TRANSFORMER dual antiplatelet therapy (DAPT) significantly improved dynamic prediction of ischaemic and bleeding events in patients receiving DAPT following percutaneous coronary intervention (PCI), according to a new study from Florida, USA.
Researchers developed Transformer-DAPT, a transformer based deep-learning survival framework designed to provide patient-specific risk estimates across clinically relevant time intervals during the first year after percutaneous coronary intervention (PCI).
The model was trained using electronic health records from 29,032 patients and externally validated in a separate cohort of 19,173 patients.
Current approaches to guiding DAPT duration rely largely on rule-based risk scores that offer static assessments at single time points and have reported discrimination values ranging from 0.63 to 0.73.
According to the investigators, such approaches may be limited in their ability to support personalised treatment decisions throughout follow-up.
In the development cohort, Transformer-DAPT achieved time dependent-concordance index values ranging from 0.84 to 0.87 for ischaemic event prediction, and 0.81 to 0.88 for bleeding event prediction.
The model outperformed other deep-learning models, including DeepSurv and DeepHit, by 2 to 12% across evaluated intervals.
Performance remained consistent during external validation. Time-dependent concordance index values ranged from 0.74 to 0.84 for ischaemic events and 0.75 to 0.83 for bleeding events.
Although performance was modestly lower than in the development cohort, the model continued to outperform benchmark models while maintaining clinically meaningful discrimination.
Beyond discrimination performance, Transformer-DAPT demonstrated improved calibration.
For 12-month ischaemic predictions, expected calibration error improved from 0.050 to 0.016 following calibration. For bleeding predictions, expected calibration error improved from 0.188 to 0.011.
The investigators also highlighted the model’s ability to generate individualised risk trajectories.
In representative patient examples, predicted shifts from ischaemic-dominant to bleeding-dominant risk occurred before or during follow-up.
Researchers suggest that this could potentially identify time points when DAPT strategies could be reassessed within the broader clinical context.
Overall, the findings suggest that Transformer-DAPT may provide a practical framework for multi-interval risk prediction and support more personalised management of DAPT during the critical first year after PCI.
Reference
Abdelhameed A et al. Transformer-DAPT: AI-based dynamic assessment of ischemic and bleeding risks in patients on DAPT following PCI. npj Digit. Med. 2026;DOI:10.1038/s41746-026-02977-9.
Featured image: Peakstock on Adobe Stock
Each article is made available under the terms of the Creative Commons Attribution-Non Commercial 4.0 License.
