ADVERTISEMENT
Improving Oncology Acute Care Event Prediction Through Machine Learning Algorithms
Research presented at the 2023 ASCO Annual Meeting indicates that machine learning algorithms that incorporate patient-reported outcome measures could help predict acute care events among patients with cancer.
“Clinical trials have shown that collecting patient-reported outcome measures (PROMs) can improve outcomes among oncology patients,” said researchers. “However, there remains uncertainty about how to best collect and use PROMs in routine clinical care.”
The researchers designed their study to determine if combining PROMs with electronic health record (EHR) data would improve the effectiveness of machine learning algorithms in predicting patients’ risk of experiencing a 30-day acute care event.
The study included 4193 participants who had solid tumors, completed at least 1 PROMs survey, and had received curative- or palliative-intent systemic therapy between September 1, 2020, and July 31, 2022. The surveys collected data via a 12-item questionnaire based on the Common Terminology Criteria for Adverse Events measurement system and were taken through the EHR’s patient portal or on tablets in waiting rooms.
The EHR data included 176 variables describing demographic data, laboratory results, vital signs, diagnoses, medications, and prior health care encounters. Each patient encounter had a binomial outcome to specify if the patient had an acute care event (ED visit or hospitalization) in the subsequent 30 days. Data were randomly assigned to training (70%) and test (30%) sets, and 4 models were developed: EHR + PROMs, EHR alone, PROMs alone, and ‘PROMs plus’ (PROMs and minimal EHR data such as age and sex).
Patients in the study completed 20,359 PROM surveys during outpatient encounters. Within the cohort, there were 2118 30-day acute care events.
“The area under the receiver operating curves (AUCs) were 0.84 (95% CI 0.81 to 0.86) for the EHR + PROMs model, 0.82 (95% CI 0.80 to 0.84) for the EHR alone model, 0.67 (95% CI 0.65 to 0.69) for the PROMs alone model, and 0.79 (95% CI 0.77 to 0.82) for the PROMs plus model,” researchers said.
The most important variables in the EHR alone and EHR + PROMs models were last albumin, white blood cell count, and hemoglobin. The most important variables in the PROMs plus model were age, treatment site, patient-reported pain, and patient-reported fatigue.
The study findings suggest that incorporating PROMs into machine learning models may help to predict the risk of acute care events and improve model performance. Researchers emphasized that the model that merged PROMs with a limited number of easily reachable EHR variables fairly matched the accuracy of models supported by the entire EHR dataset.
“These models demonstrate the predictive value of PROMs in oncology patients and they lay the foundation for future interventions aiming to reduce acute care events among high-risk patients,” said researchers.
Reference:
Roberts TJ, McGuire J, Temel JS, et all. Developing machine learning algorithms incorporating patient reported outcome measures to predict acute care events among patients with cancer. J Clin Oncol. 2023;41(suppl 16; abstr 1509). doi:10.1200/JCO.2023.41.16_suppl.1516