Improving External Prediction in New Clusters with Mixed-Effects Models: Implications for Multicenter Studies

Prediction models developed from multicenter data are often applied to new centers. A linear mixed model (LMM) captures between-center heterogeneity through random effects. However, a new center's random effect is unknown, so the LMM predicts from fixed effects alone, as does a standard linear model (LM) that ignores clustering. The two models thus differ only in how fixed effects are estimated, and it is unclear whether the LMM predicts better. We decompose the prediction error in a new center into fixed-effect estimation error, between-center variation, and residual variation. Because the last two components are identical for both models, any LMM advantage must come from more precise fixed-effect estimates. In simulations across diverse designs and random-effect structures, the LMM generally achieved lower mean squared error than the LM. Relative gains increased with the number of predictors, between-center heterogeneity, and imbalance in center sizes, and decreased as training data grew. Modeling a random slope yielded additional gains that increased with between-center variation in the predictor effect. In a stress test using held-out eICU hospitals selected for strong outcome heterogeneity, a random-intercept LMM predicted ICU length of stay more accurately than an LM. These findings clarify the conditions under which mixed models can improve prediction in new centers and can inform model development for multicenter studies.

Authors

Publication Details

Journal
arXiv (Cornell University)
Published
2026-11-18
DOI
https://doi.org/10.48550/arxiv.2611.24929
Primary Topic
Statistical Methods and Bayesian Inference
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Improving External Prediction in New Clusters with Mixed-Effects Models: Implications for Multicenter Studies

Myeonggyun Lee, Hyungrok Do, Jiawa Zhang
arXiv (Cornell University)
Statistical Methods and Bayesian Inference
preprint

Improving External Prediction in New Clusters with Mixed-Effects Models: Implications for Multicenter Studies

Myeonggyun Lee, Hyungrok Do, Jiawa Zhang
preprint en

Abstract

Prediction models developed from multicenter data are often applied to new centers. A linear mixed model (LMM) captures between-center heterogeneity through random effects. However, a new center's random effect is unknown, so the LMM predicts from fixed effects alone, as does a standard linear model (LM) that ignores clustering. The two models thus differ only in how fixed effects are estimated, and it is unclear whether the LMM predicts better. We decompose the prediction error in a new center into fixed-effect estimation error, between-center variation, and residual variation. Because the last two components are identical for both models, any LMM advantage must come from more precise fixed-effect estimates. In simulations across diverse designs and random-effect structures, the LMM generally achieved lower mean squared error than the LM. Relative gains increased with the number of predictors, between-center heterogeneity, and imbalance in center sizes, and decreased as training data grew. Modeling a random slope yielded additional gains that increased with between-center variation in the predictor effect. In a stress test using held-out eICU hospitals selected for strong outcome heterogeneity, a random-intercept LMM predicted ICU length of stay more accurately than an LM. These findings clarify the conditions under which mixed models can improve prediction in new centers and can inform model development for multicenter studies.

arXiv (Cornell University)
Statistical Methods and Bayesian Inference
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon