Long-term effect of machine learning–triggered behavioral nudges on serious illness communication and end-of-life outcomes among patients with cancer: A randomized clinical trial.

Abstract

109 Background: Early serious illness conversations (SICs) between oncology clinicians and patients are associated with improved mood, quality of life, and quality of end-of-life (EOL) care. Yet, most patients with cancer die without a documented SIC. We report on pre-specified 40-week SIC and EOL outcomes from a stepped-wedge randomized clinical trial (NCT03984773) testing the impact of clinician-directed behavioral nudges to prompt SICs among patients with cancer at high risk of mortality based on a machine learning algorithm. Methods: Our sample consisted of patients with cancer receiving care at one of 9 tertiary or community-based medical oncology clinics between June 2019 and April 2020. We identified high-risk patients using a prospectively validated electronic health record machine learning algorithm to predict 6-month mortality. The intervention consisted of: (1) Weekly emails comparing individual oncologists’ SIC rate relative to peers; (2) Weekly lists of forthcoming encounters with high-risk patients; and (3) Opt-out text messages to prompt SICs before high-risk patient encounters. Clinics were randomized in stepped-wedge fashion to receive the intervention in 4-week intervals through week 16, when all clinics received the intervention. Patients were followed through week 40. The primary outcome was SIC rates for all and high-risk patients. EOL outcomes among decedents were based on ASCO/NQF guidelines and included death in the hospital, intensive care unit admission within 30 days of death, receipt of systemic therapy within 14 days of death, hospice enrollment prior to death, and hospice length of stay. Intention-to-treat analyses were adjusted for clinic and wedge fixed effects and clustered at the oncologist-level. Results: The sample consisted of 20,506 patients and 41,021 encounters. 1,324 (6.5%) patients died by the end of follow-up. Among high-risk patients, the unadjusted SIC rate was 3.4% (59/1754) in the control period and 13.5% (510/3765) in the intervention period and remained >12% throughout follow-up. In adjusted analyses, the intervention was associated with an increase in SICs (adjusted odds ratio 2.09, 95% CI 1.53-2.87, p<0.001) and a decrease in systemic therapy at the end of life, relative to control (6.8% [72/1066]) vs 9.3% [24/258], adjusted odds ratio 0.27, 95% CI 0.12-0.63, p=0.002). There were no differences between control and intervention patients in hospice enrollment or length of stay, inpatient death, or EOL ICU utilization. Conclusions: In this randomized trial, a machine learning-based behavioral intervention led to a sustained increase in serious illness communication and reduction in EOL systemic therapy among outpatients with cancer. Machine learning and behavioral nudges can lead to long-lasting improvements in cancer care delivery. Clinical trial information: NCT03984773.