University of Southern California Dissertations and Theses

Attrition in a longitudinal drug use prevention study

(USC Thesis Other)

Attrition in a longitudinal drug use prevention study

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content ATTRITION IN A LONGITUDINAL DRUG USE PREVENTION STUDY By Zhihong Fan A Thesis Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree MASTER OF SCIENCE (Applied Biometry and Epidemiology) December, 1999 ©1999 Zhihong fan Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UMI Number: 1417211 INFORMATION TO USERS The quality of this reproduction is dependent upon the quality of the copy submitted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleed-through, substandard margins, and improper alignment can adversely affect reproduction. In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted. Also, if unauthorized copyright material had to be removed, a note will indicate the deletion. ® UMI UMI Microform 1417211 Copyright 2004 by ProQuest Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code. ProQuest Information and Learning Company 300 North Zeeb Road P.O. Box 1346 Ann Arbor, Ml 48106-1346 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UNIVERSITY OF SOUTHERN CALIFORNIA THE GRADUATE SCHOOL. UNIVERSITY PARK LOS ANGELES. CALIFO RNIA 9 0 0 0 7 This thesis, written by .?h±honjg_Fan__..... .................... under the direction of hEE. Thesis Committee, and approved by all its members, has been pre sented to and accepted by the Dean of The Graduate School, in partial fulfillment of the requirements for the degree of Master of Science in Applied Biometry/Epidemiology Dsm D / i t f November 30, 1999 THESM COMM! Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Acknowledgments This research was funded by grants from National Institute on Drug Abuse to Dr. Mary Ann Pentz (DA03468, DA03976). I wish to express my gratitude to Dr. Mary Ann Pentz. As the principal investigator of the Midwestern Prevention Project, she provided me excellent data set for me to work on. Special thanks are extended to Dr. James Dwyer who rendered enormous statistical assistance. I also thank Dr. Stanley Azen, and Dr. Kiros Berhane for their helpful comments and suggestions in the formulation of this thesis. I would also like to acknowledge the efforts of all the staffs for the Midwestern Prevention Project (MPP) who made this research possible. Portions of this thesis were presented at Norris Cancer Center, grand round poster session, July 20th, 1999. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table of Contents Acknowledgements.................................................................................................................. ii Table of Contents......................................................................................................................iii List of Tables........................................................................ iv Abstract.......................................................................................................................................v Introduction.................................................................................................................................1 Methods.....................................................................................................................................5 Study Overview.................................................................................................................5 Participants.........................................................................................................................5 Measures........................................................................................................................... 6 Statistical Analyses...........................................................................................................7 Results .................................................................................................................................... 10 Discussion.................................................................................................................................15 Conclusion............................................................................................................................... 19 References ...............................................................................................................................20 iii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. List of Tables Table 1 The number of subjects in tracking sample present in different waves.............. 6 Table 2 Summary of research questions and modeling approaches ...............................9 Table 3 Logistic regression using drug use at the last follow-up as response variables 10 Table 4.1 Estimates of effects of gateway drug use at 1991 and intervention conditions on retention during 5-year follow-up.................................................................. 12 Table 4.2 Estimates of effects of gateway drug use at 1987 and intervention conditions on retention during 5-year follow-up.................................................................12 Table 5 Demographic effects on retention.......................................................................13 Table 6......Interactions of cigarette use by condition, school type by condition ............. 13 Table 7 Comparison of different models on rank correlation for assessing predictive ability.....................................................................................................................14 iv Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Abstract Attrition impacts the effectiveness of preventive interventions as well as the external validity of efficacy analyses. This study investigated the predictors of attrition/retention during a longitudinal prevention study, especially drug use status at baseline and demographic variables; examined whether program outcomes (drug use status at last follow-up evaluation) were related to retention/attrition during follow-up, and whether there was differential attrition between the intervention conditions. The results indicated that seventh graders (compared to sixth graders), non-whites, males, low SES participants were more likely to be missing during follow-up (OR<l,y»<0.05). Cigarette users and marijuana users were more likely to be missing than nonusers (0R<1, p<0.05). There was differential attrition among intervention conditions by school type and by cigarette use status. The results of this study underscore the need for routine analyses of the effects of attrition on the validity of evaluations of drug use prevention programs. Keywords: Attrition; dropout; longitudinal study; prevention; adolescents; substance abuse. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Introduction Drug use prevention intervention projects are usually longitudinal designs, as the assessment of the long-term impact of programmatic treatment is better understood when examined over relatively long time periods (e.g., Botvin et al., 1980; Evans et al., 1981; Flay et al., 1983, 1989; Schaps, 1986; Snow, 1992; Pentz, 1994). Participant attrition is one of the major problems in longitudinal studies. It is important to understand whether bias is associated with those who do not complete the study, and whether an appropriate strategy should be taken to reduce attrition bias in evaluating program effects. Although large-scale prevention research usually use well-thought-out experimental designs to control internal and external validity, attrition is not controlled by the experimental design (Hansen et al., 1985). Therefore, statistical analyses of attrition biases have implications for the generalizability of evaluation results, as well as the design of preventive interventions. There are basically two types of attrition, intervention attrition and research attrition (Lauby et al., 1996). An examination of the characteristics of those who do and do not complete the intervention is needed to assess whether those most at risk were able to take full advantage of the program. In addition to intervention completion, differential attrition with regard to research data collection needs to be examined in any program evaluation to assess whether the measurement of outcome effects have been distorted by selection biases. The implication of these two types of attrition is different. Information about attrition from the intervention may be useful to those wishing to replicate the intervention, while the information about the attrition from research data collection is of Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. particular interest to those who intend to collect long-term follow-up data. In this study, we examined the effects of attrition with regard to research data collection. Attrition is a large source of threat to internal and external validity of a prevention intervention study. Many studies have been conducted to explore the mechanism of attrition (Little, 1995; Deluchi & Bostrom, 1999) and effects of dropouts (Biglan et al., 1985; Fitz et al., 1989; Lesaffre et al., 1996; Siddiqui et al., 1996). Internal validity is threatened by differential attrition among intervention conditions. Differential rates of attrition among conditions suggested a possible attrition artifact which will interfere with interpretation of outcome results. For example, if drug users are more likely to drop out of an intervention group at a disproportionately high rate, program effects may be spuriously significant. In contrast, if drug users drop out of control groups at high rates, the true effectiveness of a program might be masked by creating an outcome in which intervention and control groups are ostensibly not different from each other or even obtaining a negative program effect (Hansen et al., 1985). External validity is threatened to the extent that the retainers throughout an evaluation are different from dropouts. Most prevention studies only reported attrition rates and failed to report the impact attrition had on the study. Thus, the generalizability of these studies may be limited. Several studies evaluated attrition in drug use prevention programs and found that heavy smokers were more likely to drop out of the treatment group than out of the control group (Josephson & Rosen, 1978). Study dropouts also reported greater alcohol and marijuana use than the adolescents who remained (Severson & Ary, 1983). Thus, the researchers found that the efficacy of the prevention program was evaluated with a 2 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. sample that under-represented heavy smokers, alcohol and marijuana users at follow-up, creating the impression that the program was successful in reducing smoking rates when, in fact, it may not have been. Similarly, Hansen et al. (1985) analyzed two longitudinal studies of junior high and high school students and found that attrition had adverse effects on external validity, as students who used more alcohol and cigarettes were more likely to drop out of both studies. Examination of the characteristics of the dropouts will allow researchers to adopt proper strategies to adjust for the bias due to dropouts, to minimize the dropout rate in future prevention studies. In this paper we utilize data from an ongoing longitudinal drug abuse prevention study, the Midwestern Prevention Project (MPP) (Pentz et al., 1989). The specific aims of this paper are: (1) to examine if program outcomes (drug use status at last follow-up evaluation) are related to retention/attrition during follow-up; (2) to investigate the predictors of attrition, especially drug use status at baseline as well as demographic factors; and (3) to examine whether there was differential attrition in intervention conditions. The first two aims address issues related to external validity. The third addresses issues were related to internal validity (Hansen et al., 1985). The study hypotheses are: 1. The participants who completed less numbers of measurements are more likely to be drug-users at the last follow-up evaluation; 2. Drug users at baseline are less likely to be retained during longitudinal follow-up. 3 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3. Demographic characteristics such as school type, grade, Socio-Economic Status (SES), ethnicity and gender are significant predictors of attrition/retention during longitudinal follow-up. 4. There is differential attrition among intervention conditions (program vs. control). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Methods Study Overview The Midwestern Prevention Project (MPP) is a longitudinal controlled intervention trial for community-based drug abuse prevention in adolescents that was conducted in Indianapolis beginning in 1987. The major objectives were to reduce prevalence rates of use of gateway drugs - tobacco, alcohol, and marijuana - in early adolescence and to reduce use of other illicit substances in later years. The intervention was conducted in grade six or seven. The research design was a 2x2 factorial design, varying intervention condition (drug prevention program vs. standard health education) and grade of intervention (sixth or seventh grade). The unit of assignment for the initial intervention was school (middle or junior high). In 1991, a tracking sample was randomly selected and followed up to five times (waves) once every 12 -16 months through 1998. Participants A total of 5,378 students from 57 middle or junior high schools were enrolled in the study. At baseline, 51% were male, 78% were white, with a mean age of 13 years (range 12-14 years), 21% of the sample were in sixth grade (middle schools), 79% in seventh grade (junior high schools). The tracking sample consisted of a total of 1,201 participants. In the tracking sample, 684 (57%) of the tracking sample participated in all five waves, and 517 (43%) were not surveyed at one or more of these waves. Table 1 shows the number of students present at each tracking wave and the distribution of numbers of completed waves. 5 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 1. The number of participants present in different waves during follow-up (total N=1201) 91-92 92-93 93-94 94-95 96-98 Retainers 1116 1052 1024 836 885 Retention rate(%)a 92.9 87.6 85.3 69.6 73.7 The total number of waves completed by each subject 0 1 2 3 4 5 Frequency 48 27 83 136 223 684 % of total samplea 4.0 2.3 6.9 11.3 18.6 57.0 Note. a The percentages were calculated using N=1201 as the denominator. Measures Measure o f retention Retention status was defined as the number of waves (out of 5) completed by the participants during the period 1991 through 1998. Some participants completed all five waves, some missed one (or more than one) survey with various patterns. The participants were not simply categorized as completers or dropouts. Thus the measure of retention is an ordinal scale rather than a dummy variable. Predictors/correlates o f retention The variables examined as predictors/correlates of retention included intervention conditions (program versus control group), demographic variables, and drug use status at baseline 1987, at tracking baseline 1991, and at the last follow-up evaluation. Demographic variables included school type (public vs. private), grade (6th or 7th at baseline), race (non-white vs. white), SES (father’s nonprofessional or professional occupation, low vs. high), and gender (male vs. female). Drug use variables included monthly, weekly and daily cigarette use, monthly and weekly alcohol use, monthly drunkenness, monthly and weekly marijuana use. The 7- 6 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. point ordinal scales for the items were dichotomized as 0=no use and l=use. Heavy marijuana use was defined as use of marijuana two or more times last week. Statistical Analyses Logistic regression was conducted using drug use at last follow-up evaluation as a dichotomous response variables and the number of measurements completed (0-5) during follow-up as the explanatory variable, controlling for demographics (e.g., school type, grade, race, SES and gender) and drug use at baseline 1987. We assumed parallelism of effects of the ordinal explanatory variable (Dwyer et al., 1989). Cumulative ordinal logistic regression (Armstrong & Sloan, 1989) was used to model the logit of the number of measurements completed (0-5) as a function of demographic factors, group assignment, and drug use at baseline (1987) and at the beginning of tracking (1991). The ordinal logistic regression model is given by: Logit(yj) = In( /. 1(1- Yj)) = #o + A , X i + A>^2 + ---+ A A ’ where y) ■ = tii+. . ,+nj are cumulative probabilities of being in one of the first j levels. In (•) is the natural logarithmic function, and 9 0 j and Pij, p 2 j, •••, Ppj are unknown parameters, and W, x it • •• Xv are covariates. The cumulative ordinal logit model is then specified by the constraints P n = p !2= ... = P i, and p 2i= p 22= ... = p 2. The intercepts, 0 O j, however, will differ. The parameters of the model were estimated using maximum likelihood methods. Differential effects of baseline drug use status or other covariates within intervention conditions were investigated by allowing interactions between baseline drug use status or demographic covariates and intervention conditions in ordinal logistic regression models. 7 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Four indices of rank correlation (c, Somers’ D, Goodman-Kruskal Gamma, Kendall’s Tau-a) are used to assess the predictability of the logistic models. Let t denotes pairs of different responses, which are categorized as concordant (nc), discordant pairs ( ;nd) and tied (t -nc - nd). N is the total number of observations in the data set. c = (nc + 0.5(t - nc - nd)) / 1 Somers' D = (nc = nd) / 1 Goodman - Kruskal Gamma = (nc - nd) / (nc + nd) Kendall's Tau - a = (nc - nd) / (0.5N (N - 1)) The alteration of the predictive ability across a series of models could be observed by the change of those indices. Table 2 shows the analyses conducted in this study. Research question 1, 2 and 3 corresponded to hypothesis 1, 2 and 3. For Analysis 2, two baselines were considered, i.e., baseline at 1987 and tracking baseline at 1991. Hypothesis 4 was tested throughout the previous three sets of analyses and significant effects were examined further under each intervention condition. All analyses were conducted using SAS Version 6.12 (SAS institute, 1990). Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 2. Summary of research questions and modeling approaches Research Questions Logistic Models 1. Does retention/attrition predict drug use? 2. Are drug users at baseline or at the time of tracking less likely to be retained? 3. Are there demographic influences on retention/ attrition? Drug use at last follow-up =COMPLETE+drug use (at 1987)+Group+Demo+wave of last follow-up COMPLETE=Drug use (at 1987)+ Group + DrugxGroup+Demo+DemoxGroup COMPLETE=Drug use (at 1992)+ Group + DrugxGroup+Demo+DemoxGroup COMPLETE=Oemo+Group+DemoxGroup 4. Are there interactions of drug use by condition, school type by condition on retention/ attrition? Test by intervention condition: control/program COMPLETE=Drug Use+Demo COMPLETE=School type+ other Demo 5. Are predictive ability of these different models comparable assessed by rank correlations in SAS? 1). COMPLETE=Demos 2). COMPLETE=Demos + Group 3). COMPLETE^ Demos + DemosxGroup 4). COMPLETE=Demo + Demox Group + Drug +Drugx Group Note. COMPLETE =number of waves completed (1-5); Group= intervention condition (program vs. control); Drug use= use cigarette/alcohol/marijuana (yes or no); Demos = demographic variables, including school type (public vs. private), grade (grade 7 vs. grade 6 at baseline), race (white vs. non-white), SES (high vs. low), and gender (male vs. female). 9 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. RESULTS The analysis was first done to determine whether dropouts were more likely to be drug users (Research Question 1). The drug use outcomes tested included gateway drugs such as cigarette, alcohol, and marijuana. The maximum likelihood estimates of covariates-adjusted odds ratios (ORs) and 95% CIs are presented in Table 3. Use of cigarettes was associated with lower retention (all ORs<l,/K0.01). Interestingly, heavy drinking, as measured by past month drunkenness, was associated with higher retention [OR=1.27, 95%CI=(1.02, 1.61),/?<0.05]. Response variables OR (95%CI) Cigarette use in the past month (yes/no) 0.75(0.60,0.93)** Cigarette use in the last week (yes/no) 0.70(0.56,0.86)** Cigarette use in the past 24 hours (yes/no) 0.63(0.51,0.78)** Alcohol use in the past month (yes/no) 1.04(0.83,1.29) Alcohol use in the past week (yes/no) 1.05(0.85,1.32) Being drunk in the past month (yes/no) 1.27(1.02,1.61)* Marijuana use in the past month (yes/no) 1.01(0.80,1.30) Marijuana use in the last week (yes/no) 0.96(0.74,1.27) Heavy marijuana use(use twice or more in the last week) 0.95(0.72,1.29) (yes/no) Notes. * p<0.05 **_p<0.01 OR=Odds Ratio; CI= Confidence Interval a The predictors were numbers of completed waves and other covariates. The covariates included intervention condition, school type, grade, race, SES, gender and which wave the drug use outcome was obtained. Table 4.1 summarized the analyses as to whether drug users at the time of tracking are more likely to drop out (Research Question 2). After adj ustment for the covariates and interactions, a significant effect on retention due to cigarette use and marijuana use was found (ORs<l,/?<0.05). That is, cigarette users and marijuana users were less likely to be retained during follow-up. Using 1987 data as the baseline revealed 10 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. similar trends, although the effect of marijuana use failed to reach significance due to small numbers of marijuana users at baseline 1987 (Table 4.2). In addition, there was significant interaction of group by weekly cigarette use at baseline on retention [OR=2.39, 95%CI (1.00, 5.80),/?<0.05], This interaction was examined further in Table 6. There were no significant associations between baseline alcohol use and retention. To examine the sole contribution of demographic variables in predicting retention/attrition (Research Question 3), logit models were constructed incorporating intervention condition, demographics and condition by demographics interaction (Table 5). Participants from higher grade (Grade 7 at baseline), lower SES, non-whites, and males were more likely to drop out during follow-up (p<0.10). Table 4.1, 4.2, & 5 also indicated that there was no differential attrition among intervention conditions evaluated by main effect analysis. However, Table 6 shows somewhat significant condition by school type interaction [OR=0.55, 95%CI (0.30, 1.01),y><0.10]. There were no evidence of interactions between condition and other demographics. The interactions of cigarette use by condition, school type by condition on retention were further investigated by testing the effect of cigarette use and school type in control and intervention group separately (Table 6). The results indicated a significant condition x school type, condition x cigarette use interaction. In intervention group, the participants from public schools were more likely to drop out during follow-up, relative to those in private schools [OR=0.44, 95% Cl (0.29, 0.67),/K0.01], In control group, the cigarette users were less likely to be retained (OR<l,/?<0.10). 11 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 4.1. Estimates of effects of gateway drug use at 1991 and intervention conditions on retention during 5-year follow-upa Drag use at 1992 Conditional OR (95%CI) Drug Group Drag x Group Cigarette use in the past month (yes/no) Cigarette use in the last week (yes/no) Cigarette use in the past 24 hours (yes/no) Alcohol use in the past month (yes/no) Alcohol use in the past week (yes/no) Being drunk in the past month (yes/no) Marijuana use in the past month (yes/no) Marijuana use in the last week (yes/no) Heavy marijuana use(use twice or more in the last week) (yes/no) 0.56(0.38,0.83)* 0.47(0.32,0.71)** 0.43(0.29,0.67)** 0.75(0.51,1-10) 0.76(0.45,1.30) 0.79(0.50,1.28) 0.53(0.33,0.85)** 0.48(0.27,0.89)* 0.75(0.36,1.65) 1.07(0.33,3.52) 1.01(0.31,3.34) 1.00(0.31,3.30) 0.99(0.30,3.23) 1.03(0.32,3.33) 0.56(0.15,2.15) 1.11(0.35,3.59) 1.09(0.34,3.52) 1.03(0.32,3.32) 0.97(0.55,1.69) 1.03(0.58,1.85) 0.94(0.51,1.74) 0.77(0.45,1.35) 1.14(0.53,2.44) 1.03(0.52,2.05) 1.20(0.61,2.40) 1.17(0.48,2.89) 0.98(0.33,2.89) Notes. * p<0.05 **/><0.01 OR-Odds Ratio; CI= Confidence Interval aThe model was defined as COMPLETE=Drug+Group+DmgxGroup+ Demo+ Demox Group. COMPLETE =number of measurements completed (1-5); Group= intervention condition (program vs. control); Drug= use cigarette/alcohol/marijuana (yes or no); Demo represents school type, grade, race, SES, and gender. Table 4.2. Estimates of effects of gateway drug use at 1987 and intervention conditions on retention during 5-year follow-upa Drug use predictors Conditional OR (95%CI) Drug Group DrugxGroup Cigarette use in the past month (yes/no) Cigarette use in the last week (yes/no) Cigarette use in the past 24 hours (yes/no) Alcohol use in the past month (yes/no) Alcohol use in the past week (yes/no) Being drunk in the past month (yes/no) Marijuana use in the past month (yes/no) Marijuana use in the last week (yes/no) Heavy marijuana use(use twice or more in the last week) (yes/no) 0.44(0.28,0.70)** 0.33(0.19,0.59)** 0.38(0.14,1.09)+ 0.55(0.23,1.35) 1.46(0.38,7.02) 1.08(0.41,3.06) 0.59(0.20,1.85) 0.65(0.14,3.43) 0.38(0.06,3.10) 0.98(0.34,2.81) 1.01(0.35,2.92) 0.99(0.34,2.85) 0.93(0.32,2.68) 1.00(0.35,2.89) 0.95(0.33,2.74) 0.96(0.33,2.78) 1.00(0.35,2.90) 1.02(0.35,2.95) 1.63(0.84,3.20) 2.39(1.00,5.80)* 3.66(0.6,8.23) 1.82(0.44,7.99) 0.40(0.03,5.12) 0.69(0.16,3.04) 0.31(0.03,3.02) 1.12(0.05,36.37) b Notes. + /><0.10 * /?<0.05 **/><0.01 OR=Odds Ratio; CI= Confidence Interval a The model was defined as COMPLETE=Drug+Group+DrugxGroup+ Demo+DemoxGroup. COMPLETE =number of measurements completed (1-5); Group= intervention condition (program vs. control); Drug= use cigarette/alcohol/marijuana (yes or no); Demo represents school type, grade, race, SES, and gender. b Due to small number of heavy marijuana users at baseline, the maximum likelihood estimate could not be calculated for this term. 12 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 5. Demographic effects on retention Predictors/interactions Conditional OR (95% Cl) Group (program=l, control=0) 0.91 (0.32, 2.61) School type(public=T, private=0) 0.79 (0.51, 1.21) Grade (grade7=l, grade6=0) 0.54 (0.33, 0.85)* Race (white=l, non-white=0) 2.36 (1.63, 3.40)** SES (high=l, low=0) 1.40 (0.98, 2.01)+ Gender (male=l, female=0) 0.50 (0.36, 0.69)** School typex Group 0.55 (0.30, 1.01)+ GradexGroup 1.48 (0.76, 2.89) Racex Group 1.06 (0.61, 1.82) SESxGroup 0.89 (0.53, 1.49) Gender x Group 1.45 (0.91, 2.32) Notes. + jKO.10 *p<0.05 ** p<0.01 OR=Odds Ratio; CI= Confidence Interval Table 6. Interactions of cigarette use by condition, school type by condition Control variables Models Conditional OR (95% Cl) Interactions of cigarette use by condition Control (n=610) Complete=monthly cigarette use+ Dem oa Complete=weekly cigarette use+ Demo Complete=daily cigarette use+ Demo 0.44 (0.28, 0.69)** 0.32(0.18, 0.58)** 0.38(0.18, 0.58)+ Control (n=591) Complete=monthly cigarette use+ Demo Complete=weekly cigarette use+ Demo Complete=daily cigarette use+ Demo 0.72 (0.44, 1.19) 0.79 (0.41, 1.56) 1.39 (0.37, 6.66) Interaction of school type by condition Control (n=610) COMPLETE=School type, other Demo 0.79(0.51,1.20) Program (n=591) COMPLETE=School type, other Demo 0.44 (0.29, 0.67); Notes. + p<0.10 */><0.05 ** p<0.0\ OR=Odds Ratio; CI= Confidence Interval a Demo include school type, grade, race, SES and gender. The last task of this study is to compare the predictability of different models (Research Question 4). Table 7 shows the rank correlations from ordinal logistic regressions using various combinations of predictors. There were no dramatic di fferences of predictive ability assessed by four indices of rank correlations provided by logistic procedure among four models. From the point of efficiency of the model, this implied that adding group assignment and baseline drug use status did not gain much additional predictability of retention in this study. 13 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 7. Comparison of different models on rank correlations for assessing predictive ability Models Indices of rank correlation8 Somer’s D Gamma Tau-a C 1. Demob only 0.27 0.29 0.16 0.63 2. Demo + Group 0.27 0.28 0.16 0.63 3. Demo+ DemoxGroup 0.29 0.30 0.17 0.64 4. Demo + DemoxGroup + 0.29-0.34 0.30-0.34 0.17- 0.18 0.64- 0.67 Drug0 +Drugx Group 8 Logistic procedure computes the above four indices of rank correlation for assessing the predictive ability of a model. See SAS/STAT User’s guide Vol. 2 PP1091 for formulas. Demo represents school type, grade, race, SES, gender. c Drug included monthly, weekly or daily use of cigarette, alcohol and marijuana, entered one drug use variable at a time. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Discussion This analysis demonstrates the usefulness of the ordinal logistic regression procedure in analyzing the factors influencing attrition in longitudinal school-based drug use prevention studies. The dropout process varied from individual to individual in this study. Participants may miss one (or more than one) survey and return later during follow-up. Because it is desirable to obtain as many as possible measures for each subject, the number of waves completed was used as the ordinal scale response variable in the ordinal logistic model (Armstrong & Sloan, 1989). An alternative approach would have been to use proportional hazards model. However, because it was difficult to define time-to-dropout, this approach was not appropriate for this study. According to Little (1995), the dropout mechanism in this longitudinal study could be categorized as both covariate-dependent and outcome-dependent. The results indicated that seventh graders (versus sixth graders), non-whites, males, low SES participants were all significant attrition predictors. Cigarette users and marijuana users were likely to show higher levels of attrition than nonusers. The effects of attrition on internal and external validity of the study are discussed as follows. Internal validity Attrition rates in intervention conditions did not differ significantly across time in this study, evaluated by main effect analysis. However, the interaction analysis demonstrated that public school participants in the intervention group were less likely to be retained during follow-up (refer to Table 6). This result indicates that it might be more appropriate to evaluate the program effect by school type. 15 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The other important interaction was cigarette use by condition. While cigarette users tended to drop out, the cigarette users in the control group were especially more likely to do so. In other words, the intervention group was more likely to retain cigarette users than the control group. This is inconsistent with some studies which suggested that heavy smokers were more likely to drop out of the intervention group than out of the control group (Josephson & Rosen, 1978). Differential rates of attrition of cigarette users among conditions may mask true program effectiveness and cause spurious decay of program effects over time. It is not certain why the school type by condition interaction occurred. A potential explanation is that the implementation of the program in private school was better than that in public school. One indicator of the success of a program is the proportion of individuals recruited into the intervention who accomplish the entire requirement (Pentz et al., 1990; Lauby, et al., 1996). Yet we focused on research attrition instead of implementation attrition in this study. Whether better quality of implementation resulted in high retention in private school intervention group during follow-up is questionable. External validity Results from these analyses show that the generalizability of the study is limited largely to drug nonusers, high SES participants, whites, and females. Why are drug users more likely to drop out of studies than non-users? Possible explanations are: 1) drug users might move more frequently and have generally less stable family lives than nonusers, therefore making it difficult to find and measure them; 2) drug use may cause acute medical problems which may be reflected in more absenteeism; and 3) users may view answering questions about their behavior to be 16 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. aversive. Possible efforts to eliminate the problems include devising less aversive data collection techniques, making second efforts to catch missing participants due to absenteeism (Hansen et al., 1985). The depletion of users in tracking cohort leads to less power to detect the program effects on cessation or reduction (Hansen et al., 1985; Resnicow & Botvin, 1993; Delucchi & Bostrom, 1999). The drug use prevalence evaluated with the tracking data revealed that the drug use increased sharply from 88 (grade 6 or 7) to 94, but remained stable or fluctuated slightly during 94 through 98 (the data are available by request). Although one could speculate that the drug use prevalence approached plateau and started to decline to some extent at about age 20, this might be spurious due to the overrepresentation of drug users in dropouts. Moreover, it was found that the discrepancy of drug use outcome between intervention conditions became blurred across time. Inadequate statistical power due to gradual loss of drug-users during follow-up might contribute to the appearance of a decay of prevention effects (Resnicow & Botvin, 1993). When examining whether dropouts show different patterns across time on the drug use outcome variables than retainers, it was found that heavy drinking, as measured by monthly drunkenness at the last follow-up, was associated with higher retention. This is contradictory with other studies. For example, Severson & Ary’s (1983) study demonstrated that dropouts reported greater alcohol use than the adolescents who remained. However, when alcohol use at tracking baseline was used as predictors of retention, the trend was consistent with other studies, albert it did not reach statistical significance. 17 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Knowledge about the characteristics of the dropouts will allow us adopt certain strategies to reduce attrition bias in evaluating program effects. Leigh et al.(1993) proposed a procedure to adjust for attrition bias in longitudinal studies. Firstly, a probit or logistic regression is fitted to predict the probabilities that participants retain or dropout during the follow-up from the baseline social and demographic information. Then, an instrumental variable which is formed from the probability is included as an additional covariate in subsequent analyses to adjust for the attrition bias. Siddiqui et al. (1996)’s proportional hazard regression analysis demonstrated that the probability of attrition were not only a function of time-invariant variables (such as gender and race) but also a function of treatment condition and immediate consequences of the program. This study was generally consistent with Siddiqui et al’s findings. While baseline demographics contributed most to the prediction of the probability of retention/attrition, drug use status (either baseline or outcome) and intervention condition were also meaningful correlates of retention/attrition. The findings of this study have important implications for program planners and evaluators. First, it implicated that the drug abuse prevention program effects may be underestimated since drug users were less likely to be retained, and in case of cigarette users, more were missing from the control group than the intervention group. In terms of evaluation efforts, perhaps a relaxed criterion for significance or a weight might be used to correct for underestimated program effects. Second, additional efforts may be required to retain or recapture users in order to expose them to any type of prevention program. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Conclusion This analysis revealed that (1) the participants who completed less numbers of measurements are more likely to be drug-users at the last follow-up evaluation; (2) drug users at baseline are less likely to be retained during longitudinal follow-up; and (3) there was differential attrition among intervention conditions by school type and by cigarette use status. It is essential that future evaluations of drug use prevention programs include assessments of the effects of attrition on the internal and external validity of the studies and that appropriate intervention and statistical weighting strategies are taken accordingly to reduce the attrition bias. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. References Armstrong, B.G., & Sloan, M. (1989). Ordinal regression models for epidemiologic data. American Journal o f Epidemiology, 129(1), 191-203. Biglan, A., Severson, H., Ary, D., Faller, C., Gallison, C., Thompson, R., Glasgow, R., & Lichtenstein, E. (1985). Do smoking prevention programs really work? Attrition and internal and external validity of an evaluation of a refusal skills training program. Journal o f Behavioral Medicine, 10(2), 159-171. Botvin, G., Eng, A., & williams, C.(1980). Preventing the onset of cigarette smoking through life skills training. Preventive Medicine, 9, 135-143. Delucchi, K., & Bostrom, A. (1999). Small sample longitudinal clinical trials with missing data: a comparison of analytic methods. Psychological Methods, 4(2), 158-172. Dwyer, J. H., MacKinnon, D. P., Pentz, M.A., Flay, B.R., Flansen, W.B., Wang, EYI, & Johnson, C.A. (1989). Estimating intervention effects in longitudinal studies. American Journal o f Epidemiology, 130(4), 781-95. Evans, R. I., Rozelle, R. M., Maxwell, S. E., Raines, B. E., Dill, C.A., Guthrie, T.J., Henderson, A.H., & Hill, P.C. (1981). Social modeling films to deter smoking in adolescents: results of a three-year field investigation. J. Appl. Psychol, 66, 399-414. Fitz, D. & Tryon, W. (1989). Attrition and augmentation biases in time series analysis: Evaluation of clinical programs. Evaluation and program Planning, 12, 259- 270. Flay, B.R., D ’Avemas, J.R. Best, J.A. Kersel, M.W., & Ryan, K.B. (1983). Cigarette smoking: Why young people do it and ways of preventing it. In McGrath, P., and Firestone, P. (eds), Pediatric and Adolescent Behavioral Medicine, Springer-Verlag, New York. Flay, B.R., Koepke, D., Thomson, S. J., Santi, S., Best, J. A., & Brown, K. S. (1989). Six-year follow-up of the first Waterloo smoking prevention trial. Am J Public Health;79(10), 1371-6. Hansen, W. B., Collin, L. M., Malotte, C.K., Johnson, C. A., & Fielding, J. E. (1985). Attrition in prevention research. JBehav Med. 8(3): 261-75. 20 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Josephson E., & Rosen M. A. (1978). Panel loss in a high school drug study. In: Kandel D, editor. Longitudinal research on drug use: empirical findings and methodological issues. Hemisphere, Washington D.C. Lauby, J., Kotranski, L., Feighan, K., Collier, K., Semaan, S., & Halbert, J. (1996). Effects of intervention attrition and research attrition on the evaluation of an HIV prevention program. Annual of Drug Issues, 26 (3), 663-677. Leigh, J. P., Ward, M. M., & Fries, J. F. (1993). Reducing attrition bias with an instrumental variable in a regression model: results from a panel of rheumatoid arthritis patients. Stat Me, 12, 1005-18. Lesaffre, E., Molenberghs, G., & Dewulf, L. (1996). Effect of dropouts in a longitudinal study: an application of repeated ordinal model. Statistics in Medicine, 15, 1123- 1141. Little, R. J. (1995). Modeling the drop-out mechanism in repeated-measures studies. Journal o f the American Stististical Association, 90,1112-1121. Pentz, M. A., Dwyer, J. H., MacKinnon, D. P., Flay, B. R. Hansen, W. B., Wang, E. Y., & Johnson C. A. (1989). A multicommunity trial for primary prevention of adolescent drug use. JAMA, 261, 3259-3266. Pentz, M. A., Trebow, E. A., Hansen, W. B., MacKinnon, D. P., Dwyer, J. H., Johnson, C. A., Flay, B. R., Daniels, S., & Corback, C. (1990). Effects of program implementation on adolescent drug use behavior. Evaluation Review, 14(3), 264- 289. Pentz, M. A. (1994). Adaptive evaluation strategies for estimating effects of community- based drug abuse prevention programs. Journal o f Community Psychology, CSAP special issue, 26-51. Resnicow, K., & Botvin, G. (1993). School-based substance use prevention programs: Why do effects decay? Preventive Medicine, 22, 484-490. SAS Institute, Inc. 1990. SAS User’s guide: Statistics, version 6, 4th Ed. Cary, NC, USA. Severson, H. H., & Ary, D.V. (1983). Sampling bias due to consent procedures with adolescents. Addict. Behav, 8, 433-437. Schaps, E., Moskowitz, J. M., Malvin, J. H., & Schaeffer, G. A. (1986). Evaluation of seven school-based prevention programs: A final report on the Napa project. International Journal o f Addictions, 21, 1081-112. 21 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Siddiqui, O., Flay, B. R., Phil, D., & Hu, F. B. (1996). Factors affecting attrition in a longitudinal smoking prevention study. Preventive Medicine, 25, 554-560. Snow, D.L., Tebes, J.K., Arthur M.W., & Tapasak, R.C. (1992). Two-year follow-up of a social-cognitive intervention to prevent substance use. J. Drug Education, 22(2), 101-114. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Abstract (if available)

Abstract Attrition impacts the effectiveness of preventive interventions as well as the external validity of efficacy analyses. This study investigated the predictors of attrition/retention during a longitudinal prevention study, especially drug use status at baseline and demographic variables

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Rates of cognitive decline using logitudinal neuropsychological measures in Alzheimer's disease

PDF

Longitudinal changes in physical activity and physical fitness: associations with blood pressure

PDF

Comparison of predicting accuracy of neural networks for censored survival data using generalized Receiver Operating Charactaristic (ROC)-C-Index method

PDF

A descriptive analysis of medication use by asthmatics in the Children's Health Study, 1993

PDF

Endometrial cancer in Asian migrants to the United States and their descendants

PDF

Cluster analysis of p53 mutational spectra

PDF

Street connectivity and childhood obesity: a longitudinal, multilevel analysis

PDF

Descriptive epidemiology of thyroid cancer in Los Angeles County, 1972-1995

PDF

Development and evaluation of standardized stroke outcome measures in a population of stroke patients in rural China

PDF

Extent, prevalence and progression of coronary calcium in four ethnic groups

PDF

Family history, hormone replacement therapy and breast cancer risk on Hispanic and non-Hispanic women, The New Mexico Women's Health Study

PDF

Association between latchkey status and smoking behavior in middle school children

PDF

Exploring the association of number of cigarettes smoked and confidence to quit smoking in Korean American emerging adults: a multilevel modeling approach

PDF

A pilot survey of medical abortion knowledge and practices among obstetrician/gynecologists and family practitioners in Los Angeles County

PDF

Methodological approaches to assessing diurnal cortisol rythms in epidemiological studies: how many salivary samples are necessary?

PDF

Native American ancestry among Hispanic Whites is associated with higher risk of childhood obesity: a longitudinal analysis of Children’s Health Study data

PDF

Comparisons of metabolic factors among gestational diabetes mellitus probands, siblings and cousins

PDF

Polymorphism of CYP2E1 gene and the risk of lung cancer among African-Americans and Caucasians in Los Angeles County

PDF

An intervention and program evaluation to determine the effectiveness of public health reforms on primary prevention practices by chiropractic interns

PDF

A cohort study of air-pollution and childhood obesity incidence

Asset Metadata

Creator Fan, Zhihong (author)

Core Title Attrition in a longitudinal drug use prevention study

School Graduate School

Degree Master of Science

Degree Program Applied biometry and epidemiology

Publication Date 12/09/2020

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag Adolescents,attrition,dropout,longitudinal study,OAI-PMH Harvest,Prevention,substance abuse

Language English

Contributor Digitized by ProQuest (provenance)

Advisor Azen, Stanley (committee chair), Berhane, Kiros (committee member), Pentz, Mary Ann (committee member)

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c89-406017

Unique identifier UC11666511

Identifier 1417211.pdf (filename),usctheses-c89-406017 (legacy record id)

Legacy Identifier 1417211.pdf

Dmrecord 406017

Document Type Thesis

Rights Fan, Zhihong

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 au...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA