Development of the Birmingham Lung Improvement Studies (BLISS) prognostic score for COPD patients in primary care: data from the Birmingham COPD cohort

Abstract

Prognostic scores could be used to guide management of COPD patients and reduce risk of hospital admission but existing scores do not perform well enough and are not practical for primary care. Using data from the Birmingham primary care COPD cohort we developed and internally validated the new BLISS prognostic score from 23 candidate variables. 1558 patients on COPD registers of 71 GP practices and 331 newly-identified patients from a linked case-finding trial were included and their self-reported and clinical data were combined with routine hospital episode statistics. Primary outcome was the record of at least one respiratory admission within 2 years of cohort entry. The model was developed using backward elimination. Missing data were imputed using chained equations. Discrimination and calibration were assessed. Bootstrapping was used for internal validation. Median (min, max) follow up was 2.9 years (1.8, 3.8). 6 variables were retained in the final model: age, CAT score, respiratory admissions previous 12m, BMI, diabetes, FEV1% predicted. After adjustment for optimism, the model performed well in predicting 2yr respiratory admissions (c statistic=0.75 (95%CI 0.72, 0.79). The BLISS score showed better performance in predicting respiratory admissions than existing published scores. All 6 variables are readily available in primary care records or would be easy to collect, and a simple computer programme could calculate the score. Important next steps are external validation, proposing/evaluating a model of use to guide patient management and exploration of the best ways to implement the score in primary care practice.