Reframing COPD Instability Under GOLD 2026: A Bayesian Multi-Outcome Retrospective EHR Analysis in Primary Care

Authors:
Open Access
Publisher website
Google Scholar
Abstract
Background/Objectives: COPD instability is heterogeneous; GOLD 2026 lowers the prior-year threshold to ≥1 moderate or severe exacerbation. We assessed whether this low-threshold criterion behaves as a high-sensitivity operational signal in primary-care EHRs. Methods: A retrospective multicenter same-window EHR pilot study in two Spanish primary-care centers (n = 106). Predictors and six binary endpoints were aggregated over the same 12-month window: any exacerbation, high-risk history, severe hospitalization, SABA dispensing, SAMA dispensing, and any rescue dispensing. We fitted Bayesian multi-outcome hierarchical logistic models with patient-level random intercepts, cross-endpoint partial pooling, regularizing priors, and missingness indicators. Robustness used prespecified scenarios, 10-fold ELPD cross-validation, and high-missingness exclusion. Results: Any exacerbation occurred in 53/106 patients; high-risk history in 25/106; hospitalization in 16/106; and any rescue dispensing in 65/106. Diagnostics were stable, and posterior predictive checks supported marginal adequacy. Heart failure showed the clearest positive pattern across exacerbation-defined endpoints; reliever-dispensing endpoints showed a distinct care-pathway-sensitive pattern. No scenario improved out-of-sample adequacy. High-missingness exclusion preserved directionality in 120/120 overlapping pairs; the median |ΔlogOR| was 0.061; and 119/120 remained within ±log(1.25). Conclusions: GOLD 2026 “any exacerbation” behaved as a high-sensitivity operational signal in an endpoint-operating-point sense, not as a homogeneous phenotype. Findings are within-window associations, not causal, medication-effect, or prospective prediction estimates; external validation is required.

This publication has 61 references indexed in Scilit: