Development and internal validation of a multimorbidity index that predicts healthcare utilisation using the Canadian Longitudinal Study on Aging.

OBJECTIVES

We aimed to develop and internally validate a measure of multimorbidity burden using data from the Canadian Longitudinal Study on Aging (CLSA).

DESIGN

Data from 40 264 CLSA participants (52% men) aged 45-85 years (a mean of 63 years) were analysed. We used logistic regression models to predict overnight hospitalisation in the last 12 months in the development dataset (random two-thirds of the total) and used these to construct 10 multimorbidity indices (5 models, each treated with and without an age interaction term). Thirty-five chronic conditions were considered for inclusion in these models, in addition to age and sex. We assessed predictive and convergent validity for these 10 different multimorbidity indices in the validation dataset (remaining one-third of the total).

RESULTS

The absolute count of chronic conditions plus an interaction with age, displayed strong calibration properties, outperforming other candidate indices. Discrimination was modest for all of the indices that we internally validated, with C-statistics ranging from 0.66 to 0.68. The indices showed weak correlations (ie, convergent validity) with satisfaction with life, functional disability and mental health (absolute Pearson's correlation coefficients ranging from 0.11 to 0.30) but generally moderate correlations with self-rated general health (0.32-0.45).

CONCLUSIONS

We investigated alternative methods to measure the multimorbidity burden of individuals, tailored to the CLSA. Our findings show that an absolute count of conditions, along with an age interaction term, has the strongest calibration for overnight hospitalisation in the last 12 months. The utility of an age interaction term in measuring multimorbidity burden may be applicable to the study of chronic disease in cohorts other than the CLSA.