Abstract
INTRODUCTION: Long-term care (LTC) residents require extensive assistance with daily activities due to physical and cognitive impairments. Medical treatment for LTC residents, when not aligned with residents' wishes, can cause discomfort without providing substantial benefits. Predictive models can equip providers with tools to guide treatment recommendations that support person-centred medical decision-making. This study protocol describes the derivation and validation of time-to-event predictive models for (1) permanent loss of independence in physical function, (2) permanent severe cognitive impairment and (3) time alive with complete dependence for those with disability starting from the date of onset. METHODS AND ANALYSIS: We will use population-based administrative health data from the Institute for Clinical Evaluative Sciences of all LTC residents in Ontario, Canada, to construct the derivation and internal validation cohorts. The external validation cohort will use data from LTC residents in Alberta, Canada. Predictors were identified based on existing literature, patient advisors and expert opinions (clinical and analytical). We identified 50 variables to predict the loss of independence in physical function, 58 variables to predict the loss of independence in cognitive function and 36 variables to predict the time spent in a state of dependence. We will use time-to-event models to predict the time to loss of independence and time spent in the state of disability. Full and reduced models (using a step-down procedure) will be developed for each outcome. Predictive performance will be assessed in both derivation and validation cohorts using overall measures of predictive accuracy, discrimination and calibration. We will create risk groups to present model risk estimates to users as median time-to-event. Risk groups will be externally validated within the Alberta LTC cohort. ETHICS AND DISSEMINATION: Ethics approval was obtained through the Bruyère Research Institute Ethics Committee. Study findings will be submitted for publication and disseminated at conferences. The predictive algorithm will be available to the general public.