Individual-Level Predictors of Nonparticipation and Dropout in a Life-Skills HIV Prevention Program for Adolescents in Foster Care

Abstract

Purpose

To identify individual-level characteristics of foster care adolescents who are more likely to not participate in, and drop out of, a life-skills HIV prevention program delivered over 8 months.

Methods

Structured interviews were conducted with 320 foster care adolescents (15–18 years). Logistic regression and survival analyses (Cox Proportional Hazards Regression) determined the influence of demographics, HIV sexual risk behaviors, substance use, mental health problems, and other individual-level risk factors on nonparticipation and dropout.

Results

Older age and having vaginal intercourse without a condom were significant predictors of nonparticipation. Older age and marijuana use significantly increased the hazard of dropping out of the program.

Conclusions

Foster care adolescents at increased risk for HIV infection were more likely to never participate in and drop out of the program. To improve initial and ongoing participation, HIV prevention efforts for adolescents in foster care should be tailored to individual-level HIV risk behaviors and incorporate early and ongoing engagement and retention strategies.

Introduction

The effectiveness of HIV prevention efforts targeted to vulnerable populations, such as adolescents in foster care, is often threatened by attrition, because those at greatest risk may not initially participate or drop out of the programs. Attrition is also problematic for community-based organizations, policy makers, evaluators, and researchers because it weakens the internal and external validity of the program evaluation, making it difficult to determine the extent to which the program is effective. For example, attrition can result in systematic demographic, psychological, or behavioral differences between study completers and those who drop out, leading to selection bias in the final sample and limiting the external validity of the study (Kalichman, 1998; Duan, Braslow, & Weisz, 2001; Song et al., 2006). Such participation bias may also lead to an overestimation of intervention impact on subsequent behavior change when the outcome may be attributable to motivations, perceptions, and other characteristics unique to those who complete the program. Validity is even more threatened when participants and dropouts selectively differ in regard to intervention-targeted risk activity levels prior to the study (Siddiqui, Flay, & Hu, 1996). Additionally, internal validity is challenged when there is differential dropout between treatment and comparison groups. Thus, conclusions drawn from outcome evaluations of HIV prevention programs for vulnerable populations should be qualified by carefully examining attrition and dropout rates (Rutledge et al., 2002).

It is important to identify individual-level characteristics that are correlated with nonparticipation in and dropout from HIV prevention, as at least 20% of individuals who enroll in HIV prevention counseling interventions never show up and about 50% of those interested in and eligible to enroll drop out (DiFranceisco et al., 1998). Younger age, gender, race, substance use, and higher levels of sexual risk behavior have been associated with failure to return for results in HIV testing and counseling (Molitor et al., 1999). In a study on the effects of intervention attrition on the evaluation of an HIV prevention program for adult out-of-treatment drug users, Lauby and colleagues (1996) found that participants who lived alone and did not receive welfare benefits were more like to drop out. Those who ever had an STD or reported needle risk behaviors were more likely to complete the intervention. DiFranceisco and colleagues (1998) identified differences between completers and early dropouts (attended no sessions but assessed at baseline) in two small group HIV intervention trials with gay/bisexual men and severely mentally ill adults. Younger age was associated with early dropout in both samples. Predictors among gay/bisexual men included involvement in an exclusive sexual relationship, minority ethnicity, injection drug use, and higher perceived severity of AIDS. Predictors among severely mentally ill included being less knowledgeable about safer sex methods and holding positive expectancies for condom use. No studies, however, have examined the risk factors for nonparticipation and dropout in HIV prevention among adolescents in foster care.

On any given day, as many as 250,000 adolescents can be found residing in the care and custody of the United States foster care system (U.S. Department of Health and Human Services, 2006). Adolescents in foster care have been identified as a vulnerable population at increased risk for HIV infection (Auslander et al., 2002; Auslander et al., 1998). Empirical data over the last several decades indicate that adolescents in foster care are more likely than those in the general population to experience mental health and substance use problems, engage in risky sexual behaviors (Thompson & Auslander, 2007; Thompson & Auslander, 2011), not complete high school, be unemployed, and be dependent on adult public assistance (Courtney & Dworsky, 2004). Many find themselves without health care, involved with the criminal justice system, or parents within two years after exiting care (McDonald et al., 1995; Courtney et al., 2001; McMillen & Tucker, 1999). These adolescents generally lack social and financial support and are poorly prepared for employment and independent living (Courtney & Dworsky, 2005). These factors, unfortunately, may influence their ever attending or staying in an HIV prevention programs as well.

However, the majority of evaluations of HIV prevention programs that have addressed attrition have been conducted among adult populations, most of whom were receiving substance abuse or psychiatric treatment. Additionally, few HIV prevention efforts that target adolescents in foster care have been developed and evaluated. Moreover, no study to date has examined the individual-level factors associated with nonparticipation (i.e., enrolling in but never attending) and dropout (i.e., beginning but not completing) in HIV prevention programs for adolescents in foster care. Adolescents in foster care experience multiple individual-level risk factors that place them at increased risk for HIV and it is hypothesized that some of these same risk factors will also influence their participation and dropout in behavioral change programs to reduce HIV risk behaviors. Therefore, to gain a better understanding of why some foster care adolescents do not initially participate after agreeing to (i.e., nonparticipation) and why some who do begin the program stop attending (i.e., drop out), this study addressed the following research questions: 1) What are the individual-level factors that predict the nonparticipation of foster care adolescents, who are eligible and agree to participate, in a life-skills HIV prevention program delivered weekly over 8 months? and 2) What are the individual-level factors that predict the dropout of foster care adolescents who participate in at least one session of the program?

Methods/Strategies/Intervention Applications

Subjects and Procedures

The present study utilized longitudinal data (baseline and 8-month post-test assessments) collected from a total of 351 adolescents in foster care who were randomized to a life-skills HIV prevention program (experimental condition) or the usual care standard independent living skills program (ILP), both of which were group based. The adolescents were referred to the study by foster or biological parents, residential counselors, or by child welfare workers. No adolescents self-referred to the program, although this was an option. Because the usual care condition (ILP) was a well established federally funded program administrated by the state Children’s Division of the Department of Social Services, adolescents were aware of this program and have been traditionally very eager to attend. At the time of the referral to the study, all adolescents were also aware of the experimental condition, the randomization process, the length of the program, and the differences between the two conditions. Sessions were held in two locations: the usual care ILP program was held in the local Children’s Division office. The experimental expanded program was held in the building affiliated with the research project. Subjects in both conditions received transportation and were delivered in the early evening after school hours. Adolescents were eligible for the study if they were 15 –18 years of age and were able to interact appropriately with others in a group setting. Four adolescents were excluded from the study for severe behavioral problems, as they were seen as being incapable of participating without seriously disrupting the group process. The Institutional Review Board at Washington University in St. Louis approved all procedures. Informed consent was obtained from legal guardians and written assent was provided by adolescents before baseline interviews were completed. Structured baseline and 8-month posttest interviews were conducted with the adolescents by trained MSW students. Each adolescent was paid five dollars for participating in each of the pretest and posttest interviews.

Of the 351 subjects who were participating in the overall study, 60.4% (n = 212) self-identified as African American, and 30.8% (n = 108) self-identified as Caucasian. Additionally, 31 subjects responded to the question regarding race (“what race do you identify as?”) as bi-racial (n=29), American Indian (n=1), or Asian/Pacific Islander (n=1). This “other” category was excluded from this analysis because there were too few subjects in each ethnic category to analyze separately, and we chose to avoid ethnic lumping (Fontes, 1995). In addition, the other category was dropped because they did not significantly differ from the retained cases on age, gender, nonparticipation, dropout or any of the key independent variables in the study. Thus, the final sample for this analysis was comprised of 320 foster care adolescents aged 15 to 18 years (M = 16.3, SD = .84), with over half (53.8%, n = 172) females, and 46.2% (n = 148) males. The racial composition of the sample is consistent with the racial composition of adolescents in the child welfare in this region. Subjects in the experimental and usual care conditions did not differ in terms of demographics (age, gender, race) and study outcomes (nonparticipation, dropout). Thus, the two conditions combined were used to determine predictors of nonparticipation and dropout.

Description of the Interventions

Adolescents randomly assigned to the usual care ILP received 28 weekly group sessions over an 8-month period. The life skills training sessions focused on a variety of topics that prepared adolescents for independent living upon leaving state custody, such as decision making, communication, community resources, housing, job and career planning, money management, health issues, legal issues, and prevention of sexually transmitted infections (including HIV) and pregnancy. Incentives for the usual care subjects included payment for attendance to each session ($5 per session), completing homework assignments ($8 per homework), and for graduation of the program (levels ranged from $50 for a minimum of 3 absences to $500 for perfect attendance and completion of all assignments). Adolescents were paid monthly and there were no restrictions on their use of the money that they earn.

Adolescents randomly assigned to the experimental condition received the same 28 life skills training sessions as the usual care condition, however they also received 4 additional sessions of training in HIV prevention skills from a cognitive-behavioral approach, educational savings accounts, and individual educational planning sessions over 8 months. For this study, only the group sessions were used to examine predictors of participation and dropout. The content for the additional HIV prevention sessions was adapted from programs for troubled adolescents that have previously been shown to be effective for reducing HIV risky behaviors (Barthlow, Horan, & DiClemente, 1995; Rotheram-Borus et al., 1991; St. Lawrence et al., 1995). Additionally, the experimental condition emphasized future life options as a way to enable troubled adolescents to develop an internal sense of meaning and purpose, create a more hopeful orientation towards their future, and the means and motivation to reduce risky behaviors. This was accomplished by adding six individual educational planning sessions, and educational savings accounts. A similar incentive structure as the usual care condition was used for the experimental condition, except the main difference was that adolescent payments were automatically placed in a savings account, and for each dollar that they saved, they were provided with a one dollar match deposited in their accounts at the program graduation and at the follow-up interview. An underlying premise of the program was that adolescents who actively think and plan for their future, and perceive that they have more life options, will be more likely to protect their future and engage in fewer HIV risk behaviors (Auslander et al., 1998).

Variables

Nonparticipation and Dropout

Records of attendance were maintained for each group session by group facilitators and utilized to calculate nonparticipation and dropout. For nonparticipation, adolescents who were eligible and agreed to participate but never attended any group sessions were coded as “1” and considered nonparticipants. Those who attended at least one group session were coded as “0” and considered participants, as the evaluation of this program utilized an intent-to-treat model and examined dosage as a variable in the outcome analysis. For dropout, program attendance records were used to determine the number of sessions that each adolescent attended and at which point each adolescent dropped out. To address the difference in number of sessions between conditions (28 vs. 32), number of sessions completed was recalculated to percentage of sessions completed. These operationalizations of study outcomes are most consistent with, and comparable to, those used in other attrition research (Rutledge et al., 2002).

Demographics

Demographics were assessed at baseline and recoded for analyses. Age was recoded as “0” for 15–16 years (middle adolescence) and “1” for 17–18 years (older adolescence). Race (“what race do you identify as?”) and gender (“what gender do you identify as?”) were dummy-coded as “0” for African American and “1” for Caucasian and “0” for female and “1” for male, respectively.

Substance Use

Selected items from the alcohol and other drug use sections of the Diagnostic Interview Schedule for Children-Revised Version (DISC-R), initially developed by Costello and colleagues (1984), were utilized to measure whether the adolescents used alcohol, marijuana, amphetamines, barbiturates, other tranquilizers, heroin, cocaine, hallucinogens, and inhalants in the 6 months prior to assessment. Inter-rater reliability and validity for the DISC-R have been reported to be high in previous research (Shaffer et al., 1996).

Mental Health Problems

The Youth Self-Report (YSR) of the Child Behavior Checklist (CBCL) was utilized to assess mental health problems (Achenbach, 1991). The YSR is designed to assess the emotional and behavioral problems of adolescents (11–18 years) in a standardized format and is comprised of 118 items concerning the psychological functioning of the adolescent over the preceding 6 month period. The following 8 YSR mental health subscales were calculated: withdrawn; somatic complaints; anxious/depressed; social problems; thought problems; attention problems; delinquent behavior; and aggressive behavior. Subscales were recoded as “0” for not meeting criteria for respective mental health problem and “1” for meeting borderline-clinical cut-off score for mental health problem. Subscales are empirically derived, correspond to well-established clinical symptoms, and demonstrate significant associations with DSM diagnostic categories (Achenbach, 1991; Weinstein, Noam, Grimes, Stone, & Schwab-Stone, 1990). Criterion validity has been established by the ability of the YSR to discriminate between referred and nonreferred youth (partialling for demographic effects) on the basis of quantitative scales (Achenbach, 1991). Additionally, the YSR (specifically, externalizing and internalizing mental health problem scales) was utilized for generalizability purposes, as the majority of studies that have measured mental health problems among adolescents in foster care have utilized these scales (Landsverk, 1999).

Child Maltreatment

Type of child maltreatment, including emotional neglect, emotional abuse, physical neglect, and physical abuse, were measured by the Childhood Trauma Questionnaire (Bernstein & Fink, 1998). To create emotional and physical maltreatment subscale scores, 20 Likert-type items (5 per subscale), scored from never true (“1”) to very often true (“5”), were aggregated and totaled. Based upon recommended cut-off scores, subscale total scores were re-coded as “0” for not having experienced a specific maltreatment and “1” for having experienced the maltreatment and then collapsed into two variables: histories of emotional neglect or abuse; and histories of physical neglect or abuse. Three “yes” and “no” format interview items, adapted from those developed by Russell (1986), were utilized to assess histories of sexual abuse: 1) ever been forced to touch someone else’s private parts against his or her wishes; 2) ever been touched against his or her wishes; or 3) ever experienced vaginal sex, anal sex, or oral sex against his or her wishes. If the adolescent responded “yes” to any one of the three questions, they were coded “1” for having been sexually abused; those who answered “no” to all three questions were coded “0” for having not experienced sexual abuse.

School and Peer Behaviors

Participants were asked a series of “yes” and “no” questions to ascertained if the adolescents had ever run away from home or placements, if they had been suspended or expelled since the seventh grade, and if they had skipped school without permission at least once in the past year. Adolescents also reported if they had engaged in at least one physical fight with another student in the last year, if they had verbally or physically fought with a teacher in the past year, and if they had failed or repeated a grade since seventh grade. Additionally, they were asked if they had friends who drank alcohol at least once a week and if they had friends who used marijuana or other substances.

HIV Risk Behaviors

HIV risk behaviors were assessed by the reported frequency of performing 7 sexual behaviors without a condom in the last 2 months: vaginal intercourse; receptive anal intercourse; oral sex; vaginal, anal, or oral sex while under the influence of alcohol or other drugs; and trading vaginal, anal or oral sex for money, drugs, or shelter. Males were additionally asked their frequency of engaging in insertive anal intercourse. Three items assessed use of unclean needles for IV drug use, for ear or body piercing, and for tattooing. Individual risk items were coded “1” if a behavior was endorsed and “0” if not. A time frame of 2 months was chosen because of evidence that shorter time frames result in increased reliability of data (Kauth, St. Lawrence, & Kelly, 1991). Because many of the risk behaviors were not endorsed by adolescents in the 2-month time frame, for the present analyses, two variables were computed to measure HIV risk behaviors: 1) having engaged in any one of the HIV risk behaviors (“1” for yes, “0” for no); and 2) having engaged in vaginal intercourse without a condom (“1” for yes, “0” for no).

Data Analysis

Descriptive statistics were calculated to determine the extent of sample nonparticipation and dropout, as well as to describe baseline demographics, HIV sexual risk behaviors, substance use, mental health problems, and other risk factors. To examine the independent associations of demographics, HIV sexual risk behaviors, substance use, mental health problems, and other risk factors with nonparticipation, a series of bivariate logistic regression analyses were conducted. Variables significant at the bivariate level were then entered into a final logistic regression model to determine their collective influence on nonparticipation. To examine the differences in average dropout by demographics, HIV sexual risk behaviors, substance use, mental health problems, and other risk factors, multiple independent samples t-tests were performed. Survival analysis using a Cox Proportional Hazards Regression model was then used to assess the magnitude of the effect of factors significant at the bivariate level on the risk of dropping out over the course of the intervention. Cox Proportional Hazards Regression allows for the simultaneous inclusion of risk factors and covariates in the model (Fisher & Lin, 1999) and provides the ability to determine whether the event is associated with these hypothesized predictors, as well as the magnitude of these relationships.

Results

Predictors of Nonparticipation

Findings indicated that 47.5% (n = 152) of the adolescents who completed baseline interviews (and assented to be in the program) never participated in any of the intervention sessions. Table 1 displays rates of nonparticipation according to adolescent demographics, HIV sexual risk behaviors, substance use, mental health problems, and other risk categories, with frequencies presented for sample demographics and those endorsing HIV sexual risk behaviors, substance use, mental health problems (above borderline-clinical cut-off scores), and other risk factors. Chi-square analyses indicated that nonparticipation was significantly associated with being older, engaging in vaginal intercourse without a condom, using alcohol, using marijuana, and running away from home or foster placements.

Table 1.

Rates of nonparticipation by demographics, HIV sexual risk, substance use, mental health problems, and other risk factors (N=320)

Variables	Rates of Nonparticipation
	n	%
Demographics
Race
African American (n=212)	94	44.3
Caucasian (n=108)	58	53.7
Gender
Female (n=172)	83	48.3
Male (n=148)	69	46.6
Age
15–16 (n=187)	78	41.7
17–18 (n=133)	74	55.6*
HIV Sexual Risk Behavior (Yes)
Any Risk Behavior (n=153)	70	45.8
Vaginal Sex without Condom (n=57)	39	68.4**
Substance Use (Yes)
Alcohol (n=126)	69	54.8*
Marijuana (n=114)	63	55.3*
Amphetamine (n=13)	9	69.2
Barbiturates (n=7)	5	71.4
Other Tranquilizers (n=5)	3	60.0
Heroin (n=2)	1	50.0
Cocaine (n=6)	4	66.7
Hallucinogens (n=16)	10	62.5
Inhalants (n=5)	4	80.0
Mental Health Problems (> Borderline-Clinical Cut-off Scores)
Withdrawn (n=31)	16	51.6
Somatic Complaints (n=34)	14	41.2
Anxious/Depressed (n=29)	17	58.6
Social Problems (n=25)	12	48.0
Thought Problems (n=44)	25	56.8
Attention Problems (n=46)	22	47.8
Delinquent Behavior (n=70)	38	54.3
Aggressive Behavior (n=34)	15	44.1
Other Risk Factors (Yes)
Emotionally neglected/abused (n=243)	117	48.1
Physically neglected/abused (n=246)	116	47.2
Sexually abused (n=106)	66	62.3
Ran away from home/placement (n=169)	90	53.3*
Suspended/expelled from school (n=231)	114	49.4
Skipped school (n=150)	78	52.0
Fought with other students (n=90)	47	52.2
Fought with teachers (n=98)	52	53.1
Failed any classes/repeated a grade (n=208)	97	46.6
Friends drink alcohol (n=232)	110	47.4
Friends use marijuana/drugs (n=258)	126	48.8

Significant Chi-Square:

^*p < .05;

^**p < . 01;

^***p < .001

To further examine the independent and simultaneous effects of demographics, HIV sexual risk behaviors, substance use, mental health problems, and other risk factors on nonparticipation, unadjusted and adjusted logistic regression analyses were employed. Findings of the unadjusted (independent) effects of risk factors on nonparticipation indicated that adolescents who reported engaging in vaginal intercourse without a condom were almost three times as likely as those who did not to never participate in the program, with older adolescents, those who used alcohol and marijuana, and those who ran away from home or foster placements being slightly more than one and a half times as likely as their respective counterparts to not participate in the program (see Table 2). After adjusting for the simultaneous effects of significant variables, two factors – being older (17–18 years) and engaging in vaginal intercourse without a condom – remained significant predictors of adolescent nonparticipation. Older adolescents were over one and a half times as likely as younger (15–16 years) adolescents to never participate in the program (Adjusted Odds Ratio [AOR] = 1.68); those who engaged in vaginal intercourse without a condom were almost two and a half times as likely as their counterparts to be nonparticipants (AOR = 2.46; see Table 2).

Table 2.

Unadjusted and adjusted odds ratios (with 95% confidence intervals) for effects of demographics, HIV sexual risk, substance use, mental health problems, and other risk factors on nonparticipation

	Nonparticipation (N = 320) 0=part, 1=nonpart		Nonparticipation (N = 320) 0=part, 1=nonparty
	OR	95% CI	AOR	95% CI
Demographics
Race (AA=0, C=1)	1.46	0.91, 2.32	-----	-----
Gender (F=0, M=1)	0.94	0.60, 1.45	-----	-----
Age (15–16=0, 17–18=1)	1.75*	1.12, 2.75	1.68*	1.06, 2.69
HIV Sexual Risk Behavior			-----	-----
Any Risk Behavior	0.87	0.56, 1.36	-----	-----
Vaginal Sex without Condom	2.84**	1.55, 5.24	2.46**	1.29, 4.71
Substance Use			-----	-----
Alcohol	1.62*	1.03, 2.54	1.33	0.78, 2.28
Marijuana	1.62*	1.02, 2.57	1.09	0.62, 1.90
Amphetamine	2.58	0.78, 8.56	-----	-----
Barbiturates	2.82	0.54, 14.77	-----	-----
Other Tranquilizers	1.67	0.28, 10.14	-----	-----
Heroin	1.11	0.07, 17.84	-----	-----
Cocaine	2.24	0.41, 12.43	-----	-----
Hallucinogens	1.90	0.67, 5.36	-----	-----
Inhalants	4.51	0.50, 40.83	-----	-----
Mental Health Problems			-----	-----
Withdrawn	1.20	0.57, 2.52	-----	-----
Somatic Complaints	0.76	0.37, 1.56	-----	-----
Anxious/Depressed	1.63	0.75, 3.53	-----	-----
Social Problems	1.02	0.45, 2.31	-----	-----
Thought Problems	1.54	0.81, 2.93	-----	-----
Attention Problems	1.03	0.55, 1.93	-----	-----
Delinquent Behavior	1.43	0.84, 2.43	-----	-----
Aggressive Behavior	0.86	0.42, 1.76	-----	-----
Other Risk Factors			-----	-----
Emotionally neglected/abused	1.14	0.68, 1.92	-----	-----
Physically neglected/abused	0.97	0.57, 1.66	-----	-----
Sexually abused	1.65	0.98, 2.78	-----	-----
Ran away from home/placement	1.67*	1.07, 2.61	1.53	0.96, 2.44
Suspended/expelled from school	1.43	0.86, 2.38	-----	-----
Skipped school	1.47	0.94, 2.29	-----	-----
Fought with other students	1.31	0.81, 2.14	-----	-----
Fought with teachers	1.39	0.86, 2.25	-----	-----
Failed any classes/repeated a grade	0.92	0.58, 1.47	-----	-----
Friends drink alcohol	1.06	0.65, 1.75	-----	-----
Friends use marijuana/drugs	1.45	0.81, 2.59	-----	-----

^*p < .05;

^**p < . 01;

Predictors of Dropout

Findings indicated that 52.5% (n = 168) of adolescents who completed baseline interviews and assented to participate in the program attended at least one group session. Of those, 13.7% (n = 23) completed 1–8 sessions, 8.9% (n = 15) completed 9–16 sessions, 16.1% (n = 27) completed 17–24 sessions, and 61.3% (n = 103) completed 25 or more sessions. Bivariate analyses were conducted to identify which risk factors were associated with program dropout. Results indicated that age (t=2.37, df=100.01, p=.020), alcohol use (t=2.06, df=166, p=.041), marijuana use (t = 3.66, df = 79.21, p = .000), delinquent behavior (t = 2.32, df = 40.45, p = .025), having been suspended or expelled (t = 2.62, df = 116.55, p = .010), having friends who use alcohol (t = 2.36, df = 114.08, p = .020), and having friends who use drugs (t = 3.07, df = 77.86, p = .003) were significantly related to dropping out of the program. Therefore, this set of variables was entered into the Cox Proportional Hazards Regression model and were found to be significantly related to the hazard of dropping out (p = .007). Analyses to examine the individual effects of risk factors found two variables, age (p = .040) and marijuana use (p = .021), to be significant predictors of dropping out of the program. The final parsimonious model (see Table 3) was significant (p < .000), with both age (p = .026; B = .597) and marijuana use (p < .000; B = .432) remaining as significant predictors of dropout. As age increased, the hazard of dropping out of the program increased by 40.3%. Marijuana use increased the hazard of dropping out of the program by 56.8%.

Table 3.

Results of Cox proportional hazards regression to predict program dropout

Risk Factor Variables	Model with Covariates (N = 168)		Parsimonious Model (N = 168)
	Hazard Ratio	95% CI	Hazard Ratio	95% CI
Age	.607*	.378, .976	.597*	.379, .939
Alcohol use	.884	.524, 1.490	-----	-----
Marijuana use	.524*	.302, .908	.432***	.274, .682
Delinquent Behavior	.834	.465, 1.497	-----	-----
Suspended/expelled from school	.838	.474, 1.481	-----	-----
Friends drink alcohol	1.082	.577, 2.028	-----	-----
Friends use marijuana/drugs	.716	.333, 1.540	-----	-----

^*p < .05;

^**p < . 01;

^***p < .001

Discussion

Study findings identified older age, using marijuana, and engaging in vaginal sex without a condom as the most significant predictors of attrition for a longitudinal life skills and HIV prevention program for adolescents in foster care. Adolescents who were slightly older (i.e., 17–18 years vs. 15–16 years) or engaging in vaginal intercourse without a condom were most likely to never participate in the program, even after completing the baseline interview and agreeing to participate. Older adolescents (who did initially participate) and those who used marijuana were more likely than their counterparts to drop out of the program. That is, older adolescents, who were soon to be exiting the foster care system and would have likely benefited from a life skills program, were the most likely to never participate. Moreover, even if older adolescents did initially participate, they (as were adolescents who used marijuana) were still more likely to drop out of the prevention program as group sessions progressed. Likewise, adolescents who were engaging in unprotected vaginal sex (the primary HIV risk behavior targeted by the program), and who would have likely benefited from the prevention messages and risk reduction behavioral skills training provided by the program, never attended any sessions.

To better understand the finding that older adolescents were more likely to both never participate and drop out of the program (even with the restricted age range of the sample), one should consider that many of the older adolescents in the study were in the process of transitioning from foster care to independent living and more likely to be actively disengaging from the child welfare system than their younger counterparts. Developmentally, they were also transitioning from adolescence to young adulthood and likely to be seeking autonomy and individuation. Therefore, independent living and HIV prevention programs for foster care adolescents will need to implement active engagement and retention strategies with older adolescents to address their reluctance to initially or consistently participate. Further, such efforts are likely to be better received and more effective if they are developmentally appropriate and reinforcing of individual decision making.

Although marijuana use was not a significant predictor of nonparticipation, it was the strongest predictor of dropping out of the program. Perhaps, adolescents who used marijuana did so as a coping mechanism to avoid acknowledging their risk behaviors or because of their fears about leaving foster care and living independently. Likewise, marijuana use may have impeded their cognitive abilities related to decision-making and planning for their future. Regardless, independent living and HIV prevention programs for adolescents in foster care should target those who use marijuana and implement active engagement and retention strategies with them to address their readiness to change risk behaviors or reluctance to consistently participate. Programs should also incorporate evidence based approaches, such as motivational interviewing techniques (Miller & Rollnick, 2002) and a harm reduction framework (Masterman & Kelly, 2003), to address marijuana use and ambivalence about making behavioral changes, and provide information on the association between foster care, substance use, HIV risk, and housing stability to make the program relevant to their situation.

The finding that adolescents who engaged in vaginal intercourse without a condom (the primary HIV risk behavior targeted by the program) were most likely never participate in the program suggests that these adolescents may not be at individual stages of readiness to either acknowledge or consider changing their risk behaviors. To motivate such adolescents to participate in similar programs, information from group analyses of baseline demographic and risk factors could be used to target prevention information, skill training, and program activities to the common risks and personal characteristics of given participants. Individual baseline data could be utilized to create an idiosyncratic risk profile for each participant, allowing for a personalization of risk which may motivate them to consider participating in the program and/or changing their risk behaviors. Again, the incorporation of motivational interviewing techniques would be appropriate because they are geared towards individuals who are less interested in changing their behavior, do not require a lifetime commitment to change, and are not confrontational nor judgmental (Miller & Rollnick, 2002; Masterman & Kelly, 2003).

Several unmeasured structural factors may further explain the high rates of initial nonparticipation among the adolescents in this study. One possible explanation is the time delay of up to one month that occurred between baseline interviews and the first group session. During this delay, adolescents may have lost interest, experienced scheduling problems due to agency or extracurricular activities, and/or changed foster placements, thus making it more difficult to participate in the program. Another possible explanation for the high rate of initial nonparticipation is that the adolescents may have believed that the program would be similar to the experience of completing the hour long baseline interview needed for the program evaluation, instead of the more desirable role play activities and skills training exercises that characterized the sessions. External factors could threaten the validity of study findings based solely in individual factors.

To reduce the structural barriers that may have led to nonparticipation and dropout among the foster care adolescents, programmatic strategies may involve providing an introductory group session held before the baseline interview to provide adolescents with an overview of program topics, activities, and expectations and how such a program may be of value to them. More explanation could be provided of the importance of their role in evaluation and designing relevant programs for other adolescents like them. Additionally, mail, telephone, e-mail, and text contacts can be used to keep adolescents interested and connected to the upcoming program between the baseline interview and the initial session.

Conclusions

Findings from this study suggest that life skills and HIV prevention programs for adolescents in foster care that are implemented over a prolonged period of time may increase participation and decrease dropout if they are further targeted and tailored to older adolescents, those who use marijuana, and those who are already engaging in unprotected vaginal intercourse. Programs could provide information about the association between substance use and HIV risk and incorporate strategies to address these co-occurring behaviors. Engagement and retention efforts could be proactively implemented, starting with participant screening and continuing on an ongoing basis. Motivational interviewing techniques and harm reduction strategies could be incorporated, as they are promising approaches to resolving adolescent ambivalence to behavior change and assisting adolescents in personalizing prevention messages.

Study findings also support the notion that adolescents in foster care who are at greatest risk for HIV infection may be those who choose not to participate or who drop out of the program. The effectiveness of HIV risk reduction programs depends largely upon the ability of such programs to recruit and retain participants who engage in their targeted risk behaviors yet may be reluctant to initially or consistently participate in such interventions. Further long-term group efforts to address the independent living needs and HIV risk behaviors with these and other vulnerable adolescent populations should address structural barriers to program participation and incorporate motivational techniques to address individual risk behaviors (e.g., marijuana use, vaginal sex without a condom) and readiness to change such behaviors. To improve initial and ongoing participation, HIV prevention efforts for adolescents in foster care should be tailored to individual HIV risk behaviors and incorporate early and ongoing engagement and retention strategies.