A machine learning-based behavioral intervention can increase serious illness conversations (SICs) and reduce the use of systemic therapy at the end of life among patients with cancer, new research suggests.

The intervention did not have an impact on hospice enrollment, length of hospice stay, inpatient death, or ICU use at the end of life. These findings were published in JAMA Oncology

The study (ClinicalTrials.gov Identifier: NCT03984773) included 20,506 cancer patients and 41,021 patient encounters. The machine learning algorithm was used to identify 5520 patients with a high risk of death (10% or higher) within 180 days. 


Continue Reading

The study intervention included weekly emails to clinicians comparing their SIC rates against peers’ rates, a weekly list of upcoming encounters with high-risk patients, and opt-out reminder texts to clinicians on the morning of these visits with high-risk patients. After a 4-week control period in which all patients remained in usual care, there was a 16-week rollout period for the intervention and 24 weeks of follow-up. 

Across all patient visits, the unadjusted rate of SICs was 1.3% in the control period and 4.4% in the intervention period. Among high-risk patients, the unadjusted SIC rate was 3.4% for the control period and 13.5% for the intervention period. 

In an adjusted analysis, the intervention was associated with a significant increase in the rate of SICs for all patients (adjusted odds ratio [aOR], 2.09; 95% CI, 1.53-2.87; P <.001), high-risk patients (aOR, 2.62; 95% CI, 1.84-3.72; P <.001), and patients who were not high risk (aOR, 2.07; 95% CI, 1.52-2.82; P <.001). 

In addition, the use of systemic therapy within 14 days of death was significantly less likely during the intervention period than during the control period for all patients (7.5% vs 10.4%; aOR, 0.25; 95% CI, 0.11-0.57; P =.001). 

When patients were divided by risk, there was a numeric difference in the use of systemic therapy between the intervention and control periods. However, statistical significance was not reached for high-risk patients (7.2% vs 8.3%; aOR, 0.37; 95% CI, 0.13-1.04; P = .06) or for patients who weren’t high risk (6.3% vs 12.9%; aOR, 0.84; 95% CI, 0.44-1.61; P =.61).

Likewise, the intervention did not have a significant impact on hospice enrollment, length of hospice stay, inpatient death, or ICU admission within 30 days of death.

The researchers noted that this study was done in a single health system, so the results may not be generalizable to other systems. Still, the team concluded that “machine learning-based interventions can lead to long-lasting improvements in cancer care delivery.”

Disclosures: Some study authors declared affiliations with biotech, pharmaceutical, and/or device companies. Please see the original reference for a full list of disclosures.

Reference

Manz CR, Zhang Y, Chen K, et al. 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. JAMA Oncol. January 12, 2023. doi:10.1001/jamaoncol.2022.6303