Predicting mortality after hospitalisation for COPD using electronic health records

Area of Science:

• Pulmonary Medicine
• Health Informatics
• Biostatistics

Background:

• Limited prognostic models exist for hospitalized chronic obstructive pulmonary disease (COPD) patients, often based on small, specialized cohorts.
• Electronic Health Records (EHRs) offer a valuable resource for developing more representative prognostic models.
• EHR data may lack some variables used in traditional prognostic models.

Purpose of the Study:

• To develop a prognostic model for 90-day mortality in patients hospitalized with COPD using EHR data.
• To identify key predictive variables from routine EHR data.
• To create a practical risk-assessment tool for clinical use.

Main Methods:

• Retrospective cohort study of 17,973 unplanned COPD hospitalizations (2011-2017) using linked EHR, prescribing, and laboratory data.
• Exclusion of patients with known lung cancer.
• Development of a parsimonious multivariable model to predict 90-day mortality.

Main Results:

• Twelve variables from routine EHRs significantly predicted 90-day mortality.
• Key predictors included age, sex, hospitalization length, prior cancer/dementia, oxygen/digoxin prescriptions, neutrophil/lymphocyte ratio, and serum chloride, urea, creatinine, and albumin.
• The model demonstrated excellent calibration and reasonable discrimination (AUC: 0.806).

Conclusions:

• A limited set of variables from routine EHRs effectively predicts 90-day mortality post-COPD hospitalization.
• The developed risk calculator can aid in service evaluation, clinical management, and patient stratification for research.
• This approach enhances the utility of EHR data for prognostic modeling in COPD.