Development and Internal Validation of a Prediction Model for Major Cardiovascular and Respiratory Events in Chronic Obstructive Pulmonary Disease: Nationwide Primary Care Electronic Medical Records Cohort Study.

Background: Cardiovascular diseases are prevalent in individuals with chronic obstructive pulmonary disease (COPD), but current cardiovascular risk assessment models are not optimised for COPD. We aimed to develop a prediction model for the 10-year risk of major adverse cardiovascular and respiratory events (MACRE) in patients with COPD. Methods: We used nationwide primary care electronic health records from individuals with COPD, aged ≥40 years in 2011 without prior myocardial infarction in the UK Optimum Patient Care Research Database. Practices were randomly divided at the practice level into derivation (80%) and validation (20%) datasets. The primary composite outcome (MACRE) consisted of myocardial infarction, coronary revascularization, heart failure, severe COPD exacerbation, and all-cause mortality. Multivariable Cox regression was used to derive the model using the derivation dataset, with the least absolute shrinkage and selection operator used for variable selection and shrinkage. Performance was evaluated using the validation dataset over prediction horizons of five and 10 years. Results: Among the 122,077 patients included (98,959 in the derivation set; whole cohort: 47.9% women and mean age 69.3 (SD 11.2) years), cardiometabolic risk factors were prevalent, and most had moderate (53.4%) or severe (24.7%) COPD. Over a median follow-up of 10.5 [interquartile range: 4.2-12.4] years, MACRE occurred in 50.2% of the validation set. Sixty-one predictor variables constituted the model, which demonstrated good-to-excellent discrimination and satisfactory calibration across prediction horizons (AUROC 0.78 at five years and 0.82 at 10 years, Brier score of 0.18 at five years and 0.16 at 10 years) in the validation set. Conclusion: A model derived using electronic medical records predicts MACRE in COPD with high discrimination and satisfactory calibration across medium and longer-term prediction horizons. Its utility to inform trial enrolment and clinical decisions requires further study.