Racial disparities in mitral clipping: artificial intelligence-supported cohort analysis of 219 patients
Background:
Racial disparities are prevalent and preventable contributors to cardiac readmission which face mounting Medicare penalties. This is the first machine learning-augmented analysis of racial disparities for guideline-directed therapy and outcomes for mitral clipping.
Methods:
Mitral clip patients from a single high-volume quaternary academic medical center in Houston, Texas, USA, were prospectively enrolled in this cohort analysis from 11/8/11-10/29/18. Machine learning (neural network)-guided multivariable regression assessed outcomes in addition to admission and discharge guideline-directed medications. Forward and backward stepwise regression aided covariate selection.
Results:
The mean age for the 219 subjects meeting criteria was 75.67 (SD 9.90), 116 (48.74%) were female, 181 (75.73%) were white, 128 (53.56%) had commercial insurance and 91 (38.08%) had Medicare, and 100 (42.02%) had HFrEF. After fully adjusting for BMI, prior revascularization, and atrial fibrillation among HFrEF patients, non-white race had significantly lower odds of being discharged on beta blockers (OR 0.20, p=0.009). Adjusting for the same variables in addition to HFrEF among atrial fibrillation patients, non-white race had significantly higher odds of being discharged on DAPT with anti-coagulation (OR 4.65, p=0.021). There were no other disparities in outcomes with equivalent rates of device failure, discharge destination other than home, length of stay, and mortality.
Conclusions:
This AI-augmented large cohort analysis suggests significant racial disparities exist in guideline-directed therapy after mitral clipping but it appears not to impact short-term outcomes. This study potentially provides a robust example of an AI-driven approach to quickly target actionable points for outcome and equity improvement.