Background

External validation of COVID-19 prognostic models showed that most performed poorly in older populations, except the 4C Mortality model. Frailty status, measured as Clinical Frailty Scale (CFS), can improve prognostication in older adults.

Objective

This study evaluated incremental value of frailty in addition to 4C Mortality model in predicting in-hospital mortality in older hospitalized COVID-19 patients.

Methods

We included individuals aged ≥70 years from the COVID-OLD cohort, hospitalized with COVID-19 in the Netherlands (February 2020 – April 2022). CFS was added to three versions of the 4C Mortality model: original, recalibrated, and revised. Predictive performance was calculated by scaled brier score, discrimination, calibration and visual distribution of predicted risks.

Results

3,067 participants were included in the study with an in-hospital mortality fraction of 31%. External validation of the 4C Mortality model demonstrated a scaled brier score of 0.03 [-0.01 to 0.06], discrimination of 0.70 [0.68 to 0.72] and a calibration slope of 0.88 [0.77 to 0.98]. Adding CFS to original 4C Mortality model marginally improved predictive performance, with scaled brier score of 0.04 [0.01 to 0.07], discrimination to 0.71 [0.69 to 0.73], and a calibration slope of 0.89 [0.79 to 1.00]. Addition of CFS to the recalibrated 4C Mortality model also marginally improved the predictive performance scaled brier score from 0.09 to 0.11, and discrimination from 0.70 to 0.71. Model revision, with and without addition of CFS, displayed limited incremental value of CFS on top of existing predictors in 4C Mortality model with a change in scaled brier score from 0.21 [0.19 to 0.25] to 0.24 [0.22 to 0.28], and discrimination from 0.79 [0.77 to 0.81] to 0.80 [0.78 to 0.81].

Conclusion

Frailty reflected by CFS demonstrated limited incremental value for in-hospital mortality prediction in addition to validated 4C Mortality model. Future research should explore other predictors that capture heterogeneity in older population.

Copyright © 2025 The Author(s). Published by Elsevier Inc. All rights reserved.