 ,4.295)
0.79 -
Not tested
0.689
(0.107 ,4.428)
0.75 -
Comorbidity Index
≥3 (Ref.) - - 0.99 - -
2
0.992
(0.144 ,6.83)
0.99 -
Clinical Stage
in situ (DCIS/LCIS) (Ref.) - - 0.14 - - 0.004
I
0.652
(0.069 ,6.149)
0.33
0.818
(0.194 ,3.457)
0.13
II
0.192
(0.028 ,1.327)
0.12
0.202
(0.074 ,0.551)
0.054
III
0.142
(0.01 ,2.074)
0.22
0.133
(0.02 ,0.869)
0.10
42
Preliminary Model Final Model
Parameters ORs (95% CI) P-value
Global P-
value ORs (95% CI) P-value
Global P-
value
HER2 Status
Positive (Ref.) - - 0.97 - -
DCIS/Not applicable
0.724
(0.061 ,8.597)
0.81 -
Negative
0.9
(0.275 ,2.948)
0.93 -
Neoadjuvant Therapy
Yes (Ref.) - - 0.55 - -
No
1.552
(0.374 ,6.438)
0.55 -
Radiotherapy
Yes (Ref.) - - 0.003 - - 0.001
No
6.103
(1.878 ,19.836)
0.003
6.169
(2.072 ,18.368)
0.001
Chemotherapy
Yes (Ref.) - - 0.70 - -
No
1.235
(0.419 ,3.645)
0.70 -
Abbreviation: Ref., reference group; OR, odds ratio; CI, confidence interval; LA County, LA County Medical Center; Keck/Norris, Keck Hospital of USC/Norris
Cancer Center.
Note: 1. The IBR rate was the proportion of patients receiving IBR among the total patients having reconstruction.
2. Wald Chi-square test was performed to compare the odds of receiving IBR among different stratifications of patients.
43
Table 6. Hazard Ratio Estimates of Non-Mixture Cure Model for Patients with Delayed
Reconstruction or Patients Who Were Censored (Univariate Analysis)
Variable Hazard Ratio (95% CI)
[1]
P-value
[2]
Institution
LAC (Ref.) 0.004
Keck/Norris 4.826 (1.674,13.909)**
Race
Other (Ref.) 0.35
Caucasian 1.310 (0.229,7.492)
African American 1.004 (0.130,7.773)
Asian 1.652 (0.285,9.565)
Hispanic 0.764 (0.151,3.857)
Insurance Status
Non-Private (Ref.) 0.48
Private 1.342 (0.588,3.066)
Age
< 50yrs (Ref.) <0.001
≥ 50yrs 3.933 (2.199,7.034) ***
BMI
Underweight/Healthy weight (Ref.) 0.94
Overweight 0.909 (0.465,1.778)
Obese 1.005 (0.507,1.992)
Currently on diabetes medication
No (Ref.) 0.44
Yes 1.413 (0.591,3.380)
Smoking Status
No (Ref.) 0.17
Yes 0.566(0.250,1.282)
Occupation Status
Unemployed (Ref.) 0.036
Employed 0.563 (0.329,0.964)*
Family History of Breast Cancer
No (Ref.) 0.64
Yes 0.876 (0.502,1.528)
Personal History of Other Cancer
No (Ref.) 0.42
Yes 1.577 (0.517,4.807)
BRCA Status
Positive (Ref.) 0.007
Negative 0.912 (0.360,2.307)
Not tested 2.224 (0.876,5.644)
Comorbidity Index
≥3 (Ref.) <.0001
2 0.322 (0.184,0.566)***
44
Variable Hazard Ratio (95% CI)
[1]
P-value
[2]
Clinical Stage
I (Ref.) 0.24
II 0.403 (0.158,1.023)
III 0.474 (0.167,1.344)
in situ (DCIS/LCIS) 0.637 (0.200,2.033)
Pathological/Surgical Stage
complete pathologic response (ypT0N0) (Ref.) 0.57
I 2.900 (0.787,10.691)
II 2.665 (0.767,9.266)
III 2.304 (0.651,8.158)
in situ (DCIS/LCIS) 2.061 (0.482,8.817)
Estrogen Receptor Status
Positive (Ref.) 0.53
Negative 0.833 (0.468,1.481)
Progesterone Receptor Status
Positive (Ref.) 0.15
Negative 1.497 (0.861,2.602)
HER2 Status
Positive (Ref.) 0.67
DCIS/Not applicable 1.127 (0.416,3.054)
Negative 1.311 (0.715,2.404)
Recurrence of breast cancer
No (Ref.) 0.52
Yes 1.358 (0.532,3.464)
Neoadjuvant Therapy
No (Ref.) 0.53
Yes 0.826 (0.453,1.506)
Adjuvant Chemotherapy
No (Ref.) 0.67
Yes 0.891 (0.524,1.517)
Radiotherapy
No (Ref.) 0.58
Yes 0.860 (0.505,1.464)
Endocrine Therapy
No (Ref.) 0.84
Yes 0.942 (0.534,1.665)
[1]
* p < 0.05, ** p < 0.01, *** p < 0.001
[2]
Wald Chi-square test was performed to compare the reconstruction incidence among different categories of patients.
45
Table 7. Hazard Ratio Estimates of Non-Mixture Cure Models for Patients with Delayed Reconstruction
or Patients Wo Were Censored (Multivariate Analysis)
Model 1
Scale (λ) constant
Shape (γ)constant
(Global P-value < 0.001)
Model 2
Scale (λ) modeled
Shape (γ)constant
(Global P-value < 0.001)
Variable
Hazard Ratio
(95% CI)
[1]
Hazard Ratio
(95% CI)
[1][2]
Institution
LAC (Ref.)
Keck/Norris 12.643 (2.477,64.532)** 12.233 (2.392,62.573) **
Insurance Status
Non-Private (Ref.)
Private 0.203 (0.051,0.808) 0.206 (0.052,0.822) *
Age
< 50yrs (Ref.)
≥ 50yrs 3.585 (1.891,6.796) *** 3.631 (1.922,6.6857) ***
Smoking Status
No (Ref.)
Yes 0.463 (0.185,1.162) 0.459 (0.182,1.155)
Occupation Status
Unemployed (Ref.)
Employed 0.567 (0.312,1.028) 0.573 (0.316,1.1040)
Clinical Stage
I (Ref.)
II 0.341 (0.123,0.946)* 0.317 (0.113,0.894)*
III 0.586 (0.185,1.856) 0.430 (0.131,1.418)
in situ (DCIS/LCIS) 0.514 (0.145,1.819) 0.538 (0.152,1.902)
Progesterone Receptor Status
Positive (Ref.)
Negative 1.904 (1.023,3.544)* 2.111 (1.106,4.028)*
Scale Parameter (λ)
LRT (Institution) - p = 0.28
LRT (Insurance Status) - p = 0.20
LRT (Age) - p = 0.77
LRT (Smoking Status) - p = 0.27
LRT (Occupation Status) - p = 0.14
LRT (Clinical Stage) - p = 0.001**
LRT (Progesterone Receptor Status) - p = 0.71
Shape Parameter (γ)
Log Likelihood -391.13 -378.82
LRT, likelihood ratio test
[1]
* p < 0.05, ** p < 0.01, *** p < 0.001
[2]
Hazard ratios when modelling the effect of clinical stage on the scale parameter at the same time.
46
Table 8. Reconstruction Probability Estimation when Controlling and Not Controlling for Censored Observatiions
Model 1
Estimate the probability of overall
reconstruction by logistic regression model
(P
R
)
Model 2
Estimate the probability of IBR by
logistic regression model
(P
IBR
)
Model 3
Estimate the probability of DBR by non-
cure fraction model
(P
DBR
)
Variable
Parameter
Estimates SE
Parameter
Estimates SE
Parameter
Estimates SE
Intercept -0.519*** 0.1153 -1.621*** 0.1487 0.901*** 0.1505
Institution 0.092 0.3111 1.190*** 0.3215 2.436*** 0.7340
Insurance Status 1.060** 0.3342 0.875** 0.3372 -1.270* 0.6311
Estimations for Reconstruction Probability
Model 1 Estimates SE
LAC + Non-Private Insurance 0.373 0.115
LAC + Private Insurance 0.632 0.341
Keck/Norris + Non-Private Insurance 0.395 0.300
Keck/Norris + Private Insurance 0.653 0.211
Model 2 + Model 3
P
R
’ = P
IBR
+ (1 - P
IBR
) × P
DBR
Estimates SE
LAC + Non-Private Insurance 0.398 N/A
LAC + Private Insurance 0.715 N/A
Keck/Norris + Non-Private Insurance 0.414 N/A
Keck/Norris + Private Insurance 0.651 N/A
Error Rate Percentage (%)
LAC + Non-Private Insurance -6.225
LAC + Private Insurance -11.551
Keck/Norris + Non-Private Insurance -4.554
Keck/Norris + Private Insurance 0.308
47
Figures
Figure 1. ROC Curves for Logistic Models.
(a) Overall reconstruction; (b) IBR.
48
Overall Reconstruction DBR + Censored
Overall Reconstruction DBR + Censored
No. of Patients (%) No. of Patients (%)
Institution Total
Recon-
struction
Censored
(%)
[1]
Mean (SE)
[2]
Total
Recon-
struction
Censored
(%)
[1]
Mean (SE)
[2]
Keck/Norris 126 73 53 (42.06) 9.2 (0.93) 57 4 53 (92.98) 20.3 (0.47)
LAC 326 125 201 (61.66) 40.6 (1.53) 270 69 201 (74.44) 37.4 (0.91)
Total 452 198 254 (56.19) 36.7 (1.37) 327 73 254 (77.68) 38.6 (0.79)
[1]
Patients who did not undergo reconstruction until the cutoff date (Oct., 01, 2014).
[2]
The mean time to reconstruction and its standard error were underestimated because the largest observation was censored and the estimation was
restricted to the largest event time.
Figure 2. Time to Reconstruction.
(a) Overall reconstruction; (b) Overall reconstruction (stratified by institution); (c) DBR + Censored; (d) DBR + Censored (stratified by
institution); (Lower table) Descriptive Analysis for time to reconstruction.
49
Supplementary Materials
Supplementary 1. Treatment Factors of Patients by Institution
No. of Patients (%)
Treatment Factors
Total
(N = 453)
LA County Medical Center
(N = 327 (72.2%))
Keck Hospital of USC /
Norris Cancer Center
(N = 126 (27.8%))
Treatments in addition to Surgery
No 48 (10.6%) 25 (7.6%) 23 (18.3%)
Yes 405 (89.4%) 302 (92.4%) 103 (81.7%)
Treatment Type
Neoadjuvant Therapy 98 (24.2%) 81 (26.8%) 17 (16.5%)
Adjuvant Chemotherapy 199 (49.1%) 156 (51.7%) 43 (41.7%)
Radiotherapy 177 (43.7%) 145 (48%) 32 (31.1%)
Endocrine Therapy 304 (75.1%) 221 (73.2%) 83 (80.6%)
Treatment Combinations
Neoadjuvant Therapy Only 7 (1.7%) 5 (1.7%) 2 (1.9%)
Chemotherapy Only 33 (8.1%) 23 (7.6%) 10 (9.7%)
Radiotherapy Only 1 (0.2%) 1 (0.3%) 0 (0.0%)
Endocrine Therapy Only 109 (26.9%) 68 (22.5%) 41 (39.8%)
Neoadjuvant + Chemo 2 (0.5%) 2 (0.7%) 0 (0.0%)
Neoadjuvant + Radiation 30 (7.4%) 27 (8.9%) 3 (2.9%)
Neoadjuvant + Endocrine 8 (2%) 7 (2.3%) 1 (1%)
Chemo + Radiation 23 (5.7%) 19 (6.3%) 4 (3.9%)
Chemo + Endocrine 65 (16%) 49 (16.2%) 16 (15.5%)
Radiation + Endocrine 12 (3%) 8 (2.6%) 4 (3.9%)
Neoadjuvant + Chemo + Radiation 5 (1.2%) 4 (1.3%) 1 (1%)
Neoadjuvant + Chemo +Endocrine 4 (1%) 3 (1%) 1 (1%)
Neoadjuvant + Radiation +Endocrine 39 (9.6%) 30 (9.9%) 9 (8.7%)
Chemo + Radiation +Endocrine 64 (15.8%) 53 (17.5%) 11 (10.7%)
All 3 (0.7%) 3 (1%) 0 (0.0%)
Abstract (if available)
Abstract
Background/Objective: This study was performed to compare the rates of breast reconstruction between a public safety‐net hospital (LA County Medical Center (LAC)) vs. a academic medical center (Keck Hospital of USC/Norris Cancer Center (Keck/Norris)), with emphasis on immediate breast reconstruction (IBR). We hypothesized that both overall reconstruction rate (the proportion of patients receiving reconstruction among the total patients having mastectomy) and the IBR rate (the proportion of patients receiving IBR among the total patients having reconstruction) were higher at Keck/Norris than LAC. We also aimed to explore the factors that were associated with overall reconstruction rate or IBR rate, and to estimate the time to reconstruction and the fraction of patients who would not undergo delayed breast reconstruction (cure fraction). ❧ Methods: We retrospectively collected all the data of patients with stage 0 – III breast cancer who received mastectomy during Jan 2006 to Dec 2013 from Keck/Norris and LAC. Univariate and multivariate logistic regressions were performed to compare the IBR rate and overall reconstruction rate between the two institutions, as well as to evaluate the correlation between IBR rate or overall reconstruction rate with various factors. Kaplan‐Meier method was used to scrutinize the time to reconstruction, and non‐mixture cure fraction model was performed for estimating the cure fraction. ❧ Results: Of the 453 patients, 327 patients were from LAC and 126 were from Keck/Norris. The odds ratio (OR) of reconstruction rate was 2.2 at Keck/Norris compared to LAC (p < 0.001) without adjustment for patient demographic factors, but was no longer significant with adjustment. The OR of IBR rate was 21.2 compared from Keck/Norris to LAC (p < 0.001) without adjustment for patient demographic factors, and was 22.3 with adjustment (p < 0.001). Based on the p < 0.25 entry criteria, factors that we kept in the multivariate overall reconstruction rate model were insurance status (p < 0.001), race (p < 0.05), age (p < 0.001), BMI (p < 0.05), smoking status (p = 0.08), occupation status (p = 0.18), personal history of other cancer (p = 0.12), progesterone receptor (PR) status (p = 0.22) and radiotherapy (p < 0.001). For IBR rate we had institution (p < 0.001), BMI (p = 0.21), clinical stage (p < 0.01) and radiotherapy (p < 0.01) in the model. The Kaplan‐Meier curves revealed that more patients underwent IBR at Keck/Norris than at LAC. Excluding the IBR patients, the average time to reconstruction at Keck/Norris was 20.3 ± 0.47 (SE) months, and was 37.4 ± 0.91 (SE) months at LAC. The estimated cure fractions were 0.71 and 0.99 at LAC and Keck/Norris, respectively. The factors that we kept in the multivariate non‐mixture cure fraction model were institution (p < 0.01), insurance status (p < 0.05), age = (p < 0.001), smoking status (p = 0.10), occupation status (p = 0.062), clinical stage (p = 0.12), PR status (p < 0.05). ❧ Conclusions: The reconstruction rate did not differ between these two institutions, but the utilization of IBR was higher at Keck/Norris than at LAC. The reconstruction rate was significantly associated with insurance status, age, BMI, occupation status, and radiotherapy, whereas the IBR rate was found to be significantly associated with institution, clinical stage and radiotherapy. When excluding IBR patients, the average time to reconstruction at Keck/Norris was 20.3 months, and was 37.4 months at LAC. Factors including institution, insurance status and PR status were found to be significantly associated with cure fraction.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
An analysis of disease-free survival and overall survival in inflammatory breast cancer
PDF
Eribulin in advanced bladder cancer patients: a phase I/II clinical trial
PDF
Racial and ethnic disparities in delays of surgical treatment for breast cancer
PDF
Pathogenic variants in cancer predisposition genes and risk of non-breast multiple primary cancers in breast cancer patients
PDF
The effect of renal function on toxicity of E7389 (eribulin) among patients with bladder cancer
PDF
Air pollution and breast cancer survival in California teachers: using address histories and individual-level data
PDF
Predictive factors of breast cancer survival: a population-based study
PDF
Racial/ethnic differences in patient experiences with health care in association with the Gleason’s score level at prostate cancer diagnosis: findings from the SEER-CAHPS data
PDF
Arm lymphedema in a multi-ethnic cohort of female breast cancer survivors
PDF
The relationship of resistin, leptin, adiponectin, and ghrelin with bone mineral density in healthy postmenopausal women: longitudinal analysis
PDF
The impact of the COVID-19 pandemic on cancer care delivery
PDF
Comparative study of laparoscopy vs. laparotomy for ovarian mass removal
PDF
Associations between isoflavone soy protein (ISP) supplementation and breast cancer in postmenopausal women in the Women’s Isoflavone Soy Health (WISH) clinical trial
PDF
Instability of heart rate and rating of perceived exertion during high-intensity interval training in breast cancer patients undergoing anthracycline chemotherapy
PDF
Cross-sectional association of blood pressure, antihypertensive medications, MRI volumetric measures and cognitive function scores in an aging population
PDF
Surgical aortic arch intervention at the time of extended ascending aortic replacement is associated with increased mortality
PDF
Machine learning-based breast cancer survival prediction
PDF
Application of random effects models to a clinical retrospective hierarchical database
PDF
The evaluation of the long-term effectiveness of zero/low fluoroscopy workflow in ablation procedures for the treatment of paroxysmal and persistent atrial fibrillation
PDF
Survival of children and adolescents with low-risk non-rhabdomyosarcoma soft tissue sarcomas (NRSTS) treated with surgery only: an analysis of 234 patients from the Children’s Oncology Group stud...
Asset Metadata
Creator
Liu, Chia-Wei
(author)
Core Title
Factors influencing the decision and timing to undergo breast reconstruction after mastectomy for breast cancer in public hospital vs. private medical center from 2007 to 2013: a retrospective co...
School
Keck School of Medicine
Degree
Master of Science
Degree Program
Biostatistics
Publication Date
04/20/2015
Defense Date
04/20/2015
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
breast cancer,breast reconstruction,Mastectomy,OAI-PMH Harvest
Format
application/pdf
(imt)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Sposto, Richard (
committee chair
), Groshen, Susan L. (
committee member
), Mack, Wendy Jean (
committee member
)
Creator Email
eternalleaver@gmail.com,liuchiaw@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c3-553067
Unique identifier
UC11297851
Identifier
etd-LiuChiaWei-3335.pdf (filename),usctheses-c3-553067 (legacy record id)
Legacy Identifier
etd-LiuChiaWei-3335.pdf
Dmrecord
553067
Document Type
Thesis
Format
application/pdf (imt)
Rights
Liu, Chia-Wei
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
breast cancer
breast reconstruction