Problem
Multimorbidity (MM), the presence of ≥2 long-term health conditions (LTCs), is associated with higher risk of all-cause mortality but there is no consensus regarding how best to measure MM.
Approach
The objective was to compare the performance of three different MM measures in predicting all-cause mortality and hospitalization. We used a general population prospective cohort, UK Biobank, N=502,616 participants aged 37-73 years. MM at baseline was classified using three different measures, using self-report and hospital episode statistics (HES): a.) number of LTCs (n=43) b.) Charlson index (CI) c.) Elixhauser unweighted index. All three measures were categorized based on scores: 0,1,2,3, ≥4. Two clinical outcomes were ascertained using electronic data linkage: all-cause mortality, all-cause hospitalization (yes/no and number of hospitalization events). Cox proportional hazards and zero inflated poisson models were used respectively to compare the association of the three MM measures with all-cause mortality and number of hospitalization events. Area Under Curve (AUC) were used to compare performance of the three MM measures in predicting all-cause mortality and first hospitalization over follow-up. All models were adjusted for sex, socio-economic status, smoking, alcohol consumption, physical activity and BMI.
Findings
N=490,179 participants were included for the analysis, after excluding participants with missing values and those lost to follow-up. At the end of follow-up (median=7 years), n=13,623 (2.7%) participants had died; n=271,259 (55.3%) participants had experienced at least one hospitalization and the total number of hospitalization events were N= 12,00,747. The simple MM count classified highest number of participants as “multimorbid” (38.7%), closely followed by Elixhauser index (33.1%), while CI classified least as multimorbid (10.4%). There was a consistent dose-response relationship between increasing levels of multimorbidity, for all 3 MM measures, and risk of all-cause mortality. The largest effect sizes were observed with CI (Hazard Ratio-HR 7.91 for a score of ≥4), followed by Elixhauser index (HR 5.37 for a score of ≥4), while modest but significant effect sizes were observed with the simple MM count (HR 3.06 for a score of ≥4). The observed AUC for predicting all-cause mortality were similar for the 3 MM measures (CI 65.8%, Elixhauser index 66.2%, simple MM count 66%). A consistent dose-response relationship was also observed between the number of hospitalization events and increasing scores for the 3 MM measures, with largest effect sizes observed for CI and the smallest for simple MM count. However, simple MM count had the best AUC for predicting the first hospitalization event (64.6%), while CI had the lowest AUC (57.2%).
Consequences
Prevalence of MM and ability to discriminate mortality and hospitalization risk varied across the three MM measures; choice of measure may depend on the purpose of use e.g. for health service planning or for targeting interventions.