Persistent racial disparities in transcatheter aortic valve replacement guideline-directed therapy and outcomes: machine learning-augmented cohort analysis of 1,995 procedures
Background:
Healthcare disparities by race are preventable drivers of increased morbidity, mortality, and cost. There are no known artificial intelligence (AI)-augmented analyses of racial disparities for both African Americans and Hispanics compared to Caucasians for both quality metrics and outcomes within a large cohort study on patients undergoing transcatheter aortic valve replacement (TAVR).
Methods:
This prospective cohort enrolled TAVR subjects at a single high-volume quaternary academic medical center in Houston, Texas, USA, from 11/8/11-10/29/18. Neural network machine learning-augmented multivariable regression was conducted on admission and discharge guideline-directed medications, in-hospital complications, and outcomes; covariate selection was informed by forward and backward stepwise regression.
Results:
Of the 1995 subjects meeting study criteria, the mean age was 78.90 (SD 9.37), 959 (48.07%) were female, 1516 (75.99%) were white, 1151 (57.69%) had commercial insurance and 694 (34.79%) had Medicare, and 546 (27.48%) had HFrEF. After fully adjusting for BMI, atrial fibrillation, prior revascularization, and STS score, non-white race had significantly longer length of stay (beta 1.19, 95%CI 0.08-2.30, p=0.036), while having non-commercial insurance significantly decreased the odds of being prescribed aspirin on discharge (OR 0.50, 95%CI 0.28-0.91, p=0.022; though there were no differences in anti-coagulation prescriptions on discharge for atrial fibrillation patients). There were no other disparities.
Conclusions:
This large AI-guided cohort analysis provides novel evidence that a large TAVR center can reduce racial disparities to mostly equitable care, while suggesting how automated analyses can detect persistent disparities in guideline-directed therapy and preventable delays in discharge for prompt identification and resolution.