Algorithms ‘highly accurate’ when using EHRs to predict mortality for chemo patients
Machine learning algorithms that use electronic health record (EHR) data can accurately predict 30-day mortality among patients beginning chemotherapy, according to research in the Journal of the American Medical Association.
Traditionally, to estimate mortality rates for patients starting chemotherapy, physicians reference randomized clinical trial data that specifically highlight mortality risk by age, gender and the type of cancer.
“Although informative, these tools provide mortality estimates for broad populations of patients and often do not accurately estimate a specific individual’s mortality,” wrote first author Aymen Elfiky, MD, MBA, of the Dana-Farber Cancer Institute in Boston, and colleagues. “Individualized decision support tools exist but require a substantial investment of time and resources; these tools require clinicians to collect and enter data not readily available in existing records, which limits the number of variables that can be used and adds complexity to workflows.”
Elfiky and colleagues sought to determine if machine learning algorithms applied to EHRs could predict short-term risk of death at the time patients begin chemotherapy.
The researchers analyzed EHR data from 26,946 patients starting more than 51,000 chemotherapy treatments over a 10-year period. Patients were classified by their primary cancer and the presence of advanced-stage disease using registry data codes for metastases. The date of death was determined by linking health data with Social Security data.
Those at high risk of 30-day mortality were accurately identified across palliative and curative chemotherapy regimens and many types of stages of cancer using only EHR data. The algorithm, the researchers found, was more accurate than predictions based on randomized clinical trials or population-based registry data.
Only 2 percent of the 9,114 patients in the validation set died within 30 days. Elfiky et al. found model predictions were accurate for all patients at an area under the curve (AUC) of 0.94. Model predictions for patients receiving palliative chemotherapy remained highly accurate, with an AUC of 0.92. Model predictions for patients receiving curative chemotherapy had an AUC of 0.98.
The researchers noted the same model performed well for prediction of 180-day mortality.
“This model was able to predict mortality with considerable accuracy despite lacking genetic sequencing data, cancer-specific biomarkers, or any detailed information about cancers beyond EHR data,” Elfiky and colleagues wrote. “This accuracy underscores the fact that common clinical data elements contained within an EHR (e.g., symptoms, comorbidities, prescribed medications and diagnostic tests) contain surprising amounts of signal for predicting key outcomes in patients with cancer.”
As a senior news writer for TriMed, Subrata covers cardiology, clinical innovation and healthcare business. She has a master’s degree in communication management and 12 years of experience in journalism and public relations.