Emergency contraceptives are used to prevent pregnancy after unprotected intercourse. In the United States, they could eliminate up to half of the 3.0 million annual unintended pregnancies. 1,2 The best-studied emergency contraceptive method is the Yuzpe regimen. Women using that regimen take 100μg of ethinyl estradiol and 50 mg of levonorgestrel within 72 hours after unprotected intercourse. They repeat the dose 12 hours later. The therapy fails around 2.0% of the time, but prevents approximately 75% of the pregnancies that might be expected in the absence of treatment. 3 Roughly eight of every 100 women having a single act of unprotected, midcycle intercourse would become pregnant if none of the 100 used emergency contraception. If all 100 took Yuzpe emergency contraception, only about two (instead of eight) would likely become pregnant. 3
First described by Albert Yuzpe, the regimen consists, as noted, of ordinary combined oral contraceptives taken for 1 day in a higher dose. 4 The regimen, however, did not result from systematic drug development efforts. Rather, Yuzpe picked a combined oral contraceptive (Ovral; Wyeth Laboratories, Philadelphia, PA) he was already using for other research on the endometrium. He selected the 72-hour cutoff based on the high-dose estrogen regimens in Europe he was trying to improve. The timing also fit with normal clinic hours, stretching from Friday until Monday morning. When his pilot study yielded what Yuzpe considered an unacceptably high failure rate (2.0%), he arbitrarily doubled the dose, asking women to repeat the same dose again after 12 hours 5 (personal communication, A. Yuzpe, August 10, 2000). Judged against the 11 subsequent trials of the Yuzpe regimen, however, the original 2.0% failure rate is quite average. 3,6
Because the rules for the Yuzpe regimen did not originate from a rigorous evidence base, it is possible that modifications could make the therapy more accessible, comfortable, and convenient without compromising efficacy. Our trial investigated two specific aspects of the regimen.
Must the Progestin Be Levonorgestrel?
Ideally, women could use any oral contraceptive postcoitally. We tested a formulation that substitutes norethindrone for levonorgestrel. Norethindrone is the progestin with the shortest serum half-life (7 hours), whereas levonorgestrel has the longest (15 hours). 7 Therefore, if norethindrone works, other progestins might also. If effective, additional formulations could expand access. The exact brand studied in Yuzpe's original trials currently has only a tiny market share, as lower estrogen oral contraceptives replace their 50-μg predecessors. Equivalent tablets specially packaged for emergency contraception are sold in some countries, but not most. Although users can substitute lower dose brands with the same active ingredients, the rules for doing so are confusing, particularly with triphasic formulations, in which tablets of different colors contain different amounts of the relevant hormones. 8 In some countries, moreover, any levonorgestrel brands are difficult or impossible to obtain. 9 Combined oral contraceptives containing norethindrone are also more widely available even in many industrialized countries. Of the 34 combined oral contraceptive brands currently approved in the United States, 8 for example, 13 (38%) contain norethindrone (or norethindrone acetate, which metabolizes to norethindrone), whereas nine (26%) contain levonorgestrel (or norgestrel).
Is the Second Dose After 12 Hours Necessary?
Yuzpe's second dose is inconvenient for women who start therapy in the afternoon. Women could theoretically wait until the evening or morning to begin treatment, but some recent evidence suggests that emergency contraceptives might work better the sooner they are started. 10 Indirect evidence from New Zealand hints that eliminating the second dose might not compromise efficacy. 11 There, a third dose yielded no improvement over the two-dose standard. (Incidentally, no studies have explored whether 12 hours is the optimal interval between doses, assuming a second dose is helpful.)
Although effectiveness was our main outcome of interest, we also designed our study to track the frequency of various side effects and to assess the modifications’ acceptability. We paid special attention to menstrual disturbances because delays can distress women worried about pregnancy. Although not designed for this purpose, our study could also check whether any of the regimens worked better the sooner they were started and whether taking them with food reduced nausea and vomiting. Because we did not randomize on these latter two criteria, though, our answers are as observational as those in all other articles on these topics. Finally, to assess external validity, we explored the extent to which trial participants represented typical clients.
MATERIALS AND METHODS
We conducted a double-masked, multicenter randomized controlled trial. 12 Between September 1997 and February 2000, emergency contraception clients presenting at one of five clinic sites in the United States (Des Moines) and the United Kingdom (Liverpool, Glasgow, and, later, Sheffield and Edinburgh) and meeting strict eligibility criteria could participate. Women age 16–45 years (18–45 in the United States) could enroll if they had regular cycles varying between 21 and 35 days in length, presented within 3 days of having unprotected intercourse that occurred between 10 days before and 6 days after the estimated day of ovulation (defined as typical cycle length minus 13) in their current cycle, were willing to abstain from further unprotected intercourse during the study cycle, could attend follow-up visits and keep a diary of side effects, and refused copper intrauterine devices. After publication of a 1998 study finding superiority of levonorgestrel alone over the Yupze regimen, 13 clinicians at some sites opted to begin presenting levonorgestrel alone as first-line therapy; there we only enrolled women who refused levonorgestrel-alone therapy. We excluded women who had used any hormonal contraception during the past 2 months, had not had two normal periods in the previous two cycles, or were breast-feeding, as well as those who tested positive using urine pregnancy tests.
Participants were randomized to receive one of three regimens: 1) standard two-dose Yuzpe, 2) a variant of two-dose Yuzpe that substituted 2.0 mg of norethindrone for the 1.0 mg of norgestrel (two tablets of the combined oral contraceptive brand Ovcon 50 [Warner Chilcott, Rockaway, NJ] instead of two tablets of Ovral) per dose, or 3) a single-dose (instead of two-dose) regimen consisting of only the first half (100 μg of ethinyl estradiol plus 1.0 mg of norgestrel) of the standard Yuzpe regimen, followed 12 hours later by a placebo. Because norgestrel contains two isomers, only one of which (levonorgestrel) is bioactive, 1.0-mg norgestrel is equivalent to 0.5-mg levonorgestrel. Population Council staff enclosed each dose of each course of therapy in opaque gelatin capsules, and then packaged these capsules in opaque envelopes labeled either “first dose” or “second dose.” After computer-generated randomization, we inserted pairs of these envelopes into larger sequentially numbered envelopes, which we then sealed. Clinic staff simply gave each enrolled woman the next envelope in the sequence at that site. Neither the clinic staff nor the women knew which regimens had been taken by which women until we broke the code at the end of the trial, although we maintained an emergency unblinding procedure throughout. (We conducted our interim analysis without breaking the code). We took care to choose gelatin capsules that would not interfere with absorption and to ensure that pills were never out of their original packaging for more than 12 months.
At the treatment visit, staff instructed women when to take the pills and how to record any side effects after either the first or the second dose onto a simple home study card. Women also recorded whether they took the pills with food, and they noted any concurrent medications.
Clinic staff scheduled a follow-up appointment for each woman 1 week after the expected start of her next menses. Follow-up interviews covered side effects, return of menses, protocol compliance, and acceptability of the therapy. Women described the timing and volume of their menses as compared with typical cycles. Women whose menses had not yet returned could choose to take a β-human chorionic gonadotropin serum pregnancy test, take another urine pregnancy test, or continue waiting for their menses, returning to the clinic for extended follow-up 1 week later. Staff telephoned or mailed questionnaires to women who missed scheduled follow-up visits. Those still not responding received home visits. In a few cases, we obtained final outcome information by writing to women's general practitioners.
For all women with positive pregnancy tests at the follow-up or the extended follow-up visits, we estimated the date of fertilization using ultrasonography, physical examination, and women's reports. We counted all pregnancies in which the probable date of fertilization fell within the study cycle as emergency contraception failures. Cycles (and pregnancies) were classified as “perfect use” only if the woman followed the protocol perfectly (started therapy during the assigned time period, took both doses, and had no further acts of unprotected intercourse after treatment).
Throughout the trial, we monitored each site approximately every 6 months, checking every study form against the original clinic chart notes. We also audited 100% of forms to double check that we had entered the correct information in the computer database. Assuming a failure rate of 2.0% for the Yuzpe regimen, and employing a Bonferroni correction for multiple comparisons to the same control, we calculated that we would have 73% power to detect a difference of two percentage points between the experimental regimens and the control, under an equivalence approach, if we enrolled 650 women in each group of the randomized study (NQuery Advisor 1.0, Boston, MA). We adapted two stopping rules from the World Health Organization (WHO). A committee comprising the first author, the lead statistician (JT), and one outside expert reviewed results from an interim analysis conducted after 1000 patients. They would have stopped the trial if 1) the point estimate of the failure rate in any experimental arm equaled or exceeded a rate of four times that in the control arm or 2) the lower bound of the confidence interval around the interim failure rate in any experimental arm exceeded twice the hypothesized failure rate of the control group.
We analyzed efficacy in two ways: failure rates (percentage pregnant) and effectiveness rates. Effectiveness rates (proportionate reductions in pregnancy risk) compare the number of pregnancies observed with the number expected, based on estimates of conception probabilities by adjusted cycle day from the published literature. 3 We performed each analysis separately for perfect-use cycles and for typical-use cycles (includes both perfect-use and imperfect-use cycles). Using StatXact-3 (Cytel Software Corp., Cambridge, MA), we computed P values for differences in failure rates based on Fisher exact tests and 95% binomial confidence intervals for failure rates. We computed P values for differences in failure rates between days 1, 2, and 3 of treatment based on Fisher–Freeman–Halton exact tests. P values for differences in effectiveness rates and 95% confidence intervals for effectiveness rates were based on z tests.
Women were asked to document side effects. We compared these across regimens. Using StatXact-3, we calculated P values for differences based on Fisher exact tests.
To help us draw inferences from our study participants to the general population of women seeking emergency contraception, we collected a small amount of basic information (age, cycle day, and the reasons that otherwise eligible women declined to participate) from all women seeking emergency contraception at the study clinics during the study's first 6 months. We compared this information with similar data from women who enrolled in our clinical trials during the same time period (unpublished data, E. Westley, C. Ellertson, K. Blanchard, A. Bigrigg, A. Webb, S. Haskell, et al, 2002).
The trial was approved by the Population Council's ethical review board. Each participating clinic also followed local procedures for ethical review. All women gave written informed consent. We followed the CONSORT guidelines for reporting the trial results. 14
We enrolled 2041 women at five centers (Table 1). Of these, 11 (four in the Yuzpe control group, two in the norethindrone group, and five in the single dose group) turned out to have been enrolled mistakenly (mainly because their cycle days of exposure were calculated wrong). None of the 11 became pregnant, and all are retained in our intention-to-treat analysis of effectiveness.
Two thirds of participants reported that burst condoms led them to need emergency contraception, whereas a further quarter had not planned to have sex and so had not used any method. About 4% reported withdrawal accidents, and the same percentage had had unprotected sex because they lacked a regular contraceptive method.
A total of 1973 women completed the study and were available for efficacy analysis (Figure 1), resulting in a loss–to–follow-up rate of 3.3%. Among women with complete outcome information, we observed 58 pregnancies, all intrauterine. Reviewing available evidence, including ultrasonograms, serial pregnancy tests, letters from the women's regular physicians, and reports from the women, we classified 43 as method failures occurring despite perfect use and 15 as user failures occurring during imperfect use. When in doubt, we classified pregnancies as method failures. Eight of the pregnancies ended in normal births. Although four additional women expressed intent to carry their pregnancies to term and another three were unsure, we lack further information about these seven outcomes. All other pregnancies ended in documented spontaneous or induced abortions.
For each study group, we tabulated pregnancies along with the relevant denominators for calculating the failure rates. Table 2 shows the perfect-use (includes only women who followed the protocol perfectly) and typical-use (includes all eligible women analyzed by intent to treat) failure rates, as well as the proportionate reductions in pregnancy. The highest effectiveness appeared among women taking the standard two-dose Yuzpe regimen, but rates among women taking the norethindrone–ethinyl estradiol and single-dose variants were only slightly and not statistically significantly worse. We tested for center effects and found none.
Treatment did not appear to be more effective the sooner it was started. We found no statistically significant differences in the failure rates for the regimens when analyzed by the interval between unprotected intercourse and start of therapy. Starting therapy on days 1, 2, and 3 seemed to result in comparable failure rates (data not shown).
Women using norethindrone–ethinyl estradiol reported side effects nearly identical to those of women using the standard two-dose Yuzpe regimen (Table 3). Women taking the single dose, however, experienced less nausea and half the vomiting (P < .001).
About 10% of women in each group reported spotting, with most saying that it occurred on 3 days or less. Particularly when they started therapy before day 12 of their adjusted cycles, women in all groups frequently reported that their next menses returned early. Analyzing all women together, the timing was strongly related to the adjusted cycle day of emergency contraceptive treatment (Figure 2).
Taking either dose of therapy within 1 hour of eating food did not affect the rates of nausea and vomiting (data not shown).
Although the standard Yuzpe regimen had the lowest failure rate of those tested in our trial, we found clear evidence that a regimen substituting more widely available brands of combined oral contraceptives containing norethindrone instead of levonorgestrel is safe and effective. Women lacking access to levonorgestrel-only or standard Yuzpe preparations (as is unfortunately the case for hundreds of millions of women around the world) could certainly use norethindrone–ethinyl estradiol brands. In fact, the norethindrone–ethinyl estradiol regimen in our trial had an effectiveness (60.0%) identical to that of levonorgestrel alone in its first trial, 15 although the two study populations from that trial and ours were of different ethnic compositions. We also consider it probable that other, perhaps all, combined oral contraception formulations, such as those containing desogestrel–ethinyl estradiol, might be safe and effective for emergency contraception.
Our results also call into question the utility of the standard Yuzpe regimen's second dose. Women taking just the first dose of the Yuzpe regimen experienced half the vomiting but showed a pregnancy rate not different, in a statistically significant sense, from that of women taking the standard two-dose regimen. Until further research can answer the question conclusively, however, the recommendation would still be to advise a second dose. Another possibility is that the two doses could be taken at the same time as one double dose. A recent multicenter study from the WHO found that taking both doses of the standard levonorgestrel regimen at one time is as effective as taking the two doses 12 hours apart, as conventionally recommended. 16
Within the 72-hour interval for starting therapy studied in our trial, we did not find that any of the study regimens worked better the sooner they were started, in contrast to results from one recent WHO trial. 10,13 An earlier meta-analysis of the results of the nine trials previously available on this topic revealed that two found a detrimental effect of delay, whereas seven did not. 17 As with the WHO trial, 13 however, ours was not the ideal design for exploring this important question. A related new WHO trial of the levonorgestrel and mifepristone methods of emergency contraception found these methods effective for up to 120 hours after unprotected intercourse. 16 As described more fully in an accompanying article, 18 we also found that the Yuzpe regimen prevents pregnancy when used after 72 but before 120 hours following unprotected intercourse.
Similarly, our data suggest that urging women to take the therapy with food lacks empiric justification, although because our study was not randomized with this question in mind, we cannot be sure. Another recent trial similarly found no support for recommending food with the pills. 19
Strikingly, women in all groups reported that their next menses returned earlier than has previously been reported. Late menses have been hypothesized to result from delayed ovulation. Our results, that menses were much more likely to return earlier than expected, are more consistent with an alternative hypothesis that emergency contraceptive pills inhibit ovulation altogether.
Most reported side effects were consistent with the published literature on the Yuzpe regimen. Not surprisingly, women taking the single-dose regimen reported the fewest side effects.
Typical emergency contraception seekers apparently face far lower pregnancy risk than do women who join emergency contraception trials (unpublished data, E. Westley, C. Ellertson, K. Blanchard, A. Bigrigg, A. Webb, S. Haskell, et al, 2002). This point is important to remember when extrapolating averted pregnancies from the clinical trial literature to the general population (eg, when estimating the global impact on unwanted pregnancy and other programwide outcomes of substituting one regimen for another). In no case, however, can all risk be ruled out for an individual woman, so in no case should a woman seeking emergency contraception be turned away.
On a methodological note, the differences between regimens arising from the method of analysis (perfect use versus typical use) are of approximately the same magnitude as the differences between regimens. Researchers should take care to collect data that permit both analyses.
We note that the power in the present study is slightly lower than would ordinarily be ideal. Although small in absolute sense (and therefore needing a large sample size) the two percentage point difference between the experimental regimens and the control that we aim to detect represents a doubling in the failure rate. Even with over 2000 participating women (making ours one of the largest trials of emergency contraception), we have only 73% power to detect a difference between the experimental and control regimens. The research dilemma is that all the tested regimens of emergency contraception are so effective that it would take truly enormous sample sizes in each arm to be able to pick up finer differences. For example, detecting a difference between failure rates of 2.0% and 2.5% with 80% power would require 12,300 women per arm. Financial constraints prevented us from fielding a trial of this size, but even if they had not, such an allocation of resources would not be reasonable on public health grounds. We settled instead for the more modest policy aim of demonstrating that other regimens, based on widely available oral contraceptive formulations, also work for times when women cannot obtain levonorgestrel or Yuzpe.
In conclusion, oral contraceptives containing norethindrone–ethinyl estradiol work approximately as well for emergency contraception as levonorgestrel–ethinyl estradiol formulations. Eliminating Yuzpe's second dose halves vomiting; it lowers efficacy somewhat, but not statistically significantly. Taking the therapy with food does not seem to reduce nausea, and the regimens were not more effective when started sooner after unprotected intercourse.