Objective: Multi-morbidity contributes to mortality and hospitalisation in COPD but it is uncertain how this interacts with disease severity in risk prediction. We compared contributions of multi-morbidity and disease severity factors in modelling future health risk using UK primary care healthcare data.

Method: Health records from 103,955 patients with COPD identified from the Clinical Practice Research Datalink were analysed. We compared Area Under The Curve (AUC) statistics for logistic regression (LR) models incorporating disease indices with models incorporating categorised co-morbidities. We also compared these models with performance of The John Hopkins Adjusted Clinical Groups® System (ACG) risk prediction algorithm.

Results: LR models predicting all-cause mortality outperformed models predicting hospitalisation. Mortality was best predicted by disease severity (AUC & 95% CI: 0.816 (0.805 - 0.827)) and prediction was enhanced only marginally by the addition of multi-morbidity indices (AUC & 95% CI: 0.829 (0.818 – 0.839)). The model combining disease severity and multi-morbidity indices was a better predictor of hospitalisation (AUC & 95% CI: 0.679 (0.672 – 0.686)). ACG derived LR models outperformed conventional regression models for hospitalisation (AUC & 95% CI: 0.697 (0.690 – 0.704)) but not for mortality (AUC & 95% CI: 0.816 (0.805 – 0.827)).

Conclusion: Stratification of future health risk in COPD can be undertaken using clinical and demographic data recorded in primary care but the impact of disease severity and multi-morbidity varies depending on the choice of health outcome. A more comprehensive risk modelling algorithm such as ACG offers enhanced prediction for hospitalisation by incorporating a wider range of coded diagnoses.