Using individual patient data to adjust for indirectness did not successfully remove the bias in this case of comparative test accuracy

Abstract

Objectives: In comparative systematic reviews of diagnostic accuracy, inconsistencies between direct and indirect comparisons may lead to bias. We investigated whether using individual patient data (IPD) can adjust for this form of bias.

Study design and setting: We included IPD of 3 ovarian reserve tests from 32 studies. Inconsistency was defined as a statistically significant difference in relative accuracy or different comparative results between the direct and indirect evidence. We adjusted for the effect of threshold and reference standard, as well as for patient-specific variables.

Results: Anti-Müllerian hormone (AMH) and follicle stimulation hormone (FSH) differed significantly in sensitivity (-0.1563, P = 0.04). AMH and antral follicle count (AFC) differed significantly in sensitivity (0.1465, P < 0.01). AMH and AFC differed significantly in specificity (-0.0607, P = 0.02). The area under the curve (AUC) differed significantly between AFC and FSH (0.0948, P < 0.01) in the direct comparison but not (0.0678, P = 0.09) in the indirect comparison. The AUCs of AFC and AMH differed significantly (-0.0830, P < 0.01) in the indirect comparison but not (-0.0176, P = 0.29) in the direct comparison. These differences remained after adjusting for indirectness.

Conclusion: Estimates of comparative accuracy obtained through indirect comparisons are not always consistent with those obtained through direct comparisons. Using IPD to adjust for indirectness did not successfully remove the bias in this case study.

Keywords: Comparative meta-analysis; Diagnostic test accuracy; Generalized estimating equation; Individual patient data; Receiver operating characteristic; Sensitivity and specificity.