Using elastic nets to estimate frailty burden from routinely collected national aged care data

Area of Science:

• Gerontology
• Biostatistics
• Health Informatics

Background:

• Frailty is a key indicator of health status in older adults.
• Accurate frailty measurement is crucial for predicting adverse health outcomes and care needs.
• Traditional frailty measures may not capture the full spectrum of health deficits.

Purpose of the Study:

• To develop a novel frailty measure using an elastic net (EN) algorithm.
• To compare the predictive performance of EN-based and cumulative deficit (CD)-based frailty measures for mortality and permanent care entry.
• To evaluate the impact of weighting on the predictive accuracy of these frailty measures.

Main Methods:

• An elastic net (EN) algorithm was applied to a large national aged care eligibility dataset (2003-2013).
• Cox proportional hazard models were used for item selection and predictive accuracy assessment (AUC).
• Models were trained on 80% of the data and validated on 20%.

Main Results:

• The EN approach identified a 178-item frailty measure, distinct from the 44-item CD measure.
• Unweighted EN and CD measures showed similar predictive performance for mortality and care entry.
• Weighted EN and CD measures demonstrated improved and statistically significant predictive capabilities, with weighted EN slightly outperforming weighted CD.

Conclusions:

• Weighted frailty measures, both EN and CD-based, enhance prediction of mortality and permanent care entry.
• While EN identifies more items, the CD-based measure offers greater interpretability.
• The study recommends utilizing weighted and unweighted CD-based frailty measures for practical application.