Correct and consistent use of oral contraceptive pills is associated with unintended pregnancy rates of as low as 0.3% during the first year. Under “real-life” conditions of use, however, failure rates have been found to approach 8%.1 Incorrect use, or “missed pills,” is considered to be the main reason for these higher failure rates over time.
An analysis of data from the European Active Surveillance Study on Oral Contraceptives study found low oral contraceptive failure rates (Pearl Index of 0.48). which suggest a high level of correct use under real-life conditions in the seven European countries involved.2,3 Data from the U.S. arm of the International Active Surveillance of Women Taking Oral Contraceptives study examined here come from a different population, enabling comparisons, with respect to a potential relation between higher rates of extreme obesity and lower contraceptive effectiveness.
Drospirenone is a progestogen with antiandrogenic and antimineralocorticoid properties. A large active postmarketing surveillance study has demonstrated that a 21-day regimen of 3 mg drospirenone and 30 microgram ethinyl estradiol (E2) can be used safely for oral contraception.3 A 24-day regimen of 3 mg drospirenone/20 microgram ethinyl E2 (24 days of active tablets followed by 4 days of inactive tablets) was recently launched. The International Active Surveillance of Women Taking Oral Contraceptives is designed to investigate the safety and effectiveness of this new regimen in a population that is representative of typical oral contraceptive users. The International Active Surveillance of Women Taking Oral Contraceptives study is a “postauthorization safety study” requested by the U.S. Food and Drug Administration and the European Medicines Agency.
The 3 mg drospirenone/20 microgram ethinyl E2 regimen of 24 days of active tablets followed by 4 days of inactive tablets (drospirenone/ethinyl E224d) may increase overall contraceptive efficacy. A recent ovulation inhibition study investigated the effects of “missed pills” (replacement of the first three pills with placebo after a completed cycle) comparing a 24-day and 21-day regimen of 3 mg drospirenone/20 microgram ethinyl E2. Results showed that suppression of ovarian activity, which resulted in decreased hormonal fluctuation, was substantially more pronounced in the regimen of 24 days of active tablets followed by 4 days of inactive tablets than the regimen of 21 days of active tablets followed by 7 days of inactive tablets (drospirenone/ethinyl E221d).4–7 Therefore, it is conceivable that a progestogen with a long half-life8 and a regimen that shortens the pill-free interval, such as the 24-day regimen of 3 mg drospirenone/20 microgram ethinyl E2, will further increase contraceptive efficacy during “typical use.”
This article estimates real-life effectiveness of oral contraceptive pills in the United States by progestogen, length of pill-free interval, body mass index (BMI, calculated as weight (kg)/[height (m)]2), and other factors while focusing on the effect of progestogens with a long half-life (such as drospirenone) and on 24-day oral contraceptive pills regimens.
MATERIALS AND METHODS
The International Active Surveillance of Women Taking Oral Contraceptives study is a large, prospective, controlled, noninterventional, long-term cohort study with active surveillance performed in the United States and several European countries. The study is supervised by an independent Safety Monitoring and Advisory Council. In the United States, approval from the Western Institutional Review Board was received in July 2005. Participants in the United States were enrolled between August 2005 and July 2008, whereas the European arm of the study began recruitment in October 2008 as a result of later market introduction in Europe. The main outcome of interest is the occurrence of cardiovascular events (ie, venous thromboembolism, myocardial infarction, and cerebrovascular accidents). The secondary objective is the safety and effectiveness of drospirenone/ethinyl E224d under real-life conditions. Sample size has been calculated to sufficiently address both outcomes. The study is powered (1−β=0.9; α=0.05) to detect 0.8-fold contraceptive failure rate for drospirenone/ethinyl E224d compared with other oral contraceptives and 0.6-fold contraceptive failure rate for relevant subanalyses.
Recruitment into the study in the United States was conducted via a network of 5,219 gynecologists in private practice and clinics in all U.S. states representing metropolitan as well as rural areas. Participating women are new users of oral contraceptives, either first-time users, recurrent users after a break in contraception or users who have switched to a different type of pill. Potential participants are informed of the noninterventional study by their physicians only after the decision to prescribe an oral contraceptive has been made, thereby ensuring that the noninterference character of the study is maintained. There are no specific medical inclusion or exclusion criteria. Study participation is voluntary and written informed consent is required. Information about the potential study participation risks is provided in a nondirective, standardized manner.
Baseline data on current state of health and potential risk factors for oral contraceptive use are obtained via a self-administered questionnaire. Participants complete the baseline questionnaire at their gynecologists' offices, with the opportunity to clarify questions with their gynecologists and the study nurses. Subsequent follow-up questionnaires are mailed to the participants biannually for up to 5 years after baseline. The questionnaires collect information on regular and continued oral contraceptive use and the occurrence of pregnancy during oral contraceptive use. All women who report an unintended pregnancy while using their oral contraceptive are interviewed in detail to assess potential reasons for oral contraceptive failure, including whether any pills might have been missed or incorrectly administered.
To prevent loss to follow-up, a comprehensive four-level follow-up procedure is pursued. In the first instance, up to two reminder letters are mailed to women who do not return the follow-up questionnaire (level 1). In the absence of a completed questionnaire, multiple attempts are made to contact the participants, their relatives, friends, and physicians via phone (level 2), and searches in national and international telephone and address directories were initiated (level 3). Finally, level 4 activities include searches in death registries and commercial address registries, as well as official address search through governmental administrations. The same procedure yielded in the European Active Surveillance study3 a loss to follow-up rate of 2.4%; interim results here indicate that a similarly low loss to follow-up rate will be achieved in the European arm of the International Active Surveillance of Women Taking Oral Contraceptives study. The final loss to follow-up in the United States will be between 5% and 10%.
In agreement with current guidelines on the clinical investigation of steroid contraceptives in women that were published by the European Medicines Agency,9 contraceptive failure rates are described by Pearl Index and life-table analysis. Pregnancies were diagnosed by β-hCG measurements and transvaginal ultrasonography. Calculation of contraceptive failure in this study is based on all confirmed unintended pregnancies that occurred while using an oral contraceptive (numerator). The denominator is the total oral contraceptive exposure during the same time period. To calculate the Pearl Index in this study, the number of pregnancies in the relevant cohort is divided by the number of months of exposure in that cohort, and then multiplied by 1,200. Ninety-five percent confidence limits for Pearl Index are calculated using methods from Gerlinger.10
However, the comparison of exposures on the basis of the Pearl Index might be misleading. For example, the Pearl Index assumes a constant contraceptive failure rate over time. That is an incorrect assumption, given that contraceptive failure is more likely to occur in the most fertile women and the women who use an oral contraceptive for a long time are, on average, of lower fertility (“attrition of susceptible”), and self-administered contraceptive methods have better effectiveness in more experienced users. Therefore, the presented analysis focuses on life-table estimate methods (rate of contraceptive failure for each time interval, eg, months).
Predefined potential confounders were chosen (age, BMI, smoking, parity, and education). This selection accords the prognostic factors that were included in the statistical model of the European Active Surveillance study.2 Hazard ratios (HR) are calculated for the variables age, BMI, smoking status, parity, level of education, type of progestogen, and oral contraceptive regimen.
Of special interest is the association between contraceptive effectiveness and an oral contraceptive regimen with a short pill-free interval that uses a progestogen with long half-life and BMI. There have been several studies suggesting lower oral contraceptive effectiveness with increasing BMI,11,12 and the potential influence of obesity has triggered considerable discussion. The analysis of the European Active Surveillance dataset mentioned showed a statistically significant correlation between BMI and an increased rate of contraceptive failure for only one of the five progestogens in the study, with sufficiently large sample sizes up to a BMI of 30. This U.S. arm of the International Active Surveillance of Women Taking Oral Contraceptives study seeks to find results for higher BMI figures (35 or more). It therefore analyzes and discusses BMI categories of less than 20.0, 20.0 to 24.9, 25.0 to 29.9, 30.0 to 34.9, and 35 or more. Inferential statistics for contraceptive failure are based on Cox regression models using the statistical software packages StataES8 and SAS 9.
RESULTS
The presented analysis of this study is based on 52,218 U.S. women (64.9% non-Hispanic white women, 14.9% Hispanic women, 14.3% non-Hispanic African-American women, 2.2% Asian women, 1.2% Native American women, and 2.5% other race or ethnicity) who started use of a new oral contraceptive after study entry and who signed informed consent forms, yielding 99,382 woman-years of observation. The corresponding loss to follow-up rate for these U.S. women was 7.1%. The results on safety outcomes and a detailed description of the baseline demographic characteristics of the study population are reported separately. A summary of the prognostic factors relevant for oral contraceptive effectiveness is outlined and shown in Table 1. Analysis is performed for three cohorts: drospirenone/ethinyl E224d, drospirenone/ethinyl E221d, and oral contraceptive containing other progestogens (“other oral contraceptives”). In addition, subanalysis of the other oral contraceptives cohort included the 24-day and 21-day regimens of oral contraceptives containing norethisterone acetate/ethinyl E2. Overall, the baseline demographic characteristics were similar across cohorts. A total of 10,032 women (19.2%) were oral contraceptive starters (first ever users) and 42,186 women (80.8%) were recurrent users. The mean age of study participants is 26.3, with drospirenone/ethinyl E221d users being slightly younger (25.7 years) compared with users of drospirenone/ethinyl E224d (26.4 years) and other oral contraceptives users (26.3 years). Mean weight and mean BMI are similar across the three cohorts. Mean BMI for the total study is 26.3 and as follows for the three individual cohorts: 25.8 (drospirenone/ethinyl E224d), 26.8 (drospirenone/ethinyl E221d), and 26.4 (other oral contraceptives). Overall, about 23.1% of the US study participants were obese. At study entry, 43.5% of the U.S. study participants were nulliparous, 16.5% were current smokers, and 29.5% had graduated from college.
;)
Table 1: Baseline Characteristics of Study Participants
Presented analyses of unintended pregnancies are primarily based on “as treated” data, an approach that accounts for product changes and shifts from oral contraceptive use to no contraception. Analysis of contraceptive failure is based on a total oral contraceptive exposure of 73,269 woman-years. Overall, 1,634 unintended pregnancies have been reported, of which 229 (14.0%) were reported for conditions of use that gave no reason to indicate noncompliance despite intensive questioning by investigators, and 1,405 (86.0%) were reported for conditions in which indications of noncompliant use were found. A comprehensive analysis of potential reasons for contraceptive failure showed that 46.3% of the women with unintended pregnancies forgot to take the pill on time, 21.1% used antibiotics, and 9.5% reported an episode of diarrhea or vomiting.
Contraceptive failure compared with age followed a biphasic pattern; the highest rate of contraceptive failure was observed in the 20- to 24-year group, and a marked decline was observed in women older than age 30 years (Fig. 1). Cox regression analysis showed an HR of 0.2 (95% confidence interval [CI] 0.2–0.3) for contraceptive failure in women 35 years and older compared with those younger than 35 years old.
Fig. 1.:
Oral contraceptive pill failure by age: point estimates and 95% confidence intervals.Dinger. Real-Life Oral Contraceptive Pill Effectiveness. Obstet Gynecol 2011.
Contraceptive failure rates adjusted for age, parity and educational level showed a slight increase with higher BMI (Fig. 2). An adjusted HR of 1.5 (95% CI 1.3–1.8) for contraceptive failure in women with a BMI 35 or more compared with less than 35 was found.
Fig. 2.:
Oral contraceptive pill failure by body mass index: point estimates and 95% confidence intervals.Dinger. Real-Life Oral Contraceptive Pill Effectiveness. Obstet Gynecol 2011.
Parity and a low educational level were associated with a higher contraceptive failure rate; the age-adjusted HR for parous women was 2.5 (95% CI 2.3–2.8), and women who graduated from college had an age-adjusted HR of 0.5 (95% CI 0.4–0.6). The crude HR for current smoking was 1.4 (95% CI 1.2–1.6). However, adjustment for age and educational level yielded a HR of 1.1 (95% CI 1.0–1.3).
The overall Pearl Index calculated in this study is 2.2 (95% CI 2.1–2.3). The Pearl Index values for drospirenone/ethinyl E224d, drospirenone/ethinyl E221d, and other oral contraceptives are 1.6 (95% CI 1.4–1.9), 2.2 (95% CI 1.8–2.6), and 2.6 (2.4–2.7), respectively.
Life-table estimates of the rate of contraceptive failure ranged from 3.0% (95% CI 2.8–3.2) after the first study year of oral contraceptive use to 6.2% (95% CI 5.8–6.2) after the third study year. The corresponding results for the three cohorts of interest (Table 2 and Fig. 3) are 2.1% for drospirenone/ethinyl E224d, 2.8% for drospirenone/ethinyl E221d, and 3.5% for other oral contraceptives after the first year of oral contraceptive use. Further subanalyses showed lower life-table estimates of contraceptive failure rates (Fig. 4) for drospirenone/ethinyl E224d compared with norethisterone/ethinyl E224d (2.1% and 3.5% after the first year, respectively), as well as for drospirenone/ethinyl E221d compared with norethisterone/ethinyl E221d (2.8% and 4.7% after the first year, respectively). This is also true for the direct comparison of the 24-day regimens of drospirenone and norethisterone with the respective 21-day regimens.
Table 2: Life-Table Estimates of the Rate of Contraceptive Failure After 1, 2, and 3 Years of Oral Contraceptive Use
Fig. 3.:
Life-table estimates of contraceptive failure associated with the use of 24-day regimens of drospirenone and ethinyl E2, 21-day regimens of drospirenone and ethinyl E2, and other oral contraceptive pills.Dinger. Real-Life Oral Contraceptive Pill Effectiveness. Obstet Gynecol 2011.
Fig. 4.:
Life-table estimates of contraceptive failure. Comparison of two 24-day and two 21-day regimens of drospirenone and ethinyl E2 (triangle) and norethisterone acetate and ethinyl E2 (square), respectively.Dinger. Real-Life Oral Contraceptive Pill Effectiveness. Obstet Gynecol 2011.
Cox regression analysis that included age, BMI, parity, smoking, and education as covariates yielded an adjusted HR of 0.7 (95% CI 0.6–0.8) for drospirenone/ethinyl E224d compared with 21-day regimens of other progestogens. An intention-to-treat analysis yielded almost identical results (HR 0.7, 95% CI 0.6–0.8). Sensitivity analyses using additional potential confounders showed almost identical risk estimates. In addition, direct comparisons of drospirenone/ethinyl E224d compared with drospirenone/ethinyl E221d (HR 0.8), drospirenone/ethinyl E224d compared with norethisterone/ethinyl E224d (HR 0.7), drospirenone/ethinyl E221d compared with norethisterone/ethinyl E221d (HR 0.7), and norethisterone/ethinyl E224d compared with norethisterone/ethinyl E221d (HR 0.8) showed statistically significant lower HR for drospirenone or a 24-day regimen or both. To consolidate these results, several stratified analyses of drospirenone/ethinyl E224d compared with 21-day regimens of other progestogens were performed (Table 3). Use of drospirenone/ethinyl E224d was associated with a lower risk of contraceptive failure for nulliparous and parous women, women who did and did not graduate college, current smokers and nonsmokers, women 25 years and older and women younger than 25 years, as well as for women with a BMI of less than 25 and 25 or more.
Table 3: Risk of Contraceptive Failure for a 24-Day Regimen of Drospirenone/Ethinyl Estradiol Compared With the 21-Day Regimen of Other Progestogens, Stratified by Prognostic Factors
DISCUSSION
This study investigated contraceptive failure in a large U.S. cohort of oral contraceptive users. A direct comparison of our U.S. results with the European based European Active Surveillance study3—a study that used the same methodology—showed an approximately four-fold higher contraceptive failure rate in the United States, which confirms results from previous studies.13–15 Using strict criteria our results also indicated noncompliant oral contraceptive use in 86% of unintended pregnancies in the U.S. cohort. Therefore, this study confirms that the very high pharmacological efficacy of oral contraceptives is substantially reduced under real-life conditions in the United States. However, improvement of compliance and reduction of application errors (eg, by specific counseling) could reduce U.S. contraceptive failure rates substantially.
The study shows that contraceptive failure by age follows a biphasic pattern (Fig. 1), which is in line with findings that female fecundity peaks between 20 and 30 years of age,16–18 and it is in line with results from the European Active Surveillance study.2 As expected, the study demonstrates that parous women are more likely to become pregnant during oral contraceptive use, independent of the progestogen or regimen, than nulliparous women.19,20 Our results also indicate that a low educational level is a similarly strong prognostic factor for contraceptive failure.
The European Active Surveillance study reported low failure rates without a statistical significant association to obesity. However, it was not possible to assess whether this nonassociation would also be seen in populations with a higher proportion of women with World Health Organization class II and class III obesity (BMI 35 or more). A higher contraceptive failure rate and higher prevalence of obesity in the United States allowed us to analyze this question using the International Active Surveillance of Women Taking Oral Contraceptives data. We were able to show a slight, but nevertheless statistically significant, increase in contraceptive failure rates with increasing BMI (Fig. 2). One possible explanation is that oral contraceptives might be less forgiving of imperfect use among obese women.21 Obesity might influence the efficacy of oral contraceptives and women's fertility in opposite directions. For example, the high distribution volume associated with class II and III obesity may leads to lower drug concentrations and subsequently may reduce ovarian suppression. However, obesity is associated with anovulation22 and it may additionally influence sexual behavior. Our study does not provide data that would allow estimation of the effect of these factors. What matters in the end is the sum of these factors, which translates into a slight increase in the contraceptive failure rate. In consideration of the rather low quantitative effect of this effect, however, we feel that typical oral contraceptives are adequately dosed for women who are overweight or obese. Independent of this assessment, prescribing physicians might consider progestin-only or nonhormonal contraceptive methods for women who are massively obese because combined oral contraceptive use in these women is associated with a substantially increased risk of cardiovascular side effects.3
Absolute contraceptive failure rates are influenced by several factors. The 2002 cycle of the U.S. National Survey of Family Growth showed a contraceptive failure rate of 7.7% during the first 12 months of use.1 This is more than double the failure rate found in our study. The major quantitative factors that could explain this difference are the definition of person-time in the denominator and differences in the study population. In addition, the National Survey of Family Growth methods differ from the International Active Surveillance of Women Taking Oral Contraceptives methods with regard to the frequency of data collection and the definition of unintended pregnancy. Further methodological differences between the International Active Surveillance/European Active Surveillance and the National Survey of Family Growth methodologies are discussed elsewhere.2
Because the International Active Surveillance of Women Taking Oral Contraceptives study is observational, it is conceivable that our results are influenced by residual confounding or bias, which never can be entirely eliminated in this study design and thus represents a limitation. The ability to infer causation or to exclude substance-specific risk therefore is also limited.23 Although known potential confounders were documented at baseline, adjustment cannot be performed for unknown confounders. Education was used as a proxy for socioeconomic status. It is conceivable that this does not fully account for socioeconomic differences between cohorts. Furthermore, there may be selection bias associated with the readiness to participate in a long-term study, although in this case it is unlikely to have a differential effect on the risk estimates.
Potential limitations of this study should not detract from the strengths of the presented analysis. First, International Active Surveillance of Women Taking Oral Contraceptives is a large, prospective, active cohort surveillance study. Because the study is noninterventional in nature and specific inclusion and exclusion criteria are not applied, participants are probably representative of “typical” users of oral contraceptives. The intensive follow-up procedure pursued in International Active Surveillance of Women Taking Oral Contraceptives yields a low rate of loss to follow-up. It therefore may be assumed that the study results are not biased by substantial under-reporting of outcomes of interest and that the study is, in general, methodologically valid.
The presented results on the differences between drospirenone/ethinyl E224d, drospirenone/ethinyl E221d, and other oral contraceptives are not unexpected. The long half-life of drospirenone (30 hours)8 combined with a 24-day regimen provides (partially) suppressive progestogen levels even in the pill-free interval. It is therefore plausible that a 24-day regimen of drospirenone or any other progestin with a long half-life yields better effectiveness than a 21-day regimen of drospirenone or a 24-day regimen of a progestogen with a short half-life under conditions of imperfect use. The results were consistent across multiple strata of relevant prognostic factors of contraceptive failure and indicate that the pharmacokinetic properties of the progestogen and the short pill-free interval contribute to the observed differences between drospirenone/ethinyl E224d and conventional 21-day regimens of other progestogens. Given the limitations of observational research, the results of the International Active Surveillance of Women Taking Oral Contraceptives study alone could not establish strong evidence of the superiority of drospirenone/ethinyl E224d compared with progestogens with shorter half-life or 21-day regimens or both. However, taking the missed pill study4 and the International Active Surveillance of Women Taking Oral Contraceptives results together, it is likely that the stronger ovarian suppression with drospirenone/ethinyl E224d translates to a lower contraceptive failure rate under conditions of typical oral contraceptive use.
REFERENCES
1. Kost K, Singh S, Vaughan B, Trussell J, Bankole A. Estimates of contraceptive failure from the 2002 National Survey of Family Growth. Contraception 2008;77:10–21.
2. Dinger JC, Cronin M, Möhner S, et al. Oral contraceptive effectiveness according to body mass index, weight, age, and other factors. Am J Obstet Gynecol 2009;201263.e1–9.
3. Dinger JC, Heinemann LA, Kühl-Habich D. The safety of a drospirenone-containing oral contraceptive: final results from the European Active Surveillance study on Oral Contraceptives (EURAS-OC) based on 142,475 women-years of observation. Contraception 2007;75:344–54.
4. Klipping C, Duijkers I, Trummer D, Marr J. Suppression of ovarian activity with a drospirenone-containing oral contraceptive in a 24/4 regimen. Contraception 2008;78:16–25.
5. Spona J, Elstein M, Feichtinger W, Sullivan H, Lüdicke F, Müller U, et al. Shorter pill-free interval in combined oral contraceptives decreases follicular development. Contraception 1996;54:71–7.
6. Sullivan H, Furniss H, Spona J, Elstein M. Effect of 21-day and 24-day oral contraceptive regimens containing gestodene (60 microg) and ethinyl estradiol (15 microg) on ovarian activity. Fertil Steril 1999;72:115–20.
7. Willis SA, Kuehl TJ, Spiekerman AM, Sulak PJ. Greater inhibition of the pituitary–ovarian axis in oral contraceptive regimens with a shortened hormone-free interval. Contraception 2006;74:100–3.
8. Krattenmacher R. Drospirenone: pharmacology and pharmacokinetics of a unique progestogen. Contraception 2000;62:29–38.
9. European Medicines Agency Committee for Medicinal Products for Human Use. Guideline on clinical investigation of steroid contraceptives in women, July 27, 2005. Available at:
http://www.emea.europa.eu/pdfs/human/ewp/051998en.pdf. Retrieved April 14, 2010.
10. Gerlinger C, Endrikat J, van der Meulen EA, Dieben TO, Düsterberg B. Recommendation for confidence interval and sample size calculation for the Pearl Index. Eur J Contracept Reprod Health Care 2003;8:87–92.
11. Brunner Huber LR, Hogue CJ. The association between body weight, unintended pregnancy resulting in a livebirth, and contraception at the time of conception. Matern Child Health J 2005;9:413–20.
12. Holt VL, Scholes D, Wicklund KG, Cushing-Haugen KL, Daling JR. Body mass index, weight, and oral contraceptive failure risk. Obstet Gynecol 2005;105:46–52.
13. Darney P. Safety and efficacy of a triphasic oral contraceptive containing desogestrel: results of three multicenter trials. Contraception 1993;48:323–37.
14. Ferguson H, Vree ML, Wilpshaar J, Eskes TK. Multicenter study of the efficacy, cycle control and tolerability of a phasic desogestrel-containing oral contraceptive. Eur J Contracept Reprod Health Care 2000;5:35–45.
15. Grubb GS, Archer DF, Constantine GD. Differences between the United States and Europe in clinical trials of hormonal contraceptive efficacy. Obstet Gynecol 2008;111:63S.
16. Hamilton BE, Sutton PD, Ventura SJ. Revised birth and fertility rates for the 1990s and new rates for Hispanic populations, 2000 and 2001: United States. Natl Vital Stat Rep 2003;51:1–94.
17. Larsen U, Yan S. The age pattern of fecundability: an analysis of French Canadian and Hutterite birth histories. Soc Biol 2000;47:34–50.
18. Wood JW. Fecundity and natural fertility in humans. Oxf Rev Reprod Biol 1989;11:61–109.
19. Trussell J. Methodological pitfalls in the analysis of contraceptive failure. Stat Med 1991;10:201–20.
20. Howe G, Westhoff C, Vessey M, Yeates D. Effects of age, cigarette smoking, and other factors on fertility: findings in a large prospective study. Br Med J (Clin Res Ed) 1985;290:1697–700.
21. Trussell J, Schwarz EB, Guthrie K. Obesity and oral contraceptive pill failure. Contraception 2009;79:334–8.
22. Friedman CI, Kim MH. Obesity and its effect on reproductive function. Clin Obstet Gynecol 1985;28:645–63.
23. Susser M. What is a cause and how do we know one?. A grammar for pragmatic epidemiology. Am J Epidemiol 1991;133:635–48.
