Testing Reproducibility

Abstract

Poor reproducibility of diagnostic criteria is seldom acknowledged as a cause for low precision in clinical research. Also very few clinical reports communicate the levels of reproducibility of the diagnostic criteria they use. For example, of 11–13 original research papers published per issue in the ten last 2004 issues of the journal Circulation, none did, and of 5–6 original research papers published per issue in the ten last 2004 issues of the Journal of the American Association only 1 out of 12 did. These papers involved quality of life assessments, which are, notoriously, poorly reproducible. Instead, many reports used the averages of multiple measurements in order to improve the precision of the instruments used without further comment on reproducibility. For example, means of three blood pressure measurements, means of three cardiac cycles, average results of morphometric cell studies from two examiners, means of five random fields for cytogenetic studies were reported. Poor reproducibility of diagnostic criteria is, obviously, a recognized but rarely tested problem in clinical research. Evidence-based medicine is under pressure due to the poor reproducibility of clinical trials (Julius 2003; Cleophas and Cleophas 2003). As long as the possibility of poorly reproducible diagnostic criteria has not been systematically addressed, this very possibility cannot be excluded as a contributing cause for this. The current chapter reviews simple methods for routine assessment of reproducibility of diagnostic criteria/tests. These tests can answer questions like (1) do two techniques used to measure a particular variable, in otherwise identical circumstances, produce the same results, (2) does a single observer obtain the same results when he/she takes repeated measurements in identical circumstances, (3) do two observers using the same method of measurement obtain the same result.