Some Tests Based on the Profile Likelihood Estimator for Testing Homogeneity of Diagnostic Odds Ratios in Meta-analysis

Abstract

© 2016 The Authors. This research aims to propose some modified tests "χ2PLE1" and "χ2PLE2" based on the profile likelihood estimator for testing homogeneity of diagnostic odds ratios in meta-analysis and compare their performances with the conventional tests of QWLS, QMH and χ2Con. According to the performance in terms of type I error rates under H0 and power of tests under H1, Monte Carlo simulation with R language was applied. The results found that all of tests cannot control type I error rates when sample sizes are small (nDi, nHi≤5), regardless of study size (k). However, for k ≥ 16 in combination with nDi, nHi ≤ 50, three tests (χ2PLE1, QWLS, χ2Con) can control type I error rates in almost all situations. In addition, the profile test (χ2PLE1) performs best with highest power when nDi, nHi = 50,100 for k ≥ 16, while conventional tests of QWLS and χ2Con perform well with the same power as the profile test (χ2PLE1) when nDi, nHi = 500 for k ≥ 16. Therefore, the χ2PLE1 is recommended to be used when k ≥ 16 in combination with nDi, nHi = 500.