Combining Charlson and Elixhauser scores with varying lookback predicated mortality better than using individual scores.
Pritchard E., Fawcett N., Quan TP., Crook D., Peto TE., Walker AS.
OBJECTIVE:To investigate variation in the presence of secondary diagnosis codes in Charlson and Elixhauser comorbidity scores and assess whether including a one-year lookback period improved prognostic adjustment by these scores individually, and combined, for 30-day mortality. STUDY DESIGN AND SETTING:We analysed inpatient admissions from 01-Jan-2007 to 18-May-2018 in Oxfordshire, UK. Comorbidity scores were calculated using secondary diagnostic codes in the diagnostic-dominant episode, and primary and secondary codes from the year before. Associations between scores and 30-day mortality were investigated using Cox models with natural cubic splines for non-linearity, assessing fit using Akaike Information Criteria. RESULTS:The one-year lookback improved model fit for Charlson and Elixhauser scores vs using diagnostic-dominant methods. Including both, and allowing non-linearity, improved model fit further. The diagnosis-dominant Charlson score and Elixhauser score using a 1-year lookback, and their interaction, provided the best comorbidity adjustment (reduction in AIC: 761 from best single score model). CONCLUSION:The Charlson and Elixhauser score calculated using primary and secondary diagnostic codes from 1-year lookback with secondary diagnostic codes from current episode improved individual predictive ability. Ideally, comorbidities should be adjusted for using both the Charlson (diagnostic-dominant) and Elixhauser (one-year lookback) scores, incorporating non-linearity and interactions for optimal confounding control.