Become an AAHPM Member to view PC-FACS
Design and Participants: This secondary analysis of a stepped-wedge cluster randomized trial evaluated the impact of a 40-week behavioral intervention designed to increase clinician-initiated SICs (vs standard care) on EOL spending in a cohort of decedents. At nine oncology practices at a large academic institution, a machine-learning algorithm identified patients as 180-day mortality high-risk. Inflation-adjusted costs for acute care, office/outpatient care, intravenous systemic therapy, other therapy, long-term care, and hospice were abstracted from institutions’ accounting systems. A two-part model evaluated impacts on spending, first using logistic regression to model zero vs nonzero spending and second using generalized linear mixed models with gamma distribution and log-link function to model last 180-days-of-life mean daily spending (MDS). Models were adjusted for clinic and wedge fixed effects, and clustered at the oncologist level.
Results: Patients (N=1,187) were aged median 68 years (IQR=16) and were 27% Black or non-White ethnicity; 38% had SICs. Intervention was associated with lower unadjusted MDS vs control ($377.96 vs $449.92; adjusted mean difference, −$75.33; 95% CI=−$136.42 to −$14.23), translating to $13,747 total adjusted savings per decedent and $13 million cumulatively across all intervention decedents. Compared to the control, intervention incurred lower MDS for systemic therapy (adjusted difference, −$44.59; P=.001), office/outpatient care (−$9.62; P=.001), and other therapy (−$8.65; P=.04), but not for acute care, long-term care, or hospice. For patients with SICs, MDS decreased $37.92 following the first SIC.
Commentary: In this study, an AI algorithm identified oncology patients who were at high risk for 180-day mortality and alerted their oncologists through weekly emailed lists of upcoming encounters with high-risk patients, texts on the morning of the encounters, and peer-comparisons of SIC rates. While findings suggest an association between the intervention (SICs) and lower MDS, the difference in SICs between intervention and control groups was small (32.6% vs 39%). Only a minority of patients had SICs with their oncologists, even with prognostic prompting. Together, the patients who had documented SICs in this study actually had less difference in their MDS than did the larger cohort of patients whose oncologists were receiving the algorithm-derived prognostic information.
Bottom Line: Machine learning–derived prognostic information provided to oncologists is associated with decreased healthcare spending at EOL in patients with a short prognosis; this decrease in spending is less pronounced for patients who had SIC with their oncologists.
Reviewer: Brittany R. Gatta, MD, Vanderbilt University Medical Center, Nashville, TN
References:
1. Paladino J, Bernacki R, Neville BA, et al. Evaluating an intervention to improve communication between oncology clinicians and patients with life-limiting cancer: a cluster randomized clinical trial of the serious illness care program. JAMA Oncol. 2019;5:801-809. doi:10. 1001/jamaoncol.2019.0292.
2. Wright AA, Zhang B, Ray A, et al. Associations between end-of-life discussions, patient mental health, medical care near death, and caregiver bereavement adjustment. JAMA. 2008;300:1665-1673. doi:10.1001/jama.300.14.1665.
3. Bernacki RE, Block SD. Communication about serious illness care goals: a review and synthesis of best practices. JAMA Intern Med. 2014;174:1994-2003. doi:10.1001/jamainternmed.2014.5271.
4. Brinkman-Stoppelenburg A, Rietjens JAC, van der Heide A. The effects of advance care planning on end-of-life care: a systematic review. Palliat Med. 2014;28:1000-1025. doi:10.1177/0269216314526272.
Source: Patel TA, Heintz J, Chen J, et al. Spending analysis of machine learning–based communication nudges in oncology. NEJM AI. 2024;1(6):10.1056/aioa2300228. doi:10.1056/aioa2300228.
Access this article on PubMed.