Abstract
Objective: The objective of this study was to develop and validate a mortality risk index from multimorbidity using pharmaceutical dispensing data.
Design: The P3 (Pharmaceutical Prescribing Profile) mortality risk index was created (development n=2,331,645) using pharmaceutical dispensing records for the last 12 months for long-term conditions. β coefficients from a Cox proportional hazards model for mortality provided component scores for 30 medication categories. Index validity was tested (validation n=1,000,166) for risk of mortality and overnight hospitalization over 1 year, and predictive ability calculated for the P3 index relative to the hospital admission-based Charlson and M3 indices (all models adjusted for age/sex).
Setting: This study was carried out in the setting of routine health data sources for the New Zealand adult general population, for an index date of January 1, 2012.
Results: The P3 index performed equivalently to Charlson for 1-year mortality risk [c-statistics=0.920 and 0.921, respectively; difference=-0.001; 95% confidence interval (CI): -0.004, 0.001]; P3 outperformed Charlson for overnight hospitalization risk (c-statistics=0.712 and 0.682; difference=0.029; 95% CI: 0.028, 0.031). Adding P3 to a model already containing the M3 index led to only marginal improvement for mortality (difference in c-statistics=0.004; 95% CI: 0.002, 0.005) but some improvement for hospitalization risk (difference in c-statistics=0.020; 95% CI: 0.018, 0.021).
Conclusions: The P3 index provides an appropriate alternative to measures like the Charlson and M3 index when analysts only have access to pharmaceutical dispensing data for determining multimorbidity. The P3 index had a performance advantage over Charlson when analyzing risk for overnight hospital admissions.