Developing a comorbidity index for the CLSA: application to neurologic conditions

The presence of multiple chronic conditions in the same individual (i.e. comorbidity) has become increasingly common. Much of this increasing burden, especially for diseases of the brain, is driven by our aging population. Currently, the Canadian Longitudinal Study on Aging (CLSA) captures information regarding multiple comorbidities but there is no standardized and scientifically proven method of summarizing their overall burden in an individual. The goal of the proposed study is to develop a summary measure of comorbidity within the CLSA. Once this has been developed in one portion of the CLSA, it will then be applied to other portions of the CLSA to confirm that it performs well. We will carry out sub-analyses to examine what effect age, sex, and language may have on how this algorithm performs. Finally, we will assess how this newly developed summary measure performs among individuals with a disease of the brain.

We completed the development and validation of several alternative methods to predict healthcare utilization in the CLSA, including model selection techniques such as LASSO (least absolute shrinkage and selection operator), Bayesian Model Averaging, and AIC-based (Akaike information criterion) selection. Final index performances were compared using measures of discrimination, calibration, and modelling. We also studied convergent construct validity by checking for cross-sectional correlations between measures of multimorbidity, disability, and satisfaction with life. Ultimately, discrimination for all models was very modest, but when including an interaction term between the absolute sum of chronic conditions and age, the index that emerged had the strongest calibration between average predicted and observed probability of hospitalization. These findings will help to inform the future study of multimorbidity in the CLSA, as well as other study cohorts.