In his autobiography Archie Cochrane describes his experiences as a prisoner of war in Salonica, Hildburgausen, Elsterhorst and Wittenberg-am-Elbe, where as a fully qualified physician he was responsible for the care of the British, French, Polish, Yugoslav and Russian soldiers kept in these camps, treating several hundreds of men with famine oedema, dysentery, diphtheria, hepatitis and tuberculosis’. It was these experiences which made him realise that most of the available treatments were useless, and some actually harmful, and that none of the therapies he had employed had been justified by scientific evidence of their efficacy. After the war he therefore abandoned clinical medicine, becoming one of the founders of British epidemiology and the foremost champion of the randomised trial in the evaluation of the effectiveness of medical interventions.
Archie Cochrane's renunciation of clinical medicine was to have far-reaching implications, for it resulted in the initiative for the conduct of randomised trials passing from practising physicians to epidemiologists. This has had enormous benefits, for randomised trials have resulted in effective treatments for diseases which hitherto were killers—streptomycin for tuberculosis' and Salk vaccination to prevent poliomyelitis3,4 are two examples. More recently, several large multi-centre trials have confirmed the effectiveness of thrombolytic therapy in myocardial infarction in which it is now beyond question that mortality is reduced by about one-quarter, and by nearly one half if streptokinase is combined with aspirin5–9. The organisation, design and analysis of large multicentre trials have become so sophisticated that the randomised trial has become a science in itself, together with its offspring, meta-analysis; so much so that the pronouncements of specialists in randomised trials have all the authority of the Pythoness at Delphi.
But large multicentre trials in obstetrics have failed to produce such unequivocal answers as the trials in tuberculosis, poliomyelitis and myocardial infarction. The use of low dose aspirin in the prevention of proteinuric pre-eclampsia and intrauterine growth retardation is an example. From the results of smaller trials from single centres it was hoped that antiplatelet therapy with low dose aspirin would prevent these disorders10–22, but this expectation has been disappointed by larger multicentre trials23–29. It has been suggested that the apparent benefit of aspirin in the smaller trials was due to the nonreporting of small trials of aspirin with negative results10. The misleading effects of such nonreporting have been shown in the meta-analysis of magnesium sulphate as adjunctive therapy for myocardial infarction, in which there was an apparent substantial benefit30,31; this benefit was refuted by the larger ISIS-4 trial in which magnesium sulphate did not influence the outcome32.
It may therefore be supposed from the multi-centre trials that low dose aspirin is of little value in the prevention of proteinuric pre-eclampsia and intrauterine growth retardation, but there is another explanation for their negative results. In the design of the thrombolytic trials in myocardial infarction there were two important aspects which encouraged physicians to recruit patients to the trial: clinical uncertainty of the value of streptokinase, and trial procedures which were as simple as possible, with a minimum of data collection and forms to complete. They were pragmatic trials which attempted to measure the effects of treatment in actual clinical practice. And this approach was successful, for it allowed the recruitment of very large numbers of patients, such that a clear answer was obtained concerning the value of streptokinase in myocardial infarction. In the large low dose aspirin trials the same design was adopted, where obstetricians were encouraged to enter women where there was clinical uncertainty of the value of aspirin, and the trial procedures were kept as simple as possible; very large numbers were recruited.
But it was this pragmatic approach which may have confounded the trials. In the thrombolytic trials the entry criteria stipulated the occurrence of chest pain with electrocardiographic evidence of myocardial infarction; in the low dose aspirin trials the entry criteria were purely clinical: clinical perceptions of risk of proteinuric pre-eclampsia or intrauterine growth retardation. In the CLASP trial women were eligible if they had a history of pre-eclampsia, growth retardation, chronic hypertension, renal disease, or other risk factors such as maternal age, family history, or multiple pregnancy28. None of these eligibility criteria was rigorously defined. In the ECPPA trial, which was designed to determine whether low dose aspirin is effective in women at particularly high risk of adverse outcomes associated with pre-eclampsia, the entry criteria were chronic hypertension, primigravidity, diabetes, renal disease, or a history of pre-eclampsia or growth retardation. None of these eligibility criteria was rigorously defined, apart from primigravidity.
These ill-specified entry criteria may have resulted in large numbers of women being recruited to the trials who had little chance of developing proteinuric pre-eclampsia or intrauterine growth retardation, since a clinical perception of risk is an ineficient definition of eligibility. Thus the thrombolytic trials may have produced an unequivocal result because nearly every patient had sustained a myocardial infarction; the low dose aspirin trials may have produced equivocal results because hardly any women were destined to develop proteinuric pre-eclampsia or growth retardation.
Support for this notion is seen in the results of the trials. In the CLASP trial the definition of proteinuric pre-eclampsia required hypertension and proteinuria of 1 + on stick testing during pregnancy28, while the ECPPA trial defined proteinuric pre-eclampsia as ‘the development of hypertension and the detection of proteinuria in the urine after randomisation’29. The incidence of proteinuric pre-eclampsia was 7.6% in the CLASP trial and 6.1% in the ECPPA trial. The maternal and fetal prognosis in proteinuric pre-eclampsia is associated with the degree of proteinuria; proteinuria of 1+ on stick testing represents mild proteinuria with a good prognosis for mother and baby. By not specifying the degree of proteinuria in their definitions these trials may have classified many women with relatively benign pre-eclampsia as ‘proteinuric pre-eclampsia’, and so the outcome of real interest, proteinuric pre-eclampsia with proteinuria of severe degree, may have occurred with a lesser frequency than 7%. The results may therefore have been influenced by the dilution of the trials by so many women at low risk of the development of clinically important proteinuric pre-eclampsia.
It has been suggested that the results of the smaller aspirin trials have been confounded by the nonreporting of small trials with negative results, but this interpretation ignores the large treatment effect of low dose aspirin in these trials. Analysis of the twelve trials with fewer than 200 women shows that aspirin will reduce the risk of important proteinuric pre-eclampsia by about 80%, a huge reduction, and a result quite at odds with the CLASP and ECPPA trials, which were consistent with an increase in the risk of proteinuric pre-eclampsia with low dose aspirin. If the large treatment effect of aspirin in the smaller trials was due to the nonreporting of negative trials, there would have to be a very large number of small negative trials, even if some of these unreported trials showed a treatment effect with aspirin in the opposite direction.
It is more likely that the magnitude of the treatment effect in the small trials is genuine, and the reason for the difference is the entry criteria. In the trial by Beaufilsl1 the entry criteria were confined to previous stillbirth, previous intrauterine growth retardation defined as a birthweight below the third centile, or hypertension without proteinuria in the current pregnancy (incidence of proteinuric pre-eclampsia in the control group 13%); Wallenburg's trial12 was confined to women with an abnormal angiotensin sensitivity test (incidence of proteinuric pre-eclampsia in the control group 30%); McParland's trial19 was confined to women with abnormal Doppler utero-placental velocity waveforms (incidence of proteinuric pre-eclampsia in the control group 19%); Schiff's trial16 was confined to women with a positive roll over test (incidence of proteinuric pre-eclampsia in the control group 23%); and in this issue (pages 625–629) the trial by Bower et al.33 was confined to women with a diastolic notch on Doppler examination of uteroplacental velocity waveforms (incidence of proteinuric pre-eclampsia in the control group 40%).
On pages 603–607 of this issue Lilford et al.34 give a more detailed analysis of the use of low dose aspirin in the prevention of proteinuric pre-eclampsia. Figure 1 of their article shows the results of sixteen randomised trials of low dose aspirin, where the effect of aspirin in preventing proteinuric pre-eclampsia is plotted against the size of trial. Because of statistical sampling error one would expect the odds ratios of very small trials to be scattered widely, the scatter becoming narrower as the trials become larger, the whole plot forming the shape of an inverted funnel. In Figure 1 the right half of the inverted funnel is missing, which is interpreted by the authors as being due to the nonreporting of negative trials. But Figure 1 is also consistent with the alternative explanation of the smaller trials having different entry criteria. Inspection of Figure 1 of only the trials with fewer than 200 women show that they form an excellent inverted funnel, suggesting that the skewed distribution of the points in the whole of Figure 1 is due to the heterogeneity of the trials.
The entry criteria for the smaller trials therefore identified women at the very highest risk of proteinuric pre-eclampsia, identified as high risk not according to clinical perceptions of risk, which are inefficient, but according to the results of fundamental investigations of the pathophysiology of pre-eclampsia, which are efficient predictors of proteinuric pre-eclampsia and intrauterine growth retardation. The apparently negative results of the large multicentre trials of aspirin are misleading, for the trials may have been swamped by women at low risk of proteinuric pre-eclampsia, such that any benefit of low dose aspirin in women at genuine high risk of proteinuric pre-eclampsia may have been unnoticed. The large multicentre trials may even be harmful, since their negative messages may dissuade obstetricians from prescribing low dose aspirin, even to those women at genuine high risk of proteinuric pre-eclampsia. Similar problems are seen with the MRC-RCOG Cervical Cerclage Trial, where the entry criteria based on clinical perceptions of risk of premature labour allowed the recruitment of women at low risk of premature labour, such that any benefit of cervical cerclage in genuine cervical incompetence may have been unnoticed33. The Collaborative Eclampsia Trial is the only large multicentre trial in obstetrics to produce a clear result: this is because the entry criterion, eclampsia, is clinically unequivocal, and so virtually every woman in the trial had sustained an eclamptic seizure34.
The main difficulty with multicentre trials in obstetrics is that too much attention has been paid to design and analysis and too little to entry criteria, which should depend on an understanding of the pathophysiology of the disease. In the past obstetricians and gynaecologists have been mesmerised by considerations of randomisation, minimisation, stratification, trial hypotheses, allocation concealment, baseline characteristics, calculations of trial size, calculations of odds ratios, and calculations of confidence intervals; now they have been introduced to these mysteries, and no longer fear them, for the concepts are not difficult and the mathematics not troublesome. Obstetricians and gynaecologists should take the initiative in the design and analysis of randomised trials in obstetrics and gynaecology, for their knowledge of pathophysiology learned in the laboratory and at the bedside is necessary to establish entry criteria for randomised trials. Obstetricians and gynaecologists should hesitate before asking the women in their care to participate in large multicentre pragmatic trials where the entry criteria are a clinical perception of risk. Instead they should concentrate on smaller explanatory trials in which treatments are intended to alter disease processes; only then can a multicentre trial based on an important clinical outcome be justified. Thus since it is known that low dose aspirin can alter the pathophysiology of proteinuric pre-eclampsia, it is then justified to perform a multicentre trial concentrating on women with abnormal uteroplacental Doppler velocity waveforms where the primary outcome may be perinatal mortality; since these women have a high risk of proteinuric hypertension only a few centres will be required.
For our part the BJOG: An International Journal of Obstetrics & Gynaecology welcomes the submission of observational studies describing the pathophysiology of disease in obstetrics and gynaecology, and of explanatory trials designed to investigate treatments to alter these patho-physiological processes. Archie Cochrane would approve of this approach, for the suffering he encountered in Hildburghausen and Elsterhorst taught him the importance not only of performing randomised trials, but of performing randomised trials in people who actually have disease.