Objectives: to describe an innovative algorithm to classify the general population, in homogeneous groups of severity and complexity of disease and real needs, by using three dimensions: health, frailty, and disability.
Design: retrospective cohort study.
Setting and participants: the study includes the population covered by the Agency for health protection of Metropolitan Area of Milan (3,4 million of habitants). We identified two cohorts of residents: the first at 01.01.2015 and the second at 01.01.2016, classified in four different and mutually exclusive groups based on health and social data of the previous year.
Main outcome measures: we estimated prevalence by age of the four main groups and we studied the transition, observed among groups, from 2015 to 2016. The algorithm validation was performed using non-conditional logistic regression+R14 models to estimate the association with total mortality with increasing levels of severity through the odds ratio (OR) and corresponding 95% confidence intervals (95%CI). The model performance, i.e., its predictive power and calibration, was evaluated by means of C-index and Hosmer-Lemeshow test, respectively.
Results: a total of 19% of subjects is healthy (group A); 41.6% has non-specific access to the health regional system (group B); 17% is a vulnerable (group C); and 22% has a chronic condition (group D). Combining chronic conditions with the frailty level, we classified population into subgroups. The risk of death within a year increases linearly in relation with increasing complexity of the health category and frailty level, with a grow of estimates from 0.83 to 135.6, using the healthy subjects as reference. The evaluation of the overall predictive power of the model, calculated by the C-index, shows a value of 0.94. The calibration of the model evaluated using the Hosmer-Lemeshow test returns a value of 327.2 (χ2 8 df, p-value <0.0001), underestimating the expected in the first three deciles and overestimating in the last deciles.
Conclusions: the algorithm classify general population in homogeneous groups allowing to develop taking care models allocating health resources based on the real needs of patients.