Emergency contraception with levonorgestrel: one hormone better than two.

new method has arrived, we know more about how to optimise treatment, and new distribution channels are opening up. Combined with the political will, these advances should significantly reduce the burden of unwanted pregnancies. One of the important new ingredients is the WHO trial demonstrating the superior efficacy of levonorgestrel used on its own; the pregnancy rate with levonorgestrel alone was a third that of the comparison group using the traditional Yuzpe regime.1 The study was the largest of its kind, with almost 2 000 participants, had a secure method of randomisation, and was blinded to all participants. Two study groups were thus created which differed only in the type of emergency contraception used, thereby minimising the possibilities for random error and bias. The result, however, has been questioned on a number of accounts. The Yuzpe regime did not appear to perform particularly well in the trial and this might have influenced the comparison.2,3 But the method used to estimate the ‘true efficacy’ of emergency contraception, or proportion of pregnancies prevented, has limitations, although, as it was applied equally to both arms of the trial, it should not have biased the result. In the absence of placebo controlled trials, external controls, derived from published studies in which no contraception was used, are used to provide an estimate of the number of ‘expected pregnancies’, based on fecundity on each day of the menstrual cycle.4–6 However, these calculations assume that the day of ovulation can be accurately determined from retrospective reporting of day of the last menstrual period and cycle length. A recent study found a poor correlation between menstrual data and hormonal status on the day of emergency treatment.7 Not surprisingly, studies of emergency contraception frequently find pregnancies occurring when none are expected, using what is, in effect, the rhythm method to estimate risk of pregnancy. In the WHO study the external controls used were older and some desired pregnancy, while almost half of the women in the trial had accidents with a barrier contraceptive, often with a spermicide. Other differences from the external controls are likely. Applying these externally derived estimates to randomised trials opens too many doors through which error can enter into otherwise well-conducted studies. Because of its unreliability, this attempt to correct for a possible imbalance in risk of pregnancy should be dropped from randomised trials which will distribute women of equal risk of pregnancy to each arm, if the numbers are large enough, as indeed they were in the recent WHO trial. We cannot tell the so-called ‘true effectiveness’ of each method, but we can tell reliably if one is better than the other, and by how much precisely the sort of information required for clinical decision making. The opportunity to determine the true efficacy of Yuzpe was lost as the early workers, as is common, felt it unethical to conduct a placebo controlled trial.8 Had one been organised by the doubters, the value of the Yuzpe regime should have been apparent and the uptake of emergency contraception greater. Archie Cochrane lamented the way in which medical interventions become established without a randomised trial being conducted, so that it is then regarded as unethical for a trial to be set up. The only way to avoid this, in his opinion, is to randomise the first patient whenever a new treatment is introduced.9 In family planning we have many practices which have not been assessed, having taken what Muir Gray calls, ‘the evaluation bypass’. Caution in accepting the WHO trial results has also been advised because it has been said that fewer women in the Yuzpe arm met the study criteria for ‘correct use’.3 This is not supported in the data. Similar numbers in both groups used the method correctly, and amongst these women the Yuzpe pregnancy rate was halved in the levonorgestrel arm. The pregnancy rate following both treatments was higher following incorrect use, but again the progestogen method was superior; indeed, the benefits of levonorgestrel were even greater following incorrect use. It appears that the traditional Yuzpe method is less tolerant to variations in the regimen, a common experience in everyday practice. Unknown to the investigators, a few women were already pregnant at the time of treatment, and more of these were in the Yuzpe group. However, removing these women from the analysis does not effect the comparative advantage of levonorgestrel. It is interesting that the part of the WHO study that has been given most credence3,10 is the result with the lowest validity the shorter the coitus to treatment interval, the lower the pregnancy rate.1 It is based on non-experimental observations, albeit from a randomised trial. Clinical practice has been heavily influenced by an earlier pooling of observational studies of the Yuzpe regimen, of variable quality, which found a similar pregnancy rate regardless of the day of treatment.11 Adding the WHO data to the earlier studies of the Yuzpe method gives a combined pregnancy rate for Days one to three of 1.7, 2.8 and 2.6, respectively. The result is statistically significant (p = 0.05), but not when adjusted for the effects of study site (p = 0.1). There is a high degree of heterogeneity across studies, and the loss to follow-up rates are as high as 24%, so it is questionable if the pooling exercise is justified. Given the small number of pregnancies reported in the studies, any unrecorded pregnancies could have a considerable impact on the results. The rigorous methods of the WHO trial, and the low loss to follow-up rate (1.4%), probably avoided some of the biases in the earlier studies, but in the absence of randomisation to different periods of delay it remains possible that the more fertile women presented later for treatment. So how do we advise women? Our previous advice that delay in treatment was a secondary concern was based on more questionable data. The close relationship in the WHO study between interval to treatment and pregnancy rate is impressive. It appears that intuition the sooner the treatment the better is probably correct. We were right to say that it is not just for the ‘morning after’,12 but it may