In a groundbreaking advances for personalized cancer treatment, researchers at Moffitt Cancer Center have combined patient-reported results with data from wearable sensors to develop innovative machine learning models that predict emergency treatments undergoing systemic therapy for non-small cell lung cancer (NSCLC). This pioneering study, published recently in JCO Clinical Cancer Informatics, shows that integrating multidimensional health data sources can significantly improve our ability to predict patients at an increased risk of treatment-related complications requiring emergency medical procedures.
Systemic therapies, including chemotherapy, immunotherapy, and targeted agents often result in severe toxicity that requires emergency intervention. Traditionally, predictions of such treatment complications have relied heavily on clinical and demographic data. Current research leverages the power of Bayesian network modeling, an explanatory form of machine learning, to synthesize diverse data streams and generate dynamic risk profiles that adapt as new information becomes available.
The core of this study is the integration of patient-reported results (PRO), capturing subjective symptoms and quality of life indicators directly from affected individuals. These self-report data provide a nuanced perspective on the burden and functional status of symptoms that clinicians may not be fully captured during routine visits. Alongside the professionals, wearable sensors such as Fitbit devices continuously monitor physiological parameters such as heart rate and sleep quality, allowing the capture of subtle physiological changes that precede clinical degradation.
This study involved 58 patients undergoing systemic treatment with NSCLC who were monitored longitudinally using wearable devices and completed a standardized questionnaire that captured symptoms trends and well-being. By adopting the Bayesian network model, researchers were able to elucidate complex interdependencies between clinical variables, self-reported symptoms, and sensor-derived metrics. Unlike the “black box” machine learning approach, Bayesian networks provide interpretability and provide clinicians with transparent inferences about how different factors contribute to increased risk.
Comparative analyses reveal that models that incorporate wearables and patient-reported data significantly outweigh traditional clinical-only models in distinguishing between high and low-risk patients at emergency visits. The addition of continuous physiological monitoring data enhanced early detection of subtle changes predicting treatment complications, maintaining these signals within patient subjective experience. This multimodal modeling approach presents an important step towards more aggressive and personalized oncology care that can reduce adverse events before escalation.
According to Dr. Brian D. Gonzalez, lead author of the study and researcher of outcomes and behavior at Moffitt's Department of Health, the integration of patient-generated health data and machine learning tools has actionable intelligence for clinicians. “Our ambition is to provide real-time, interpretable predictions that encourage timely clinical interventions and ultimately improve patient outcomes and minimize costly hospitalizations,” he explained. This fusion of technology and patient involvement provides a paradigm shift from responsiveness in cancer treatment to preventive health care management.
Co-author Dr. Yi Luo, an expert in the Faculty of Mechanical Sciences at Moffitt, highlighted the important role of clinical trust and explainability in the development of adoption. Unlike opaque prediction algorithms, Bayesian networks reveal how variables such as sleep disorders, symptoms severity, laboratory findings, and vital signs influence risk. This transparency not only encourages shared decision-making, but also helps coordinate interventions based on drivers underlying toxicity risk, thereby enhancing precision medicine.
Although this study was conducted in a single institution with a modest patient cohort, the promising results highlight the potential for broader applications. These models could be extended to incorporate additional data layers such as molecular tumor profiles, or larger multicenter patient populations could further improve predictive accuracy and generalizability. Future directions include validating these approaches in real-world clinical workflows, integrating machine learning insights into electronic health records for streamlined use.
This study exemplifies the growth trends in oncology to leverage “big data” and artificial intelligence to combat the complexities of cancer treatments. By bridging the gap between subjective patient experience and objective physiological monitoring, this study strengthens a more holistic, predictive approach to managing treatment-related risks. As wearable device technologies become more ubiquitous and patient engagement tools evolve, these integrated predictive models could revolutionize not only lung cancer care but also management of diverse chronic diseases.
Supported by the National Institutes of Health, this study reflects interdisciplinary collaborations ranging from oncology, behavioral sciences and computational modeling. Innovative methodologies provide a timely response to subtracting clinical challenges and provide a blueprint for future research aimed at leveraging multidimensional health data to improve cancer treatment delivery. As machine learning continues to mature in healthcare, explanability and patient-centricity will continue to be of paramount importance in translating data insights into meaningful clinical impacts.
In conclusion, the integration of patient-reported results and wearable sensor data into the Bayesian network model represents a major advance in predicting the urgent care needs of NSCLC patients receiving systemic therapy. This approach not only improves predictive performance, but also ensures transparency that promotes clinical trust, potentially transforming ways clinicians predict and respond to treatment-related toxicities. The possibility of intervention early and personalizing care pathways tells us a new era in oncology. There, technology and human experience converge to improve outcomes and the quality of life of patients.
Research subject: people
Article Title: Predict emergency care visits in patients receiving systemic therapy for non-small cell lung cancer using Bayesian networks
News Release Date：September 15, 2025
Web reference:
Articles from JCO Clinical Cancer Informatics
Moffit Cancer Center Lung Cancer
reference:
doi:10.1200/cci-24-00315
Image credits: Moffitt Cancer Center
keyword: Machine Learning
Bayesian Network Modeling for Cancer Therapy Risk Assessment to Improve Quality of Life in Cancer Patients Health Data Integrated Cancer Patient Learning Learning Oncological Patient Report Results for Oncological Cancer Treatment Advances Toxic Wearable Sensors in Health Monitoring
