Person and Environment in HIV Risk Behavior Change Between Adolescence and Young Adulthood Arlene Rubin Stiffman, PhD Peter Dore, MA Renee M. Cunningham, PhD Felton Earls, MD

This article explores how personal and environmental variables influence change in human immunodeficiency virus (HIV)-related risk behaviors between adolescence and young adulthood. Repeated interviews with 602 youths from 10 cities across the United States provide the data These interviews first occurred in 1984-1985 and 1985-1986 when the youths were adolescents and were repeated again in 1989-1990 and 1991-1992 when they were all young adults. A longitudinal multivariate analysis shows that 31% of the variance in HIV risk behaviors by inner-city young adults is predicted by a combination of adolescent risk behaviors, personal variables (suicidality, substance misuse, antisocial behavior), environmental variables (history of child abuse, poor relations with parents, stressful events, peer misbehavior, number of AIDS prevention messages), and interactions between variables (number of neighborhood murders with child abuse, number of neighborhood murders with substance misuse, and unemployment rates with antisocial behavior).

preventive interventions, relying largely on methods designed to have an impact knowledge and attitudes, and directed to youths, are being mounted on local, regional, and national levels. Early in the epidemic, the potential for educational intervention seemed high because studies showed that levels of knowledge among young people were low.’ Since then, education programs have been implemented virtually worldwide.2 Recent studies reveal higher levels of knowledge among young adult men,3 IV drug users,4 and teenagers. However, this knowledge does not appear to be related to diminishing risk behaviors. In earlier articles, the authors reported that change in HIV risk behaviors by inner-city young adults was not related to knowledge, exposure to information, counseling, or acquaintance with those dying of AIDS.6.7.8 These findings echo other studies that found no change,9012 inadequate levels of change,3.4.I3.t4 or low Extensive on

maintenance of changed Arlene Rubin Stiffman is an associate professor and Peter Dore is a statistician at the George Warren Brown School of Social Work, Washington University, St. Louis. Renee M. Cunningham is a post-doctoral fellow in the Department of Psychiatry at Washington University, St. Louis. Felton Earls is a professor in the Department of Behavioral Science, Harvard University School of Public Health.

Address reprint requests to Arlene Rubin Stiffman, PhD, George Warren Brown School of Social Work, Washington University, St. Louis, MO 63130. Telephone: (314) 935-6685; fax: (314) 935-7508. Research for this project was funded by the Robert Wood Johnson Foundation and National Institutes of Mental Health Grant No. 1RO1MH45118-1RO 1MH45118-01 and NIMH Grant No. 1R24MH50857-01A1.

211

212

Those few studies that found substantial change rely on intensive intervention and skill development 15.16 However, due to the great expense in mounting such programs and the time commitment required of attendees, only a small proportion of individuals needing such interventions can be exposed. Therefore, the incumbent task is to identify other factors, potentially amenable to intervention on a broad basis, that contribute to change in HIV risk behaviors among youths. Many studies examine one or two different factors that are associated with HIV risk behaviors. Most can be categorized as personal variables, but some focus on the environment. These researched variables include youths’ personal and social problems;7.8 other risk behaviors;3.11.14.17 psychological distress;&dquo; certain cultural and social norms (such as expectations that males will have multiple partners);’8-&dquo; alcohol misuse, drug misuse, depression, and suicidaIity;6,7,8.11.2o and low self-esteem.21 Most research on HIV risk behaviors has been atheoretical, but recently researchers have examined the utility of various theoretical approaches. However, despite considerable research on HIV-related health behaviors, there is still no tested, accepted, theoretical framework or directional model for understanding HIV risk behavior, particularly for high-risk youths .21.1 Our research began in 1984-1985 as a two-wave demographic study to examine the outcome of medical care.24 Following this phase, we selected subsamples at varying degrees of risk for HIV infection and followed them over two additional waves of data collection. As the various waves of interviews were obtained and the data . analyzed, it became clear that we were finding that the determinants of mental health and risk behaviors could be characterized as either personal or environmental in nature .7 We also found that, when we examined these categories together, a combination of personal and environmental variables were always better predictors of the dependent variable in question than either category was alone. It was at this point that we began to recognize the necessity of including both personal and environmental variables in explaining health related behaviors. This article explores how well our data support a general person/ environment framework. A number of theoretical models attempt to predict health or risk behaviors. The Health Belief Model25 focuses on outcome expectancies and suggests that individuals engage in a costlbenefit analysis. The theories of reasoned action26 and planned behavior7 state that changes in outcome expectancies and their values, self-efficacy, and intentions (motivation) are linked to changes in behavior. Protection/motivation theory&dquo; and conflict theory29 use similar notions along with an emphasis on coping responses. The fact that most health theories and most health interventions focus on the person while ignoring the importance of the environment has been a major concern.3O A number of factors may serve to attenuate the explanatory predictive power of theories that focus only on person or only on environment; this is especially true for HIV risk behaviors in high-risk youths.3t,32 On the personal level, many HIV risk behaviors occur in conjunction with alcohol or drug use, which impair judgment 33 Many HIV risk behaviors also serve potent habitual or physiological drives.~ Youths may have coexisting beliefs or values concerning sexual behavior or condom use.’ Also, many of the theories presume that health is a valued concern or goal, which, in the case of depressed or suicidal youths, may not be true. At the environmental level, pressure by partners influences sexual and IV needlesharing behaviors,’ and risk behaviors may be undertaken for nonhealthy reasons such as the desire to conform to norms to obtain companionship or approval. Finally, the behaviors may be inhibited by economic or environmental factors (e.g., poverty limits the ability to purchase or obtain condoms). Many of the personal or environmental factors are particularly salient during the transition from adolescence to young adulthood. Indeed, that transition period is thought

213

particularly critical one for the movement to responsible young adulthood, but few studies of change in any risk behaviors between the two periods.35 There are virtually no studies that look at a range of personal and environmental factors that might explain changes in HIV risk behaviors between adolescence and young adulthood. A review of the current state of knowledge about theories and predictors of risk behavior in youths prompted our interest in exploring frameworks that include both person and environment in determining change in HIV-related risk behavior during the adolescent-young adulthood transitional period. to be there

a

are

We were able to find two models/theories that use the notions of both person and environment in relationship to health risk behaviors: social cognitive theory34 and an

ecological mode1.30.36 Social cognitive theory&dquo; posits that behavior, personal determinants (cognitive, affective, and biological), and environmental influences all function as interacting determinants of each other.3; McLeroy and his colleagues call for using an ecological model to explain health behavior by including the connection between environment and behavior. However, he does not specify if the effect is additive or interactive, &dquo;

or

.

both.

Despite the fact that our data were not designed to test any of the above theories, much of our data still closely approximate the variables proposed by person/environment theorists. However, we are not testing all of the variables posited in the theories. For instance, Bandura specified many concepts, including outcome expectations, values, beliefs, and self-efficacy for which we have no specific measures. Likewise, McLeroy posits environment as comprising, in part, organizational or communal factors that we have not measured. We are testing the general idea of person and environment as explanations of HIV risk behavior. We reviewed the literature to see which personal or environmental determinants other researchers have conceptualized as important in predicting risk behavior, risk behavior change, and in testing available theories. Then we selected variables from our existing data set that operationalized those concepts. The youths’ personal determinants comprise those variables usually included in health behavior models: self-esteem, mental health problems, perceived vulnerability, and knowledge. The youths’ environmental determinants are one portion of those that McLeroy and his colleagues3° posit as having equal importance when understanding health behavior. For conceptual convenience, we categorize our environmental variables as relating to three areas: the family (e.g., support and maltreatment), the society (e.g., social activities and the problem behaviors of friends), and the community (e.g., rates of HIV infection and rates of unemployment). We posit, like both McLeroy and Bandura, that the joint effect of variables from both person and environment influences risk behaviors, and we posit, like Bandura, that the variables interact with one another. Our study is the first that examines how this combination of person and environment explain change in HIV risk behavior and is the first to examine that during the adolescent-young adulthood transition. ’

METHOD

Design Detailed information,

gathered in repeated interviews with the same 602 youths in and 1984-1985, 1985-1986, 1989-1990, provides a history of change in HIV-rclated behaviors from adolescence to young adulthood. high-risk

214

Sample The 602 young adults who constituted the study sample were selected from a larger sample of 2,787 inner-city youths who were first interviewed as a part of an evaluation of a multisite program initiated by the Robert Wood Johnson Foundation.24.37 That larger study was designed to develop health clinics that would consolidate the delivery of medical, social, and mental health services to adolescents and young adults living in communities characterized by high rates of adolescent pregnancy, homicide, suicide, and substance abuse. The sample of 2,787 youths was drawn from all patients between the ages of 13 and 18 who were attending each of the clinic programs during a 5-month period in 1984-1985. The programs providing consolidated services were located in Boston, Chicago, Indianapolis, Jackson, Mississippi, New Haven, Dallas, and Los Angeles. Those providing only more traditional health services were located in St. Louis, Buffalo, and New Orleans. Further details concerning sample selection and differences between the clinics are described in other articles. 24.38 The subsample of 602 youths presented herein was drawn as a stratified random sample from the participants in the larger study. To obtain this sample, all youths were divided into three groups based only on their self-reports (from either of the two waves of interviews during their adolescence) of risk factors for contracting HIV infection (see Stiffman and Earls3’ for an explanation of the rationale). The high-risk group (n = 59) engaged in prostitution, intravenous drug use, male homosexual or bisexual behavior, or had had a sexually transmitted disease associated with genital ulcers or sores at either the time 1 or 2 interviews. Moderate-risk subjects (n = 390) had had more than six sexual partners in the preceding year or had had nonulcerative forms of sexually transmitted diseases at either the time 1 or 2 interview. We characterized the remaining subjects (It = 1,962) as being at low risk for HIV infection. For the stratified random sample, we tried to reinterview all high-risk youths, 60% of moderate-risk youths, and 12% of the low-risk group. We were able to locate and reinterview only 71% (it = 42) of the high-risk individuals, but 92% of both the randomly selected moderate {n = 236) and low-risk ,

(n

=

324) subsamples. interview, the youths’ social security numbers and names and addresses of

At each

family membcrs and friends were obtained to facilitate future tracing. Survey Research Associates of Baltimore, Maryland, a firm that specializes in tracing hard-to-reach populations, did the tracing and interviewing. In 6 of the 10 cities, the interviewers at each wave were the same individuals, so no attempt was made to blind them as to the youths’ prior behavior. The youths selected for the subsample (n = 602) did not differ significantly from the rest of the youths (r: = 2,787) in age, sex, race, or in the employment status of their parents: 76% were female, 24% male; 70% were Black, 24% White, 6% other; 6% were professional/managerial,11 % white-collar, 49% blue-collar, 34% semiskilled workers or laborers. Their average age was 16.2 years in 1984 and 21.2 by early young adulthood in 1989-1990.

Instruments The current interview yielded data that could be classified as Personal Determinants, Environmental Determinants, or HIV Risk Behaviors. See Table 1 for a list of

215

the variables, their characteristics, and coefficients.

sources

citing reliability, validity,

or

published

Person

.

Under personal determinants, we asked about the subjects’ knowledge concerning HIV infection or AIDS. The questions came from the AIDS questionnaire used by the National Center of Health Statistics.39 We asked about mental health. We assessed symptoms for several specific psychiatric problems, such as youths’ depression, anxiety, post-traumatic stress, suicidality, substance abuse (alcohol or drug), and antisocial behavior. The questions for all symptoms except antisocial behavior came from the National Institutes of Mental Health’s Diagnostic Interview Schedule (DIS).40.41.42 We counted all symptoms categorized as positive for a mental health problem after using the DIS probe to eliminate physical’illness or medication effects as the source of the symptom. The symptoms for antisocial behavior came from the Diagnostic Interview for Children and Adolescents (DICA).43 We also asked about accident proneness (number of injuries), general physical health (poor to excellent), demographics (race, age, gender, parent employment levels), achievements in work or school, career and educational goals, and educational level. Eftvironntent Under environmental determinants, we asked about three groups of variables: family, society, and community. Family variables pertain to the youth’s family setting. These include marriage or partnerships, children, relationships with parents 44 mentally ill family members,45 and history of child mistreatment .46 Society-level variables pertain to the youth’s relationship with nonfamily members, and social experiences or activities. They include traumatic experiences, life events, peer/friend misbehaviors,&dquo; relationships with peers, social activities, and employment. The Diagnostic Interview Schedule section on posttraumatic stress yielded counts of the number of experiences that induced traumatic reactions 4’ We also used a summary score for stressful life events from the Diagnostic Interview for Children and Adolescents. 43 The events included experiences of illness, poverty, household violence, family death, homelessness, parental separation/divorce, and unemployment. Construct reliability of this scale is evidenced by its association with depression and suicidal ity,37.44 relationship problems p4’ and-AIDS risk behavior.’ To measure peer behaviors, the youths rated how many of their friends (none, a few, about half, most, or all of their friends) had trouble with the police, used drugs or marijuana, were both unemployed and out of school, or drank alcohol almost every day. A summary score from these five questions indicates peer misbehavior. These peer misbehaviors have been correlated with other problem behaviors such as violence4’ and change in AIDS risk

behavior.’

Community-level variables pertain to aspects of the surrounding community and its To tap communal variables, we directly asked the youths about neighborhood drug dealing and murders (which are indicators of neighborhood crime, violence, and deterioration), and about the number of different sources of AIDS information to which resources.

x

&dquo;0 a

M u

r.5

&dquo;2

::J 4

St) E E! 11)

r_ ...... o

Z

~ 216

14

S a 0. E ;pl In

II

x (I)

ei 217

218

they

were

exposed (from pamphlets, radio, TV, etc.). Finally,

we

added aggregate

information about city unemployment rates (which are indicators of surrounding poverty and lack of opportunity) during the youths’ young adulthood ’43 and rates of AIDS from the HIV/AIDS Surveillance Report.’9

HIV Rlsk Behaviors

The 1989-1990 interview repeated questions from the interviews in the years 19841985 and 1985-1986, but added more items relevant to risk for HIV infection .17 The young adults were asked about their high-risk drug and sexual behavior and their partners’ risk behaviors. We calculated the number of HIV-related risk behaviors engaged in during young adulthood by summing from the following list: unprotected male homosexual or bisexual behavior, prostitution, intravenous drug use, six or more sexual partners in one year, irregular use of condoms, sharing of unclean IV drug paraphernalia, and choice of a partner who was known to be at risk (due to unprotected male homosexual activity, IV drug use, HIV seropositivity, or prostitution).6 We used the same information to calculate the number of different types of HN related risk behaviors engaged in during adolescence, aggregating information obtained during the two time periods (19841985 and 1985-1986). However, for adolescence, we lacked information on two variables : partner behavior and cleaning of IV drug paraphernalia. For each subject, we calculated the sum of the different risk behaviors engaged in during adolescence (19841985 and 1985-1986) and the sum of the risk behaviors engaged in during young adulthood (1989-1990).

ANALYSES The data on the number of risk behaviors were analyzed through linear regression. In this article we present the analyses showing how environmental and personal situations occurring in the year prior to the 1989-1990 interview affect change in risk behaviors from adolescence to young adulthood. All data in this article focus on behavior in early young adulthood in 1989-1990, and all analyses reported in this article include only the youths who participated in all of the first three waves of interviews. (Note that we also analyzed behavior change from adolescence to a 1991-1992 interview, but we do not report those findings because (1) the results were parallel in their support for the person/environment framework,’and (2) the adolescent-to-early-young-adulthood transitional span is the most critical period to those planning preventive intervention with

adolescents.) There are a number of different methods possible for examining changes One must choose a method based on the appropriateness of the data for the method and the anticipated audience. Path analyses models and structural linear equation modeling both require continuous, normally distributed data. Further, both require multiple measures of each latent variable. Our data contained many dichotomous variables and, often, only one measurc of a latent variable. Therefore those methods were not available to us. Another method is to develop a difference score by subtracting scores at time 1 from scores at time 2. To do this requires the same algorithm at both times, and our algorithm differed slightly from wave to wave. Therefore, we chose to use a variant of analysis of covariance

219

by using scores at time 2 as the dependent variable while controlling for scores at time 1, and using a general linear model. This method has several advantages for this article: (1) It controls for regression to the mean effects, which are particularly problematic with a truncated range of possible scores, (2) since the instruments or scores at time 1 and time 2 are based on slightly different algorithms, it compensates for these differences,St.s2. (3) it accommodates both dichotomous data and interactions, and (4) it is a standard method generally understood by a broad range of trained clinicians and researchers from a variety of backgrounds. To examine contributors to change in risk behaviors, we, therefore, used multiple regression analyses, where young adult risk behaviors are predicted while controlling for adolescent risk behavior. Nominal or ordinal variables were dummy coded for the regression analyses. The variables from the personal and environmental determinants were examined individually as predictor variables, and then the model as a whole was examined multivariately. We examined all possible interactions between personal and environmental variables and between family, society, and community variables, provided they were significant at the bivariate level. A problem may arise from using a stratification variable in the sample selection that . is similar to the dependent measure. Oversampling from the higher risk groups could lead to a change in slope at the upper ends of the distribution. To test for this we weighted the subjects by risk category, assigning weights of 1.4 (59/42) to the highest risk group, 1.65 (390/236) to the moderate risk group, and 6.1 (19b21324) to the low-risk group. (The numbers in parentheses refer to the n of original sample divided by the n of the current sample.) We then ran test models on the weighted sample and compared the parameter estimates to the parameter estimates of the unweighted sample. Typically the mean difference between the parameter estimates for a model was only 0.01, indicating that the stratification did not cause a bias in the slope of the regression models and that using unweighted data in our method of analysis was appropriate.

RESULTS

Description of the Dependent Variable-Risk Behaviors Youths averaged 1.2 risk behaviors during young adulthood, with a range of 0 to 5 and standard deviation of 0.6. The distribution had a skewness of 2.3 and a kurtosis of 7.6. (Although the distribution is skewed and the response model has a large tail, the normality of the residuals, not the normality of the dependent variable, is an assumption of regression models, and we meet that assumption.) a

Individual Regressions We were interested in how each variable contributed to risk behavior and in how each contributed to change in risk behavior. First, we ran a series of regressions with HIV risk behaviors at young adulthood as the dependent variable to compute zero-order relationships (see Table 2, column 1). Then we ran a series of multiple regressions controlling for risk behavior in adolescence to examine the determinants of change in risk behavior (see Table 2, column 2).

220

Table 2. Associations of Independent Variables With Risk Behaviors in Young Adulthood and Change in Risk Behaviors

a. Change is computed by using regressions controlling for behavior at adolescence. *p < 0.05. **p < 0.001. ***p < 0.0001.

to

predict behavior

at young

adulthood, while

Personal Determinants We found that each of the mental health variables contributed unique significant variance to both HIV risk behaviors in young adulthood and to change in HIV risk behavior since adolescence. In terms of change in risk behaviors, substance misuse contributed 13% of the variance, suicidality 8%, depression 5%, antisociallillegal behavior 3%, anxiety 2%, and posttraumatic stress 2%. General physical health contributed 1% of the variance, and accident proneness (measured as the number of injuries in the la5t year) contributed 4%. Most important for preventive intervention was the finding that knowledge about HIV or AIDS and about prevention of HIV infection was unrelated to change in HIV-related risk behaviors. Specifically, young adult knowledge was not associated with changes in risk behavior between adolescence and young adulthood. This strongly suggests that information alone is not a potent factor in risk change. (The youths were, largely,

221

knowledgeable about HIV infection and AIDS. Of the 25 knowledge questions, youths averaged scores of approximately 21 correct answers [range 9-25]. In fact, between 80% and 90% selected the correct responses for 19 or more items.) Neither the demographic variables (age, gender, race, or parent employment), nor ’achievements, nor educational level contributed significantly to change in HIV risk =

behaviors.

Environmental Determinants

Family-Level Variables. A history of having been abused contributed 4% of the variance in change in HIV risk behavior; having been reared withlby mentally ill family members contributed 2%; and poor relations with parents contributed 4%. Having a stable marriage or partnership and having children did not contribute to change in risk behaviors. Society-Level Variables. Anumber of the social variables contributed to change in HIV risk behaviors. Ongoing life stressors contributed 8% of the variance; friends’ misbehaviors 7%; traumatic experiences 1 %; and support from friends 1%. Social activities (sports, church, classes, hobbies) did not contribute to change in risk behavior. Community-Level Variables. The youths’ self-reports concerning neighborhood drug dealing and neighborhood murders each contributed 1 % of the variance of change in risk behaviors. The unemployment rates and the rates of AIDS in the youths’ cities did not contribute to change in HIV risk behaviors. The number of different sources of information to which young adults were exposed predicted 1% variance in change in risk behaviors. Youths were exposed to an average of five different sources of information, with a range of 1 to 17 sources. However, youths who reported more sources of information were more likely to increase their risk behaviors. Perhaps those engaging in high-risk behavior found the different sources more pertinent and thus were more likely to remember them. Or perhaps their high-risk behavior led service providers to give them more information. Grand Regression Model

looking at each predictor variable in separate linear regressions as reported exploratory multiple regressions, which included both personal and environmental variables and interactions between variables. The final grand regression model, including only those variables that predicted unique variance in change in HIV risk behaviors, explained 31% of the total variance, F(14, 520) 16.6, p < 0.oo0l (see After

above,

we ran

=

Table 3). Adolescent risk behaviors explained a unique 8% of the variance in young adult risk behaviors. However, we are not interested in this variable as a predictor, but rather, as a control. By entering this first, all other variables then explain that portion of the variance in young adult risk behaviors that is not accounted for by adolescentrisk behaviors, or phrased differently, they explain the change in young adult risk behaviors since adolescence. The personal variables of suicidality, substance misuse, and antisocial behavior explained a unique 7%, 9%, and 1 %, respectively, of the total variance.

222

Table 3.

Grand Regression Analysis, Predicting Change in AIDS Risk Behaviors From

Adolescence to Young Adulthood

NOTE: Model

R2= 0.31, F(14, 520} = 16.6,p < 0.0001. r computed by dividing the Type III sum of squares by the 1)’pe I model sum of squares. This gives the variance when controlling for all other variables in the model. b. r~ is computed for the noninteractions without the interactions in the model, and then computed a.

for the interactions with the full model. *p < 0.05. **p