Development and validation of a 5-year mortality prediction model using regularized regression and Medicare data.

Area of Science:

• Health Services Research
• Biostatistics
• Gerontology

Background:

• De-implementing low-value care for patients with limited life expectancy is difficult.
• Accurate identification of patients with limited life expectancy is crucial for healthcare stakeholders.
• Routinely collected healthcare data can be leveraged for mortality prediction.

Purpose of the Study:

• To develop and validate a claims-based prediction model for 5-year mortality.
• To identify populations with limited life expectancy using healthcare data.
• To improve the identification of patients for whom de-implementation of low-value services may be appropriate.

Main Methods:

• A cohort of Medicare beneficiaries aged 66+ was identified in 2008.
• Regularized logistic regression (LASSO) was used to develop a 5-year mortality prediction model from claims data.
• Model performance was validated against the Gagne comorbidity score.

Main Results:

• The final model included demographics, 161 indicators of comorbidity, and function.
• The model demonstrated excellent discrimination (c-statistic 0.825) and calibration.
• The LASSO model improved 5-year mortality classification compared to the Gagne score.

Conclusions:

• A feasible, claims-based model for predicting 5-year mortality was developed using regularized regression.
• The model showed excellent performance and improved mortality classification.
• This tool can enhance healthcare research and quality evaluation for patients with limited life expectancy.