Baseline and longitudinal measures of continuity of primary care and COPD-related service utilisation: a cohort study...

To compare baseline, cluster-based and trajectory-based classifications of continuity of primary care (COC) and to examine their associations with chronic obstructive pulmonary disease (COPD)-related hospitalisation and emergency department (ED) visits.

Population-based retrospective cohort study.

National Health Insurance claims data from South Korea (Korean National Health Insurance Service), 2014-2021.

92 977 patients aged ≥40 years newly diagnosed with COPD between 2015 and 2016, with at least four ambulatory primary care visits during the exposure period.

The primary outcome was the incidence of COPD-related hospitalisation; the secondary outcome was the incidence of COPD-related ED visits. Continuity of primary care was quantified using the Continuity of Care Index (COCI) and classified using three approaches: (1) a baseline COCI-based method, (2) a cluster-based method using K-means clustering and (3) a trajectory-based method using group-based trajectory modelling. Fine-Gray subdistribution hazard models, accounting for death as a competing risk, were used to estimate adjusted HRs (aHRs).

Across all three classification approaches, low continuity of care was consistently associated with an increased risk of COPD-related hospitalisation (aHR 2.43, 95% CI 2.22 to 2.66 for baseline COCI; 2.26, 95% CI 2.06 to 2.47 for cluster-based; 2.73, 95% CI 2.49 to 2.99 for trajectory-based) and ED visits (aHR 2.32, 95% CI 2.04 to 2.63 for baseline COCI; 2.55, 95% CI 2.25 to 2.88 for cluster-based; 3.48, 95% CI 3.05 to 3.96 for trajectory-based). Effect estimates were consistently strongest for the trajectory-based approach and weakest for the baseline COCI approach. Agreement among the three classification methods was substantial (overall 87.2%; Fleiss' kappa 0.78).

Baseline COCI classification offers a pragmatic and robust measure of COC, producing effect estimates directionally consistent with those derived from more complex longitudinal approaches; however, observed transitions between continuity groups over time indicate that reliance on baseline classification alone may attenuate effect estimates when continuity patterns change. Trajectory-based methods may yield additional insight in populations with greater temporal variability in care patterns but entail substantially higher data and modelling requirements.