Frailty-aware extreme gradient boosting for high-dimensional clustered time-to-event outcomes

Survival data arising from multiple centres or studies frequently exhibit cluster-level heterogeneity that is not accounted for by observed covariates. High-dimensional prognostic modelling therefore requires methods that can jointly accommodate such unobserved heterogeneity, nonlinear covariate effects, and principled variable selection. We propose mixed-effects extreme gradient boosting (MEXGB), a hybrid approach that integrates a relaxed-lasso Cox component with Newton-boosted decision trees and log-normal cluster-level frailties. Frailty parameters are estimated through a Poisson representation of the Cox model using cross-fitted risk predictions, and a calibration slope is subsequently applied to rescale the fixed-effect log-hazard contribution. Under explicitly stated assumptions, we derive theoretical guarantees that characterise the loss in predictive accuracy attributable to either ignoring clustering or enforcing a misspecified linear log-hazard structure. In simulation studies involving up to 2,000 predictors, MEXGB demonstrated improved predictive accuracy compared with the strongest competing method, particularly in settings where alternative approaches either neglected clustering structure or imposed an incorrect functional form. Performance was comparable to that of a correctly specified frailty model, and MEXGB attained the lowest integrated Brier score under a common procedure for constructing individual survival curves. In applications to a herd study and to multi-study breast- and ovarian-cancer cohorts, MEXGB achieved the highest mean predictive accuracy among the evaluated methods.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-20
DOI
https://doi.org/10.21203/rs.3.rs-11789696/v1
Primary Topic
Statistical Methods and Inference
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Frailty-aware extreme gradient boosting for high-dimensional clustered time-to-event outcomes

Nada Abdullah Alharbi, Samuel Adewale Aderoju, Oyebayo Ridwan Olaniran, Saheed Ajibade Kunle et al.
4 links
Research Square
Statistical Methods and Inference
preprint

Frailty-aware extreme gradient boosting for high-dimensional clustered time-to-event outcomes

Nada Abdullah Alharbi, Samuel Adewale Aderoju, Oyebayo Ridwan Olaniran, Saheed Ajibade Kunle, Ayodeji Idowu, Abdulhameed Bolakale
preprint en

Abstract

Survival data arising from multiple centres or studies frequently exhibit cluster-level heterogeneity that is not accounted for by observed covariates. High-dimensional prognostic modelling therefore requires methods that can jointly accommodate such unobserved heterogeneity, nonlinear covariate effects, and principled variable selection. We propose mixed-effects extreme gradient boosting (MEXGB), a hybrid approach that integrates a relaxed-lasso Cox component with Newton-boosted decision trees and log-normal cluster-level frailties. Frailty parameters are estimated through a Poisson representation of the Cox model using cross-fitted risk predictions, and a calibration slope is subsequently applied to rescale the fixed-effect log-hazard contribution. Under explicitly stated assumptions, we derive theoretical guarantees that characterise the loss in predictive accuracy attributable to either ignoring clustering or enforcing a misspecified linear log-hazard structure. In simulation studies involving up to 2,000 predictors, MEXGB demonstrated improved predictive accuracy compared with the strongest competing method, particularly in settings where alternative approaches either neglected clustering structure or imposed an incorrect functional form. Performance was comparable to that of a correctly specified frailty model, and MEXGB attained the lowest integrated Brier score under a common procedure for constructing individual survival curves. In applications to a herd study and to multi-study breast- and ovarian-cancer cohorts, MEXGB achieved the highest mean predictive accuracy among the evaluated methods.

Research Square
University of Ilorin (NG), King's College London (GB), Taibah University (SA), Babcock University (NG), Kwara State University (NG)
Good health and well-being
Statistical Methods and 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