Developing and validating complex behavioural outcome measures
Developing and validating complex behavioural outcome measures
Abstract and Keywords
Interventions developed to reduce the risk of sexually transmitted infections (STIs) have typically focused on changes in discrete behaviours. However, simple behaviours alone may not be directly related to risk of infection. This chapter examines the types of behavioural measures commonly used, and explores the development of complex measures. It considers the utility of each with reference to data from randomized controlled trials (RCTs) that have utilized both biological and behavioural measures.
Interventions developed to reduce the risk of infection with sexually transmitted infections (STIs) have typically focused on changes in discrete behaviours. However, simple behaviours alone may not be directly related to risk of infection. The purpose of this chapter is to examine the types of behavioural measures commonly used, and to explore the development of complex measures. The utility of each will be considered with reference to data from randomized controlled trials (RCTs) that have utilized both biological and behavioural measures.
Many STI/HIV risk-reduction interventions, tested in RCTs, have successfully changed aspects of condom-use behaviour in one or both genders.1–8 Change in partner-related behaviours, such as partner number and partner type, have been reported less frequently, and have met with limited success.1, 2, 6 Few RCTs have used infection as an outcome measure,1–4, 8–10 and only two successfully reduced the rate of new infections.1, 2, 11 Data from both of these trials have been used to examine the relationship between infection and behaviour measured by responses to interview questions.1, 12–16 This type of analysis can provide insight into which behaviours were modified by the intervention, and the extent to which these changes influenced infection rates. Specifically, Project RESPECT found minimal association between infection and behavioural measures,16 whereas Project SAFE,1, 12–14 using more comprehensive measures incorporating context, found more significant relationships.
It is important to gain insight into the complex relationship between infection and behaviour to learn: (a) which behaviours or combinations of behaviours must be changed in order to reduce the rate of new infection; and (b) meaningful ways to measure sexual behaviours so that they reflect actual (p.138) likelihood of infection. This is especially important because many STI/HIV prevention studies rely on behavioural outcomes, primarily those pertaining to condom use, to evaluate intervention efficacy. The purpose of this chapter is to examine the relationship between infection and: (a) simple and complex measures of commonly used behavioural variables; (b) less commonly used behavioural measures; and (c) complex measures, considered simultaneously. Discussion relies primarily on data from Project SAFE,1, 12, 14 but also refers to results from Project RESPECT.2, 16, 17 This chapter suggests that links between measured behaviours and biology are clarified when behaviours are purposefully examined as more holistic entities and considered jointly.12, 14, 15 The use of biological and behavioural outcomes is also discussed in Chapter 8.
Project SAFE characteristics
This behavioural risk-reduction intervention was designed for Mexican- and African-American women and evaluated in an RCT, stratified by ethnicity. All 617 participants had a non-viral STI at baseline. They were screened and treated for STIs, and interviewed at baseline, and at six- and twelve-month follow-ups. Visits included physical examinations and laboratory tests. Participants could also return to the research clinic between scheduled visits if they experienced particular problems. The follow-up rate at twelve-months was 89 per cent, with a sample of 549. The biologic outcome measure was infection with gonorrhea and/or chlamydia. Study procedures are described in greater detail in the original publication.1
Data presented here are based on the subset of 477 women with complete data from both follow-up visits. Unless otherwise indicated, results are from the cumulative follow-up period zero-to-twelve months and reflect worst-case values; women were assigned to the lower-risk category cumulatively only if they were low-risk in that measure at both six- and twelve-months. Data on condom-use measures reflect behaviour during the last three months of the zero-to-six or six-to-twelve month interval, or in the case of the cumulative measure, both three-month intervals. Baseline sample characteristics and infection rates for the 477 women are similar to those previously described for the entire follow-up sample.1, 14 A more complete analysis of behavioural and biological data can be found elsewhere.14
Commonly used measures of sexual behaviour
Partner number and type
Few studies refer to the number and type of partner to explain STI rate reductions.1, 13, 16 Results reported from Project RESPECT indicate that having three (p.139) or more partners in a six-month interval was associated with infection (adjusted odds ratio = 1.9)16. In Project SAFE, various partner-number cut-offs were associated with infection. During the twelve-month study period, participants with one or no partner had an infection rate of 14.9 per cent, compared with 32.6 per cent for those with two or more partners (p < 0.001).1 Women with three or more partners had an infection rate of 37.9 per cent, compared with 17.5 per cent (p < 0.001) for those with two or fewer partners.
Project RESPECT interventions focused on consistent condom use.2 However, data were collected on partner type (main vs occasional partners), and used as a stratification variable for the number of unprotected acts,16 discussed in the condom-use section below. Project SAFE focused on a number of issues, including partner relationships. Consequently, a great deal of attention was given to partner measurements. One potential source of measurement error can be the varying definitions given to the word ‘steady’ by different participants.14 To minimize this error, Project SAFE researchers classified relationships reported to be steady ‘all or most of the time’ as steady. ‘Off-and-on’ steady relationships and any type of steady relationship reported by married women living apart from their spouses because of marital difficulties were classified as casual. This category also included occasional partners, those a woman had just met, someone who provides drugs, etc. It seemed, from initial research with similar populations, that such relationships were often high risk: partners frequently had sex with others during the ‘off’ periods and during separation. Participants often continued to have occasional sex with these partners, despite their separated status.
The results support this definition. During Project SAFE's twelve-month follow-up period, 344 women reported that they had steady partners exclusively; their infection rate was 15.4 per cent. The 104 women whose relationships were steady ‘off and on’ (i.e. casual according to the definition used), had a rate of 23.1 per cent, compared with 12.1 per cent for the 240 participants whose relationships were steady all or most of the time.
The number of partners, along with the nature of these relationships, also affects infection rates. Women with one ‘steady’ (our definition) partner had an infection rate of 11.6 per cent, vs 18.9 per cent for those with two or more steady partners. Among women with any ‘casual’ partners, infection rates were 24.2 per cent for women with one partner, and 39.7 per cent for women with two or more partners. Perceived partner fidelity also affected the likelihood of infection. To increase precision, a ‘mutual monogamy’ variable was created, conceptualized to include type of relationship, partner number, and fidelity.14 The 203 women who had one steady, monogamous sex partner, or no sex (p.140) during follow-up (n = 15) had an infection rate of 8.4 per cent compared with 31.4 per cent for everyone else (n = 274), translating into a risk ratio of 3.7. These refinements provided a more complete characterization of the relationship, and therefore the measure was more significantly related to infection.
Many RCTs rely on condom-use measures to assess efficacy, because condoms can prevent transmission of some STIs, including HIV.18 It is also relatively simple to focus on changing one aspect of sexual behaviour. Condom-use measures have included proportion of acts protected, mean number of unprotected acts, 100 per cent condom use, and no unprotected acts. The first of these is rightly noted to be very limited, because it does not consider the number of total acts; for example, 50 per cent use could mean one, or more than a hundred, unprotected acts.19, 20
In the few instances in which infection and behavioural data are simultaneously available, it has become increasingly clear that commonly employed condom-use measures alone are not reliable predictors of disease acquisition. For example, in a Baltimore clinic, men using condoms ‘all the time’ were as likely to be infected as those who never used them, or who sometimes used them.21 Re-examination of these data found no support for risk of infection varying according to level of use. The authors suggested that these anomalous results were due to systematic group differences in reporting errors, and in risk of exposure to an STI.22 Data from Project RESPECT showed that both of the counseling intervention arms were associated with some behavioural change at follow-up. Increases in ‘any condom use’ and ‘no unprotected vaginal sex’ persisted until six months, whereas group differences in ‘condom used at last sex’, ‘one or fewer sex partners’, ‘no casual partners’, and ‘no new partners’ were found only at the three-month visit.2 More importantly, changes in the number of unprotected acts, or the percentage of sexual acts without a condom, even after controlling for sex with a new partner, were not related to infection.17
Data from Project SAFE yielded similar results. Of the 462 women who had sex during the twelve-month follow-up period, 80 used condoms ‘most or all of the time,’ regarded as 75 per cent of the time. Their infection rate was 21.3 per cent, compared with 22.5 per cent for those who did not use condoms, or used them less frequently. The variable ‘mean number of unprotected acts’ was not associated with infection (Pearson's r=−0.01).14 Only 41 women had no unprotected acts; their infection rate was 14.6 per cent (p.141) compared with 22.3 per cent for everyone else (p = 0.26). The non-significant lower rate was due to zero infections among the 15 women with no sex. The 26 participants who used condoms all the time had an infection rate of 23.1 per cent, compared with 22.2 per cent for the remaining participants. Hence, these simple predictors were not appropriate indicators of infection in this study.
Although 26 women is a small sub-set, it is instructive to examine the behaviour of the three infected women who reported 100 per cent condom use. They had infections only during the first six-month interval; two at problem visits, and one at the six-month screen. They probably provided accurate reports about condom use, which referred only to the last three months of that interval. However, their responses to another question indicated that none of the three women used condoms initially following enrollment. All reported they had unprotected sex with partners for whom they could not be sure of complete treatment following the baseline infection. Other researchers2, 3, 16 have also utilized a recall period of the last three months because it covers a sufficiently long period of time to be representative, but does not present a large recall burden. However, in the absence of detailed data from other questions, it was not possible to know what had actually happened. Even had behaviour been measured every three months and achieved complete ascertainment, assessment of the effect of 100 per cent condom use would have been limited by the small number of participants who practised it.
A more useful way of conceptualizing unprotected acts is a specific cut-off point during a given interval of time.1 In project SAFE, fewer than five unprotected acts during two three-month periods more accurately discriminated the infected from the uninfected than other condom-use measures; women with five or more acts were more than twice as likely to be infected than those with fewer unprotected acts than this.1, 14 Although fewer than five unprotected acts yielded a useful cumulative measure, it was not a significant predictor of infection at zero-to-six or six-to-twelve months: the shorter-term measures were confounded by women who had very few coital acts, all of which were unprotected sex with a casual partner (see the definition above). Over time, represented by the cumulative measure, many of these participants shifted to a higher number of unprotected acts. Of interest, in this population fewer than five unprotected acts affected the likelihood of infection only among women with riskier relationships; for example, among participants with only steady partners, infection rates were 9.5 per cent and 11.0 per cent, respectively, for women with low and high frequency of unprotected acts. On the other hand, among women with any casual partners, the infection rate was 12.7 per cent for those with fewer than five unprotected acts, compared with 38.2 per cent for those with five or more (p < 0.001). Project RESPECT found similar (p.142) results:16 six or more acts with an occasional partner, but not with a steady partner, was related to infection with an odds ratio of 1.9.
Successful condom use depends not only on the consistency, but also correctness, of use.12, 14, 23 Participants who reported breakage, slippage, or other problems such as the condom becoming dislodged inside the vagina, were classified as having problematic use. Ethnographic data indicated that, in order to retain some flesh contact, some couples only applied the condom partially, occasionally resulting in its being retained inside the vagina. Consequently, women were asked if they routinely checked that the penis was completely covered; if they did not, they too were considered to have problematic use. Of interest, a negative response to this question was more strongly related to infection than were reports of slippage and breakage. Of the 141 condom-users with problems, the 59 who did not check that the penis was covered had an infection rate of 42.4 per cent, compared with 21.7 per cent for those with other problems. Thus, the addition of this dimension of condom-use behaviour to the variable ‘problematic use’ improved its explanatory power.
Results indicate that, of the 331 women with any condom use during follow-up, 42.6 per cent experienced some problematic use. Their cumulative infection rate was 30.5 per cent, compared with 16.8 per cent among those without problems. Considering only those women whose cumulative acts were protected more than 75 per cent (n = 80), 90 per cent or more (n = 59), and 100 per cent (n = 26) of the time, respective infection rates for participants with and without problems were 24.3 per cent vs 18.6 per cent, 20.8 per cent vs 17.1 per cent, and 30.0 per cent vs 18.8 per cent. Very few participants always or almost always used condoms; whereas correct use did make a difference, it did not overcome all the problems associated with appropriately classifying individuals with regard to their condom use.
To minimize these problems, a composite variable ‘unsafe sex’ was created, incorporating the number of unprotected acts, problematic usage, and relationship type.14 The variable ‘unsafe sex’ was designed to be partner-specific with respect to use/never-use of condoms, because of a belief that individuals choosing risky partners may compensate for this elevated risk by increasing condom use.19, 26 It was reasoned that never using condoms would be much higher risk with a casual rather than a steady partner, because it would reflect absence of any caution. These women and those who had a combination of five or more unprotected acts in three months as well as problematic use were assigned to ‘unsafe sex’; all others were assigned to ‘safer sex.’ A participant (p.143) with a casual partner would be considered high risk in the variable ‘mutual monogamy’, but if she used condoms with that partner, she could be considered low risk with respect to unsafe sex. Although type of relationship was included in both ‘unsafe sex’ and ‘mutual monogamy’, tests of co-linearity indicated sufficient independence.14
The 305 women who practised ‘safer sex’ had a cumulative infection rate of 13.8 per cent, whereas the172 who experienced ‘unsafe sex’ had a rate 35.5 per cent (risk ratio = 2.6). Unlike five or more unprotected acts, this measure was also significantly associated with infection at both zero-to-six and six-to-twelve months.14
Acquiring a new sex partner is assumed to increase infection risk because that individual may be an unknown entity, transmitting new infections. However, this assumption has not been adequately tested. In fact, data from Project RESPECT indicate that having a new partner is not associated with infection.16 In Project RESPECT, partner status was determined by asking participants if a given person was a new sex partner. Project SAFE, in contrast, considered a partner ‘old’ only if he was listed as a sex partner in the prior six-month interval. All other men, including ‘ex-steadies,’ were classified ‘new’ because they represented potential new exposures. The 210 participants who gained a new partner anytime during follow-up had an infection rate of 31.5 per cent, compared with 16.7 per cent for those who only had ‘old’ partners.14
However, this simple comparison is misleading. Of the women with ‘new’ partners, the 70 who waited three or more months between sex partners had an infection rate of 8.6 per cent, compared with 39.3 per cent for the 140 women who did not wait. Consequently, lack of a waiting period (i.e. either not taking three or more months to select a new partner, to have sex with him, or a combination of both) was associated with infection.14
The three-month waiting period may reflect selectivity and patience, or may be associated with reduced infection rates simply because waiting restricts the number of sexual partnerships that can be formed in one year. Using logistic regression analysis among the subset of 210 participants who had new partners, it was found that, by including the waiting period in the model (adjusted odds ratio = 5.1), neither multiple partners (i.e. one or none vs two vs three or more), nor concurrent relationships were significantly related to infection.14
(p.144) A new variable, ‘rapid partner turnover’, was created to distinguish between women who waited less than three months between sex partners from those who waited three or more months, or who did not acquire a new man. Project SAFE results indicate that the 140 women with rapid partner turnover had an infection rate of 39.3 per cent, compared with 14.2 per cent for participants with lower partner turnover (risk ratio = 2.8).14
Less commonly used variables
Avoiding sex with an untreated partner
Perhaps the best way to maximize re-infection risk is having unprotected sex with an untreated or incompletely treated partner. Whereas it is impossible to know the infection status of a partner unless he is tested, incomplete treatment appears to be an excellent proxy. In Project SAFE, women were asked at the six-month interview if they had sex with any partner before he completed treatment following the baseline infection. Responses were ‘no,’ ‘yes’ (with or without condoms), and ‘don't know’ (with or without condoms). Nurse-clinicians' notes were also used to cross-validate the responses. The 17 participants who said they did not have sex before a partner completed treatment, but who told the nurse otherwise, were considered to have had sex with an untreated man.
At six-month follow-up, 414 women had avoided unprotected sex with an untreated partner. Their infection rate was 9.4 per cent, compared with 46.0 per cent for the 63 participants who did not (risk ratio=4.9). Cumulative infection rates for these sub-sets were 16.9 per cent and 52.4 per cent, respectively (risk ratio = 3.1).12, 14 A similar measure, ‘avoiding any sex with an untreated partner,’ yielded similar results.1 Although both measures are excellent predictors of infection, having unprotected sex with an untreated partner is the better of the two, probably because it indicates exceedingly high-risk behaviour. This variable remained an indicator of infection throughout the twelve-month study period because of its strength during zero-to-six-months. In addition, some women returned to, or never left, the untreated partner. Having unprotected sex with a man thought to be untreated or incompletely treated may also reflect a propensity for risk behaviour.
No one variable, regardless of its complexity, can provide an accurate estimate of the likelihood of infection. Sexual risk consists of various dimensions of behaviour and circumstance that are interwoven. Thus, accuracy of estimates increases when variables are combined. As noted earlier, data from Project RESPECT show that fewer than six unprotected acts in a three-month interval among (p.145) women with casual partners was associated with infection, whereas it made no difference among women with only steady partners.16 Project SAFE results were similar, using a cut-off point of fewer than five acts in two three-month intervals. This outcome makes sense, in that women with steady partners presumably have more intimate knowledge of their partners' history and current behaviour, can more accurately assess risk, and thus have a smaller likelihood of exposure to infection than women with casual partners. Using a condom in a low-risk relationship is not likely to contribute much additional prevention benefit.
Combining complex variables provides even better explanation.12–14 In Project SAFE, the most powerful two-variable combination was mutual monogamy and avoiding unprotected sex with an untreated partner. As already noted, mutually monogamous women had an infection rate of 8.4 per cent, compared with 31.4 per cent for the non-monogamous. Participants who avoided unprotected sex with an untreated partner had an infection rate of 16.9 per cent, compared with 52.4 per cent for those who did not. When combined, the 182 mutually monogamous women who avoided such sex had an infection rate of 4.4 per cent, compared with 32.2 per cent for everyone else, who were high-risk on one or both of these behaviours. Pairing mutual monogamy with each of two other modifiable behaviours (i.e. low partner turnover or safer sex), yielded infection rates of 7.7 per cent in both cases.14 Participants who avoided sex with an untreated partner, and had either low partner turnover or safer sex, had infection rates of 9.2 per cent, compared with 41.3 per cent and 38.1 per cent, respectively, for everyone else.
The 232 non-monogamous women who avoided sex with an untreated man had an infection rate of 26.7 per cent, whereas the 42 women with both high-risk behaviours had an infection rate of 57.1 per cent, compared with 4.4 per cent for women with neither behaviour (risk ratio = 13.0).14 Non-monogamous women practicing safer sex had an infection rate of 23.0 per cent, similar to the rate (23.1per cent) attained by non-monogamous women with low partner turnover. These results indicate that a mutually monogamous woman can probably avoid re-infection if she makes certain that her partner is treated or, if not, avoids unprotected sex with him (infection rate of 4.4 per cent). By definition, if a couple is mutually monogamous within the limits of measurement error, neither will become re-infected following an initial infection if both partners are adequately treated before resuming unprotected sex. However, not all women experience a mutually monogamous relationship. These women, already at higher risk, must avoid several risk behaviours in order to reduce substantively their re-infection rate. But even then, these rates do not approach those of mutually monogamous women with treated partners.14
(p.146) Logistic regression results from a multivariate model incorporating the four risk behaviours discussed here, as well as douching after sex, demonstrated that approximately 70 per cent of re-infections may be predicted by these variables.14 The Project SAFE intervention helped reduce infection rates because it motivated participants to: avoid unprotected sex with an untreated partner; be more cautious in partner selection (i.e. have a lower partner turnover); achieve mutual monogamy; and/or avoid very unsafe sex.12, 14
It has become evident that sexual behaviour change necessary to reduce STI rates is exceedingly complex, dependent on a variety of factors, including: number of partners; partner selectivity; type of relationship formed; adherence to treatment protocols on the part of both partners; and correct and consistent condom use.1, 12, 14–16, 19, 23–26 This chapter has demonstrated that these factors do not all have to occur simultaneously for an individual to be protected. For example, mutually monogamous relationships do not require condom use if both partners are treated. In contrast, changing only one aspect of behaviour does not ensure protection from infection. For example, a woman who reduces her number of unprotected acts to only one or two in three months is nonetheless likely to be infected if those acts are with casual partners. Some behavioural changes are interdependent on others to prevent infection. Thus, a commonly used behavioural measure—number of unprotected acts—may not be a reliable predictor of infection if considered alone.
Even 100 per cent condom use, as reported in this chapter and by others,16, 21 does not necessarily discriminate between the infected and uninfected. This could be due to: faulty recall or reporting; incorrect use; lack of correspondence between the reporting periods for infection and behaviour; or other factors. In Project SAFE, one source of measurement error was that infection was screened routinely every six months, whereas the number of unprotected acts was based on the last three months of that interval. Asking women to report on behaviours in the last six months, or interviewing them every three months, may have produced different results. In a new trial,27 respondents are being asked about their sexual behaviour in the last three and six months, in order to achieve better ascertainment. Another approach, frequent interviewing, may unduly contribute to patient burden and would make an already costly study even more expensive. Only 51 per cent of Project RESPECT participants returned for all four follow-up visits.2 Moreover, although women in Project RESPECT were questioned every three months about their condom-use behaviour, and were screened for infection every six months, zero unprotected acts, vs everything else, was not related to infection.16 Correctness of use was not (p.147) included in this variable; perhaps it would have made a substantive difference.24 However, it is also likely that collecting condom-use data is difficult to do without introducing a great deal of measurement error. Study participants may not recall exactly when they started to use condoms all the time, or if they stopped for any reason. Some may not realize that they, or their partners, are using condoms incorrectly. In this chapter, it has been demonstrated that several women in Project SAFE who reported consistent condom use in the last three months of each follow-up interview were, nonetheless, infected: they had had unprotected sex with an untreated partner following the baseline infection. Their responses may have been different had the time frame been the last six months. However they still may not have recalled when they began consistent condom use. The question regarding sex with an untreated partner was sufficiently specific to invite correct recall.
Another approach would be to ask participants to maintain a diary of daily sexual-activity. However, there are problems with this as well. Because it takes a special type of participant to be willing to comply, the potential participant pool may become too restrictive, and thus introduce selection bias that would be difficult to measure. It is unlikely that women at high risk for infection, often living difficult and disrupted lives, would be willing or even able to maintain a diary. Timeliness and accuracy of data entry would be uncertain. Moreover, the very act of recording sexual activity and condom-use behaviour may in itself constitute an intervention, similar to food diaries for the overweight. Thus, results probably would not be generalizable to a broader population.
If the objective of a given intervention is to change a specific behaviour that may contribute to reduced infection rates, such as consistent condom use, it would be sufficient to focus on this behaviour, and to examine self-reported behaviour regarding this outcome. However, if the objective is to reduce infection rates, a more holistic approach is required. Given the complex relationship between behaviour and disease incidence, such interventions must focus on changing several behaviours. For example, it is not sufficient to increase condom use or rates of mutual monogamy, if having sex with an untreated partner is neglected. It is not surprising that, when included as a variable, having sex/unprotected sex with an untreated or incompletely treated partner is the strongest predictor of infection.1, 12, 14, 15 Mutual monogamy is not protective if the partner remains infectious. On the other hand, mutual monogamy, coupled with having a fully treated partner, is exceedingly protective. When the results of Project SAFE were first published,1 intervention efficacy was attributed to the multiplicity of behaviours addressed, from health-seeking behaviour and adherence to treatment protocols to a focus on correct, consistent condom use, and male–female relationships. Because different aspects of behaviour and circumstances interact, (p.148) it is important to influence several key factors simultaneously. It is also important to impress upon intervention participants the importance of exercising judgment in their decision-making processes. For example, if the realities of their lives make them likely to choose multiple or risky partners, it is especially important for them to be vigilant about consistent and correct condom use. Intervention participants can also learn the importance of not having unprotected sex with a partner until she knows he is not infected.
In order to develop measures of sexual behaviour that are meaningful with respect to likelihood of infection, it is critical to include biological parameters. It is important to examine behavioural measures similar to those used in Project SAFE in the context of other studies to determine if their utility is broadly based. For example, in the Project SAFE study population, the question concerning checking that the penis was fully covered was strongly related to new infection. This was not surprising, since prior exploratory work suggested that it would be useful. This may not be the case in other populations. This also speaks to the importance of knowing one's target population.
It has been shown that simple behaviours, commonly addressed and measured in isolation, are often unrelated to infection rates. The link between behaviour and biology is clarified when behavioural measures incorporate context, and are considered simultaneously. Both the development of interventions, and evaluation of their efficacy, must reflect this complex interplay among sexual risk behaviours and infection.
1. Shain, R.N., Piper, J.M., Newton, E.R., Perdue, S.T., Ramos, R., Champion, J.D., and Guerra, F.A. (1999) A randomised, controlled trial of a behavioural intervention to prevent sexually transmitted disease among minority women. New Engl J Med, 340, 93–100.
2. Kamb, M.L., Fishbein, M., Douglas, J.M., Rhodes, F., Rogers, J., Bolan, G., Zenilman, J., et al. (1998) Efficacy of risk-reduction counselling to prevent human immunodeficiency virus and sexually transmitted diseases. JAMA, 280, 1161–7.
3. The National Institute of Mental Health (NIMH) Multisite HIV Prevention Trial Group. (1998) The NIMH Multisite HIV Prevention Trial: reducing HIV sexual risk behaviour. Science, 280, 1889–94.
4. Boyer, C.B., Barrett, D.C., Peterman, T.A., and Bolan, G. (1997) Sexually transmitted disease (STD) and HIV risk in heterosexual adults attending a public STD clinic: evaluation of a randomised controlled behavioural risk-reduction intervention trial. AIDS, 11, 359–67.
5. Jemmott, J.B., Jemmott, L.S., and Fong, G.T. (1998) Abstinence and safer sex HIV risk reduction interventions for African American adolescents: a randomized controlled trial. JAMA, 279, 1529–36.
7. DiClemente, R.J. and Wingood, G.M. (1995) A randomized controlled trial of an HIV sexual risk-reduction intervention for young African-American women. JAMA, 274, 1271–6.
8. Imrie, J., Stephenson, J.M., Cowan, F.M., Wanigaratne, S., Billington, A.J., Copas, A.J., French, L., French, P.D., and Johnson, A.M. (2001) A cognitive behavioural intervention to reduce sexually transmitted infections among gay men: randomised trial. Br Med J, 322, 1451–56.
9. Branson, B.M., Barrett, D.C., Peterman, T.A., and Bolan, G. (1998) Group counselling to prevent sexually transmitted disease and HIV: a randomised controlled trial. Sex Transm Dis, 25, 553–60.
10. Orr, D.P., Langefeld, C.D., Katz, B.P., and Caine, V.A. (1996) Behavioural intervention to increase condom use among high-risk female adolescents. J Pediatr, 128, 288–95.
11. Stephenson, J., Imrie, J., and Sutton, S.R. (2000) Rigorous trials of sexual behaviour interventions in STD/HIV prevention: what can we learn from them? AIDS, 14 (suppl. 3): S115–S124.
12. Shain, R.N., Perdue, S., Piper, J.M., Holden, A., Dimmitt-Champion, J., Newton, E.R., et al. (1999) Sexual risk-reduction behaviours amenable to intervention: the importance of context. 13th Meeting of the International Society for Sexually Transmitted Diseases Research. Denver CO, USA, 11–14 July 1999, Abstract No. 001.
13. Shain, R.N., Piper, J.M., Newton, E.R., Perdue, S., Champion, J., Holden, A., et al. (1997) A controlled randomised trial of a risk-reduction intervention: behaviours contributing to reduced infection rates at 12 months' follow-up. 12th Meeting of the International Society for Sexually Transmitted Diseases Research. Seville, Spain, 19–22 October 1997, Abstract O202.
14. Shain, R.N., Perdue, S.T., Piper, J., Holden, A.E.C., Champion, J.D., and Newton, E.R. (2002) Behaviours changed by intervention are associated with reduced STD recurrence—the importance of context in measurement, Sex Transm Dis, 29 (9), 520–9.
15. Perdue, S.T., Sahin, R.N., Piper, J.M., Holden, A., Dimmitt-Champion, J., Newton, E.R., et al. (1999) Risk-reduction index for behaviours amenable to intervention. 13th Meeting of the International Society for Sexually Transmitted Diseases Research. Denver CO, USA, 11–14 July 1999, Abstract No. 001.
16. Peterman, T.A., Lin, L.S., Newman, D.R., Kamb, M.L., Bolan, G., Zenilman, J., Douglas Jr, J.M., Rogers, J., and Malotte, C.K., for the Project RESPECT Study Group. (2000) Does measured behaviour reflect STD risk? An analysis of data from a randomized controlled behavioural intervention study. Sex Transm Dis, 27 (8), 446–51.
17. Peterman, T.A., Lin, L., Newman, D., Kamb, M.L., Bolan, G., Zenilman, J., et al. (1997) Association between behaviour change and STD incidence? 12th Meeting of the International Society for Sexually Transmitted Diseases Research. Seville, Spain, 19–22 October 1997, Abstract P603.
18. De Vincenzi, I., for the European Study Group on Heterosexual Transmission of HIV. (1994) A longitudinal study of human immunodeficiency virus transmission by heterosexual partners. New Engl J Med, 331 (6), 341–6.
20. O'Leary, A., DiClemente, R.J., and Aral, S.O. (1997) Reflections on the design and reporting of STD/HIV behavioural interventions research. AIDS Educ Prev, 9, 1–5.
21. Zenilman, J.M., Weisman, C.S., Rompalo, A.M., Ellish, N., Upchurch, D.M., Hook, E.W., and Celentano, D. (1995) Condom use to prevent incident STDs: the validity of self-reported condom use. Sex Transm Dis, 22 (1), 15–21.
22. Turner, C.F. and Miller, H.G. (1997) Zenilman's anomaly reconsidered: fallible reports, Ceteris Paribus, and journalArticle hypotheses. Sex Transm Dis, 24 (9), 522–7.
23. Fishbein, M. and Pequegnat, W. (2000) Evaluating AIDS prevention interventions using behavioural and biological outcome measures. Sex Transm Dis, 27 (2), 101–10.
24. Fishbein, M. and Jarvis, B. (2000) Failure to find a behavioural surrogate for STD incidence—what does it really mean? Sex Transm Dis, 27 (8), 452–5.
25. Pequegnat, W., Fishbein, M., Celentano, D., Ehrhardt, A., Garnett, G., Holtgrave, D., Jaccard, J., Schachter, J., and Zenilman, J. (2000) NIMH/APPC workgroup on a behavioural and biological outcomes in HIV/STD prevention studies: a position statement. Sex Transm Dis, 27 (3), 127–32.
26. Richens, J., Imrie, J., and Copas, A. (2000) Condoms and seat belts: the parallels and the lessons. Lancet, 355 (9201), 400–3.
27. Shain, R.N., Piper, J.M., Perdue, S., Holden, A., Champion, J.D., and Guerra, F. (2001) Behavioural intervention reduces rates of chlamydia and/or gonorrhea among minority women at 12 and 24 months' follow-up. International Congress of Sexually Transmitted Infections, ISSTDR/IUSTI. Berlin, Germany, 24–27 June 2001, p. 80.