Meta-analysis of diagnostic test studies using individual patient data and aggregate data

Area of Science:

• Medical Statistics
• Diagnostic Test Evaluation
• Evidence Synthesis

Background:

• Meta-analysis of diagnostic tests typically uses aggregate data (AD), limiting tailored clinical results due to heterogeneity and 'average' patient focus.
• Bivariate random-effects meta-analysis (BRMA) synthesizes AD but struggles with between-study heterogeneity and generalizability.

Purpose of the Study:

• To develop and assess meta-analysis models using individual patient data (IPD) for more personalized diagnostic test accuracy results.
• To incorporate study-level and patient-level covariates to explain heterogeneity and assess patient characteristic effects on test accuracy.
• To extend models for combining IPD and AD studies.

Main Methods:

• Developed IPD models extending the BRMA framework with study-level and patient-level covariates.
• Carefully separated within-study and across-study accuracy-covariate effects to avoid confounding.
• Assessed models via simulation and applied them to 23 studies on ear thermometer accuracy for diagnosing fever in children (16 IPD, 7 AD).

Main Results:

• IPD models successfully incorporated covariates to explain between-study heterogeneity.
• Heterogeneity was partly explained by different measurement devices used in studies.
• No evidence found that infant status significantly modifies diagnostic accuracy.

Conclusions:

• IPD meta-analysis provides more tailored clinical results by accounting for patient-level covariates.
• The developed models effectively explain heterogeneity and can combine IPD with AD studies.
• Ear thermometer accuracy for fever diagnosis in children is influenced by device type but not infant status.