An empirical investigation on matching in published case–control studies

Abstract

The methodological discussion about matching when recruiting controls in case-control studies has been controversial for a long time. To delineate the impact of this discussion on the practice of matching we reviewed 266 case-control studies published in nine yearly volumes of three major epidemiological journals within the period 1955-1994. Among studies published until 1980 71.7% of the control groups were recruited by individual matching compared to 46.4% in 1994. This decline is paralleled by an increase in the application of frequency matching (from 5.0% to 26.2%). As the issue of matching is closely connected with methodological questions of the statistical analysis we also examined the type of analysis applied to the data. We found that the use of logistic regression modeling has dramatically increased during this period (from 18.4% up to 87.2%), whereas application of the traditional Mantel-Haenszel technique for estimating summary odds ratios has nearly vanished. The correct approach for individually matched data in the logistic modeling framework, the conditional likelihood technique, has been unknown in the early part of the time window of our investigation, but is even nowadays applied by only three quarters of the corresponding studies. Our literature-based investigation provides thus compelling evidence that the type of control selection and statistical analysis used in case-control studies have changed substantially during recent years.