ARAL, SEVGI O. PhD; PETERMAN, THOMAS A. MD, MSc
Centers for Disease Control and Prevention (CDC), Atlanta, Georgia
The authors thank Patricia Jackson for her outstanding support in preparing this manuscript.
Correspondence: Sevgi O. Aral, PhD, Associate Director for Science, Division of STD Prevention, CDC, 1600 Clifton Road, NE, MS-E02, Atlanta, GA 30333. Reprints: CDC, NCHSTP, Information Technology Services, 1600 Clifton Road, MS-E06, Atlanta, GA 30333.
Received for publication May 17, 2002 and accepted June 4, 2002.
IN STUDIES OF SEXUALLY TRANSMITTED DISEASE (STD) and HIV epidemiology and prevention, it is tempting to focus on the measurement of risk behaviors rather than on the measurement of STDs and HIV themselves. Particularly in intervention studies, measuring biomedical outcomes can be complicated and costly. An intervention study where the goal is to have 80% likelihood of finding a difference between intervention and control groups at P < 0.05, requires 313 persons per arm if the outcome of interest is an increase in condom use from 20% to 30%. On the other hand, if the outcome of interest is a decrease from 10% to 8% in the gonorrhea reinfection rate in an STD clinic population, 3,312 persons would be required per arm. A decrease in HIV incidence from 0.5% to 0.4% would require 72,307 persons per arm.
Measuring risk behaviors in lieu of STDs and HIV, however, is problematic. A number of observational studies that explored the relationship between behavioral and biomedical outcomes failed to show a strong relationship between behaviors and acquisition of sexually transmitted infection. 1–9 Moreover, some behavioral intervention studies also failed to show a strong relationship between behaviors and STD acquisition. 10–12 The debate surrounding the issue of discordance between behavioral and biomedical outcome measures has focused on the imperfections in behavioral outcome measurement. 1–6,13 In this issue of Sexually Transmitted Diseases, Shain and colleagues 14 present results of additional analyses of data from project SAFE. 15 They put forth the compelling argument that incorporating context into the conceptualization of behavioral measurements and considering several behaviors simultaneously may resolve the inconsistencies between behavioral and biomedical outcomes.
Current approaches to STD epidemiology recognize at least three distinct components of STD transmission dynamics: transmissibility of infection upon exposure between an infected and an uninfected person, likelihood of sexual exposure between infected and uninfected individuals, and duration of infection among infected persons. 16 Consistent with this recognition, behavioral interventions aimed at reducing STD transmission focus on: (1) behaviors related to transmission of infection between infected and uninfected partners, including condom use (or unsafe sex), sexual practices such as anal sex or dry sex, douching practices and abstaining from sex until the completion of therapy; (2) behaviors related to sexual exposure between infected and uninfected persons, including sex with infected or untreated partners, lack of mutual monogamy, rapid partner turnover, number of partners, number of new partners, and choice of partners; and (3) duration of infection or infectiousness, such as the timely seeking of healthcare and treating of partners.
Despite clear recognition of the three distinct components of STD transmission dynamics, in general, neither observational nor intervention studies of the relationship between behavioral and biomedical outcomes of STD transmission adequately differentiate among these three components. Such studies often rely on the measurement of relevant behaviors and statistical analyses as they attempt to relate behaviors to the overall outcome—acquisition of a sexually transmitted infection. While some intervention studies have incorporated aspects of the three components into the intervention 15,17 and measurement of behavior, too often these issues are not taken into consideration in study design.
Specification of the behaviors, the outcomes, and the sexually transmitted infection in question may have important implications for study design, study populations, and follow-up periods.
Transmission Probability and Behaviors
Behaviors related to transmissibility can be evaluated effectively only in discordant partners. 18–20 In project SAFE the strongest association between behavioral and biomedical outcomes (new infection) was found for persons who had sex with an untreated partner. 14 In the 1980s a number of partner studies examined the probability of heterosexual transmission of HIV. 21 Transmission probabilities of other sexually transmitted infections generally have not been examined through discordant partner studies.
One behavior that is closely related to transmission probability is condom use. Many studies of the association between use of condoms and acquisition of sexually transmitted infections have yielded paradoxical results. People who use condoms are found to be at least as likely to acquire infections as people who do not use condoms. People tend to use condoms with partners who they think are risky but not with partners who they consider to be safe. Consequently, increased condom use may be a marker either of decreased STD transmission risk or of increased likelihood of STD exposure. Outside the context of discordant partner studies, it may be almost impossible to tease out the effects of interaction between behaviors and risk, and to study the relationship between behaviors and transmissibility effectively.
Specification of the sexually transmitted infection under consideration may have implications for the required period of observation and sexual activity level of the study population. Infections caused by highly infectious pathogens such as gonorrhea may be studied in moderately sexually active discordant couples over relatively short observation periods. Infections caused by pathogens of low infectivity such as herpes may necessitate the enrollment of highly sexually active couples over longer observation periods.
Risk of STD Exposure and Behaviors
The association between behavior and exposure to infected sex partners may be effectively studied among partnered and nonpartnered persons. Behaviors of both the respondent and his/her partners are critical determinants of exposure risk. A large number of behavioral and epidemiologic indicators are used in attempts to assess partners’ infection status. Parameters including number of sex partners, number of new partners, presence of concurrent partnerships, the gap between sex partners, partners’ number of partners, and risk status of partners’ partners have all been used to assess sex partners’ infection status.
The particular sexually transmitted infection under consideration may have important implications for the study of the association between risk of STD exposure and related behaviors. The association may be easier to study when the sexually transmitted infection under consideration is moderately prevalent in a population. For sexually transmitted infections of very low prevalence, e.g., chancroid in the United States, even very large numbers of new partners or high levels of concurrency may not be associated with risk of STD exposure. Conversely, for sexually transmitted infections of very high prevalence, e.g., human papillomavirus, even very small numbers of sex partners or low levels of other risk-taking activities may be associated with risk of STD exposure. Thus, in choosing the study population it is important to consider the specific STD being studied; both the distributions of relevant behaviors and the distribution of infection are important in this context.
Measurement Error and the Relationship Between Behaviors and Biomedical Outcomes
Measurement of sexual behaviors almost by definition depends on self-reports of respondents. Consequently, sampling procedures, representativeness of the sample of respondents, response or participation rates, item specific response rates, bias in reporting and recall may all contribute to errors in measurement. 18 Systematic measurement errors in study design and analysis may have an important impact on inferences of association. The role of differential misclassification or systematic measurement error is widely recognized and, in most studies, care is taken to minimize this type of bias. 18 The effects of random measurement error, or nondifferential misclassification, on epidemiologic inference often receive less attention. 22 Nondifferential misclassification of exposures and outcomes will lead to an attenuation of the resulting measure of association.
Sexual behavior measurement involves the measurement of and cross-adjustment for several sexual behaviors that are often related to each other. Random measurement error in potential confounding variables may influence the inferences made from study results. 22 Nondifferential misclassification of a dichotomous confounding variable may lead to residual confounding and the false appearance of statistical interaction. 23 Random measurement error in confounders that are continuous variables may bias the adjusted measure of association unpredictably. 24,25 Such misclassification is of greatest concern when the exposure-disease association is relatively weak compared to the exposure-confounder and outcome-confounder relation. 26 Small random errors may have major effects on adjusted measures of association, and multivariate analyses may compound the unpredictability of the effects of misclassification. 27 Random measurement error may play an important role in studies that attempt to adjust for sexual behaviors such as numbers and types of sex partners, condom use practices, or number of sex acts. All STD studies that focus on sexual behavior have to adjust for some sexual behaviors as they attempt to measure the effects of others.
The relationship between behavior and biomedical parameters in the epidemiology of sexually transmitted infections is complex. The aggregated approach, which tends to combine behaviors related to transmission and behaviors related to exposure as “sexual risk behavior,” and all sexually transmitted infections as “biomedical outcome,” leads to the persistence of the behavioral/biomedical outcome conundrum. Perhaps the time has come for a more analytical, stratified approach.
1. 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.
2. Weir SS, Feldblum PJ. Condom use to prevent incident STDs. Sex Transm Dis 1996; 23:76–77; discussion 79–82.
3. Galavotti C, Cabral R, Beeker C. Condom use to prevent incident STDs. Sex Transm Dis 1996; 23:77–79; discussion 79–82.
4. Turner CF, Miller HG. Zenilman's anomaly reconsidered: fallible reports, ceteris paribus, and other hypotheses. Sex Transm Dis 1997; 24: 522–527.
5. Fishbein M, Jarvis B. Failure to find a behavioral surrogate or STD incidence—what does it really mean? Sex Transm Dis 2000; 27: 452–455.
6. Fishbein M, Pequegnat W. Evaluating AIDS prevention interventions using behavioral and biological outcome measures. Sex Transm Dis 2000; 27: 101–110.
7. Pequegnat W, Fishbein M, Celentano D, et al. NIMH/APPC workgroup on behavioral and biological outcomes in HIV/STD prevention studies: a position statement. Sex Transm Dis 2000; 27: 127–132.
8. Richens J, Imrie J, Copas A. Condoms and seat belts: the parallels and the lessons. Lancet 2000; 355: 400–403.
9. Schachter J. Biologic versus behavioral endpoints—the duet continues. Sex Transm Dis 2000; 27: 456–457.
10. Peterman TA, Lin LS, Newman DR, et al. Does measured behavior reflect STD risk? An analysis of data from a randomized controlled behavioral intervention study. Project RESPECT Study Group. Sex Transm Dis 2000; 27: 446–451.
11. The National Institute of Mental Health (NIMH) Multisite HIV Prevention Trial Group. The NIMH Multisite HIV Prevention Trial: reducing HIV sexual risk behavior. Science 1998; 280: 1889–1894.
12. Boyer CB, Barrett DC, Peterman TA, Bolan G. Sexually transmitted disease (STD) and HIV risk in heterosexual adults attending a public STD clinic: evaluation of a randomized controlled behavioral risk-reduction intervention trial. AIDS 1997; 11: 359–367.
13. Aral SO, Peterman TA. Measuring outcomes of behavioural interventions for STD/HIV prevention. Int J STD AIDS 1996; 7 (Suppl 2): 30–38.
14. Shain RN, Perdue ST, Piper JM, et al. Behaviors changed by intervention are associated with reduced STD recurrence: the importance of context in measurement. Sex Transm Dis 2002; 29: 520–529.
15. 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.
16. Aral SO, Holmes KK. Social and behavioral determinants of the epidemiology of STDs: industrialized and developing countries. In: Holmes KK, Sparling PF, Mardh P-A, et al, eds. Sexually Transmitted Diseases, 3rd ed. New York: McGraw Hill, 1999: 39–76.
17. Kamb ML, Fishbein M, Douglas JM, et al. Efficacy of risk-reduction counseling to prevent human immunodeficiency virus and sexually transmitted diseases: a randomized controlled trial. Project RESPECT Study Group. JAMA 1998; 280: 1161–1167.
18. Fenton KA, Johnson AM, McManus S, Erens B. Measuring sexual behaviour: methodological challenges in survey research. Sex Transm Inf 2001; 77: 84–92.
19. Aral SO, Peterman TA. STD diagnosis and treatment as an HIV prevention strategy. In: O'Leary A, ed. Beyond Condoms: Alternative Approaches to HIV Prevention. New York: Kluwer Academic/Plenum Publishers, 2002: 77–90.
20. Warner L, Newman D, Peterman TA, et al. Studying condom effectiveness for sexually transmitted disease (STD) prevention: the importance of knowing partner infection status (abstract B9B). Presented at the 2002 National STD Prevention Conference, San Diego, California, March 4–7, 2002.
21. De Vincenzi I, for the European Study Group on Heterosexual Transmission of HIV. A longitudinal study of human immunodeficiency virus transmission by heterosexual partners. N Engl J Med 1994; 331: 341–346.
22. Myer L, Morroni C. Implications of random measurement error in studies adjusting for sexual behaviours. Sex Transm Inf 2001; 77: 302.
23. Cox B, Elwood JM. The effect on stratum-specific odds ratios of non-differential misclassification of a dichotomous covariate. Am J Epidemiol 1991; 133: 202–207.
24. Wacholder S. When measurement errors correlate with truth: surprising effects of non-differential misclassification. Epidemiology 1995; 6: 157–161.
25. Brenner H. Bias due to non-differential misclassification of polytomous confounders. J Clin Epidemiol 1993; 46: 57–63.
26. Marshall JR, Hastrup JL. Mismeasurement and the resonance of strong confounders: uncorrelated errors. Am J Epidemiol 1996; 143: 1069–1078.
27. Phillips AN, Davey G. Bias in relative odds estimation owing to imprecise measurement of correlated exposures. Stat Med 1992; 11: 953–961.