The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence) Machine Learning for Equitable Healthcare Outcomes with Irene Chen - #479
Apr 29, 2021
Irene Chen, a Ph.D. student at MIT, is on a mission to ensure fair healthcare outcomes through machine learning. She discusses innovative projects like the early detection of intimate partner violence, aiming to improve patient care. Irene dives into the importance of risk stratification and the ethical challenges of AI in healthcare. She emphasizes the need for collaboration between clinicians and ML researchers to create algorithms that address disparities and enhance predictive accuracy.
AI Snips
Chapters
Transcript
Episode notes
Healthcare Data Challenges
- Healthcare data is noisy and confounded, making machine learning challenging.
- Labels can be inaccurate, data is sparse and incomplete, and the stakes are high.
Model Disparity Mystery
- Irene Chen developed a patient mortality prediction algorithm that performed poorly for certain racial groups.
- This sparked a year-long investigation into the causes of this disparity.
Risk Stratification
- Risk stratification categorizes patients by risk levels, allowing for resource allocation.
- It involves predicting probabilities of adverse events and calibrating these scores for accuracy.