Materials and Methods
This trial was conducted at eight sites in Mexico, Ecuador, Guatemala, Ghana, and the United States between September 1995 and July 1997. The study was approved by the ethical review boards at the sites in the United States and Durango, Mexico, and by the board at Family Health International, which was intended to cover the other six sites that did not have local ethical review boards. All participants gave written informed consent before study enrollment.
Healthy, sexually active women aged 18 to 35 years who had no history suggestive of subfecundity, who were at low risk for sexually transmitted diseases, and who wished to use a spermicide as their only contraceptive method were recruited. Exclusion criteria included contraindications to nonoxynol-9 use or to pregnancy, fewer than 6 weeks since last pregnancy, breastfeeding, or signs or symptoms of vaginal infection at enrollment. Before admission, women had an interview, pelvic examination, wet prep, and urine pregnancy test sensitive to 25 mIU/mL hCG (Unigold, Trinity Biotech Inc., Irvine, CA). At each site, eligible women were randomly assigned to one of the two study spermicides by opening sealed, sequentially numbered, opaque envelopes provided by Family Health International. A computer-generated randomization scheme stratified by center with randomly ordered block sizes of four and six was used. Participants received instruction on product use and were provided with diaries in which to record all menstrual periods, coital acts, and use of contraception throughout the study period. They were asked to return to the clinic if they missed a menstrual period or if they suspected pregnancy. A sufficient amount of spermicide was dispensed to last until the next scheduled appointment.
Participants were instructed to return at 4, 12, 20, and 28 weeks after admission. At each visit, the diaries were reviewed, and information was collected on correctness of product use at last use. Participants were asked about medical problems. A pelvic examination and wet prep were done whenever symptoms of vaginal irritation were reported within the previous week, and routinely at the last visit. A urine pregnancy test was performed. Questionnaires on product use and acceptability were administered at the 4-week visit and at study discontinuation. Additional spermicide was dispensed as needed.
Participants were followed up until they completed at least 28 weeks of participation or until they stopped considering the assigned spermicide to be their primary contraceptive method. Participants who discontinued early were asked to return for a pregnancy test 2 weeks after discontinuation.
In late 1996, the only planned interim data analysis showed that the 6-month probability of pregnancy in both spermicide groups was about 25%, which was considerably higher than the figure quoted to participants at enrollment (“about one in eight”). All women remaining in the trial were informed of the interim findings and asked to sign a supplementary informed consent form if they wished to remain in the study. This reconsent process was approved by all ethical review boards reviewing the study.
All pregnancies detected either by pregnancy test or by participant report were considered for this analysis. The estimated date of fertilization was determined by adding 14 days to the onset date of the last menstrual period or by using other obstetric information when available.
The sample size of the study was planned to meet the requirements of a proposed rule for the development of vaginal contraceptive drugs issued in 1995 by the United States Food and Drug Administration.6 This rule specified that clinical studies of nonoxynol-9 products should include at least 200 women who either become pregnant or complete 6 months of product use. Anticipating that 35% of participants would discontinue early for reasons other than pregnancy, we estimated that we needed to enroll 308 women per group. The intent of the study was to determine whether the film was at least as effective as the tablets. If the true 6-month pregnancy proportion in the tablet group were 12%,1 this sample size would provide 81% power to show significance with a one-sided test (α = .10) if the true pregnancy proportion was at least 19% in the film group.7 Initially, we planned to enroll 90 participants at each of seven sites. Before enrollment began, an eighth site was added, which was also asked to recruit 90 women.
Analysts were masked to spermicide group assignments until primary findings were apparent. All enrolled women were included in primary analyses. Each participant contributed to the analyses until the earliest of the following dates: the estimated date of fertilization; the date she stopped relying primarily on the assigned spermicide for contraception; the latest date her pregnancy status could be determined reliably; 183 days after admission; 31 days after the last reported coital act; or the date that the interim analysis results were released to the clinical sites. Participants were removed from the analysis on the last of these dates because of concern that the reconsent process could have led to selective discontinuation of participants who were most interested in avoiding pregnancy.
The cumulative probability of pregnancy during typical use of the spermicide was estimated using the Kaplan-Meier product-limit method.8 The primary comparison of these probabilities tested the null hypothesis that the risk of pregnancy during typical use is identical in the two groups against the one-sided alternative that the risk is greater in the film group (α = .10). This analysis used a Cox proportional-hazard regression model9 controlling for center, prior spermicide use, and other baseline variables found in preliminary analyses to be significantly related to the risk of pregnancy. The other variables considered a priori for inclusion in the model were age, years of education, previous pregnancy, and desire for additional children.
The cumulative probability of pregnancy during consistent use of the product was estimated by applying the Kaplan-Meier approach to the subset of months during which the participant reported use of the assigned spermicide and no other contraceptive at every act of intercourse. A piecewise exponential model was used to compare the risk of pregnancy in the two groups, controlling for the same variables included in the typical-use model.10
The cumulative probability of early discontinuation or loss to follow-up in each group was estimated using the Kaplan-Meier method. Group differences were tested using a two-sided log-rank test (α = .05).
Altogether 765 women were enrolled in the trial (383 in the tablet group and 382 in the film group), including 90 in Ghana, 292 in the three centers in Mexico, 91 in Guatemala, 226 in the two centers in Ecuador, and 66 in the United States. All sites except Ghana were encouraged to enroll as many participants as possible during the allotted recruitment period. The Ghana site, which initially had the most rapid recruitment rate, was asked to stop recruitment when the enrollment target of 90 participants had been reached to prevent that site from finishing the study with substantially more participants than the other sites.
Seven participants did not fulfill all admission criteria: one had had no menses since her last delivery, five had bacterial vaginosis on admission, and one did not have an examination on admission. In three cases at one center, randomization envelopes were opened before the women were fully screened for the study. (Although this premature opening of the envelopes may have resulted in selection bias, the problem involved so few participants that we do not believe that it substantially affected the conclusions of the study.) No study participants were breastfeeding at admission, were pregnant within 6 weeks before admission, had recently used hormonal contraceptives, or had any other known indication of reduced fecundity. All participants received the assigned spermicide.
Baseline characteristics of the two groups were similar (Table 1). Some characteristics differed between centers (results not shown).
The two groups contributed similar amounts of time to the pregnancy analysis (Table 2). Thirty-one participants contributed no time: four were pregnant at admission (one in the tablet group and three in the film group); 24 never returned after admission (15 in the tablet group and nine in the film group); and three provided no information on pregnancy status after admission (two in the tablet group and one in the film group). These participants' baseline characteristics did not differ substantially from those of the entire study population. The cumulative lifetable probability of early discontinuation for reasons other than pregnancy or of loss to follow-up was 29.7% in the tablet group and 26.2% in the film group (P = .32). The cumulative lifetable probability of early discontinuation varied substantially across centers (results not shown).
Coital diary data indicated that coital frequency and compliance with use of the assigned spermicide were high in both groups (Table 3). We also examined compliance as reported on questionnaires completed by participants at 4 weeks after admission, at the time of the reconsent process, and at discontinuation. These data confirmed the diary information. For example, of the 446 women with coital diaries indicating use of the assigned product at every coital act, 460 (98.7%) reported the same information on the questionnaires.
At each clinic visit, participants were asked whether they had followed specified instructions at the time of their last product use: to insert the product deep into the vagina, not to douche within 6 hours after using the product, and to insert the product within the specified time before intercourse. At least 90% of participants in each group indicated at every visit that they had followed all of these instructions.
The cumulative 6-month probability of pregnancy during typical use was 28.0% in the tablet group and 24.9% in the film group (Table 4). Within each group, the probabilities of pregnancy during typical and consistent use were similar. The planned comparison of the two groups, which controlled for center, prior use of spermicide, and desire for additional children, showed no evidence to reject the null hypothesis that the risk of pregnancy in the film users was less than or equal to the risk in tablet users in either the typical-use analysis (P = .78) or the consistent-use analysis (P = .81). Desire for additional children was significantly related to the probability of pregnancy in this model (P = .034), but prior use of a spermicide was not. No significant interactions were found between spermicide group and any covariate in this model. Post hoc two-sided hypothesis tests found no significant differences between groups (P = .44 for the typical-use analysis; P = .39 for the consistent-use analysis).
Estimates of the typical-use pregnancy probabilities varied little by subgroup, except that the probability of pregnancy was substantially lower in women who did not desire more children (Table 4). Probabilities of pregnancy also varied by center (Figure 1). Pairwise statistical comparisons controlling for prior use of spermicide and desire for additional children showed significantly different pregnancy risks between sites A and B and each of the five sites with the highest pregnancy rates (D, E, F, G, and H). Variation in reported coital frequency and compliance with spermicide use did not appear to account for these differences between centers.
Secondary analyses of pregnancy risk during typical and consistent use of the study spermicides were done excluding women with characteristics suggesting less-than-optimal baseline risk of pregnancy. Specifically, these analyses excluded two women who had abnormal preadmission menstrual periods (both in the tablet group), two who had a history of ectopic pregnancy (one in each group), and 54 who had fewer than four coital acts in the first month after admission (31 in the tablet group and 23 in the film group). Data from the remaining women were excluded from the time the following events first occurred: fewer than four coital acts in a month after the first month (n = 55); or participant was instructed to abstain from intercourse for 15 days or longer (n = 2). These analyses included data from 331 women (1178.1 woman-months of typical use) in the tablet group and 345 women (1276.1 woman-months of typical use) in the film group. Probabilities of pregnancy in these analyses were not substantially different from those in the primary analyses.
Findings were similar when the data obtained after the interim analysis results were included in the calculations (results not shown).
In this study, we found no significant difference in the cumulative life-table probabilities of pregnancy during 6 months of typical use of the tablet (28.0%) or the film (24.9%). These probabilities were considerably greater than the assumed values used for the original power calculations. Poststudy power analysis in which the assumptions were updated accordingly indicated that the study had nearly 75% power to detect a seven percentage point difference in pregnancy risk for the planned one-sided test of interest (α = .10). However, for the post hoc two-sided test (α = .05), the power to detect a difference of seven percentage points was only about 50%.
To compare the present results with those of earlier studies, we extrapolated our 6-month life-table typical-use probabilities of pregnancy to estimate 12-month probabilities. Two previous trials of barrier contraceptives showed that the average ratio of the probability of pregnancy in months 7 through 12 to the probability in months 1 through 6 is .795.11 When this ratio was applied to our data, extrapolated 12-month cumulative probabilities of pregnancy were 44.0% in the tablet group and 39.8% in the film group. These figures are higher than any previously reported 12-month life-table estimates of the probability of pregnancy during use of various nonoxynol-9 spermicide products, which range from .3% to 39.2% in 12 months.1 Only two previous studies of Conceptrol foaming tablets have been published. One study, which included 101 women in Ghana, found a 12-month life-table probability of 11.3%,12 only about two-thirds of the extrapolated 12-month probability for the tablet group in our Ghana center. The other, which included 48 women in Thailand, reported a 6-month life table probability of pregnancy of 15.2%.13 No life-table probabilities of pregnancy among film users previously have been reported.
Most older studies of spermicide effectiveness reported results in terms of a different statistic, the Pearl index, which is the number of pregnancies per 100 woman-years of use of the method. In our study, Pearl indices were 67.4 per 100 woman-years for the tablet group and 59.8 per 100 woman-years for the film group. These figures also exceed those previously reported for spermicides. The highest Pearl index for a nonoxynol-9 product, which came from a trial of Vaginal Contraceptive Film, was 55.7 per 100 woman-years (Smith M, Vessey MP, Bounds W, Warren J. C-film as a contraceptive. BMJ 1974;4:291); that trial was stopped early because the number of pregnancies was so high. However, at least 20 other spermicide studies reported considerably lower Pearl indices.1 Among these were three trials of Vaginal Contraceptive Film including 966 women, which found Pearl indices between .6 and 6.5 per 100 woman-years14,15 (Apothecus Pharmaceutical Corporation. VCF Vaginal Contraceptive Film. East Norwich, NY: Apothecus Inc., 1992).
What might account for the high pregnancy rates observed in our study? One explanation for the high Pearl indices, at least, is the inherent flaw in this statistic. Because the probability of pregnancy decreases over time (because more fecund women and poorer contraceptors become pregnant early), the Pearl index tends to decrease the longer a cohort is observed. For example, in one study of nonoxynol-9 foaming tablets (not Conceptrol), the Pearl index including all months of data was 22.5 per 100 women-years, but if only the first 6 months of product use were considered, the index was 33.6 per 100 woman-years.16 Because the present study had a relatively short duration (6 months), our Pearl indices would be expected to be higher than those found in earlier studies that followed participants for a year or longer. Because the present study also found higher life-table rates than previous studies, other explanations must also be considered.
Unlike most previous studies, our trial procedures included multiple scheduled highly sensitive pregnancy tests, which could have led us to detect pregnancies that would not otherwise have been noted, particularly pregnancies that spontaneously aborted very early. We believe that because spermicides are designed to prevent all pregnancies, it was appropriate to test participants actively for pregnancy and to include all detected pregnancies in the analysis. Furthermore, ongoing additional analyses suggest that the number of such pregnancies in our study was small and that ignoring them would not likely affect our results substantially.
Most of our study participants were young, had proven fertility, and had intercourse often. Therefore, they might have been at higher risk for pregnancy than women in previous studies. We deliberately designed our trial to enroll women at moderate to high risk of pregnancy because we believed that this was the most appropriate group in which to study contraceptive effectiveness. Approximately 86% of our participants stated that they desired more children, which might have negatively influenced motivation to avoid pregnancy.
The probability of failure to complete the planned follow-up period in our study was about 30%. In this respect, our trial appears comparable to other recent studies of barrier contraceptives, although differences in statistical methods preclude direct comparison.11,17 However, if women who left the study early were at lower risk for pregnancy than those who continued, our pregnancy probability estimates could be biased upwards.
Poor compliance with use of the spermicides could result in a high number of pregnancies. Reported compliance in our study was so high that we suspect some overreporting. The fact that the probability of pregnancy during consistent use was almost identical to that during typical use also might suggest overreporting of product use is likely, although other explanations are possible (eg, perhaps most inconsistent use occurred on cycle days when the chance of pregnancy was negligible). No independent verification of our compliance data, such as from product counts, is available. If poor compliance did contribute to the high pregnancy rates in this trial, from a public health or programmatic viewpoint, we would interpret this finding as an indication of product failure in the study population.
Several study sites were located in tropical climates, which raises the question of whether the spermicide products deteriorated during the study period. Data from the Product Quality and Compliance Laboratory at Family Health International and information from the manufacturers of the two study products indicate that these products are stable for at least 2 years even in extreme conditions of temperature and humidity. The Product Quality and Compliance Laboratory tested tablets from a lot used in the trial that had been stored at one of the sites with the highest pregnancy rates and found no apparent degradation in amount of nonoxynol-9 or in foaming ability.
One puzzling finding of our study was the wide variation in pregnancy rates between sites. The life-table probabilities of pregnancy at two sites, one in Ghana and one in Mexico, were significantly different from those at five of the six other sites. This finding was not explained by differences between sites in the one baseline characteristic shown to be related to pregnancy in this study, desire for additional children, or by differences in coital frequency or reported compliance with spermicide use. In addition, we are unaware of any differences in how the study was conducted or reported at the eight sites. Unmeasured factors or chance must account for the different site results. However, it should be noted that even at the sites in which women had the lowest probabilities of pregnancy, the figures were still higher than those reported from many previous studies of spermicidal effectiveness.
The most likely explanation for the high pregnancy rates found in this study is that these two spermicide products are not very effective for preventing pregnancy among fertile, highly sexually active women. Both products appear to be more effective than nothing. The probability of pregnancy in 6 months among women using no contraception is estimated to be 61%,1 which exceeds the upper 95% confidence limit for the typical-use probabilities for both products. Therefore, these products do have a place as a contraceptive option for women who cannot use other methods or who find other methods unacceptable. However, women who strongly desire to avoid pregnancy should be encouraged to consider alternative contraceptive methods.
1. Trussell J. Contraceptive efficacy. In: Hatcher RA, Trussell J, Stewart F, Stewart GK, Kowal D, Guest F, et al, eds. Contraceptive technology, 17th rev ed. New York: Irvington Publishers, 1998: 779–844.
2. Cates W Jr, Raymond EG. Vaginal spermicides. In: Hatcher RA, Trussell J, Stewart F, Stewart GK, Kowal D, Guest F, et al, eds. Contraceptive technology, 17th rev ed. New York: Irvington Publishers, 1998:357–69.
3. Roddy RE, Schulz KF, Cates W Jr. Microbicides, meta-analysis, and the N-9 question—where's the research? Sex Transm Dis 1998;25:151–3.
4. Potts M. The urgent need for a vaginal microbicide in the prevention of HIV transmission. Am J Public Health 1994;84:890–1.
5. Steiner M, Spruyt A, Joanis C, Glover L, Cordero M, Alvarado G, et al. Acceptability of spermicidal film and foaming tablets among women in three countries. Int Fam Plann Perspect 1995;21:104–7.
6. Federal Register. Vaginal contraceptive drug products for over-the-counter human use. Department of Health and Human Services. Food and Drug Administration, 21 CFR Part 310. 1995;60: 6892–902.
7. Lachin JM, Foulkes MA. Evaluation of sample size and power analysis of survival with allowance for nonuniform patient entry, losses to follow-up, noncompliance, and stratification. Biometrics 1986;42:507–19.
8. Kaplan EL, Meier P. Nonparametrics estimation from incomplete observations. J Am Stat Assn 1958;53:457–81.
9. Cox DR. Regression models and life tables (with discussion). J R Stat Soc Series B 1972;34:187–220.
10. Koch GG, Atkinson SS, Stokes ME. Poisson regression. In: Kotz S, Johnson NL, eds. Encyclopedia of statistical sciences, 7th ed. New York: John Wiley and Sons, 1986:32–41.
11. Trussell J, Sturgen K, Strickler J, Dominik R. Contraceptive efficacy of the Reality female condom: Comparison with other barrier methods. Fam Plann Perspect 1994;26:266–72.
12. Lamptey P, Klufio C, Smith S, Feldblum P. A comparative study of Neosampoon, Ortho vaginal tablets and Emko vaginal tablets in Accra, Ghana. Contraception 1985;32:445–54.
13. Chompootaweep S, Dusitsin N. A comparative study of the safety, effectiveness and acceptability of two foaming vaginal tablets (nonoxynol-9 versus menfegol) in Thai women. Contraception 1990;41:507–17.
14. Brigato G, Pisano G, Bergamasco A, Pasqualini M, Cutugno G, Luppari T. Vaginal topical chemical contraception with C-film. Ginecol Clinica 1982;3:77–80.
15. Iizuka R, Kobayashi T, Kawakami S, Nakamura Y, Ikeuchi M, Chin B, et al. Clinical experience with the vaginal contraceptive film containing the spermicide polyoxyethylene nonylphenyl ether (C-film study group). Jpn J Fertil Steril 1980;25:64–8.
16. Dingle JT, Teitze C. Comparative study of three contraceptive methods: Vaginal foam tablets, jelly alone, and diaphragm with jelly or cream. Am J Obstet Gynecol 1963;85:1012–22.
17. Mauck C, Glover LH, Miller E, Allen S, Archer DF, Blumenthal P, et al. Lea's shield: A study of the safety and efficacy of a new vaginal barrier contraceptive used with and without spermicide. Contraception 1996;53:329–35.