Serious illness conversations (SICs) enable patients and providers to align on a care plan, but most patients die without having one. Collaborating with the Nudge Unit, researchers at the University of Pennsylvania created a machine-learning algorithm to predict cancer patients with high mortality risk. We then prompted clinicians to have SICs by providing them with lists of high-risk patients, peer comparison feedback, and opt-out reminders.

An extended analysis showed that the intervention increased SICs fourfold for high-risk patients and also increased SICs for patients as a whole. These findings suggest that behavioral interventions that leverage machine learning to target specific patient populations can lead to long-term reductions in low-value cancer care.