Macaluso, Maurizio MD, DrPH^*; Kelaghan, Joseph MD, MPH^†; Artz, Lynn MD, MPH^*; Austin, Harland DSc^‡; Fleenor, Michael MD, MPH, FACP^§; Hook, Edward W. III MD^*; Valappil, Thamban MS^*

THE SURGEON GENERAL of the United States¹ and the Centers for Disease Control and Prevention² recommend use of male condoms for personal protection from infection. One of the national objectives published by the U.S. Public Health Service in Healthy People 2000³ is to double the use of male condoms as one step in reducing the incidence of gonorrhea and chlamydia infections. In vitro studies suggest that the latex condom provides excellent protection against a variety of sexually transmitted diseases (STDs) if it does not break or slip off during use. Although there is substantial evidence that condoms are effective in preventing HIV transmission, the degree of protection afforded by condoms against other STDs has not yet been adequately documented. Condoms are likely not to be effective if they fail during intercourse, so that measures of the rates of breakage and slippage are important indicators of condom effectiveness.

Several limitations characterize the studies that have attempted to measure condom breakage and slippage. Retrospective surveys suffer from potential inaccuracies in reporting because respondents can only estimate the total number of condoms used and are potentially inclined to recall selectively a condom break. Such studies^4–12 have reported breakage in 0.2% to 7.3% of condoms used and slippage in 1.0% to 4.4% of condoms. These findings suggest that failure of the condom to prevent exposure to semen may be as high as 10%, although most studies found total failure rates in the range of 3% to 5%. Prospective studies are rarer than retrospective studies, and many assess failure among small groups of subjects reporting on few condoms (usually 6 to 12) or on all condoms used over a short period of time. Prospective studies^13–18 have measured breakage rates from 0.0% to 5.9% and slippage rates from 0.1% to 6.6%, again suggesting that the failure rate (breakage plus slippage) may be as high as 10%. Only recent studies have begun to focus on characteristics of the users that may influence condom failure.¹⁹ Finally, most studies of breakage and slippage rates have been carried out in groups at low risk of STD infections.

We present information obtained from a large prospective study of a cohort of women at high risk who used condoms for protection against STDs. The large number of condoms used provides precise estimates of mechanical condom failure, and the associated data collected during the study allow investigation of several potential determinants of slippage and breakage.

Methods

Study Design and Procedures

The study protocol is described in detail elsewhere.²⁰ Briefly, this was a prospective follow-up study of women who attend the STD clinic of the Jefferson County Department of Health in Birmingham, AL. Patients were eligible if they were 18 to 34 years of age, neither pregnant nor planning to become pregnant within the next 6 months, and had no history of hysterectomy. Women who agreed to participate in the study were scheduled to attend the initial study visit at least 10 days after recruitment. At the initial visit, each participant was interviewed more extensively and taught to complete a detailed sexual diary including information about contraceptive use. A nurse delivered an intensive behavioral intervention promoting consistent and correct use of barrier method strategies, including joint use of a condom and vaginal spermicide product as the recommended option, followed by condoms alone if joint use was not feasible. The nurse also conducted a pelvic examination, in which specimens were obtained for STD screening. The participants then left the clinic with free supplies of their chosen barrier method and were scheduled to return for six monthly follow-up visits.

Each participant was trained to record information about all sexual activity and barrier use in a sexual diary. The diary allowed discreet encoding of a few key variables: type of sexual activity, type of protection used, if any, the initials of the partner, and whether any problems occurred during each sex act. The participants were instructed to report whether “the condom broke during intercourse” or “slipped off the partner's penis during intercourse.” No explicit distinction was made between complete and partial slippage. The definitions used are compatible with “clinical breakage” and “clinical slippage” as defined by Steiner et al.²¹ At a follow-up visit, the diary information was abstracted and encoded, and any reporting problems were discussed and resolved with the participant. Additional quality control procedures included reconciliation of condom use reported in the diary and supplies returned to the clinic.²⁰

Data Analysis

The potential determinants of condom failure evaluated in this analysis included a set of important baseline characteristics of the study participants and a set of time-dependent behaviors. Baseline characteristics included age, race, education, marital status, lifetime number of partners, number of children, alcohol and drug use, risky sex (defined as having exchanged sex for money or drugs, having had sex while drunk or high on drugs), STD history, and previous condom use. Time-dependent variables included the total number of acts of unprotected sex preceding the current act, the total number of condoms used without problems, the number of previous slips, the number of previous breaks, the number of other problems, the type of partner (regular, new, or casual) at a specific act of intercourse, and concomitant use of a vaginal spermicide.

The main outcomes evaluated were condom breakage and condom slippage, as reported by the participants in their diaries. Simple analyses were done by stratifying the number of events and the number of condoms used according to the various characteristics, and by computing average breakage and slippage rates per hundred uses. However, conventional stratified or Poisson regression analysis of rates is based on the assumption of independence among condom uses. To test whether breakages were distributed at random in the study group, the condom use experience of each woman was considered as a binomial experiment with mean equal to the number of condoms used times the average breakage rate for the entire group. Next, for each woman the probabilities of observing zero, 1, 2,…k…, up to the number of condoms used were computed and summed across the 892 women to obtain the expected distribution of women by frequency of condom breakage. Finally, the observed distribution was compared with the expected and the difference was evaluated using a goodness-of-fit chi-squared statistic. The same procedure was used to test whether slippage was randomly distributed among participants. The results of these analyses (see later) indicated that mechanical failures tended to cluster, confirming the suspicion that an analysis based on independence may lead to biased results.

A method for the analysis of repeated measurements was used to disentangle a random subject effect from the effects of potential determinants of condom failure. The SAS macro GLIMMIX (SAS Institute, Cary, NC), which implements generalized linear mixed models,²² was used to model condom failure as a binary outcome. It was assumed that within each subject, the outcome of one condom is correlated more strongly with the condoms used immediately before and after and progressively less strongly with condoms farther apart in the series. Finally, the regression coefficient estimates were used to compute the relative odds (odds ratios [OR]) of breakage and slippage and its 95% confidence interval (CI) as in a regular logistic regression model.

Results

A total of 1,122 women participated in the study. Information used in this analysis pertains to 892 women who reported using at least one condom during follow-up. The group consisted predominantly of young adult (median age, 24 years), black (90%), high-school-educated (median number of school years, 12), low-income women (Table 1). The history of sexual and reproductive behavior of the group did not necessarily indicate high-risk behavior (median age at first intercourse, 16 years; median lifetime number of partners, 6), but 70% reported at least one STD episode before entry into the study, and many reported two or more episodes. Most participants had experience using condoms, although only 23% relied exclusively on condoms for birth control, and 22% reported consistent condom use during the 30-day period preceding enrollment (results not shown in detail).

TABLE 1

The 892 women returned a total of 3,960 sexual diaries, which comprise the main source of information for this analysis. A latex condom was used in 21,852 (64%) of the 34,036 acts of vaginal intercourse reported during follow-up. In 7,244 (33%) of the acts in which a condom was used, so was a vaginal spermicide product (contraceptive foam, jelly, film, or suppositories). The group reported a total of 500 (2.3%) condom breakages, 290 (1.3%) condom slippages, and 154 (0.7%) other use problems (condom was not used throughout the intercourse or problems were encountered with the spermicide product).

Figure 1 displays the distribution of observed and expected number of women who experienced 0,1,2…k condom breakages. There was a modest excess of women who never experienced any condom breakages, a deficit of women who experienced only one breakage, and an excess of women who experienced multiple breakages. Because of the large study size, the difference between the observed and expected distributions is statistically significant (P < 0.0001), and indicates that breakages are not randomly distributed in the study group. A similar significant difference was observed for condom slippage (Fig. 2, P < 0.0001). To accommodate within-subject correlation of failures, generalized linear mixed models for repeated measurements were used to evaluate the determinants of breakage and slippage rates. We used alternative modeling strategies, including 1) a straightforward logistic regression model (i.e., assuming total independence of errors across subjects and condoms), 2) population-average models using generalized estimating equations (GEE) (i.e., no random subject effects, but correlation of errors among repeated measurements), and 3) mixed models with random subject effects and correlated within-subject errors. We also evaluated alternative error structures (e.g., autoregressive vs. compound symmetry). The estimates did not vary appreciably with the technique used, presumably because the study is so large that even in the presence of clustering of outcomes, the cluster effects are sufficiently diluted not to represent an important source of bias. The results presented in the following sections and in the tables are from population-average GEE models. We selected these estimates because 1) although events clustered within the experience of some study subjects, the fit of mixed models yielded only a few significant subject effects, and 2) within the experience of the same subject, we expected that the outcomes of condoms uses that occur close to each other (e.g., during the same week) should be more correlated than the outcomes of condom uses occurring farther apart in time (e.g., in different months of follow-up).

Fig. 1

Fig. 2

Baseline Predictors of Condom Breakage

Age younger than 30 years was associated with a twofold increase in the risk of condom breakage (Table 2) compared with age 30 years or older. The crude breakage rate among women aged 30 years or older was approximately 1/100 condom uses, whereas the crude rates for younger age categories ranged from 2.3 to 2.9/100 uses. After adjusting for other risk factors and for the correlation among repeated measurements, the relative odds estimate is approximately two in each of the younger categories. The 95% CI of the relative odds indicate that the study results are compatible with a 30% to 200% increase in breakage risk among younger couples. Black race was associated with increased risk of condom breakage (OR = 1.6, CI = 1.1-2.4), parity was associated with decreased risk (OR = 0.7, CI = 0.6-0.9). A history of engaging in high-risk sexual behavior (exchanging sex for money or drugs, having sex after drinking or using drugs) was the strongest determinant of condom breakage among the baseline characteristics (OR = 2.0, CI = 1.4-2.8). Prior use of condoms and consistency of condom use were weakly associated with a reduction in breakage rates, but the association was not statistically significant.

TABLE 2

Time-Dependent Predictors of Condom Breakage

An increasing number of sexual partners was weakly and inconsistently associated with an increase in breakage rates, whereas partner type (regular vs. new or casual) was not associated with breakage (Table 3). Concurrent use of a vaginal spermicide was associated with a modest reduction in breakage rates (OR = 0.8, CI = 0.6-1.0). The strongest time-dependent determinant of condom breakage was experience with condom use during the study. In this analysis, for each condom use, the following time-dependent covariates were computed to summarize the experience of a participant during follow up: 1) the total number of breakages before the index condom use, 2) the total number of prior slippages, 3) the total number of prior condom uses in which some other problem was reported, 4) the total number of prior condom uses in which no problem was reported, and 5) the total number of prior acts of intercourse during which a condom was not used. The sum of the five numbers at the end of follow-up is the total number of acts of intercourse reported by the participant.

TABLE 3

As compared to a woman who had experienced no prior breakages, a woman who had reported one prior breakage was approximately four times more likely to experience a new breakage (OR = 3.6, CI = 2.7-4.8), and a woman who had reported two or more prior breakages had a ninefold increase in risk (OR = 9.3, CI = 7.2-12). Multiple prior slippages were associated with a nonsignificant 30% decrease in breakage rates, whereas other problems were associated with an increase in risk of breakage.

Predictors of Condom Slippage

Condom slippage displayed a pattern of association that was substantially different from that described for condom breakage (Table 4). Most baseline characteristics had no association with slippage, with the exception that married women had a 40% increase in slippage rates as compared to other women (OR = 1.4, CI = 1.0-2.0). In addition, parity was associated with a similar increase in slippage risk (OR = 1.4, CI = 1.0-1.9). Slippage was inconsistently associated with the number of partners encountered during the study, and with the number of unprotected acts of intercourse (Table 5). Intercourse with a new or occasional partner was 60% more likely to lead to condom slippage than intercourse with the regular partner (OR = 1.6, CI = 1.1-2.1). Use of a vaginal spermicide was associated with a 20% decrease of slippage rates, although this difference was of borderline statistical significance. As for condom breakage, previous condom failures were strong predictors of slippage, whereas the number of condoms used without problems had a strong protective association. The slippage rate decreased 10 times from the category of fewer than five condoms used to the category of 30 or more condoms used without a problem.

TABLE 4

TABLE 5

Discussion

A few potential limitations are important to consider when interpreting the results from this study. Subjects did not necessarily use one brand of condom throughout the follow-up period, but chose monthly from among five brands known to appeal to this population. The brand of the condom used at each act of intercourse was not reported in the diary. Thus, our data do not allow assessing the effect of any specific brand. The available evidence suggests that factors such as manufacturing batch, time since production, and storage conditions may be more prominent determinants of condom quality than the brand. For example, in one study the variability in mechanical test results was at least as large among lots within brand as it was among brands.²³

Another important concern regarding this study is its dependence on self-report of sexual activity, condom use, and problems with condoms. This is a problem common to all studies that require data on sexual activity. We argue that although subjects may remember less accurately, or even intentionally misrepresent the frequency of unprotected sexual acts or the total number of condoms used, they would be more likely to remember and report condom breakage and slippage. The use of an instrument to record sexual activity daily was one method we used to minimize recall bias. In addition, the monthly follow-up visits dictated that even subjects who had not recorded information daily would have to recall only a relatively short period since the last follow-up visit. Subjects were interviewed during follow-up visits and provided information similar to that recorded in the diary, and these two sources of information were highly correlated (data not shown). The study protocol also included two procedures intended to encourage subjects to report information honestly. The interviewer collecting information each month was not the study nurse with whom the subject may have developed a close relationship and to whom the subject may have been inclined to report behaviors encouraged by the study protocol. In addition, study personnel encouraged the subjects to be truthful and did not berate them for any behavior.

A final problem with self-reporting is the possibility that subjects may not have been aware that a condom broke or slipped off the penis. This is a problem common to all studies that are based on self-report.

The limitations discussed earlier are offset by the considerable strengths of the study. To date, this is the largest study to measure condom breakage and slippage prospectively among women who were using condoms for protection from STDs. The large number of subjects (n = 892) reporting use of at least one condom and the large number of condoms used by these women (21,852 uses) provide a high degree of precision for the observed breakage rate (2.3%) and slippage rate (1.3%). The prospective follow-up design and the collection of condom use data with monthly diaries are additional strengths. Although undoubtedly some women recorded data on less than a daily basis, the monthly schedule for follow-up visits dictated that the recall time was relatively short. In addition, the study protocol included several components designed to encourage truthful reporting. The 6 months of follow-up allowed observation of multiple condom uses for each woman and changes in condom use over time. Finally, this study measured characteristics of condom use among a high-risk population for whom condoms are recommended for protection from STDs.

The low rates of breakage and slippage observed in this study are compatible with the relatively wide range of findings reported by most other investigators, but they are appreciably lower than some of the worst estimates. With the exception of one study that demonstrated leakage of HIV-sized particles through latex,²⁴ most in vitro studies suggest that intact latex condoms do not permit transmission of sexually transmitted organism.^25–31 Thus, the findings of a breakage rate of 2.3% and a slippage rate of 1.3% suggest that a condom user will have <4% chance of exposure to the potentially infectious secretions of a partner during protected intercourse. However, even low breakage and slippage rates provide reason for caution for those at high risk of contracting HIV from an infected partner; for such couples health providers should strongly recommend avoiding intercourse rather than relying on a condom.³² We note that even a low failure rate may translate into a high risk for individuals who are very sexually active. Also, partner type appears to have a modest but significant effect on condom failure. If intercourse with a new or casual partner is at high risk of transmission and at high risk of condom failure, the overall protective effect of condoms will be reduced. Finally, although breakage and slippage are the most commonly recognized forms of condom failure, they are not the only potential source of exposure to potentially infectious secretions during “protected intercourse.” Exposure may occur even without direct penile-vaginal contact, as when a condom is inadvertently put on inside-out, then flipped and used during intercourse. A series of specific forms of incorrect condom use may also be implicated in the failure of a condom to protect from exposure.³³

This study shows that condom breakage and slippage become progressively less common with continued use. Such a trend may be anticipated if those at highest risk of experiencing a break (e.g., younger users) are also more likely to drop out of the study early, leaving behind a sample at decreased risk of experiencing problems. Nevertheless, limiting the analysis to the women who completed all six follow-up visits did not substantially change the results (based on 338 reported condom breaks and 212 slips during 14,985 uses-data not shown), suggesting that subjects' skills improve with continued condom use and lead to reduced risk of breakage and slippage over time.

An important observation from this study is that condom breakage and slippage rates are associated with characteristics of the users, independent of experience and skill. Breakage is associated with several demographic factors (black race, younger than 30, and lower parity), and slippage is weakly associated with being married and having children.

These findings have several implications for public health policy and future research. First, it is doubtful that any of these factors directly affects the performance of condoms, but is instead a proxy marker for traits or behaviors that lead to breakage or slippage. It is important to identify and assess the relative importance of nonmodifiable traits and modifiable behavior associated with condom failure. Second, there probably is not just one standard rate for condom breakage or slippage, but different rates depending on population characteristics and the length of time they are studied. Third, forms of condom failure, such as breakage, which are usually regarded as “method failures,” appear to be a function of user characteristics and not of inherent defects in the products used. Efforts to improve condom performance should focus more on modifiable characteristics of the user. Finally, use of a spermicidal product with the condom was associated with decreased breakage and slippage, presumably because of increased lubrication. This finding contradicts the notion that spermicidal products damage the integrity of the latex, and suggests that concurrent use of a spermicide or other lubricant could result in improved condom performance. Steiner et al. found that increased lubrication reduced the risk of condom breakage, but increased the risk of slippage.¹⁴

In conclusion, condom breakage and slippage were generally low in this large observational study of women at high risk for STD. Both indices of condom failure decreased substantially with increasing condom use during the study, and both were associated with specific user characteristics. The findings of this study highlight the limitations of a rigid classification of failure according to whether it is attributable to the user or to the product, and emphasize the need for improved user motivation and training.

References