New Coding Approaches Improve Mortality Risk Models

Hands of doctor and patient meeting with computer

In a recent study in JAMA, S. James Adelstein Professor of Health Care Policy (Biostatistics) Sharon-Lise T. Normand, PhD, and colleagues examined the effectiveness of current admission indicators and diagnostic codes in CMS mortality models for acute myocardial infarction, heart failure, and pneumonia.

Currently in the CMS mortality model, when a patient is admitted to the hospital their present on admission (POA) code is entered with their current condition as well as diagnoses and procedures from the last 12 months. The ICD-9-CM codes that classify conditions are combined into groups, which raises the risk of higher-frequency codes overwhelming lower-frequency codes.

The study team hypothesized the adjustments to the POA indicator and diagnostic codes could improve discrimination of hospital-level performance. Using data from Medicare patients aged 65 or older, the study investigated leveraging POA coding and the use of individual codes versus grouped coding. The POA codes distinguished the condition at point of admission from conditions treated over the past year, as well as conditions that developed during the hospital stay. They also separated ICD-9-CM codes that were currently used in the variable groups to ensure equal representation of all codes. The best-performing models were compared to publicly reported CMS hospitalization performance measures.

The study found that these measures were associated with improved patient- and hospital-level morality risk models for patients with acute myocardial infarction, heart failure, and pneumonia.