IMplementing Predictive Analytics Towards Efficient COPD Treatments (IMPACT) Study

IMPACT

Patient / Population Intervention / Exposition Vergleich Endpunkt

In 1,111 COPD outpatient visits with 22 respirologists, a treatment recommendation from the ACCEPT exacerbation-risk prediction model improved guideline-concordant pharmacotherapy compared with usual care.

N

1.111Patienten

Design

Stepped-wedge cluster randomised trial of 22 respirologists at two Canadian teaching hospitals; all clusters started in routine practice and crossed over in random order to an on-screen decision support tool that predicts 12-month exacerbation risk with the ACCEPT model and recommends pharmacotherapy from the 2023 Canadian COPD guideline

Endpunkt

Alignment between recommended and prescribed treatment class (guideline-concordant pharmacotherapy)

Relevanz

3Ergänzend — Kontext, kein Muss-Wissen.

ErgebnisGuideline-concordant prescribing in 49.9% of visits during intervention periods vs 38.6% during control periods (adjusted OR 1.70, 95% CI 1.08-2.70). The largest gain came where triple therapy was recommended; no significant effect modification by age, sex or GOLD grade. 22 respirologists at two teaching hospitals, 1,111 outpatient COPD visits between March 2023 and June 2025.

Sadatsafavi M, Burns P, Mokhtaran M, et al. Presented at ERS Congress 2026, ALERT 1, 6 September. Not yet published.

Diskussion & Kritik

A decision support screen that shows the ACCEPT risk estimate and the guideline's drug class at the moment of prescribing moved concordance from roughly two visits in five to one in two. That is a process outcome: whether better-aligned prescribing prevents exacerbations is a registered secondary outcome, due from administrative data a year on, and the setting was 22 respirologists in two academic clinics, not the primary care where most COPD is prescribed. The stepped-wedge design shares calendar time between arms, so the estimate is adjusted for secular trends, but the interval runs close to one. Evidence that embedding a prediction model changes prescribing, not yet that it changes outcomes.