Development and validation of a claims-based index for adjusting for risk of mortality – The case of acute myocardial infarction

We developed a comorbidity index on a cohort of 162,699 Medicare beneficiaries who had an acute myocardial infarction (AMI) in 1987 and validate it on two national cohorts: (1) a cohort of 164,427 Medicare beneficiaries who had an AMI in 1988 and (2) a cohort of 10,466 patients admitted to Veterans Administration Hospitals (VAH) for AMI in 1988-1991. The impact of each sensitivity was expressed as; (1) the risk of mortality for those with the comorbidity, (2) the adjustment to the log odds for 2 year mortality and (3) the age-based likelihood of 2 year mortality. Models were validated by calculated the area under an ROC curve obtained by fitting a logistic regression model to each validation population. The two year mortality rate for 30-day survivors was approximately 30% in each of the 3 cohorts. The 5 most prevlent comorbidities coded in the developmental cohort were heart failure (34%), chronic angina (27%), minor arrythmias (25%) and uncomplicated hypertension (18%). Cancer was the most powerful predictor of 2 year mortality, impacting mortality the same as a 18.3 year age increase. Saturation (having all secondary diagnoses in the discharge summary filled) resulted in a 9.2 year age increase. Validation in the 1988 Medicare and in the Veterans Administration Hospitals cohorts resulted in areas of 73% and 72% under the respective ROC curves. Our methods can serve as a prototype for others wishing to assess comorbidity in other targeted subgroups. (February 1995)