- Combining the power of EHR data with machine learning may be the key to more accurately predicting mortality among cancer patients undergoing chemotherapy.
This finding comes from a recent JAMA study by Elfiky et al. that explored the effectiveness of applying machine learning to EHR data to predict patient’s short term risk of death when they start chemotherapy.
While chemotherapy significantly lowers the risk of recurrence in early-stage cancers and can improve survival rates and symptoms in later-stage disease, the treatment is challenging and costly for patients.
“These patients experience burdensome symptoms without many of the potential benefits of chemotherapy,” wrote researchers in the report.
Researchers set out to find a way to more accurately predict mortality risk before administering chemotherapy treatment to ensure patients that undergo the stress and burden of treatment will also reap its benefits.
In the cohort study, researchers analyzed the EHR data of 26,946 patients starting 51,774 chemotherapy regimens at Dana Farber/Brigham and Women’s Cancer Center from January 1, 2004 to December 31, 2014. Researchers identified the date of death for patients by linking their health records to their Social Security data.
The team classified patients by primary cancer and presence of distant-stage disease using registry data codes for metastases. With this information, researchers attempted to accurately predict death within 30 days of starting chemotherapy with a machine learning model based on single-center EHR data.
Ultimately, the machine learning model was able to accurately predict mortality rates despite lacking genetic sequencing data, cancer-specific biomarkers, or any detailed information beyond EHR data. Specifically, patient EHR data used in the machine learning model including symptoms, comorbidities, prescribed medications, and diagnostic tests.
“Mortality estimates were accurate for chemotherapy regimens with palliative and curative intent, for patients with early- and distant-stage cancer, and for patients treated with clinical trial regimens introduced in years after the model was trained,” stated researchers.
Researchers emphasized that EHR data contains “surprising amounts of signal for predicting key outcomes in patients with cancer.”
In addition to proving accurate, the machine learning model developed by researchers would also only minimally increase administrative burden on clinicians. The machine learning algorithm would not require manual data input from clinicians.
Instead, the algorithm could pull directly from existing patient EHRs.
“Although our algorithm was developed using a single institution’s data, its inputs are available nearly everywhere with an EHR,” wrote researchers.
“In addition, no special infrastructure is required to pull these data from an institution’s data warehouse; in the same way that today’s EHR systems pull a rich set of data from a database to present it to clinicians, an algorithm could pull and process the same data in real time using the processing power on a desktop computer,” the team continued.
The team also suggested the machine learning algorithm could potentially be designed to support EHR integration. Healthcare organizations could integrate the algorithm directly into existing health IT systems.
“Algorithmic predictions such as ours could be useful at several points along the care continuum,” wrote researchers. “They could provide accurate predictions of mortality risk to a clinician or foster shared decision making between the patient and clinician.”
By predicting short-term mortality for cancer patients, clinicians can identify patients who are unlikely to benefit from chemotherapy and instead may be better suited for early palliative care referral and advance care planning.
“For patients receiving systemic chemotherapy, an estimate of 30-day mortality risk may be a useful quality indicator of avoidable treatment-associated harm,” researchers concluded.
Leveraging EHR data to predict patient health outcomes may help providers to avoid clinical decisions that add unnecessary strain on patients for minimal benefit.