探討共病測量方法於健保次級資料之應用

Objectives: National Health Insurance databases have become an important resource for studies in health services research. It is important to measure comorbidity; however, there has been no study to investigate the performance of the various available claims-based measures of comorbidity in Taiwan. Methods: Five different measures of comorbidity including Deyo's Charlson Comorbidity Index (CCI), D-M's CCI, D'Hoore's CCI, Elixhauser and Chronic Disease Score (CDS), were chosen for investigation in this retrospective cohort study. We compared the performance of the five measures of comorbidity in predicting in-hospital and one-year mortality among patients with chronic renal disease and pneumonia. The measures of comorbidity were implemented as individual components (the presence or absence of the comorbid condition), and also as an index (weighted sum of comorbidity indicators). The measures of comorbidity were created based on 3 sources of data: the index hospitalization, the index and prior hospitalizations, and the index and prior hospitalizations as well as outpatient visits. The c-statistics of logistic regression were used to compare performance. Results: Better discrimination was achieved with the D-M's CCI or the Elixhauser method when using individual components. When measures of comorbidity were used as indices, better discrimination was achieved with the D-M's CCI. For CCI methods, patient information available from both the index and prior hospitalizations yielded the best results. For the Elixhauser method, patient information available from the index hospitalization yielded the best results when predicting in-hospital death. Adding prior outpatient data did not improve the performance of the measures. Conclusions: D-M's CCI performed best, and future investigators might consider this method. When the sample size is large enough, D-M's CCI or the Elixhauser method could be implemented as the individual components. If the sample size is small, then D-M's CCI used as an index and information from the index and prior hospitalizations would be more appropriate.