Predicting Individual Treatment Effects to Determine Duration of Dual Antiplatelet Therapy After Stent Implantation

Abstract

Background: After coronary stent implantation, prolonged dual antiplatelet therapy (DAPT) increases bleeding risk, requiring personalization of DAPT duration. The aim of this study was to develop and validate a machine learning model to predict optimal DAPT duration after contemporary drug-eluting stent implantation in patients with coronary artery disease.

Methods and results: The One-Month DAPT, RESET (Real Safety and Efficacy of 3-Month Dual Antiplatelet Therapy Following Endeavor Zotarolimus-Eluting Stent Implantation), and IVUS-XPL (Impact of Intravascular Ultrasound Guidance on Outcomes of Xience Prime Stents in Long Lesion) trials provided a derivation cohort (n=6568). Using the X-learner approach, an individualized DAPT score was developed to determine the therapeutic benefit of abbreviated (1-6 months) versus standard (12-month) DAPT using various predictors. The primary outcome was major bleeding; the secondary outcomes included 1-year major adverse cardiac and cerebrovascular events and 1-year net adverse clinical events. The risk reduction with abbreviated DAPT (3 months) in the individualized DAPT-determined higher predicted benefit group was validated in the TICO (Ticagrelor Monotherapy After 3 Months in the Patients Treated With New Generation Sirolimus-Eluting Stent for Acute Coronary Syndrome) trial (n=3056), which enrolled patients with acute coronary syndrome treated with ticagrelor. The validation cohort comprised 1527 abbreviated and 1529 standard DAPT cases. Major bleeding occurred in 25 (1.7%) and 45 (3.0%) patients in the abbreviated and standard DAPT groups, respectively. The individualized DAPT score identified 2582 (84.5%) participants who would benefit from abbreviated DAPT, which was significantly associated with a lower major bleeding risk (absolute risk difference [ARD], 1.26 [95% CI, 0.15-2.36]) and net adverse clinical events (ARD, 1.59 [95% CI, 0.07-3.10]) but not major adverse cardiac and cerebrovascular events (ARD, 0.63 [95% CI, -0.34 to 1.61]), compared with standard DAPT in the higher predicted benefit group. Abbreviated DAPT had no significant difference in clinical outcomes of major bleeding (ARD, 1.49 [95% CI, -1.74 to 4.72]), net adverse clinical events (ARD, 2.57 [95% CI, -1.85 to 6.99]), or major adverse cardiac and cerebrovascular events (ARD, 1.54 [95% CI, -1.26 to 4.34]), compared with standard DAPT in the individualized DAPT-determined lower predicted benefit group.

Conclusions: Machine learning using the X-learner approach identifies patients with acute coronary syndrome who may benefit from abbreviated DAPT after drug-eluting stent implantation, laying the groundwork for personalized antiplatelet therapy.

Keywords: drug‐eluting stents; dual antiplatelet therapy; machine learning; percutaneous coronary intervention; treatment effect heterogeneity.

Figure 1. An X‐learner approach to detecting treatment effects heterogeneous to dual antiplatelet therapy (DAPT) duration strategies in the EVEREST (Severance Dual Anti‐Platelet Therapy After Drug‐Eluting Stent Implantation Cohort Registry) cohort.

A, A derivation cohort that included the One‐Month DAPT (n=3020), RESET (Real Safety and Efficacy of 3‐Month Dual Antiplatelet Therapy Following Endeavor Zotarolimus‐Eluting Stent Implantation; n=2148), and IVUS‐XPL (Impact of Intravascular Ultrasound Guidance on Outcomes of Xience Prime Stents in Long Lesion; n=1400) trials. Validation cohort: TICO (Ticagrelor Monotherapy After 3 Months in the Patients Treated With New Generation Sirolimus‐Eluting Stent for Acute Coronary Syndrome; n=3056). B, In step 1, we subsampled the population based on the treatment assigned to each patient (ie, standard or abbreviated DAPT) and then used patient variables to estimate the occurrence of each outcome event (ie, major bleeding within 1 year). In step 2, the model developed in step 1 was employed to impute the outcomes for instances where patients received treatments that they could not have received originally. The absolute risk reduction (ARR) was predicted by comparing these imputed outcomes with the observed outcomes. Following the development of the individualized DAPT (iDAPT) score in the derivation cohort during steps 1 and 2, individual treatment effects are predicted in step 3 as a weighted average of the imputed treatment effects estimated in the validation cohort. C, Survival analysis was performed on 3 outcomes between patients assigned to different DAPT maintenance strategies in patients classified as likely to benefit according to the iDAPT score calculated in the validation cohort. D, Comparison of 3 outcomes between the abbreviated and standard DAPT groups in those who benefited from abbreviated DAPT (higher predicted benefit [HPB] group: iDAPT score ≥0) as determined by the calculated iDAPT score in the validation cohort. ARD indicates absolute risk difference.

Figure 2. Application of the individualized dual antiplatelet therapy (iDAPT) score in the validation cohort.

Risk curves represent 12‐month Thrombolysis in Myocardial Infarction (TIMI) and major adverse cardiac and cerebrovascular events (MACCE) according to the iDAPT score in the validation cohort. The shaded areas around the curves indicate the 95% CIs for the fitted logistic regression curves. The histogram shows the iDAPT score distribution, with the light greenish bars denoting the higher predicted benefit (HPB) group and the light brown bars denoting the lower predicted benefit (LPB) group.

Figure 3. Kaplan–Meier estimates 3 outcomes in the validation cohort.

Absolute risk differences (ARDs) are displayed, where a positive ARD indicates a reduced risk for the red group (abbreviated dual antiplatelet therapy [DAPT]) compared with the blue group (standard DAPT). A through C, One‐year outcomes of major bleeding, net adverse clinical events (NACE), and major adverse cardiac and cerebrovascular events (MACCE) in patients who benefited from an abbreviated DAPT maintenance strategy determined by the individualized DAPT score.

Figure 4. Absolute risk difference (ARD) between abbreviated versus standard dual antiplatelet therapy (DAPT) duration strategies for major bleeding and net adverse clinical events (NACE) within the recommended treatment period for the individualized DAPT (iDAPT) and PRECISE‐DAPT (Predicting Bleeding Complications in Patients Undergoing Stent Implantation and Subsequent Dual Antiplatelet Therapy) scores.

A positive ARD indicates that the abbreviated DAPT strategy reduced the number of outcomes compared with the standard DAPT strategy. The higher predicted benefit (HPB) group was defined as having an iDAPT score of ≥0. The high‐risk group was defined as those with a PRECISE‐DAPT score ≥25. The high+moderate risk group was defined as those with a PRECISE‐DAPT score ≥18. Bars show 95% CIs.