A study evaluated the predictive value of a machine learning algorithm to forecast 180-day mortality in cancer patients in the outpatient oncology setting.
“Machine learning (ML) algorithms can identify patients with cancer at risk of short-term mortality to inform treatment and advance care planning. However, no ML mortality risk prediction algorithm has been prospectively validated in oncology or compared with routinely used prognostic indices.”
The study included cancer patients with outpatient oncology visits between March 1, 2019, and April 30, 2019. The machine learning algorithm was employed to predict 180-day mortality risk between four and eight days ahead of the patient encounter, which took place at either a tertiary practice (n=1) or general oncology practice (n=17). Patients aged 18 years or older were eligible for inclusion. They were excluded if their appointment was scheduled after the creation of weekly predictions and if their evaluation only took place in benign hematology, palliative care, or rehabilitation practices.
The machine algorithm that predicted mortality was trained using “retrospective data from a subset of practices,” the researchers shared. The main outcome of the study was mortality within 180 days of the visit, with the area under the curve receiver operating characteristic curve (AUC) used as the primary performance metric.
A total of 24,582 patients were seen during the study period; 1,022 (4.2%) died during the 180-day period following their appointment. Among the total cohort, the median (interquartile range) age was 64.6 (53.6–73.2) years, and nearly two-thirds (n=15,319; 62.3%) were female. For the whole study group, the AUC was 0.89 (95% confidence interval [CI], 0.88 to 0.90). The AUC varied by disease at the tertiary practice, ranging from 0.74 to 0.96. The predetermined mortality risk threshold to separate high- versus low-risk patients was 40%. From this benchmark, the observed 180-day mortality in the high-risk group was 45.2% (95% CI, 41.3% to 49.1%) and the low-risk group was 3.1% (95% CI, 2.9% to 3.3%).
The authors observed that “Integrating the [machine learning] algorithm into the Eastern Cooperative Oncology Group and Elixhauser comorbidity index–based classifiers resulted in favorable reclassification (net reclassification index, 0.09 [95% CI, 0.04-0.14] and 0.23 [95% CI, 0.20-0.27], respectively).”
The research was reported in JAMA Oncology.
The authors of an invited commentary said that the study “represents an important early step in health care [machine learning], though there remains a long road ahead for the field.”