Studies that measure the role of male condoms in preventing STI transmission by comparing users with nonusers may underestimate the method's effectiveness because of differences between users and nonusers that are difficult to measure.1 In a case-crossover analysis using data from women visiting an STI clinic in the U.S. state of Alabama, in which each participant served as her own control, consistent use of the method was associated with a significant reduction in the incidence of gonorrhea and chlamydia, particularly when condoms neither broke nor slipped off. A cohort analysis based on the same data set, however, in which patterns of condom use and infection status were compared across individuals, showed no reduction in risk associated with consistency of use.

Participants were 18-34-year-old women attending the clinic between 1992 and 1995, who were neither pregnant nor planning to conceive within the next six months. At their initial study visit, the women underwent STI testing, received an intervention that promoted consistent and correct use of condoms and spermicides, completed a behavioral interview, and learned how to use a diary to record their sexual activity and use of barrier methods. Participants were scheduled for six monthly follow-up visits at which they discussed their diary entries with project staff, completed additional interviews, were again tested for STIs and received a six-week supply of their chosen barrier method.

Both analyses examined the incidence and predictors of gonorrhea and chlamydia in the one-month intervals between follow-up visits. The case-crossover analysis compared intervals in which no infection was diagnosed with intervals in which either infection was detected in the same woman (and included only women who had both kinds of intervals), thus making each woman her own control and eliminating potential bias from unmeasured factors that do not change over time. The cohort analysis also compared intervals with and without a diagnosis, but included all women, regardless of whether they became infected during follow-up.

Most of the 1,122 women who enrolled in the study were black (89%), were younger than 25 (53%), had no more than a high school education (70%) and were neither married nor living with a partner (89%). All but 9% had had sex in the month before entering the study; most (68%) had had only one partner during that time. Thirty-one percent of participants tested positive for gonorrhea or chlamydia at study entry.

The analyses are based on data from 919 participants who made at least one follow-up visit and reported on both their sexual activity and their frequency of condom use (categorized as consistent, or 100% use, with neither breakage nor slippage; consistent with breakage or slippage; inconsistent; or no use). For the case-crossover analysis, the researchers used data on 228 intervals in which gonorrhea, chlamydia or both were diagnosed and 743 matched intervals from the same women in which no infection was diagnosed. The cohort analysis included 245 intervals with a diagnosis (all of those from the case-crossover analysis plus 17 that were ineligible for that analysis) and 3,896 intervals in which no infection was detected.

In the case-crossover analysis, the risk odds ratio derived from conditional logistic regression indicated that the likelihood of infection with gonorrhea or chlamydia was significantly lower during intervals in which a woman had consistently used condoms than during intervals in which she had never used them (odds ratio, 0.5). Additionally, the association was stronger for intervals characterized by consistent use with no breakage or slippage than for consistent use with either of these problems. This analysis also revealed that a diagnosis was significantly more likely in an interval when a woman had had multiple partners than in an interval when she had had only one (1.8), and that the likelihood of infection during a given interval increased significantly with the number of unprotected sex acts (i.e., occasions of nonuse of condoms or use with slippage or breakage).

By contrast, using unconditional logistic regression, the cohort analysis showed no difference in the incidence of infection by consistency of condom use and no trend toward greater risk of infection with an increase in unprotected sex. Having multiple partners was once again associated with an increased risk of infection (risk odds ratio, 2.1), as were being younger than 25 (1.5) and receiving a diagnosis of chlamydia or gonorrhea at enrollment in the study (1.5).

Given the strengths of their study—notably, the use of two analytic approaches, including one designed to "circumvent unmeasured confounding and reduce its impact"—the researchers contend that "epidemiologic studies of condom effectiveness are probably confounded by unmeasured differences between users and nonusers." Moreover, they conclude that "the likely result of such confounding is underestimation of the effectiveness of condoms."—D. Hollander


1. Warner L et al., Application of the case-crossover design to reduce unmeasured confounding in studies of condom effectiveness, American Journal of Epidemiology, 2005, 161(8):765-773.