The performance of three mortality risk-adjustment comorbidity indices in a community epilepsy cohort

Abstract

Mortality risk-adjustment comorbidity indices are an efficient means of controlling for the important confounding effect of somatic and psychiatric comorbidities in observational mortality studies. We carried out an external validation study and compared the performance of the Charlson, Elixhauser and Epilepsy-specific (ES) indices using the National General Practice Study of Epilepsy, a community-based prospective cohort of 558 people with incident epilepsy followed for 23.3 years (median). The minimum and maximum crude mortality rates were similar between the three indices, but mid-range Elixhauser scores predicted lower rates relative to the two other indices. Two of the stratified Charlson Kaplan-Meier survival probability curves crossed, and a low Elixhauser score was associated with a counterintuitive increase in mortality. Each comorbidity index was a significant predictor of mortality in the Cox proportional hazards models, although there was evidence that the unadjusted Charlson regression model violated the proportionality assumption. Harrell's c-statistics were >0.87 in all adjusted models. All three indices performed well, but there is evidence that the ES index may be more discriminating and have a better model fit than the Charlson or Elixhauser indices in a community-based clinical cohort of people with epilepsy.