Resolving discrepancies between a meta-analysis and a subsequent large controlled trial

Context A recent mete-analysis found calcium supplementation to be highly effective in preventing preeclampsia but a large National Institutes of Health trial (Calcium for Preeclampsia Prevention [CPEP]) found no risk reduction due to calcium in healthy nulliparous women. Objectives To resolve discrepancies between the results of the meta-analysis and the CPEP trial and to assess the role of effect heterogeneity in the discrepancies. Data Sources Literature search of English-language articles published prior to July 10, 1997, the date of publication of the CPEP trial, using MEDLINE and by a manual search of bibliographies of published articles. Study Selection Trials were included if they reported data on preeclampsia and calcium supplementation. Fourteen trials were systematically evaluated for differences in study design and patient populations. One trial was excluded because its results were reported after publication of the major CPEP results. Data Extraction The sample size and number of subjects who developed preeclampsia in the calcium supplementation group vs a control group were recorded and analyzed on an intent-to-treat basis. Each author independently extracted the data. Data Synthesis Substantial heterogeneity existed across trials (P = .001). After stratifying studies by the presence of a placebo-controlled group and by high-risk and low-risk populations, the conclusions of the meta-analysis of placebo-controlled trials enrolling a low-risk population (relative risk, 0.79; 99% confidence interval, 0.44-1.42; P = .30) were compatible with the conclusions of the CPEP trial that calcium supplementation does not prevent preeclampsia in healthy nulliparous women. In contrast, the data implied a strong beneficial calcium effect (relative risk, 0.19; 99% confidence interval, 0.08-0.46; P = .001) in healthy high-risk subject populations. However, only 225 women were analyzed and because of inconsistent data, these results remain equivocal. Conclusions Further studies are needed to establish the efficacy of calcium for preeclampsia prevention in healthy high-risk populations. A single summary measure does not adequately describe the findings of a meta-analysis when the observed effects in individual studies differ substantially. In such settings the primary focus should be to identify and incorporate pertinent covariates that reduce heterogeneity and allow for optimum treatment strategies.