Machine learning in critical care: the role of diabetes and age in acute coronary syndromes

Background Patients with diabetes and non-ST elevation acute coronary syndrome (NSTE-ACS) have an increased risk of mortality and adverse outcomes following percutaneous coronary intervention (PCI). Purpose We aimed to investigate the impact of early, within 24 hours PCI compared with only routine medical treatment on clinical outcomes in a large international cohort of patients with NSTE-ACS and diabetes. Methods We identified 1,250 patients with diabetes and NSTE-ACS from a registry-based population between October 2010 and April 2016. The primary endpoint was 30-day all-cause mortality. The secondary endpoint was the composite outcome of 30-day all-cause mortality and left ventricular dysfunction (ejection fraction <40%). We undertook analyses to explore the heterogeneity of treatment effects using meta-classification (MC) algorithms followed by propensity score matching and inverse-probability-of-treatment weighting (IPTW) from a landmark of 24 hours from hospitalization. Results Of 1,250 NSTE-ACS first-day survivors with diabetes (median age 67 years; 59%, men), 470 (37.6%) received early PCI and 780 routine medical treatment. The overall 30-day all-cause mortality rates were higher in the routine medical treatment than the early PCI group (6.3% vs. 2.5%). The prediction results of the MC algorithms accounted for only one interaction term that was statistically significant: age ≥65 years. After propensity-matched analysis as well as IPTW, early PCI was associated with reduced 30-day all-cause mortality in the older age (OR: 0.35; 95% CI: 0.14 to 0.92 and 0.43; 95% CI: 0.21 to 0.86, respectively), whereas younger age had no association with the primary endpoint. Similar results were also obtained for the secondary endpoint. Conclusions Among patients with diabetes hospitalized for NSTE-ACS, an early, within 24 hours, PCI strategy is associated with reduced odds of 30-day mortality only for patients aged 65 years or over. MC algorithms provide accurate identification of treatment effect modifiers.