You are here:
Evaluating the consistency of heterogeneous results: important determinants of inconsistency in epidemiological evidence
Citation:
Glenn, B., E. RadkeFarabaugh, AND A. Kraft. Evaluating the consistency of heterogeneous results: important determinants of inconsistency in epidemiological evidence. NAS Evidence Integration Workshop, District of Columbia, Washington, June 03 - 04, 2019.
Impact/Purpose:
The evidence on associations of health effects with chemical exposures that have an extensive research history often includes studies with heterogeneous results, complicating the ability to draw conclusions about whether a hazard is indicated. This case study discusses the impact of bias and other aspects of studies that could influence associations with health effects, and illustrates an approach to analyzing their impact using examples from systematic reviews conducted in the IRIS Program. While the specific determinants may vary, sensitivity analyses that examine potential determinants of heterogeneity in study results are essential to analyses of evidence consistency as part of the integration of evidence in systematic reviews.
Description:
The examples of using sensitivity analysis to evaluate inconsistency in evidence will be presented in a poster to illustrate the value of analyzing the impact of bias and other aspects of studies that could influence the magnitude and direction of associations with health effects in epidemiological studies. The consistency of a set of results is examined via forest plots presenting effect estimates (e.g., risk ratios, odds ratios) stratified by ratings for the domains that comprise the IRIS study evaluation tool including participant selection, exposure, outcome, confounding, analysis and sensitivity. Additional factors are analyzed including exposure (low vs high) and overall study confidence. The case examples include studies from a variety of exposure settings (i.e., population-based and occupational studies) that appear to have considerable heterogeneity across studies for specific outcomes. However, when the effect estimates are stratified by exposure level and setting, and overall confidence in the exposure-outcome association, a more consistent pattern emerges.