AI Tool Predicts Unplanned Care and Symptom Burden in Cancer Survivors

Unplanned emergency visits and hospitalizations remain common in cancer survivorship, when routine clinical contact often tapers while new symptoms emerge. These events reflect unmet needs and disrupt continuity of care. Identifying survivors most likely to deteriorate would enable earlier, targeted support and closer monitoring. To help address this challenge, researchers have developed AI models that combine electronic health records and patient‑reported outcomes to forecast risk of acute care use and worsening symptom burden after treatment.

The machine learning approach was developed at Sylvester Comprehensive Cancer Center, part of the University of Miami Miller School of Medicine (Miami, FL, USA), and published May 26, 2026, in JCO Clinical Cancer Informatics. The models use electronic health records (EHRs) and patient‑reported outcomes (PROs) to estimate near‑term risk for emergency department visits, hospitalizations and elevated symptom burden. The work emphasizes proactive survivorship planning rather than reactive responses to crises, while highlighting survivorship as a dynamic, ongoing process.

The system converts routinely collected clinical data and patient-reported outcomes (PROs) into risk signals that clinicians can interpret. Recent clinical activity provided the strongest signal for predicting acute events such as emergency visits and hospitalizations. For forecasting symptom burden, longer-term trends were more informative, while adding PROs substantially improved performance compared with clinical data alone. The system’s emphasis on interpretability also showed which factors contributed to an individual’s risk score and how those factors changed over time.

Researchers analyzed data from more than 25,000 survivors followed over three years. When the models flagged the highest-risk 10% of patients, that group accounted for roughly half of subsequent unplanned care events and elevated symptom episodes. Incorporating patient-reported outcomes (PROs) nearly doubled model performance compared with using clinical variables alone. The project was conducted in collaboration with the University of Miami’s Frost Institute for Data Science and Computing.

Although not intended to immediately change clinical practice, the findings point to a path for earlier, more personalized survivorship support. The team plans to refine and validate the models across broader survivor populations and explore how EHR‑ and PRO‑driven risk stratification could be integrated into survivorship standards of care. Earlier identification of high‑risk survivors could facilitate targeted symptom management, psychosocial support and closer monitoring.

“This is about shifting from reactive to proactive survivorship care. If we can identify patients who are more likely to struggle, we can begin to align supportive resources earlier and more effectively,” said Frank J. Penedo, Ph.D., director of Sylvester’s Survivorship and Supportive Care Institute and the study’s senior author.

“Our long-term goal is to ensure that survivorship care keeps pace with advances in treatment. That means using data not only to describe outcomes, but to anticipate them, so we can more proactively support patients in the years after cancer,” said Akina Natori, M.D., MSPH, an oncologist in the Division of Medical Oncology at Sylvester and first author of the study.

Related Links
Sylvester Comprehensive Cancer Center