Forecasting the course of disease is a challenging art. However, solely focusing on prognoses for cancer patients doesn’t account for how patients define what constitutes meaningful ends to their lives. One way of eliciting preferences is engaging in serious illness conversations (SICs), a structured discussion that gauges patients’ understanding, goals, and wishes. Recent research demonstrates that this approach helps to improve quality of care and lower healthcare spending for individuals with terminal disease. The question is, what can be done to increase these conversations in practice and can automated computing be part of the solution?

In a new JAMA Oncology study, LDI Senior Fellow Ravi Parikh and colleagues tested whether nudges to clinicians via email and text, prompted by a machine learning algorithm could increase the number of serious illness conversations with cancer patients in a randomized controlled trial. The machine learning predictive algorithm classified cancer patients’ six-month mortality risk, and this information determined when a prompt might be helpful.  

Machine Learning Nudges Increased Serious Illness Conversations

In a cohort of 20,056 patients with cancer spread out across nine University of Pennsylvania clinics, the rate of SICs was 13.5% in the treatment (nudge) group versus 3.4% in the control group (usual care). The number of SICs amongst all patients increased in the intervention, but much more among high-risk patients who were targeted by the nudge.

How Can AI Improve End of Life Care?

While clinicians and patients may wonder how machine learning could be useful in treating a serious condition, Parikh’s study shows that these algorithms can provide useful information to clinicians to help guide care. The program they tested did not remove clinicians from decisions and care but instead was used to promote deeper discussions with patients about goals of care.  Long-term, the hope is that by promoting more serious illness conversations it can improve end-of-life care and reduce interventions that may not be consistent with patients’ end-of-life goals. 

While the authors acknowledge that a computer algorithm should not define what counts as end-of-life care, nor what individual patients deem as a desirable clinical outcome, it can aid physicians in the delivery of high value end-of-life care.

The article, “Long-term Effect of Machine Learning–Triggered Behavioral Nudges on Serious Illness Conversations and End-of-Life Outcomes Among Patients With Cancer: A Randomized Clinical Trial” was published January 12, 2023 in JAMA Oncology. Authors include Christopher R. Manz, Yinchen Zhang, Kan Chen, Qi Long, Dylan S. Small, Chalanda N. Evans, Corey Chivers, Susan H. Regli, C. William Hanson, Justin E. Bekelman, Jennifer Braun, Charles L. Rareshide, Nina O’Connor, Pallavi Kumar, Lynn M. Schuchter, Lawrence N. Shulman, Mitesh S. Patel, and Ravi B. Parikh


Samuel Schotland

Sam Schotland

Policy Analyst

More on Improving Health for Older Adults