Heterogeneity of Treatment Effect of ABC Pathway–Based Care in Older Atrial Fibrillation: Insights from the mAFA-II and MIRACLE-AF Trials Using a Causal Machine Learning Approach

Integrated care based on the Atrial fibrillation Better Care (ABC) pathway improves outcomes in patients with atrial fibrillation (AF), but whether treatment effects differ across older patients remains unclear. To evaluate heterogeneity of treatment effect (HTE) of ABC pathway-based care and assess its transportability across healthcare settings using causal machine learning (ML). Individual participant data from the cluster-randomised mAFA-II and MIRACLE-AF trials were analysed, including 3,209 patients aged ≥65 years. Three causal ML models were developed in mAFA-II and externally validated in MIRACLE-AF. The primary outcome was a composite of all-cause mortality, ischaemic stroke, intracranial haemorrhage, heart failure, acute coronary syndrome, or major bleeding. The uplift random forest demonstrated the best performance and identified meaningful HTE. In the mAFA-II testing cohort, ABC pathway-based care was associated with a lower risk of the primary outcome in patients predicted to benefit, whereas no benefit and possible harm were observed in those predicted to derive harm (P-interaction=0.002). In external validation, benefit was preserved in the predicted benefit group but not in the predicted harm group, although treatment discrimination was attenuated. Patients predicted to derive limited benefit had greater clinical complexity, including higher comorbidity burden and CHA2DS2-VASc scores. Sex, heart failure, and hypertension were the strongest predictors of treatment effect. ABC pathway-based integrated care demonstrates clinically meaningful treatment heterogeneity among older patients with AF. These findings support more personalised implementation of integrated care and highlight the potential of causal ML to guide precision management in AF. MAFA-II trial (ChiCTR-OOC-17014138 || http://www.chictr.org.cn/, registered on 26 December 2017), MIRACLE-AF trial (NCT04622514 || https://clinicaltrials.gov/, registered on 10 November 2020).