A Nonparametric Random-Effects Meta-Analysis Model for Diagnostic Accuracy Studies with Multiple Thresholds of Quantitative Biomarkers

We introduce a nonparametric random-effects model for the meta-analysis of a series of diagnostic accuracy studies of a quantitative biomarker that reported sensitivity-and specificity-values at multiple thresholds of the biomarker. The model is based on the observed numbers of cases and controls in between cutoff-values of the biomarker.

Observed numbers of cases and controls within studies were modeled using multinomial-normal or multinomial-dirichlet distributions. Parameters of our new model were Estimated using an MCMC-algorithm in a Bayesian framework. We provide code to run our method within the R statistical software. With our new model two example datasets were analyzed and compared to the method implemented in the R package diagmeta.

With two example datasets our approach gave comparable results as the method implemented in diagmeta. Results are illustrated using estimated density and cumulative distribution function of the biomarker and associated credibility intervals. Transformations thereof such as ROCcurve, area under the ROC curve, and the Youden-index curve are also estimated together with credibility intervals.

We developed a new model for meta-analysis of diagnostic studies evaluating multiple thresholds of a quantitative biomarker. The new model provided comparable results as an existing method but with less assumptions.

Keywords:Diagnostic test; Meta-analysis; Multiple thresholds

Abbreviations:AUC: Area Under the Curve; CI: Common random Intercept; CS: Common random Slope; DICS: Different random Intercepts and Common random Slope; DIDS: Different random Intercepts and Different random Slopes; ELF: enhanced liver fibrosis; FENO: fractional exhaled nitric oxide; HA: hyaluronic acid; JAGS: Just Another Gibbs Sampler; MCMC: Multi Chain Monte Carlo; PIIINP: Procollagen III N-terminal Propeptide; ROC: Receiver Operator Curve; TIMP: tissue inhibitor of metalloproteinase-1