New AI Model Can Predict Unplanned Hospitalization During Cancer Therapy
New research shows that an AI model, which analyzes wearables data, could be used to predict the odds of an unplanned hospitalization among those undergoing cancer therapy.
Source: Getty Images
By - US-based researchers have created a new artificial intelligence (AI) model that uses data from consumer wearables, like daily step counts, to determine health outcomes during cancer therapy and predict the likelihood of unplanned hospitalization.
Unplanned hospitalizations are a significant issue for patients undergoing cancer therapy. Among those receiving outpatient radiation or chemoradiation therapy, 10 to 20 percent end up needing acute care that requires a visit to the emergency department (ED) or an unplanned hospital stay.
Researchers created an AI model to assist providers by identifying and addressing this risk early.
“If you can anticipate a patient’s risk of unplanned hospitalization, you can change how you support them through their cancer treatments and reduce the likelihood that they will end up in the ED or hospital,” said Julian Hong, MD, senior author of the study and an assistant professor of radiation oncology and computational health sciences at the University of California, San Francisco (UCSF), in a press release.
For the study, researchers gathered data for 214 patients from three prospective clinical trials, all of whom wore fitness trackers for several weeks while receiving chemoradiation therapy. Among the trial participants, the most common types of cancer were head/neck and lung, with 30 and 29 percent of participants battling those conditions, respectively.
Using data from the wearable devices, such as daily step counts, researchers created an elastic net-regulated logistic regression model. In addition to daily step counts, the researchers leveraged metrics like the relative changes to a person’s week-by-week averages.
After validating the model, researchers determined that it could predict the odds of a patient requiring hospitalization the following week, and it significantly outperformed the model without step counts.
“The step counts immediately preceding the prediction window ended up being generally more predictive than clinical variables. The dynamic nature of the step counts, the fact that they're changing every day, seems to make them a particularly good indicator of a patient's health status,” said Hong.
Further, researchers intend to perform a more precise validation of the model, as well as an examination of other metrics collected by wearables and how they can be used in a clinical setting.
The use of AI to perform health risk predictions has become increasingly common.
For example, in April, researchers from Johns Hopkins University created an AI tool that uses raw images of patient hearts and demographic data to predict cardiac arrest. Researchers found that the tool had the potential to support clinical decision-making.
Another study from June described how clinical risk prediction models for sepsis, delirium, and acute kidney injury showed the ability to perform well when used in live clinical workflows.
