OBJECTIVES: Summary health-state measures are essential for risk-adjustment in comparative effectiveness or performance measurement studies and to inform policy-makers. The SNIIRAM database provides morbidity information for each beneficiary of the main health insurance scheme, representing over 57 million individuals, through a set of algorithms combining ICD-10 diagnoses and pharmacy data to identify specific conditions. In this study, we developed a multimorbidity measure predictive of 2-years mortality.

METHODS: All beneficiaries aged 65 years or older in December 31, 2013 with complete follow-up until December 31, 2015 (n=7,672,111) were randomly divided into a training (n=3,836,056) and a validation population (n=3,836,055). We used age, gender and selected indicator morbidity variables as predictors for 2-years mortality in multivariate logistic regression analyses, in the training population. Predictors were selected according to face validity of their identifying algorithm, number of events, univariate association with mortality and principal components analysis within disease categories. To derive a weighted index from the adjusted regression coefficients, we applied a scoring rule where each additional point reflected risk associated with a 5-years age increase. We assessed the predictive performance of the computed index in the validation population, using discrimination and calibration measures.

RESULTS: Adjusted Odds Ratios [95% confidence interval] for the fifteen selected conditions ranged from 1.35 [1.31-1.38] for Depression, 1.36 [1.32-1.40] for Rheumatic or connective tissue diseases, to 3.89 [3.72-4.07] for End-Stage Renal disease, 4.06 [4.01-4.11] for Cancer. Score values were monotonically related to mortality. The weighted index had a higher discrimination (c-statistic=0.825) than age and gender (c-statistic=0.761) or count of all-56 available morbidity variables (c-statistic=0.790). Within each score value, predicted probability was included in the 95% confidence interval of observed mortality proportion.

CONCLUSIONS: We computed a performant score to study multimorbidity in the French context. Such predictive measures could be externally validated in large medico-administrative databases with routinely collected morbidity information.