Development of a risk-adjusted in-hospital mortality prediction model for community-acquired pneumonia: A retrospective analysis using a Japanese administrative database

Abstract

Background: Community-acquired pneumonia (CAP) is a common cause of patient hospitalization and death, and its burden on the healthcare system is increasing in aging societies. Here, we develop and internally validate risk-adjustment models and scoring systems for predicting mortality in CAP patients to enable more precise measurements of hospital performance. Methods: Using a multicenter administrative claims database, we analyzed 35,297 patients hospitalized for CAP who had been discharged between April 1, 2012 and September 30, 2013 from 303 acute care hospitals in Japan. We developed hierarchical logistic regression models to analyze predictors of in-hospital mortality, and validated the models using the bootstrap method. Discrimination of the models was assessed using c-statistics. Additionally, we developed scoring systems based on predictors identified in the regression models. Results: The 30-day in-hospital mortality rate was 5.8%. Predictors of in-hospital mortality included advanced age, high blood urea nitrogen level or dehydration, orientation disturbance, respiratory failure, low blood pressure, high C-reactive protein levels or high degree of pneumonic infiltration, cancer, and use of mechanical ventilation or vasopressors. Our models showed high levels of discrimination for mortality prediction, with a c-statistic of 0.89 (95% confidence interval: 0.89-0.90) in the bootstrap-corrected model. The scoring system based on 8 selected variables also showed good discrimination, with a c-statistic of 0.87 (95% confidence interval: 0.86-0.88). Conclusions: Our mortality prediction models using administrative data showed good discriminatory power in CAP patients. These risk-adjustment models may support improvements in quality of care through accurate hospital evaluations and inter-hospital comparisons.