In the domain of cancer treatment, the detrimental effects of concurrent chemoradiotherapy (CRT) often lead to treatment disruptions and hospitalisations, posing challenges to patient care and escalating healthcare expenses. However, a recent randomised trial employing machine learning (ML) techniques has shown promising results in predicting unplanned medical visits during radiotherapy, thus guiding supportive care effectively.
In a study published today in JAMA Oncology, researchers delve into the potential of wearable devices and artificial intelligence to predict adverse events during CRT. By utilising data from three prospective trials involving patients with various solid tumours, researchers aimed to develop and validate ML models to forecast unplanned hospitalisations based on daily step counts collected from wearable devices.
The study, conducted from June 2015 to August 2018, included 214 patients undergoing curative-intent CRT. Participants wore study-provided wearable devices, and their daily step counts were recorded from approximately one week before CRT initiation through treatment completion. ML models were trained to predict hospitalisations in the week following a given prediction day, using data from the preceding two weeks.
Results demonstrated the efficacy of ML approaches, particularly elastic net–regularised logistic regression (EN), in predicting hospitalisations during CRT. The EN model showed the best performance on validation, with a receiver operating characteristic area under curve (ROC AUC) of 0.83. Notably, incorporating only step count data into the EN model yielded comparable results to models incorporating clinical features, suggesting the potential of simple activity metrics in predicting adverse events.
The study underscores the importance of leveraging patient-generated health data, such as daily step counts from wearable devices, in enhancing predictive care strategies. By accurately identifying patients at high risk for unplanned hospitalisations, ML approaches could enable timely interventions to mitigate adverse events and improve clinical outcomes.
Moving forward, ongoing trials such as NRGF-001 aim to validate and further refine predictive models based on activity data, paving the way for more precise and patient-centric cancer care strategies. As wearable technology continues to evolve, its integration with advanced analytics holds immense promise in revolutionising healthcare delivery and improving patient outcomes.
