International Journal of Technology Assessment in Health Care, 8:Suppl. 1 (1992), 40-45. Copyright © 1992 Cambridge University Press. Printed in the USA.
RANDOMIZED CONTROLLED TRIALS IN THE EVALUATION OF ANTENATAL CARE Murray W. Enkin McMaster University, Hamilton
Abstract Many of the practices carried out during antenatal care improve the well-being of mother or baby and reduce the burden of adverse perinatal outcome. Other practices have either not been evaluated or have been shown to be ineffective. Evidence from randomized clinical trials provides the best evidence about the effectiveness of these practices.
The concept that antenatal care requires evaluation is relatively new. The aims of antenatal care —a healthy mother and baby—are so self-evidently good that it seems heretical to question the means used to try to reach them. As the British Social Security Committee report put it in 1980, "While we unhesitatingly accept the often reiterated claim of antenatal care as a means of reducing perinatal and neonatal mortality, what exactly antenatal care consists of, and how it works, has been less clear to us" (6). The pattern of modern antenatal care has remained unchanged for three-quarters of a century. In recent decades a variety of investigations, procedures, prescriptions, and proscriptions have been added to this pattern. As antenatal care has evolved, there has been a growing tendency to apply to all women measures that are of unquestionable value to only a minority. A great deal of expense, time, and effort goes into the application of these measures. Health care providers, those who pay for the services, and women who receive them have begun to question their value. In what ways and in what circumstances can antenatal care be expected to lead to an improvement in outcome over and above what might be predicted from our knowledge of the natural robustness of human pregnancy? The variation in antenatal care practices from country to country, from community to community, from institution to institution, and from practitioner to practitioner is enormous, considering the similarity in the objectives of care. We see this variation, for apparently well women, in the differences in the frequency of antenatal visits required, in the methods used for risk screening and scoring, and for the evaluation of fetal well-being. Well-intentioned physicians and midwives cannot agree as to whether these women should receive routine iron and vitamin supplements, or whether a vaginal examination should be carried out at each antenatal visit. For women with identified problems, the variation in care practices is even greater. How can we decide which policies of care will be least likely to result in an adverse 40
Evaluation of randomized controlled trials
outcome? Is hospitalization and bed rest beneficial for women with multiple pregnancies? Does cervical cerclage reduce the risk of preterm delivery for those women? Will it be of help to women with a previous mid-trimester loss? Should corticosteroids be used for lung maturation when preterm birth is expected? For women with diabetes who are insulin-dependent, will best results be obtained with tight, very tight, or moderate control of blood glucose levels? For the woman with a breech presentation, should external version be attempted preterm, when the procedure is easier; at term, when the presentation is more likely to persist; or not at all? The large variation in care policies among well-intentioned care providers reflects a collective uncertainty about the effectiveness and safety of the various elements that constitute antenatal care. (Perhaps even more pernicious than the collective uncertainty is the certainty expressed by some individuals, when the best available evidence is inconclusive or contrary. "It's not what we don't know that causes the most trouble, it's what we know, that ain't so!") False inferences about the effects of care will result in some women and babies being denied effective forms of care, while others are offered care that is either ineffective or actually harmful. It is, therefore, vitally important that the evidence on which care decisions are made be as strong as possible. Sometimes, but rarely, past experience will provide a sufficient basis for valid assessment of the effects of care. This will only be the case when the results are dramatically different, and often not even then. Comparisons with so-called "historical controls"—comparing results in women who receive the new form of care with those who received other care in the past — can be grossly misleading. To cite a well-known, tragic example, the results of studies using historical controls led to the widespread belief that administration of diethylstilbesterol during pregnancy would result in a dramatic decrease in the risk of miscarriage and perinatal death. Subsequent randomized comparisons showed that the postulated benefits did not occur; later follow-up studies demonstrated the serious adverse effects of the medication on the offspring of the women who received it (4). Case control studies and nonrandomized, matched concurrent controls represent attempts to reduce biases in selection of the control group. Sometimes these are the only forms of studies that are feasible, and represent the best evidence that we are likely to get. Causal inferences based on such weaker studies, however, are often insecure. Sometimes their conclusions are supported, and sometimes not supported, by subsequent, better controlled studies. No amount of matching using information about known confounding factors can ever eliminate the effect of unrecognized confounding variables. There is only one certain way to overcome the bias that results from women at different prior risk selectively receiving one of the alternative forms of care being compared. This is to conduct a prospective experiment in which randomization—the play of chance — is used to decide which of the alternative forms of care a particular woman will receive. Randomization not only controls for selection biases known to be important, it is the only known way to control for unknown selection biases. A randomized controlled trial is thus the only form of evaluation of alternative forms of care in which we can be reasonably confident that an unbiased comparison has been made.
WHAT DATA ARE AVAILABLE FROM RANDOMIZED TRIALS? Data from randomized trials are available to address many questions about antenatal care. The Oxford Database of Perinatal Trials (1) currently lists 693 published trials of prophylaxis or therapy in antenatal care, as well as 104 unpublished, planned, or INTL. J. OF TECHNOLOGY ASSESSMENT IN HEALTH CARE 8:SUPPL. 1, 1992
41
Enkin Table 1 . Controlled Trials of Antenatal Care Listed on the Oxford Database of Perinatal Trials (1)a Published
Unpublished
Planned/ ongoing
Overviews
14 18 35 65 28 208 385 6 114 23 38
4 2 4 3 6 4 23 0 12 7 2
4 3 13 6 8 5 25 1 11 6 6
1 4 6 2 2 7 46 5 8 6 6
Clinical screening and diagnosis Laboratory tests Radiology/ultrasound Special diagnostic procedures Routine care in pregnancy Nutrition interventions Drugs and medication Surgery in pregnancy Counseling, education Hospitalization in pregnancy Miscellaneous interventions in pregnancy
' Some papers listed may be additional reports of the same trial. Some trials may be listed under more than one intervention compared.
ongoing trials. The results of trials of similar interventions have, wherever possible, been combined into 75 overviews. The database also lists 122 published trials of antenatal diagnostic procedures, along with 10 unpublished trials, 20 ongoing or planned trials, and 10 overviews. Table 1 lists the fields of interventions covered by these trials. I have selected trials of three different types of procedures to illustrate the role of randomized trials as a guide to effective antenatal care: anti-smoking interventions; bed rest and hospitalization for multiple pregnancies; and the use of corticosteroids for pulmonary maturation. After a brief discussion of these, I address the more complicated role of randomized controlled trials in evaluation of diagnostic tests. Anti-Smoking Interventions (7)
About 20% of pregnant women who smoke quit by the time of their first antenatal visit. Only a small proportion (
RANDOMIZED CONTROLLED TRIALS IN THE EVALUATION OF ANTENATAL CARE Murray W. Enkin McMaster University, Hamilton
Abstract Many of the practices carried out during antenatal care improve the well-being of mother or baby and reduce the burden of adverse perinatal outcome. Other practices have either not been evaluated or have been shown to be ineffective. Evidence from randomized clinical trials provides the best evidence about the effectiveness of these practices.
The concept that antenatal care requires evaluation is relatively new. The aims of antenatal care —a healthy mother and baby—are so self-evidently good that it seems heretical to question the means used to try to reach them. As the British Social Security Committee report put it in 1980, "While we unhesitatingly accept the often reiterated claim of antenatal care as a means of reducing perinatal and neonatal mortality, what exactly antenatal care consists of, and how it works, has been less clear to us" (6). The pattern of modern antenatal care has remained unchanged for three-quarters of a century. In recent decades a variety of investigations, procedures, prescriptions, and proscriptions have been added to this pattern. As antenatal care has evolved, there has been a growing tendency to apply to all women measures that are of unquestionable value to only a minority. A great deal of expense, time, and effort goes into the application of these measures. Health care providers, those who pay for the services, and women who receive them have begun to question their value. In what ways and in what circumstances can antenatal care be expected to lead to an improvement in outcome over and above what might be predicted from our knowledge of the natural robustness of human pregnancy? The variation in antenatal care practices from country to country, from community to community, from institution to institution, and from practitioner to practitioner is enormous, considering the similarity in the objectives of care. We see this variation, for apparently well women, in the differences in the frequency of antenatal visits required, in the methods used for risk screening and scoring, and for the evaluation of fetal well-being. Well-intentioned physicians and midwives cannot agree as to whether these women should receive routine iron and vitamin supplements, or whether a vaginal examination should be carried out at each antenatal visit. For women with identified problems, the variation in care practices is even greater. How can we decide which policies of care will be least likely to result in an adverse 40
Evaluation of randomized controlled trials
outcome? Is hospitalization and bed rest beneficial for women with multiple pregnancies? Does cervical cerclage reduce the risk of preterm delivery for those women? Will it be of help to women with a previous mid-trimester loss? Should corticosteroids be used for lung maturation when preterm birth is expected? For women with diabetes who are insulin-dependent, will best results be obtained with tight, very tight, or moderate control of blood glucose levels? For the woman with a breech presentation, should external version be attempted preterm, when the procedure is easier; at term, when the presentation is more likely to persist; or not at all? The large variation in care policies among well-intentioned care providers reflects a collective uncertainty about the effectiveness and safety of the various elements that constitute antenatal care. (Perhaps even more pernicious than the collective uncertainty is the certainty expressed by some individuals, when the best available evidence is inconclusive or contrary. "It's not what we don't know that causes the most trouble, it's what we know, that ain't so!") False inferences about the effects of care will result in some women and babies being denied effective forms of care, while others are offered care that is either ineffective or actually harmful. It is, therefore, vitally important that the evidence on which care decisions are made be as strong as possible. Sometimes, but rarely, past experience will provide a sufficient basis for valid assessment of the effects of care. This will only be the case when the results are dramatically different, and often not even then. Comparisons with so-called "historical controls"—comparing results in women who receive the new form of care with those who received other care in the past — can be grossly misleading. To cite a well-known, tragic example, the results of studies using historical controls led to the widespread belief that administration of diethylstilbesterol during pregnancy would result in a dramatic decrease in the risk of miscarriage and perinatal death. Subsequent randomized comparisons showed that the postulated benefits did not occur; later follow-up studies demonstrated the serious adverse effects of the medication on the offspring of the women who received it (4). Case control studies and nonrandomized, matched concurrent controls represent attempts to reduce biases in selection of the control group. Sometimes these are the only forms of studies that are feasible, and represent the best evidence that we are likely to get. Causal inferences based on such weaker studies, however, are often insecure. Sometimes their conclusions are supported, and sometimes not supported, by subsequent, better controlled studies. No amount of matching using information about known confounding factors can ever eliminate the effect of unrecognized confounding variables. There is only one certain way to overcome the bias that results from women at different prior risk selectively receiving one of the alternative forms of care being compared. This is to conduct a prospective experiment in which randomization—the play of chance — is used to decide which of the alternative forms of care a particular woman will receive. Randomization not only controls for selection biases known to be important, it is the only known way to control for unknown selection biases. A randomized controlled trial is thus the only form of evaluation of alternative forms of care in which we can be reasonably confident that an unbiased comparison has been made.
WHAT DATA ARE AVAILABLE FROM RANDOMIZED TRIALS? Data from randomized trials are available to address many questions about antenatal care. The Oxford Database of Perinatal Trials (1) currently lists 693 published trials of prophylaxis or therapy in antenatal care, as well as 104 unpublished, planned, or INTL. J. OF TECHNOLOGY ASSESSMENT IN HEALTH CARE 8:SUPPL. 1, 1992
41
Enkin Table 1 . Controlled Trials of Antenatal Care Listed on the Oxford Database of Perinatal Trials (1)a Published
Unpublished
Planned/ ongoing
Overviews
14 18 35 65 28 208 385 6 114 23 38
4 2 4 3 6 4 23 0 12 7 2
4 3 13 6 8 5 25 1 11 6 6
1 4 6 2 2 7 46 5 8 6 6
Clinical screening and diagnosis Laboratory tests Radiology/ultrasound Special diagnostic procedures Routine care in pregnancy Nutrition interventions Drugs and medication Surgery in pregnancy Counseling, education Hospitalization in pregnancy Miscellaneous interventions in pregnancy
' Some papers listed may be additional reports of the same trial. Some trials may be listed under more than one intervention compared.
ongoing trials. The results of trials of similar interventions have, wherever possible, been combined into 75 overviews. The database also lists 122 published trials of antenatal diagnostic procedures, along with 10 unpublished trials, 20 ongoing or planned trials, and 10 overviews. Table 1 lists the fields of interventions covered by these trials. I have selected trials of three different types of procedures to illustrate the role of randomized trials as a guide to effective antenatal care: anti-smoking interventions; bed rest and hospitalization for multiple pregnancies; and the use of corticosteroids for pulmonary maturation. After a brief discussion of these, I address the more complicated role of randomized controlled trials in evaluation of diagnostic tests. Anti-Smoking Interventions (7)
About 20% of pregnant women who smoke quit by the time of their first antenatal visit. Only a small proportion (
