Machine learning algorithm demonstrates ability to appropriately identify cancer patients for serious illness conversations

1. Patel and colleagues conducted a secondary economic analysis of a trial that assigned high-risk cancer patients to serious illness conversations (SICs).

2. A machine learning algorithm effectively identified patients with high mortality risk. SICs for these patients were associated with reduced healthcare spending.

Evidence Rating Level: 2 (Good)

Study Rundown: Advanced cancer patients often receive treatment incompatible with their preferences at the end of their lives. However, serious illness conversations (SICs) are rarely initiated despite their ability to improve quality of life and mood and decrease healthcare use. In a randomized controlled trial (RCT), a machine learning (ML) algorithm identified cancer patients with a 10% or higher risk of 180-day mortality. Then, these patients were assigned randomly to an SIC intervention group and a standard-of-care control group. Patel and colleagues conducted a secondary analysis of this trial, using mean total and daily healthcare spending during the last six months of life as primary outcomes. Secondary outcomes included the mean spending during the final three and one month of life. They analyzed 1,187 patients and found that mean daily healthcare spending in the final six, three, and one month of life was lower in the intervention than in the control group. The savings were present for systemic and outpatient therapies. This study demonstrated that ML algorithms can predict patient mortality and prompt physicians to initiate SICs, resulting in significant cost savings and better goal-concordant care.

Click here to read the study in NEJM AI

Relevant Reading: Economics of Artificial Intelligence in Healthcare: Diagnosis vs. Treatment

In-Depth [randomized controlled trial]: An ML algorithm was trained using 26,525 adult cancer patients and predicted 180-day mortality using demographic variables, Elixhauser comorbidities, and laboratory and select electrocardiogram data. Then, the algorithm was calibrated to 10% or higher risk and identified eligible patients for the RCT. The intervention included SIC reminders to clinicians before patient encounters, and clinicians’ SIC rates were compared to their peers. Patel and colleagues’ secondary analysis included 1,187 patients from the RCT and gathered spending information per patient from the hospital accounting system. The primary outcomes were mean total and daily healthcare spending during the last six months of life. The secondary outcomes were the mean daily spending during the final three and one month of life. All outcomes were stratified by visit types (e.g., acute care, outpatient). Compared to the control group, the intervention group demonstrated a cost saving of $75.33 (95% CI, -$136.42 to -$14.23) for mean daily spending in the last six months of life. With visit type stratification, the savings were significant for systemic therapy and outpatient care. The total cost savings in the final six months were $13,747 (95% CI, -$24,897 to -$2,598). Additionally, the intervention group had reduced spending for both secondary outcomes ($431.80 vs. $473.20 for the last three months, and $814.46 vs. $947.18 in the final month). While the authors found significant savings, the results were limited by sampling from a singular academic health system and significant baseline differences between the participants of the two groups.

Image: PD

©2024 2 Minute Medicine, Inc. All rights reserved. No works may be reproduced without expressed written consent from 2 Minute Medicine, Inc. Inquire about licensing here. No article should be construed as medical advice and is not intended as such by the authors or by 2 Minute Medicine, Inc.