Secondary Logo

Journal Logo


Baseline Infection With a Sexually Transmitted Disease Is Highly Predictive of Reinfection During Follow-up in Malagasy Sex Workers

Turner, Abigail Norris PHD*; Feldblum, Paul J. PHD; Hoke, Theresa Hatzell PHD

Author Information
doi: 10.1097/OLQ.0b013e3181d70a03
  • Free

Mathematical models of sexually transmitted infection (STI) transmission emphasize the role of “core” groups: those with elevated numbers of new, multiple, or concurrent partnerships.1 Sex workers have been identified as a critical core group,2,3 and both high prevalence and high incidence of STIs in sex workers have been documented.4,5 While planning for a large-scale randomized trial in Madagascar to assess the STI-prevention efficacy of a diaphragm plus candidate vaginal microbicide in female sex workers, we analyzed data originally collected in a randomized trial of male and female condom promotion in the same population.6,7 We sought to identify characteristics of sex workers most strongly associated with acquisition of incident STI, to target the future trial's recruitment efforts to the population of women most likely to experience these outcomes. In particular, we wanted to assess whether prevalent STI at baseline was associated with increased risk of incident STI during the 18-month follow-up period.


Our data source is the “Impact of Female and Male Condom Promotion Trial” conducted in Madagascar from 2001–2003. Detailed study methods and results have been published.6,7 Briefly, 1000 female sex workers in Antananarivo and Tamatave, Madagascar were enrolled in a randomized trial that compared the effect of 2 counseling strategies for male and female condom promotion. For 6 months, control women received peer-based counseling (the standard of care) on male condom use, while the intervention group received peer-based counseling supplemented with clinic-based counseling. From months 7 through 18, women were counseled on use of both male and female condoms. Every 2 months the participants underwent interviews documenting their sexual behavior. At baseline and the 6-, 12-, and 18-month visits they also received STI testing (Chlamydia trachomatis [CT], Neisseria gonorrhoeae [GC], and Trichomonas vaginalis [TV]). GC and CT testing was by ligase chain reaction on urine specimens (Abbott LCx Probe System, Abbott Laboratories, Abbott Park, IL) and TV testing was by InPouch culture (BioMed, San Jose, CA). All participants were treated presumptively for gonococcal and chlamydial infection with ciprofloxacin and azithromycin, respectively, at baseline and at 6- and 12-month visits. At those visits, treatment was given 1 week later for trichomoniasis (metronidazole) for women testing positive. At the 18-month visit, the participants were treated in accordance with national guidelines for syndromic management of STIs.8 In addition, when women reported STI signs or symptoms at interim visits (2, 4, 8, 10, 14, and 16 months), they were examined and treated syndromically in accordance with national guidelines, but not tested.


We conducted several analyses to measure the association between prevalent STI at baseline and incident STI during the follow-up period. First, we examined the unadjusted Kaplan-Meier plot9 and log-rank test,10 comparing STI-free survival time by baseline STI status. Second, we computed unadjusted incidence rates (IRs) for each STI, separately for women with prevalent STI at baseline and those who were STI free at baseline. Third, we used Cox proportional hazards models to estimate unadjusted and adjusted hazard ratios (HRs) and 95% confidence intervals (CIs) comparing the hazard of incident STI for women with and without STI at baseline.9

We restricted this retrospective cohort analysis to women who completed at least 1 follow-up visit with valid STI results and had nonmissing STI data at baseline. The exposure for each analysis was baseline STI status, modeled separately as individual STIs and as a combined variable capturing any chlamydial, gonococcal, or trichomonal infection at baseline (including women with more than one infection). Our outcome measures similarly captured each STI separately as well as a combined “any STI” outcome.

The outcome for each analysis was the first incident infection with each of the STIs under investigation. An infection was considered the first incident infection if the participant had tested negative for that STI at all previous follow-up visits. For each analysis, women were censored after the first incident infection with the specific STI under consideration.

Manual backward elimination was used to build the Cox models. We began with fully adjusted models controlling for site, randomization group, age, marital status, sex with nonpaying partners in the past week, the proportion of sex acts protected by condoms with clients in the past month, and the number of clients in the past week. We assessed the proportional hazards assumption by creating interactions between each variable and time, and used Cox tests11 to evaluate the significance of the interaction term. Site and randomization group were retained in all models because of a priori hypotheses about their importance. Covariates were retained only if removal resulted in ≥10% change in the association between prevalent STI at baseline and incident STI during follow-up. Any variable identified as a confounder in any of the model-building processes was retained as an adjustment variable for all models. All statistical analyses were conducted in SAS (Version 9.2, Cary, NC).

A total of 1000 women enrolled in the trial, of whom 90.1% (n = 901) returned for the 6-month follow-up visit, 86.3% (n = 863) returned at 12 months, and 81.8% (n = 818) returned at 18 months.

Nearly one-quarter of participants (n = 224, 22.4%) had gonococcal infection at baseline, 147 (14.7%) had chlamydial infection, and 356 (35.6%) had trichomonal infection. More than half (n = 519, 51.9%) had at least 1 of the 3 infections at baseline. Sex workers with prevalent STI at baseline tended to be younger (<25 years), from Antananarivo, and to identify as single. About half of the women, regardless of baseline STI status, reported using a condom at the last sex act with a client. Of women who reported sex with a nonpaying partner, 80.6% did not use a condom at last sex with that partner (Table 1).

Selected Demographic Characteristics by Baseline Sexually Transmitted Infection Status

STI IRs during follow-up were considerably higher among sex workers with a prevalent STI detected at baseline, considering both individual and combined STI outcomes. Among women with CT at baseline, the IR for CT during follow-up was 50.5 per 100 person-years (PY) (95% confidence interval [CI]: 38.7, 62.4); the IR for chlamydial infection for women without CT at baseline was 13.7/100 PY (95% CI: 11.5, 16.0). The IR for gonococcal infection for women with GC at baseline was 49.8/100 PY (95% CI: 40.1, 59.4) versus 20.0/100 PY (95% CI: 17.0, 22.9) for women without GC at baseline. The IR for trichomonal infection for women with TV at baseline was 89.6/100 PY (95% CI: 78.4, 100.8) compared with an IR of 36.8/100 PY (95% CI: 32.2, 41.4) for women without TV at baseline. Considering the 3 infections together, the IR for women with an STI at baseline was 109.5/100 PY compared to 45.5/100 PY for women without an STI at baseline (Table 2).

Unadjusted Incidence Rates for Incident Sexually Transmitted Infection During Follow-up, by Baseline Sexually Transmitted Infection Status

The unadjusted Kaplan-Meier plot showed that women who had CT, GC, or TV at baseline were significantly more likely to acquire CT, GC, or TV during follow-up (log-rank P < 0.0001). Cox models showed similar results (Table 3). In unadjusted models, compared to women who were STI-negative at baseline, women with CT, GC, or TV at baseline had increased risk of acquisition of any incident STI during follow-up (HR: 2.7, 95% CI: 2.2, 3.1). They had similarly and significantly increased risk of incident CT, GC, and TV modeled separately. HRs for homologous reinfection were generally stronger than HRs using the combined “any STI” exposure variable (Table 3). For example, women with CT at baseline had a risk of incident CT during follow-up nearly 4 times as high as women without CT at baseline (HR: 3.9, 95% CI: 2.9, 5.2). No variable violated the proportional hazards assumption and only age, randomization group, and site were retained as covariates in final multivariable models; other variables considered for adjustment did not meet the criteria for retention in the models. The adjusted estimates were largely unchanged from unadjusted ones (Table 3).

Unadjusted and Adjusted Hazard Ratios for the Effect of Baseline Sexually Transmitted Infection on Incident Sexually Transmitted Infection During Follow-up

Female sex workers who had prevalent chlamydial, gonococcal, or trichomonal infection at enrollment into a randomized trial in Madagascar were 2 to 4 times as likely to become infected during follow-up, compared to women without STIs at baseline. The finding that STI-infected individuals have high rates of reinfection is not new,12–14 and these results underscore the need for intensive intervention with individuals testing positive for STIs. One conceivable intervention to combat a potential infection reservoir is testing and treatment of primary, nonpaying partners, with whom condom use is known to be infrequent.7 However, although many women reported sex with nonpaying partners in the past week, adjustment for both condom use and sex with nonpaying partners did not measurably change our estimates. Interventions targeting the male clients of female sex workers have had some limited success,15–17 but the logistical challenges of locating, recruiting, and retaining clients for research studies are not trivial. Another possible intervention in this population may be periodic presumptive therapy, as was built into this trial, although at more frequent intervals. A trial of female sex workers in Nairobi found that monthly oral administration of 1 g of azithromycin versus placebo significantly decreased the incidence of CT, GC, and TV.18 However, even while the cost of medications continues to decrease, the operational, financial, and (where sex work is criminalized) legal challenges associated with such programs are substantial.

Surprisingly, our multivariable analyses did not reveal any strong confounders of the associations of interest (aside from age, which we previously identified as strongly predictive of incident STI in this population19). Our findings agree with recent secondary analyses of four microbicide trials, in which the only consistent, significant factors associated with incident STI during follow-up were age and baseline STI status.20

Our analyses have limitations. Most significantly, many measures of potentially confounding variables were self-reported and therefore susceptible to reporting biases. Other studies, including one using the same study participants analyzed here, that attempted to validate self-reported unprotected sex by testing for prostate-specific antigen in vaginal specimens found poor agreement.21 Although our main exposure and outcome variables were not self-reported, our STI detection methods do not have 100% sensitivity and 100% specificity, so some women's exposure or outcome status may have been misclassified. STI treatment failure could also lead to infections being erroneously identified as incident, when they were truly lingering baseline infections.

Aside from the higher STI incidence among women with STIs at baseline, the absolute IRs in all women in this study population were exceedingly high, nearly 50 infections per 100 PY among those without STI at baseline and more than double that rate for women with STI at baseline. This is despite provision of condoms, safer sex counseling, and repeat STI testing and treatment. True STI incidence may have been even higher than the rates observed in our study. We asked women at each visit (including interim visits) whether they had sought treatment for STI symptoms since the previous visit. Between 11.1% and 19.8% of participants, depending on the visit, reported at interim visits that they had had STI symptoms. A subset of these (between 9.3% and 18.7% of those reporting symptoms) said they had taken medication to treat STI symptoms since the previous visit. If women had actually been STI infected and also received effective therapy between visits, the true STI incidence may have been even higher than we report here.

To our knowledge STI incidence of other high-risk groups in Madagascar has not been reported in the literature, although STI prevalence among rural men in Madagascar (combined prevalence of GC, CT, and Mycoplasma genitalium of 11%) and STD clinic patients in Antananarivo (prevalence of GC, CT, and TV of 32%, 40%, and 19%, respectively) has been documented.22–23 The study described earlier of Nairobi sex workers reported IRs in the placebo group of 13/100 PY, 15/100 PY, and 20/100 PY, for GC, CT, and TV, respectively.18 To our knowledge, the IRs among these Malagasy sex workers are among the highest reported in the literature. Clearly, STI prevention efforts used during this study were not sufficient, and alternative approaches for reducing risk are needed.


1. Anderson R. Transmission dynamics of sexually transmitted infections. In: Holmes KK, Sparling PF, Mårdh P, et al, eds. Sexually Transmitted Diseases, 3rd ed. New York, NY: McGraw-Hill, 1999:25–37.
2. Mustikawati DE, Morineau G, Nurhayati, et al. Sexual risk taking, sexually transmitted infections and HIV prevalence among four “high-risk” occupational groups of Indonesian men. Sex Transm Infect 2009; 85:391–396.
3. Gibney L, Saquib N, Metzger J, et al. Human immunodeficiency virus, hepatitis B, C and D in Bangladesh's trucking industry: Prevalence and risk factors. Int J Epidemiol 2001; 30:878–884.
4. Paris M, Gotuzzo E, Goyzueta G, et al. Prevalence of gonococcal and chlamydial infections in commercial sex workers in a Peruvian Amazon city. Sex Transm Dis 1999; 26:103–107.
5. Shahmanesh M, Cowan F, Wayal S, et al. The burden and determinants of HIV and sexually transmitted infections in a population-based sample of female sex workers in Goa, India. Sex Transm Infect 2009; 85:50–59.
6. Feldblum PJ, Hatzell T, Van Damme K, et al. Results of a randomised trial of male condom promotion among Madagascar sex workers. Sex Transm Infect 2005; 81:166–173.
7. Hoke TH, Feldblum PJ, Damme KV, et al. Randomised controlled trial of alternative male and female condom promotion strategies targeting sex workers in Madagascar. Sex Transm Infect 2007; 83:448–453.
8. Behets FM, Rasolofomanana JR, Van Damme K, et al. Evidence-based treatment guidelines for sexually transmitted infections developed with and for female sex workers. Trop Med Int Health 2003; 8:251–258.
9. Kaplan EL, Meier P. Nonparametric estimation from incomplete observations. J Am Stat Assoc 1958; 53:457–481.
10. Selvin S. Statistical Analysis of Epidemological Data, 3rd ed. New York, NY: Oxford University Press, 2004.
11. Kleinbaum DG, Klein M. Survival Analysis: A Self-Learning Text, 2nd ed. New York, NY: Springer, 2005.
12. Burstein GR, Zenilman JM, Gaydos CA, et al. Predictors of repeat Chlamydia trachomatis infections diagnosed by DNA amplification testing among inner city females. Sex Transm Infect 2001; 77:26–32. PMCID: 1758317.
13. De P, Singh AE, Wong T, et al. Predictors of gonorrhea reinfection in a cohort of sexually transmitted disease patients in Alberta, Canada, 1991–2003. Sex Transm Dis 2007; 34:30–36.
14. Miller PJ, Law M, Torzillo PJ, et al. Incident sexually transmitted infections and their risk factors in an Aboriginal community in Australia: A population based cohort study. Sex Transm Infect 2001; 77:21–25. PMCID: 1758324.
15. Barrington C, Latkin C, Sweat MD, et al. Talking the talk, walking the walk: Social network norms, communication patterns, and condom use among the male partners of female sex workers in La Romana, Dominican Republic. Soc Sci Med 2009; 68:2037–2044.
16. Lau JT, Wan SP, Yu XN, et al. Changes in condom use behaviours among clients of female sex workers in China. Sex Transm Infect 2009; 85:376–382.
17. Lowndes CM, Alary M, Labbe AC, et al. Interventions among male clients of female sex workers in Benin, West Africa: An essential component of targeted HIV preventive interventions. Sex Transm Infect 2007; 83:577–581.
18. Kaul R, Kimani J, Nagelkerke NJ, et al. Monthly antibiotic chemoprophylaxis and incidence of sexually transmitted infections and HIV-1 infection in Kenyan sex workers: A randomized controlled trial. JAMA 2004; 291:2555–2562.
19. Pettifor AE, Turner AN, Van Damme K, et al. Increased risk of chlamydial and gonococcal infection in adolescent sex workers in Madagascar. Sex Transm Dis 2007; 34:475–478.
20. Feldblum PJ, Lie C-C, Weaver MA, Van Damme L, Halpern V, Microbicide Research Group. Baseline factors associated with incident HIV and STI in four microbicide trials. Sex Transm Dis (in press).
21. Gallo MF, Behets FM, Steiner MJ, et al. Prostate-specific antigen to ascertain reliability of self-reported coital exposure to semen. Sex Transm Dis 2006; 33:476–479.
22. Harms G, Matull R, Randrianasolo D, et al. Pattern of sexually transmitted diseases in a Malagasy population. Sex Transm Dis 1994; 21:315–320.
23. Leutscher P, Jensen JS, Hoffmann S, et al. Sexually transmitted infections in rural Madagascar at an early stage of the HIV epidemic: A 6-month community-based follow-up study. Sex Transm Dis 2005; 32:150–155.
© Copyright 2010 American Sexually Transmitted Diseases Association