Using harmonised results of different tests for a single biomarker in test accuracy meta-analysis

Presentation video:




Oral session: Diagnostic test accuracy review


Tuesday 22 October 2019 - 11:00 to 12:30


All authors in correct order:

Vali Y1, Lee J1, Zafarmand MH1, Boursier J2, Bossuyt PM1
1 Department of Clinical Epidemiology, Biostatistics & Bioinformatics, Amsterdam UMC, Amsterdam, The Netherlands
2 Hepato-Gastroenterology Department, Angers University Hospital, HIFIH Laboratory, UPRES EA3859, Angers University, France
Presenting author and contact person

Presenting author:

Yasaman Vali

Contact person:

Abstract text
Background: evaluating the performance of a biomarker can be challenging when different tests exist for measuring the same marker. Along with other obvious sources of heterogeneity in systematic reviews of diagnostic test accuracy (DTA) studies, this can further influence and confound the results of a meta-analysis.

Objectives: we here propose a strategy to combine multiple tests to measure the same marker in a single meta-analysis. We apply this strategy to a meta-analysis of DTA studies of the Enhanced Liver Fibrosis (ELF) test, used in non-alcoholic fatty liver disease patients.

Methods: our systematic search in five databases identified nine studies. Two different ELF tests were proposed, each using a different formula, expressed on a different scale. We initially conducted two separate meta-analyses, accounting for the multiple thresholds (diagmeta package in R). We then 1) evaluated, in a separate study of 502 samples, the presence of a linear relationship between the results of the tests. We 2) used the regression equation to obtain harmonised test results and 3) performed a single meta-analysis, combining the results from all nine studies.

Results: six studies used one formula (Siemens) and three used another (Guha). The first meta-analysis of the six studies resulted in an 'optimal' threshold (maximum Youden) of 9.19 (8.85 to 9.55), for a sensitivity of 0.77 (95% confidence interval (CI) 0.63 to 0.87) and a specificity of 0.73 (95% CI 0.57 to 0.85). After checking the linearity (R2: 0.995) and mapping the results on the same scale (Figure 1), a meta-analysis of all nine studies was possible. This resulted in an 'optimal' threshold of 7.63 (4.44 to 10.82) for a sensitivity of 0.88 (95% CI 0.59 to 0.99) and a specificity of 0.72 (95% CI 0.33 to 0.94; Figure 2).

Conclusions: our three-step method allows the combination of multiple tests of the same marker in a single meta-analysis, facilitating the interpretation of the accuracy of using specific thresholds.

Patient or healthcare consumer involvement: combining studies of multiple tests of the same marker in a single meta-analysis allows a synthesis of all the available evidence, and an informed selection of the threshold. This allows evidence-based, medical decision making, eventually leading to improved patient outcomes.