Background: Persistent uncertainty remains regarding assessments of patient comorbidity based on administrative data for mortality risk adjustment. Some models include comorbid conditions that are associated with improved mortality while other models exclude these so-called paradoxical conditions. The impact of these conditions on patient risk assessments is unknown.
Objective: To examine trends in the prevalence of conditions with a paradoxical (protective) relationship with mortality, and the impact of including these conditions on assessments of risk adjusted mortality.
Methods: Patients age 65 and older admitted for acute myocardial infarction (AMI) or coronary artery bypass graft (CABG) surgery during 1994 through 2005 were identified in Medicare Part A files. Comorbid conditions defined using a common algorithm were categorized as having a paradoxical or non-paradoxical relationship with 30-day mortality, based upon regression coefficients in multivariable logistic regression models.
Results: For AMI, the proportion of patients with one or more paradoxical condition and one or more non-paradoxical condition increased by 24% and 3% respectively between 1994 and 2005. The odds of death for patients with one-or-more paradoxical comorbidities, relative to patients with no paradoxical comorbidity, declined from 0.69 to 0.54 over the study period. In contrast, the risk associated with having one or more non-paradoxical comorbidities increased from 2.66 to 4.62 for AMI. This pattern was even stronger for CABG. Risk adjustment models that included paradoxical comorbidities found larger improvements, in risk-adjusted mortality for AMI and CABG, over time than models that did not include paradoxical comorbidities.
Conclusion: The relationship between individual comorbid conditions and mortality is changing over time, with potential impact on estimates of hospital performance and trends in mortality. Development of a standard approach for handling conditions with a paradoxical relationship to mortality is needed.
Public Domain.