On June 7, a new article published in JCO Clinical Cancer Informatics (JCO CCI), “Applied Informatics Decision Support Tool for Mortality Predictions in Patients With Cancer,” reports on a prediction model the authors developed to identify patients with high mortality risk prior to the start of treatment regimens.
The authors used patient electronic health records and their newly developed predictive tool to estimate the probability of mortality for an individual patient as his or her next treatment is being considered. The tool—which uses a decision-tree algorithm—considered more than 400 predictive features, including albumin levels, changes in weight, demographics, diagnoses, laboratory results, treatments, and vital signs.
Using machine learning, the authors ‘trained’ the algorithm to predict the 60-, 90-, and 180-day mortality of each patient. The model produced predictions with accuracies of 94.9%, 93.3%, and 86.1%, respectively. The algorithm then became part of the predictive tool, which is available online. A clinician can enter the desired time horizon and cancer type along with answers to adaptive questions, and the tool will provide an estimate of mortality risk.
Although the authors acknowledge needed updates to include cutting-edge treatments and therapies, they believe this tool in its current form could prove useful in providing decision support to all oncologists.
JCO CCI is an online-only, peer-reviewed, interdisciplinary journal that publishes clinically actionable research and important new clinical hypotheses based on biomedical informatics methods and processes applied to cancer-related electronic health record, registry, clinical trial, and omics data and information. JCO CCI publishes original research, reviews, and editorials on biomedical informatics studies relevant to cancer treatment, care delivery, outcomes, and prevention. The publication continuously seeks papers that advance the development, interpretation, and clinical application of cancer informatics research.