Artificial Intelligence to Search for Abnormalities in Ambulatory Cancer Patients

During treatment, cancer patients may experience side effects related to their disease but also to the different treatments they receive.

Currently, adverse effects and toxicities are well codified in the oncology community, notably via the NCI CTCAE criteria.

Unlike objective data such as a blood sample or a CTscan, a major bias in patient assessment is the subjective assessment of the physician or its team at a given time, which may not reflect the overall situation (for better or worse). Several studies had already highlighted the discrepancies between medical and patient data collection.

Self-assessment of symptoms is one way to overcome this bias. Moreover, there are now a large number of solutions that allow to perform these self-assessments at home.

Thanks to these tools, there are now two situations, the scheduled evaluation (before a chemotherapy treatment, or after a surgical procedure for instance) and the unscheduled situations, where it is the patient himself who can trigger an evaluation form.

These new evaluation methods also allow to take a quality of life approach. Patient-reported outcomes (PROs) is now a valid evidence-based assay to detect patient's symptoms and therefore provide helpful clinical information to healthcare providers.

The goal of this study is to go one step further than the previous PROs studies and evaluate the ability to train a machine learning algorithm to detect at-risk situations and lay the foundation for a viable solution for future prospective and randomized trials.