The prediction accuracy of dynamic mixed-effects models in clustered data

Area of Science:

• Biostatistics
• Health Informatics
• Clinical Epidemiology

Background:

• Clinical prediction models struggle with clustered data due to heterogeneity in outcomes and covariate effects.
• Standard models like generalized linear mixed-effects models lack accuracy in novel clusters.
• Dynamic mixed-effects models were proposed to improve generalizability by incorporating past predictions.

Purpose of the Study:

• To evaluate the prediction accuracy gains from dynamic mixed-effects models in clustered data.
• To assess the robustness of dynamic models to volume-outcome relationship misspecification.
• To compare dynamic models against static models for clinical prediction.

Main Methods:

• A simulation study was conducted to quantify prediction accuracy improvements.
• Dynamic mixed-effects models were compared to static models.
• The impact of volume-outcome relationship misspecification and updating frequency was examined.

Main Results:

• Dynamic mixed-effects models demonstrated substantial improvements in prediction accuracy across various conditions.
• These models were consistently superior to static models.
• Dynamic models showed robustness to misspecification of the volume-outcome relationship and updating frequency.

Conclusions:

• Dynamic mixed-effects models significantly enhance prediction accuracy in clustered settings.
• They offer a valuable alternative to static models for improving generalizability of clinical predictions.
• Implementation may involve logistical considerations but offers substantial benefits.