This paper represents an effort to elucidate two key results from the Seropositive Urban Men's Intervention Trial (SUMIT) study. First, the intervention failed to achieve a significant decrease, relative to a comparison, in the number of men reporting unprotected insertive anal intercourse (UIAI) with partners of negative or unknown HIV status, the behavior carrying the highest risk of transmission of the virus to others. Second, a significant decline in this behavior was observed in both treatment arms post-intervention .
The SUMIT intervention, like many other behavioral interventions, sought to affect several potential mediators that we had found in formative research to be correlated cross-sectionally with behavior and that we thought would be responsible for intervention effectiveness. As shown in Fig. 1, mediators were assessed at all assessment points, but were only expected to change among those receiving the intervention, which was designed specifically to change them. Mediation analysis provides a strategy for identifying which intervention components account for resulting changes in behavior [2,3]. Mediation analysis involves, first, identifying which potential mediators were changed by the intervention, and second, assessing the effect of mediator inclusion in models of treatment condition and outcome. However, this approach is only appropriate when a significant effect of the intervention has been achieved relative to a comparison group. A common finding in behavioral intervention to reduce HIV risk, however, is that both treatment groups reduce their risk behavior and the intervention effect is itself non-significant [4–7]. Such a result is difficult to interpret: it may be caused by non-specific factors accruing to study participation, such as assessment reactivity; items on mediator scales may have stimulated thought among men in both intervention arms, producing an effect of the assessment on behavior; or it may reflect socially desirable responding (‘telling study staff what they want to hear’). As shown in Fig. 1, these factors would accrue to both intervention conditions. When an intervention is ineffective but a trend towards change is observed in both groups, analyses in which the data are treated as longitudinal may provide some clues as to whether self-reported behavior change actually occurred, and what factors may account for behavior change independent of the treatment condition [8,9].
Because the intervention was not effective in changing the sexual risk outcome, the second step of recommended analysis by Baron and Kenny  was not performed. However, as we observed a reduction in this behavior in both treatment arms, we wished to conduct a longitudinal analysis to identify factors associated with behavior change, irrespective of which intervention participants attended. All participants completed an assessment survey that included scales to measure each of a number of potential mediators derived from formative research with the study population and behavioral theory, particularly social cognitive theory [10,11]. Social cognitive theory asserts that skills and self-efficacy (self-confidence) for behavior change, as well as outcomes that are expected when behavior is changed, are necessary for an intervention to be effective. Our mediators are examples of these relevant to the population and outcomes, as described below in the Measures section.
In this paper, two sets of analyses are presented. In the first, we perform the first step of a standard mediation analysis to identify effects of the intervention. Second, we compare men reporting the discontinuation of UIAI with HIV-negative or unknown-status partners with those initiating this behavior.
For a description of the eligibility criteria and procedures for the trial, see Wolitski et al. . The main outcome paper  presents sociodemographic information for the 811 trial participants (randomly assigned from a total of 1168 assessed at baseline).
All participants completed an audio-computer assisted assessment at baseline, and at 3 months and 6 months after the interventions had ended. Retention was 85% (n = 693) at the 3-month follow-up and 90% (n = 729) at 6 months.
A general description of the intervention conditions is given in Wolitski et al. . In the Measures section are included descriptions of the elements of the intervention specifically with regard to the mediators. The comparison condition was designed specifically not to address the mediators, and consisted of an informational question-and-answer session on prevention-related issues. The comparison condition was thus likely only to have increased knowledge, which has been shown to be ineffective in changing behavior . Figure 1 depicts the study design in terms of mediators and non-specific factors.
At each timepoint, men reported their sexual behavior during the preceding 90 days. Reports were given separately for main partners and non-main partners of HIV-positive status, HIV-negative status, unknown status, and female partners. Participants were first asked whether the behavior occurred; then how many times it occurred with a condom; then how many times it occurred without a condom. Of the behaviors, only whether the participant reported UIAI with HIV-negative or unknown-status partner(s) is relevant to the present paper.
To identify men whose behavior changed over time, we computed the change in status of transmission risk UIAI with HIV-negative or unknown-status partners from baseline. Participants who changed from ‘yes’ at baseline to ‘no’ after the intervention were classified as risk discontinuers. Those who changed from ‘no’ to ‘yes’ were the risk initiators. These are the comparison groups for the behavior change analyses.
Partner serostatus assumptions
A scale was developed for the study to assess the degree to which participants made assumptions about potential sex partners’ HIV status. The tendency to assume the seroconcordance of sex partners of unknown HIV status was found to be prevalent in formative research . The eight-item scale used a response format ranging from ‘strongly disagree’ (1) to ‘strongly agree’ (5); the item average became the scale score. Example items are: ‘I can usually tell if a guy is HIV positive without asking him’; ‘A guy who lets me fuck him without a condom is probably HIV positive’; and ‘If a guy is over 40 in this city, it is reasonable to assume that he is HIV positive’. Coefficient alpha for the scale was 0.78 among the baseline sample (n = 1163). This issue was addressed in the intervention in session five: ‘Is he or isn’t he? that is the question’, in which this tendency was addressed in a game in which participants guessed the serostatus of men in photos (often incorrectly). This activity occurred during a session concerning the importance of serostatus disclosure to sex partners.
Perceived peer norms
Three items assessed men's beliefs about the behaviors engaged in by their HIV-positive peers with non-main partners. Men were asked ‘How many of the HIV-positive men you know (do the behavior) with their non-main partners who are HIV negative or whose status they do not know?’. The response format ranged from ‘none’ (1) to ‘all or almost all’ (5); the item average became the scale score. The behaviors were ‘get blow jobs without condoms’, ‘get fucked without a condom’, and ‘fuck (their partners) without a condom’. Of the 756 men reporting HIV-negative or serostatus-unknown non-main partners, 740 completed this scale. Coefficient alpha for the scale was 0.67 among the baseline sample. The primary way in which peer norms were influenced in the intervention was via small-group discussions throughout the intervention (i.e. by hearing the views of similar others in the groups).
Self-evaluative outcome expectancies
Two types of outcome expectancies were assessed in this study. One was self-evaluative expectancies following safe or unsafe sexual behavior; this has been associated with risk behavior in previous studies . It consisted of six items with five-point scales (1–5) of agreement, from ‘strongly disagree’ to ‘strongly agree’; the item average became the scale score. Examples of items from this scale are: ‘Fucking somebody without a condom makes me feel bad about myself’ and ‘I feel good about myself when I have safer sex’. The coefficient alpha for this scale was 0.84 in the baseline sample (N = 1157).
This construct, along with the related personal responsibility construct described below, were addressed in session four: ‘Dating and accountability’ in which participants were encouraged to use their ‘power to protect’ others from HIV infection.
Hedonistic outcome expectancies for condom use
We assessed whether condom use was associated with the expectation of a reduction of pleasure during sex. This scale consisted of three items with the same response format as the self-evaluative outcomes scale. An example item from this scale is: ‘Condoms can make me lose my hard on’. This scale's coefficient alpha was 0.74 in the baseline sample (N = 1160). In the intervention, this factor was addressed in a ‘dating game’ in which potential ‘dates’ were interviewed and asked how they would make condoms more fun.
Condom use self-efficacy
Self-efficacy was assessed for safer sex with main partners and non-main partners in separate scales. In each, six items concerned insertive, and six receptive, anal sex. Given our research focus, only the six-item non-main partner scale for insertive anal sex was used in the present analysis, and this was only asked of subjects with non-main HIV-negative or serostatus-unknown partners. This scale used a response format with five levels (1–5) of efficacy, from ‘absolutely sure I cannot’ to ‘absolutely sure I can’. Item averages became the scale scores. Example items from the insertive scale (used here) are: ‘When I want to fuck a non-main partner who is HIV negative or whose status I do not know, I can use a condom even when he doesn’t know I’m positive’ and ‘… I can use a condom even if I’ve met someone I really want to like me’. Coefficient alpha for this scale was 0.83 (n = 747) in the baseline sample of participants with an HIV-negative or unknown-status non-main partner. Self-efficacy for condom use was addressed in the intervention primarily through discussions regarding how to negotiate it (practice in the mechanics of condom use was assumed to be in place for this population; a previous mediation analysis for a different study found this to be an unhelpful element ).
The personal responsibility scale was developed by the investigators , and was found to be associated with risk behavior in formative research . The scale contained six items with a five-point response format from ‘strongly disagree’ (1) to ‘strongly agree’ (5); the item average became the scale score. Example items are: ‘I feel responsible for protecting my partners from HIV’; ‘HIV-positive men have a special obligation to have safe sex with men who are negative or do not know their HIV status’; and ‘It's very important for me to use condoms to protect my sex partners from HIV’. Coefficient alpha for the scale at baseline was 0.84 (N = 1163). This construct is similar to ‘self-evaluative outcome expectancies’ and was addressed in the same intervention activities.
Participants completed three subscales of the Brief Symptom Inventory . The six items comprising the anxiety scale, the seven from the depression subscale, and five comprising the hostility subscale were used in the present analyses. Respondents indicated how much each symptom had bothered them during the past week on a five-point scale (1–5) ranging from ‘not at all’ to ‘extremely’. Item values were averaged to create the scale score. Example items from the anxiety subscale are ‘suddenly scared for no reason’ and ‘feeling tense or keyed up’. Items from the depression subscale include ‘feeling blue’ and ‘feeling worthless’. Examples of hostility items include ‘getting into frequent arguments’ and ‘having urges to break or smash things’. Coefficient alphas for these subscales in our baseline sample were 0.86 for anxiety (N = 1163), 0.89 for depression (N = 1163) and 0.77 for hostility (N = 1165). We predicted that negative affect would decrease during the intervention as participants gained the support of the group and guidance on safer sex. One session, ‘riding the roller coaster’ dealt specifically with mood changes and the challenge of coping with ever-changing developments in HIV treatment.
Sexual compulsivity was assessed with a shortened version of a scale that has been standardized with both gay men and HIV-positive individuals [18,19], and was correlated with sexual risk at baseline in the present study . This was a six-item scale (those with the highest correlations with the 10-item original) with items such as: ‘My desires to have sex have disrupted my daily life’. Responses described how ‘like me’ the items were on a four-point scale; the item average became the scale score. This scale yielded a coefficient alpha of 0.87 in the baseline sample of the present study. The intervention did not address this factor directly; encouraging practice in ‘cooling’ sexual arousal (thinking about baseball; taking a cold shower) would be the appropriate way to do so.
Respondents indicated which of several non-injected drugs they had used within the previous 90 days. These were speed/amphetamines, barbiturates/tranquilizers, cocaine, ecstasy, ketamine, marijuana, poppers, gamma hydroxybutyrate and crystal. In addition, one other drug name could be written in. Then they were asked which of several drugs they had injected during the previous 90 days. The drug use score was simply the sum of the number of non-injected drugs checked, plus an additional point if any drug had been injected. Drug use was addressed in the intervention in a session called ‘designer drugs: some boys just want to have (too much) fun’. This session included a humorous ‘drag drug parade’, in which characters representing different drugs paraded down the aisle, with the aim of winning the designation ‘biggest mess’. Each character described the drug's effects (positive and negative) and why people use it. Their dress and behavior also served to stigmatize drug use. Then participants discussed an audiotaped narrative of someone who had planned to avoid risk failing because he was high.
Intercorrelations between mediators (Table 1) were assessed using Pearson correlation coefficients. The effect of the intervention on the mediators (first stage mediation analysis; Table 2) was determined using simple linear regression models. The mediator (at 3 and 6 months) was regressed on the dichotomous treatment indicator, while controlling for the baseline value. Significance was determined at the 0.05 level. In comparing the association between change in risk and change in mediators (Table 4 for univariate and Table 5 for multivariate), logistic regression was used to model the probability of discontinuing risk (i.e. success) as a function of each mediator change score. Mediator change scores were computed as 3-month follow-up value minus baseline value (or 6-month minus baseline). The odds ratio (OR) can be interpreted as the increase in odds of discontinuing risk (versus initiating risk) for a unit increase in the mediator change score. For those mediators whose higher values correspond to more negative states, an OR greater than 1.0 suggests that a larger increase in mediator change (i.e. greater change to a negative state) is associated with an increased odds of discontinuing risk. In contrast, for those mediators whose higher values correspond to a more positive state (i.e. self-evaluative outcome expectancies, personal responsibility, self-efficacy), an OR greater than 1.0 suggests that a larger increase in mediator change (i.e. greater change to positive state) is associated with an increased odds of discontinuing risk. All logistic models included effects for city, cohort (group of participants recruited in the same wave; associated with secular change), age (continuous), education (high school degree or not), and race/ethnicity (black/Hispanic or not). Because of small sample sizes, however, the subset analyses among men reporting having sex with an HIV-negative or status-unknown non-main partner in the past 90 days could only adjust for city. Finally, backward elimination methods were used to identify final models with an exit level at P > 0.10.
Intercorrelations among the mediators at baseline
Table 1 gives the intercorrelations among the mediators at baseline among trial participants (N = 811). In general, correlations were modest, with the exception of intercorrelations among Brief Symptom Inventory distress variables and the correlation between the related constructs of personal responsibility and self-evaluative outcome expectancies. Almost all correlations were significant (as expected by SCT, some of whose theoretical relationships are discussed elsewhere ).
Changes in mediators across time by treatment condition
Table 2 presents levels of the mediators at baseline, 3 months, and 6 months by intervention condition among trial participants. Statistically significant intervention effects on the mediators at follow-up, adjusting for baseline mediator levels, are also indicated in this table. At the 3-month follow-up, significant differences in the level of mediator by intervention condition were observed in four mediators. Compared with men in the comparison condition, those who had received the intervention displayed a reduced tendency to make HIV-status assumptions (P = 0.01), more favorable hedonistic outcome expectancies regarding condom use (P = 0.05), reduced anxiety (P = 0.02), and reduced depression (P = 0.04). There were no significant intervention effects on any mediator levels at the 6-month follow-up.
Differences between risk discontinuers and initiators: univariate results
At the 3-month follow-up, 87 men (12.6%) were classified as risk discontinuers (that is, reported UIAI at baseline but not at follow-up) and 57 men (8.2%) were classified as risk initiators (that is, reporting UIAI at follow-up but not at baseline). Likewise, 84 men (11.5%) were classified as risk discontinuers and 56 men (7.7%) were classified as risk initiators at the 6-month follow-up. Mediator levels for the 90 days before the follow-up for men discontinuing versus initiating transmission risk are presented in Table 3. In order to detect significant differences between risk discontinuers and risk initiators, we used logistic regression modeling to measure the strength of association between the direction of the behavior change and the magnitude of each mediator change separately. These results are presented in Table 4; significant OR are indicated by the confidence intervals and P values. In these analyses, we adjusted for city, age, education and race. At 3 months post-intervention, risk discontinuers were significantly more likely to have had greater decreases in serostatus assumption and a greater improvement in hedonistic beliefs about condoms. More specifically, at 3 months, we can say that the odds of discontinuing risk was reduced by an estimated 60% for each unit increase in HIV-assumption (i.e. towards a less desirable state). Put another way, we can say that the odds of discontinuing risk was 2.5 times greater as serostatus assumption decreases by one unit (i.e. towards a more desirable state). Although these changes remained significant at the 6-month timepoint, three additional significant relationships emerged. Risk discontinuers, relative to risk initiators, were significantly more likely to have greater increases in depression, a greater improvement in self-evaluative outcome expectancies and greater increases in personal responsibility to protect sex partners. These latter results suggest that risk discontinuers, more often than risk initiators, internalized self-standards for protecting others from their HIV infection.
Differences between risk discontinuers and initiators: multivariate results
The final multiple logistic regression models of these results, based on a backwards elimination process, are presented in Table 5. Even when adjusting for all significant mediators, significant and independent effects were found for HIV-status assumptions, self-evaluative outcome expectancies, hedonistic outcome beliefs regarding condoms, personal responsibility, and depression.
Examination of risk changers based on study period
The 6-month data include both men whose behavior changed between baseline and 3 months, and those whose behavior changed between 3 and 6 months. We examined these groups separately to assess the relative contribution of each to the 6-month results for self-evaluative outcome expectancies and personal responsibility. The levels of these mediators for the two groups of men are graphed in Figs 2 and 3, respectively. The number of men changing risk status from baseline to 3 months only was 88, the number changing from 3 to 6 months only was 53, and the number changing status during both periods was 52. Significance tests performed on mediator levels for the two risk change groups at 6 months revealed only one significant effect, that for personal responsibility from 3 to 6 months (P < 0.05; all other P > 0.20). However, all of the changes were in the same direction: risk initiators reduced their levels of self-evaluative outcome expectancies and personal responsibility, relative to risk discontinuers, at the 6-month timepoint. This suggests that men changing risk status during both study intervals contributed to the 6-month results.
These results suggest that only some of the factors that the intervention was designed to change, the mediators, were actually changed by it. Two of these were indices of distress (anxiety and depression) and were reduced by the intervention relative to the comparison condition. The other two, HIV status assumption-making and hedonistic expectancies regarding condom use, were derived from behavior change theory, and in fact distinguished risk discontinuers from risk initiators in the participant group as a whole. Clearly, though, these changes, which were small in magnitude, were not sufficient to produce a significant intervention effect.
The behavior change analyses indicate that two important correlates of behavior change to reduce transmission risk were not affected by the intervention: self-evaluative reactions to transmission risk behavior (OR 4.8; Table 5) and the belief that HIV-positive men are responsible for protecting their partners from infection (OR 2.4; Table 5). These two factors are conceptually very similar, one involving affective, and the other cognitive, aspects of holding a moral stance toward one's behavior as it might affect others. Both reflect the engagement of moral and altruistic processes in determining behavior. Interestingly, whereas they are only moderately correlated, r = 0.51 (Table 1), both exert a strong and independent influence on long-term behavior. It may be that the assessment procedure, which involved responding to numerous items from these scales, initiated cognitive processing with the potential to affect behavior both among the intervention and the control participants. It may also be the case that the intervention had differential results on different men, increasing perceived responsibility for some but decreasing it for others. This might have been the result of different norms emerging in different intervention groups. Still another possibility is that the tendency to hold, or to internalize, other-protective standards, is a personal factor that is relatively immune to the influence of interventions. Finally, men's beliefs about personal responsibility may be post-hoc justifications for unsafe behavior.
The finding that risk discontinuers had greater depression than risk initiators at 6 months (Table 3) and were more likely to have increased depression over time is difficult to interpret. It may be that individuals who are prone to depression are also more likely to experience guilt and self-punitive cognitions about transmission risk behavior, and are more likely to internalize new standards for behavior. In turn, giving up pleasurable intimate behaviors may contribute to increased depression. Depression is also sometimes associated with a lower libido , although its effect on sexual behavior is unclear . The null result for self-efficacy may have been a result of the condom-use focus of this scale; condom use is not necessarily relevant to our target behavior. That the intervention did not affect drug use is not surprising given the small amount of time devoted to this in the intervention.
The weak effects of the intervention on targeted behaviors may thus be attributable to the study team's failure to design an intervention that was effective in enhancing personal responsibility for protecting partners from HIV, and its self-evaluative effects. Assuming that these are malleable factors, how might intervention efforts be more successful in achieving this aim? One possibility is that when individuals have already knowingly put others at risk of contracting HIV, the task of incorporating a new self-standard is more difficult because it requires an admission of past error. If this were so, intervention efforts should be more effective among newly diagnosed men. In the present study, we looked at the correlation between time since diagnosis, on the one hand, and the altruism-related mediators (self-evaluative outcome expectancies and personal responsibility beliefs) and changes in these mediators, on the other. None of the correlations was significant (Ps > 0.10).
Another is that interventions may be more effective if they are not fun and playful in tone, as that of SUMIT was. The group format may have militated against the incorporation of changing standards, as it is likely that at least one individual in each group voiced alternative, incompatible attitudes.
As the baseline correlates paper appearing in this special issue  used many of the same mediators, it is tempting to compare results. As stated above, cross-sectional correlations are frequently used to identify possible mediators to address in intervention. The outcome variable used in that paper, the proportion of sex acts protected by condoms, is quite different from that used here. Furthermore, a number of variables used in that paper are not changeable by intervention (e.g. race, childhood sexual abuse) and thus were not included here. Nevertheless, several of the same mediators were associated with sexual risk in both analyses. These include self-evaluative and hedonistic expected outcomes of safer sex, and distress variables (with depression being associated with less safe sex in the other paper). On the other hand, self-efficacy and drug use were correlated with safe sex in the other paper, but were neither changed by the intervention nor predicted the direction of sex risk response in the present intervention. Whenever possible, longitudinal and intervention studies should be consulted, along with correlational studies (and theory, of course) to identify mediators for interventions.
It should also be noted that, whereas several theoretically derived potential mediators were examined here, there may be critical intervention components that we did not consider or measure. In a previous mediation analysis for an effective intervention , whereas mediators similar to those tested here accounted for significant variance in intervention outcomes, another significant and substantial portion of variance was not accounted for, leaving unanswered the question of how best to improve future interventions.
One note of concern that these data raise is the applicability of delivering to HIV-positive populations interventions that were designed for and tested with HIV-negative individuals. In the United States, there is currently great encouragement for interventions to be delivered to seropositive populations, and because few rigorously evaluated interventions for positive individuals exist, many community-based organizations are ‘tailoring’ interventions that were initially designed for HIV-negative individuals and delivering them to HIV-positive individuals. It will be important that this tailoring consider the possibly fundamental differences in determinants of behavior change in the two populations. Whereas HIV-negative individuals are likely to be motivated to practise safe sex by self-interest, it appears that HIV-positive individuals are likely to be motivated by altruistic concern for the welfare of others.
In general, the most significant methodological finding from these analyses is the importance of assessing theoretical mediators in intervention research. When interventions are effective, mediation analysis can provide information on which aspects were the ‘active ingredients’ and which may be superfluous. This information can in turn be used to fine-tune interventions and make them more cost-effective. When an intervention fails to achieve its objective, an analysis of change in mediators can help investigators to understand which intervention aims they failed to achieve. Furthermore, an analysis of theoretical correlates of behavior change can suggest, as they did here, ways to improve subsequent interventions.
The authors would like to thank an anonymous reviewer for thorough and insightful comments on the paper. The authors would also like to acknowledge the following people who contributed to this research: Bill Woods, Perry Halkitis, Robert Remien, David Bimbi, Rich Wolitski, Tim Matheson, Byron Mason, Carmen Mandic, Bonnie Faigeles, Nick Alvarado, Robert Hays, Andrew Nelson Peterson, Eric Rodriguez, Paul Galatowitsch, Michael Marino, Aongus Burke, Michael Stirratt, Eric Martin, Gloria Abitol and Caroline Bailey.
Sponsorship: Research on SUMIT was funded by the Centers for Disease Control and Prevention through cooperative agreements with New Jersey City University (UR3/CCU216471) and the University of California, San Francisco (UR3/CCU916470).
1. Wolitski RJ, Gómez CA, Parsons JT, and the SUMIT Study Team. Effects of a peer-led behavioral intervention to reduce HIV transmission and promote serostatus disclosure among HIV-seropositive gay and bisexual men. AIDS 2005; 19(suppl. 1):S99–S109.
2. Baron RM, Kenny DA. The moderator-mediator variable distinction in social psychological research: conceptual, strategic, and statistical considerations. J Personal Social Psychol 1986; 51:1173–1182.
3. O'Leary A, Rutgers PI, and the NIMH Multisite HIV Prevention Trial Group. Social cognitive theory mediators of behavior change in the NIMH Multisite HIV Prevention Trial. Health Psychol 2001; 20:369–376.
4. Coyle SL. Women's drug use and HIV risk: findings from NIDA's Cooperative Agreement for Community-based Outreach/Intervention Research Program. Women & Health 1998; 27:1–18.
5. Nyamathi A, Flaskerud J, Leake B, Dixon E, Lu A. Evaluating the impact of peer, nurse case-managed, and standard HIV risk-reduction programs on psychosocial and health-promoting behavioral outcomes among homeless women. Res Nursing Health 2001; 24:410–422.
6. O'Leary A, Ambrose T, Raffaelli M, Maibach E, Jemmott LS, Jemmot JB, et al. Effects of an AIDS risk reduction program on sexual risk behavior of low-income STD patients. AIDS Educ Prevent 1998; 10:483–492.
7. Patterson TL, Shaw WS, Semple SJ. Reducing the sexual risk behaviors of HIV+ individuals: outcome of a randomized controlled trial. Ann Behav Med 2003; 25:137–145.
8. Celentano DD, DiIorio C, Hartwell T, Kelly J, Magana R, Maibach E, et al. Predictors of sexual behavior patterns over one year among persons at high risk for HIV. Arch Sex Behav 2002; 31:165–176.
9. O'Leary A, Maibach E, Ambrose TK, Jemmott JB III, Celentano DD. Social cognitive predictors of condom use behavior change among STD patients. AIDS Behav 2000; 4:309–316.
10. Bandura A. Social foundations of thought and action: a social cognitive theory. Englewood Cliffs, NJ: Prentice-Hall, Inc; 1986.
11. O'Leary A, Wolitski RJ, Remien RH, Woods W, Parsons JT, Lyles C. Psychosocial correlates of transmission risk behavior among HIV-seropositive gay and bisexual men. AIDS 2005; 19(Suppl. 1):S67–S75.
12. Wolitski RJ, Parsons JT, Gómez CA, Purcell DW, Halkitis PN, and the SUMIT Study Team. Prevention with gay and bisexual men living with HIV: rationale and methods of the Seropositive Urban Men's Intervention Trial. AIDS 2005; 19(Suppl. 1):S1–S11.
13. Jemmott LS, Jemmott JB III, O'Leary A. A randomized controlled trial of brief HIV/STD prevention interventions for African American women in primary care settings: effects on sexual risk behavior and STD incidence. Am J Public Health 2005; in press.
14. O'Leary A. Guessing games: sex partner serostatus assumptions in the SUMS. In: Halkitis, PN, Gómez C, Wolitski RJ, editors. Positive sex: the psychological and interpersonal dynamics of HIV seropositive gay and bisexual men's relationships. Washington, DC: American Psychological Association; 2005.
15. Wolitski RJ, Flores S, Bailey C, Gómez CA, Bimbi D, and the SUMIT Study Group. Do HIV-seropositive MSM believe they have a responsibility to protect others? American Public Health Association Annual Meeting. Philadelphia, PA, November 2002 [abs. 40867].
16. Wolitski RJ, Bailey CJ, O'Leary A, Gómez CA, Parsons JT. Self-perceived responsibility of HIV-seropositive men who have sex with men for preventing HIV transmission. AIDS Behav 2003; 7:363–372.
17. Derogatis LL, Lazarus L. SCL-90–R, Brief Symptom Inventory, and matching clinical rating scales. In: Maruish ME, editor. The use of psychological testing for treatment planning and outcome assessment. txv, 673pp. Hillsdale, NJ: Lawrence Erlbaum; 1994. pp. 217–248.
18. Kalichman SC, Rompa D. Sexual sensation seeking and sexual compulsivity scales: reliability, validity, and predicting HIV risk behavior. J Personal Assessment 1995; 65:586–601.
19. Kalichman SC, Rompa D. The Sexual Compulsivity Scale: further development and use with HIV-positive persons. J Personal Assessment 2001; 76:379–395.
20. Phillips RL Jr, Slaughter JR. Depression and sexual desire. Am Family Physician 2000; 62:782–786.
21. Crepaz N, Marks G. Are negative affective states associated with HIV sexual risk behaviors? A meta-analytic review. Health Psychol 2001; 20:291–299.
© 2005 Lippincott Williams & Wilkins, Inc.