Meta-Analysis Unconfounded: A Case History

Abstract

Prevention of low birth weight, of maternal complications during pregnancy, and other negative sequalae of pregnancy, has been a major focus of efforts by national health bodies and international agencies-most notably the World Health Organization (WHO)for many decades. Maternal anthropometric indicators (e.g. maternal height, pre-pregnancy weight, gestational weight gain and mid-upper arm circumference) are widely used as proxy measures of current or past maternal health and nutritional status, which bear directly or indirectly on pregnancy outcome. An extensive body of information on the interpretation of these indicators as predictors of pregnancy outcome already exists. Nevertheless, there remains an urgent need to provide sound technical advice on the usefulness and efficiency of selected anthropometric indicators for routine application in the context of primary health care, especially in settings in which there is a high mortality or severe morbidity risk to mother and infant. It is estimated by WHO that over 4 million infants die annually in the first week of life, and at least half a million women die as a result of pregnancy each year. Most of these deaths are preventable.In 1990, WHO launched an initiative to identify appropriate data sets to be used to undertake a secondary data analysis in order to provide a systematic reassessment of the available evidence. Subsequently, 25 data sets from 23 countries were identified, collated and brought together on one computer. This provides a unique opportunity to select, compare and contrast various performance measures for individual indicators. This report discusses a number of practical and technical problems associated with the secondary analysis of the individual data sets, and those issues that need to be considered in implementing a formal meta-analysis. Results from both approaches are used to illustrate the advantages and difficulties of meta-analysis.