loading page

Meta-Analysis of mutual information applied in EBM diagnostics
  • Athanasios Tsalatsanis,
  • Iztok Hozo,
  • Benjamin Djulbegovic
Athanasios Tsalatsanis
University of South Florida
Author Profile
Iztok Hozo
Indiana University Northwest
Author Profile
Benjamin Djulbegovic
City of Hope National Medical Center
Author Profile

Peer review status:IN REVISION

01 Jul 2020Submitted to Journal of Evaluation in Clinical Practice
02 Jul 2020Assigned to Editor
02 Jul 2020Submission Checks Completed
03 Jul 2020Reviewer(s) Assigned
11 Aug 2020Review(s) Completed, Editorial Evaluation Pending
11 Aug 2020Editorial Decision: Revise Major


Rationale Assessing the performance of diagnostic tests requires evaluation of the amount of diagnostic uncertainty the test reduces (i.e. 0% - useless test, 100% - perfect test). Statistical measures currently dominating the evidence-based medicine (EBM) field and particularly meta-analysis (e.g. sensitivity and specificity), cannot explicitly measure this uncertainty reduction. Mutual information (MI), an information theory statistic, is a more appropriate metric for evaluating diagnostic tests as it explicitly quantifies uncertainty and, therefore, facilitates natural interpretation of a test’s value. In this paper, we propose the use of MI as a single measure to express diagnostic test performance and demonstrate how it can be used in meta-analysis of diagnostic test studies. Methods We use two cases from the literature to demonstrate the applicability of MI meta-analysis in assessing diagnostic performance. These cases are: 1) Meta-analysis of studies evaluating ultrasonography (US) to detect endometrial cancer and 2) meta-analysis of studies evaluating magnetic resonance angiography to detect arterial stenosis. Results Results produced by the MI meta-analyses are comparable to the results of meta-analyses based on traditionally used statistical measures. However, the results of MI are easier to understand as it relates directly to the extent of uncertainty a diagnostic test can reduce. For example, a US test diagnosing endometrial cancer is 40% specific and 94% sensitive. The combination of these values is difficult to interpret and may lead to inappropriate assessment (e.g. one could favour the test due to its high sensitivity, ignoring its low specificity). In terms of MI however, the test reduces diagnostic uncertainty by 10%, which is marginal and thus the test is clearly not very useful. Conclusions We have demonstrated the suitability of MI in assessing the performance of diagnostic tests, which can facilitate easier interpretation of the true utility of diagnostic tests.