Researchers have validated a machine-learning algorithm that was integrated into an electronic health record to generate real-time, accurate predictions of the short-term mortality risk for patients with cancer, according to a recent study.
Additionally, this machine-learning algorithm outperformed other prognostic indices.
“Such an automated tool may complement clinician intuition and lead to improved targeting of supportive care interventions for high-risk patients with cancer,” the researchers wrote.
The prospective study included 24,582 patients with outpatient oncology encounters from March 2019 to April 2019. Encounters occurred at 1 tertiary and 17 general oncology practices. The machine learning algorithm predicted 180-day risk between 4 and 8 days prior to the scheduled encounter.
The area under the curve was 0.89 (95% CI, 0.88-0.90) for the entire cohort, and it ranged across disease-specific groups, the study found. For example, AUC ranged from 0.74 for neuro-oncology to 0.96 for breast oncology. There was no difference found between the tertiary center and the general oncology practices.
The researchers used a prespecified 40% mortality risk as a threshold to differentiate high- vs low-risk patients. At this threshold, the observed 180-day mortality was 45.2% for the high-risk patients compared with 3.1% for the low-risk patients.
Finally, the study integrated the machine-learning algorithm with ECOG and Elixhauser comorbidity index-based classifiers, which resulted in favorable reclassification of patients compared with either baseline classifier alone.
“When considering the implementation of a tool such as this, clinicians must consider the need to account for prognostic factors that are unique to specific cancers,” the researchers wrote. “Relatedly, clinicians may use this point-of-care tool differently with different threshold preferences for patients with stage I to III cancer — who are generally treated with curative intent — compared with patients with stage IV cancer who are generally treated with palliative intent.”
Manz CR, Chen J, Liu M, et al. Validation of a machine learning algorithm to predict 180-day mortality for outpatients with cancer. JAMA Oncol. Published online September 24, 2020. doi:10.1001/jamaoncol.2020.4331.