What is the predictive performance of a claims-based frailty indicator in Belgium?
Abstract
Background: Frailty is increasingly prevalent and is expected to rise further as the population ages. This trend highlights the need for population-level screening and monitoring to inform effective health service planning. However, standard clinical and survey-based measures are not widely available at this level. This study investigates whether Belgian Compulsory Health Insurance (BCHI) data can be used to develop and validate a reliable frailty indicator.
Methods: Data on Fried’s frailty phenotype from participants aged ≥65 years in the Belgian Health Interview Survey 2018 (n = 2,299) were linked to BCHI data. Backward stepwise logistic regression identified reimbursement for medications related to pain, depression, diabetes, and the cardiovascular and respiratory systems, along with sociodemographic characteristics and a healthcare consumption–based indicator of chronic disease status, as significant predictors of frailty. Model performance was assessed in terms of discrimination (AUC-ROC), calibration (Brier score), and classification metrics (sensitivity, specificity, positive predictive value [PPV], and negative predictive value [NPV]) using a probability threshold of 0.5.
Results: Frailty prevalence was 22.8% based on survey data and 13.1% based on model predictions. The model showed good discrimination (AUC 0.81; 95% CI 0.79–0.83) and a good calibration (Brier score 0.13). Specificity was high (94.7%), with PPV of 68.9% and NPV of 84.1%, but sensitivity was low (39.6%).
Conclusions: Health insurance claims data show potential for predicting frailty at the population level in Belgium, but low sensitivity limits their usefulness for screening. This limitation likely reflects the absence of certain frailty-related aspects in the data. Future research should examine additional predictors—particularly healthcare utilization patterns—and refine probability thresholds to improve model performance.
Key messages: • Claims data enable scalable frailty monitoring. They can track frailty at population level, but miss many frail individuals.
• Better data and model tuning are needed for policy use. Improving sensitivity is key to support effective planning and targeting of care.
Related
- Performance of Claims-based Alzheimer’s Disease and Related Dementias Algorithms in Medicare Advantage and Fee-for-Service Populations
- Predictive value of the combined cholesterol, high-density lipoprotein, glucose and frailty indices for cardiometabolic multimorbidity incidence: evidence from a national prospective cohort study
- An operational key performance indicator framework and reference data from a multi-room pencil beam scanning proton therapy center
- Toward a Predictive Model for Medical Residency Entrance Examination Performance in Brazil: A Preregistered Cross-Sectional Study and Exploratory Prototype Score Calculator
- Comparative performance of multiple predictive models constructed for budesonide nebulized formulation requirements
BibTeX
@article{heyden2026what,
title = {{What is the predictive performance of a claims-based frailty indicator in Belgium?}},
author = {JH Van der Heyden and R Charafeddine and S Demarest and N Biegel and H Bruggeman},
journal = {The European Journal of Public Health},
year = {2026},
month = oct,
doi = {10.1093/eurpub/ckag169.1889},
url = {https://doi.org/10.1093/eurpub/ckag169.1889},
note = {CC BY-NC 4.0},
}
RIS
TY - JOUR
TI - What is the predictive performance of a claims-based frailty indicator in Belgium?
AU - JH Van der Heyden
AU - R Charafeddine
AU - S Demarest
AU - N Biegel
AU - H Bruggeman
PY - 2026
DA - 2026/10/01
JO - The European Journal of Public Health
DO - 10.1093/eurpub/ckag169.1889
UR - https://doi.org/10.1093/eurpub/ckag169.1889
AB - Background: Frailty is increasingly prevalent and is expected to rise further as the population ages. This trend highlights the need for population-level screening and monitoring to inform effective health service planning. However, standard clinical and survey-based measures are not widely available at this level. This study investigates whether Belgian Compulsory Health Insurance (BCHI) data can be used to develop and validate a reliable frailty indicator. Methods: Data on Fried’s frailty phenotype from participants aged ≥65 years in the Belgian Health Interview Survey 2018 (n = 2,299) were linked to BCHI data. Backward stepwise logistic regression identified reimbursement for medications related to pain, depression, diabetes, and the cardiovascular and respiratory systems, along with sociodemographic characteristics and a healthcare consumption–based indicator of chronic disease status, as significant predictors of frailty. Model performance was assessed in terms of discrimination (AUC-ROC), calibration (Brier score), and classification metrics (sensitivity, specificity, positive predictive value [PPV], and negative predictive value [NPV]) using a probability threshold of 0.5. Results: Frailty prevalence was 22.8% based on survey data and 13.1% based on model predictions. The model showed good discrimination (AUC 0.81; 95% CI 0.79–0.83) and a good calibration (Brier score 0.13). Specificity was high (94.7%), with PPV of 68.9% and NPV of 84.1%, but sensitivity was low (39.6%). Conclusions: Health insurance claims data show potential for predicting frailty at the population level in Belgium, but low sensitivity limits their usefulness for screening. This limitation likely reflects the absence of certain frailty-related aspects in the data. Future research should examine additional predictors—particularly healthcare utilization patterns—and refine probability thresholds to improve model performance. Key messages: • Claims data enable scalable frailty monitoring. They can track frailty at population level, but miss many frail individuals. • Better data and model tuning are needed for policy use. Improving sensitivity is key to support effective planning and targeting of care.
SN - 1101-1262
SN - 1464-360X
ER -