Share this article on:

Does Measured Behavior Reflect STD Risk?: An Analysis of Data From a Randomized Controlled Behavioral Intervention Study

Peterman, Thomas A. MD, MSc*; Lin, Lillian S. PhD*; Newman, Daniel R. MA*; Kamb, Mary L. MD, MPH*; Bolan, Gail MD; Zenilman, Jonathan MD; Douglas, John M. JR. MD§; Rogers, Judy MS; Malotte, C. Kevin DrPHThe Project Respect Study Group

Original Articles

Background: Many studies measure sex behavior to determine the efficacy of sexually transmitted disease (STD)/HIV prevention interventions.

Goal: To determine how well measured behavior reflects STD incidence.

Study Design: Data from a trial (Project RESPECT) were analyzed to compare behavior and incidence of STD (gonorrhea, chlamydia, syphilis, HIV) during two 6-month intervals.

Results: A total of 2879 persons had 5062 six-monthly STD exams and interviews; 8.9% had a new STD in 6 months. Incidence was associated with demographic factors but only slightly associated with number of partners and number of unprotected sex acts with occasional partners. Many behaviors had paradoxical associations with STD incidence. After combining behavior variables to compare persons with highest and lowest risk behaviors, the STD incidence ratio was only 1.7.

Conclusion: Behavioral interventions have prevented STD. We found people tend to have safe sex with risky partners and risky sex with safe partners. Therefore, it is difficult to extrapolate the disease prevention efficacy of an intervention from a measured effect on behavior alone.

*From the National Center for STD, HIV, and TB Prevention, Centers for Disease Control and Prevention, Atlanta, Georgia; the †San Francisco Health Department, San Francisco, California; the ‡Baltimore City Health Department and Johns Hopkins University, Baltimore, Maryland; the §Colorado Department of Public Health and Environment and Denver Public Health, Denver, Colorado; the ∥New Jersey State Health Department, Newark STD Clinic, Newark, New Jersey; and ¶California State University, Long Beach, Long Beach, California

Correspondence: Thomas A. Peterman, MD, MSc, Division of HIV/AIDS Prevention, Mailstop E-46, CDC, Atlanta, GA 30333. E-mail:

Reprint requests: Technical Information Services, Mailstop E-06, CDC, Atlanta, GA 30333.

Received for publication October 14, 1999, revised January 19, 2000, accepted January 24, 2000.

MANY INTERVENTIONS HAVE been developed to change behaviors that put persons at risk for acquiring HIV via sexual contact.1–3 One study, Project RESPECT, demonstrated that brief prevention counseling sessions can reduce the incidence of new sexually transmitted infections within a year by 20%.4 A second study found three small group sessions decreased sexually transmitted disease (STD) incidence by 38%.5 Nearly all other behavior change interventions have been evaluated by considering the effect of the intervention on behavior; the effect on disease incidence has typically not been measured.6

Prevention studies that use behavior change as an outcome appear to be using behavior change as a surrogate measure for a reduction in the incidence of STD or HIV. Using HIV incidence as the outcome for a prevention study would require an enormous sample size, even in areas where HIV incidence is as high as 2 to 4%. Sexual behavior seems like a logical money saving and timesaving surrogate measure for HIV incidence because epidemiologic studies have found associations between certain behaviors and HIV infection. However, surrogate measures should be used cautiously until they are validated. Seemingly attractive surrogate measures used in other research areas have proven deceptive.7 For example, a low CD4 count is associated with progression to AIDS but the CD4 response to zidovudine therapy has not been predictive of the drug's ability to prolong life.7 How good are behaviors as surrogate measures for HIV or STD incidence? While few would doubt that, if all other things remained unchanged, increasing condom use would decrease STD incidence, it is unlikely that “all other things remain unchanged.” Although some persons increase condom use in general, others increase condom use because they have acquired a new or more risky partner. Thus, STD incidence might paradoxically increase despite an increase in condom use.

We studied data from a behavioral intervention trial that reduced the incidence of STD.4 This trial measured behavior and STD incidence at multiple points in time, allowing us to assess how well measured behavior reflected STD incidence. We hoped to identify behaviors that could be used as surrogates for STD incidence in future trials.

Back to Top | Article Outline

Materials and Methods

Project RESPECT enrolled HIV-negative heterosexuals attending large public STD clinics in five cities (Baltimore, Newark, Denver, San Francisco, and Long Beach, California).4 Participants were randomized to four arms. Three different counseling interventions were tested. All interventions were completed within four weeks of enrollment. Three of the study arms had interviews at baseline, 3, 6, 9, and 12 months and STD exams at baseline, 6, and 12 months (Figure 1). For this study we considered only the 4328 persons in the three arms with scheduled follow-up visits. Persons from these three arms were combined and evaluated to determine if behaviors were associated with STD incidence regardless of the intervention group.

Fig. 1

Fig. 1

Since many STDs were asymptomatic we considered two possible periods of observation that began and ended with an STD exam to determine if persons had acquired a new infection. All persons who had an infection diagnosed at the initial exam of the interval were treated appropriately for their infection. Persons were included in the first interval if they attended at baseline, 3, and 6 months. Persons were included in the second interval if they attended at 6, 9, and 12 months. We excluded individuals with missing questionnaires or lab visits any time during an interval except for individuals who missed a visit after they had developed an STD (an end point for that interval). Interviewers asked about behavior in the previous 3 months. Risk behavior during the 6-month interval was the average of the two interviews. For example, a person who reported one partner during the first 3-month interval and three partners during the second 3-month interval was considered to have an average of two partners per 3 months. When a person had an STD diagnosed at or before the 3-month (or 9-month) interview we used the information from that interview to represent their behavior for the 6-month interval (we did not take the average of the 3- and 6-month interviews). We counted only the first STD that an individual had in an interval.

Questions about sexual behaviors included condom use and types of partners during the 3 months preceding the interview. The number of episodes of unprotected sex was calculated by subtracting the reported number of times subjects had vaginal sex with a condom from the total reported number of times they had vaginal sex. Partner questions included number of partners; whether the partner was considered to be a “main” partner (i.e., a girlfriend, wife, or lover) or an “occasional” partner (all others); and whether this was a new sex partner.

The incident sexually transmitted infections measured for this study included chlamydia, gonorrhea, syphilis, and HIV. Chlamydia was defined as a positive Chlamydia trachomatis polymerase chain reaction from an endocervical specimen for women or a urine specimen for men. Gonorrhea was defined as a positive culture for Neisseria gonorrhoeae or (for men) gram-negative intracellular diplococci on Gram stain of a urethral swab. Syphilis was defined as a suggestive history or physical examination with supportive treponemal and nontreponemal antibody tests. HIV infection was defined as a repeatedly reactive enzyme immunoassay for HIV antibody with a positive confirmatory test.

Crude and adjusted odds ratios for risk of acquiring new STDs were computed using generalized estimating equations (GEE) using Proc GENMOD in SAS v. 6.12 (SAS Institute Inc., Cary, NC) which allows for correlated outcomes because each person could contribute the results of up to two STD tests and corresponding behavioral data. A GEE model allows for fixed and time-varying covariates. Fixed covariates in this analysis included variables such as age, study site, and gender, while time-varying covariates included variables such as number of partners and number of times they had sex without a condom. This analysis was also done using only some STDs as an outcome or using subgroups of the population. We conducted the analysis excluding chlamydia as an outcome. We confined the analysis to a single site, single race, or included only persons who did (or did not) have an infection at baseline. The results of the subgroup analyses were similar to the findings for the whole group and are not shown.

We tried to identify a combination of behavior variables from our univariate and multivariate analyses that could be followed as indicators of STD and HIV risk. To do this, we first tried to identify persons at particularly high or low risk. Persons from Baltimore and Newark (the clinics with the highest STD incidence) were used to build risk models so that we could test the models on persons from the other sites. Because we were interested in measuring the effects of changes in behavior, we examined only risk factors that the participant could control (such as use of condoms); demographic variables were not included in the risk scores.

Back to Top | Article Outline


Among 4328 subjects randomized to intervention groups with scheduled follow-up, 2879 contributed 5062 six-month periods of observation that included an STD exam before and after the period. The cumulative incidence of new STDs was 8.9% in six months (10.5% in the first period and 7.0% in the second). The incident STDs were most often chlamydia (233), gonorrhea (163), or both (30), and less often syphilis (16), HIV (4), syphilis and chlamydia (1), or HIV and gonorrhea (1).

The strongest predictors of new STDs were factors that could not easily change, including study city, race, and age (Table 1). Persons were more likely to acquire a new infection if they were: from Newark (adjusted odds ratio [ORa] = 2.3) or Baltimore (ORa = 1.6 compared to San Francisco); black (ORa = 2.2 compared to all other races); and <20 years old (ORa = 4.3 compared to persons aged ≥35). Gender was not associated with likelihood of new infection. College graduates were less likely to acquire a new STD compared to persons who did not graduate from high school (ORa = 0.51).



Some measured behaviors were slightly associated with cumulative incidence of new infection. New infection was more common among persons with 3 or more sex partners (ORa = 1.9 compared to 1 partner), but was not related to having a new sex partner. There was no association between STD incidence and the number of unprotected sex acts with the main partner. However, new infection was more common among persons who had 6 or more episodes of unprotected sex with an occasional partner (ORa = 1.9 compared to persons who had no unprotected sex with their occasional partner).

Many other variables had seemingly paradoxical associations with the incidence of new STDs (Table 2). The cumulative incidence of new infection was 8.9%. Persons were more likely than others to acquire an infection if they reported having refused sex because they were concerned about HIV infection (10.3%) or because their sex partner would not use a condom (10.7%). Persons who reported having sex while high were less likely to have a new infection (7.9%) than those who did not.



Unprotected sex was more common with low-risk partners than with high-risk partners. Among persons who had sex with both main and occasional partners, at least one episode of unprotected sex was reported more commonly with the main partner (64%) than with the occasional partner (40%). Among persons with only one partner, at least one episode of unprotected sex was more commonly reported by persons who had been with their partner for 90 days or more (74% of 2680) compared with persons who had been with their partner for 30 days or less (36% of 59).

In Newark and Baltimore, 14.0% of the participants acquired a new infection during the 6-month interval. By combining variables we were able to define a reasonably large group with a cumulative incidence of new STDs as low as 12.1% (n = 747). These persons reported only one sex partner, who was a main partner who the respondent thought was not having sex with others, and who had been a sex partner for at least 3 months). The highest risk STD incidence we could identify for a large group by combining behavioral variables was 20.8% (n = 120). These persons reported three or more sex partners and some unprotected sex with an occasional partner. The STD incidence ratio for the persons at highest risk compared with the persons at lowest risk was 1.7. We did not test this model in the other study sites because it was a weak predictor of STD even at the sites used to build the model.

Back to Top | Article Outline


While the trial showed that behavior change interventions can reduce the risk of acquiring STDs,4 our analysis suggests that the incidence of new STD was not very well reflected by the behaviors we measured despite an extensive questionnaire on sexual risk behavior. We did not identify any behaviors that (used alone) could be surrogates for STD incidence in future trials. The problem appears to be a measurement issue. The risk of acquiring a new infection depends on the type of sexual contact, and whether the partner is infected. One obvious problem in measuring this risk is not knowing if the partner was infected. Regardless of behavior, there is no risk of acquiring infection from uninfected partners. Including persons whose partners are uninfected creates a lack of precision in measuring associations of behaviors and STD risk and reduces the power of the study to detect an impact on disease rates by measuring behavior.

A more important measurement problem is confounding by different behaviors because behaviors are related to the infection status of the partner. Persons appear to be more likely to have safe sex with risky partners and risky sex with safe partners. Many people use condoms with new partners but stop using them when they are in a long-term relationship. We found persons were more likely to have at least one episode of unprotected sex with their main partner (64%) than with their occasional partner (40%). Given this conditional nature of condom use, it is possible that some persons who increased condom use were more likely to acquire an infection than persons who decreased condom use. Analyses of risk factors can only adjust for this type of confounding when the infection status of each partner is known. It is possible that more detailed information on partners would increase the ability to estimate infection status of the partners. However, researchers are unlikely to have enough information to adjust for the likelihood of infection in each partner of every person in a study.

Some individuals who acquired a new infection reported having no sex partner during the interval. This might represent reporting error, lab error, or treatment failure. A detailed review of these individuals provided few clues about the likely explanation. Project RESPECT included an extensive quality assurance protocol,4 so we expect that other studies that measure behavior and STD would have similar findings.

We had hoped that we could develop a risk score based on self-reported behaviors that would be sensitive to changes in a person's risk over time.8 A valid behavior risk score would be less costly and far easier to use than repeated measures of STD or HIV in evaluations of STD/HIV prevention interventions. However, the group we identified as being at highest risk was only about twice as likely to have acquired an infection as the group at lowest risk. While a relative risk of two indicates some correlation with STD, to be a valid surrogate change in measured behavior would have to fully capture the net effects of the intervention on STD.7

Our inability to identify a valid risk score is particularly disturbing because very few behavioral intervention studies have used STD or HIV as the outcome.1–6 Most early studies measured attitudes or knowledge while recent studies have measured behaviors. The disease prevention efficacy of interventions that have been tested using these measures is uncertain given the uncertain association of psychosocial and behavioral measures with STD or HIV.

Other studies have examined the association of STD and unprotected sex. Zenilman et al9 found little association between condom use and incident infections in a cohort of 598 persons in Baltimore. Their suggestion that self-reported condom use may be subject to substantial reporting bias led to a heated debate with authors presenting evidence that behavior has been accurately reported in other studies and has been associated with risk of infection.10,11 Studies that found associations between condom use and infection tend to be studies where condom use did not change with infection status of the partner. For example, studies of HIV-discordant couples (in which one partner is infected with HIV and the other is not) have found fewer transmissions occur among couples that report consistent condom use compared to couples who do not use condoms.12,13 A study of prostitutes in Cameroon found a dose-response relationship with more HIV infections in prostitutes reporting more unprotected sex.14 A reanalysis of Zenilman's data supported the finding of a lack of association between condom use and STD and raised the suggestion that condom use is a marker for sex with risky partners.15 Subjects may have used condoms more often with risky partners and less often or not at all with partners who they thought were unlikely to be infected. Using behavior as a surrogate for STD risk may be particularly problematic when studying persons who might vary their use of condoms with partners of varying risk.

Our study has some limitations and strengths. 1) We studied HIV-negative heterosexuals attending STD clinics. Although this is a population of interest because they are at high risk for STD and HIV, it is a bit different from persons identified in other high-risk settings. Many were attending because they had acquired an STD from their current partner, so, for them, finding a new partner might decrease their risk. 2) We measured behavior over 3-month intervals but new infection at 6-month intervals, requiring some combination of data that could introduce (probably random) error. 3) Our study included only some of the possible approaches to assessing behavior. Computer-assisted interviews might reduce social desirability bias. It is possible that other measures (e.g., information about partner treatment5 or a partner's likelihood of being infected) would more closely reflect the risk of incident STD. 4) We used STD as our outcome, even though we were most interested in preventing HIV, the STD with the highest case-fatality ratio. Trends in one STD could be quite different from trends in other STDs in general population studies.16 However, we believe randomized controlled trials of behavioral interventions will eliminate many potential confounders (such as temporal trends) and may make STD a useful surrogate for HIV. 5) A major strength of our study was the use of repeated examinations to detect STDs. In Project RESPECT persons with a scheduled follow-up exam were much more likely to be diagnosed as having gonorrhea or syphilis after 6 months than persons with passive follow-up who returned when they were symptomatic (5.6% versus 3.2% for men and 4.5% versus 1.6% for women). Many STDs are asymptomatic, so studies that do not directly measure new infection will contain errors of unknown direction and magnitude.

Efficacy evaluations of disease prevention interventions must be able to quantify the incidence of disease in different groups. The very rich data from Project RESPECT provided an unusual opportunity to assess the potential use of reported behavior as a surrogate for disease incidence. Behavior change was responsible for the 20% decrease in STD seen in the counseling arms of this randomized trial because the only difference between the arms was the behavioral intervention. However, our analysis found only weak associations between the behaviors we measured and incident infection. Persons were more likely to practice safe sex with a risky partner, and risky sex with a safe partner. The risk of infection is related to both behavior and infection status of the partner. Without knowing the infection status of each partner, we cannot reliably estimate the incidence of STD. Therefore, we conclude that it is difficult to extrapolate the disease prevention efficacy of an intervention from a measured effect on behavior alone.

Back to Top | Article Outline


1. Higgins DL, Galavotti C, O'Reilly KR, et al. Evidence for the effects of HIV antibody counseling and testing on risk behaviors. JAMA 1991; 266:2419–2429.
2. Choi K-H, Coates TJ. Prevention of HIV infection. AIDS 1994; 8:1371–1389.
3. Wolitski RJ, MacGowan RJ, Higgins DL, Jorgensen CM. The effects of HIV counseling and testing on risk-related practices and help-seeking behavior. AIDS Educ Prev 1997; 9(Suppl B):52–67.
4. Kamb ML, Fishbein M, Douglas JM Jr, et al. Efficacy of risk-reduction counseling to prevent human immunodeficiency virus and sexually transmitted diseases: a randomized controlled trial. JAMA 1998; 280:1161–1167.
5. Shain RN, Piper JM, Newton ER, et al. A randomized, controlled trial of a behavioral intervention to prevent sexually transmitted disease among minority women. N Engl J Med 1999; 340:93–100.
6. Aral SO, Peterman TA. Do we know the effectiveness of behavioral interventions? Lancet 1998; 351(suppl III):33–36.
7. Fleming TR, DeMets DL. Surrogate end points in clinical trials: are we being misled? Ann Intern Med 1996; 125:605–613.
8. Susser E, Desvarieux M, Wittkowski KM. Reporting sexual risk behavior for HIV: a practical risk index and a method for improving risk indices. Am J Public Health 1998; 88:671–674.
9. Zenilman JM, Weisman CS, Rompalo AM, et al. Condom use to prevent incident STDs: the validity of self-reported condom use. Sex Transm Dis 1995; 22:15–21.
10. Weir SS, Feldblum PJ. Condom use to prevent incident STDs (letter). Sex Transm Dis 1996; 23:76–77.
11. Galavotti C, Cabral R, Beeker C. Condom use to prevent incident STDs (letter). Sex Transm Dis 1996; 23:77–79.
12. Saracco A, Musicco M, Nicolosi A, et al. Man-to-woman sexual transmission of HIV: longitudinal study of 343 steady partners of infected men. J Acquir Immune Defic Syndr Hum Retrovirol 1993; 6:497–502.
13. de Vincenzi I. A longitudinal study of human immunodeficiency virus transmission by heterosexual partners. N Engl J Med 1994; 331:341–346.
14. Weir SS, Feldblum PJ, Zekeng L, Roddy RE. The use of nonoxynol-9 for protection against cervical gonorrhea. Am J Public Health 1994; 84:910–914.
15. Turner CF, Miller HG. Zenilman's anomaly reconsidered: Fallible reports, ceteris paribus, and other hypotheses. Sex Transm Dis 1997; 24:522–527.
16. Hamers FF, Peterman TA, Zaidi AA, Ransom RL, Wroten JE, Witte JJ. Syphilis and gonorrhea in Miami: similar clustering, different trends. Am J Public Health 1995; 85:1104–1108.
© Copyright 2000 American Sexually Transmitted Diseases Association