Using Artificial Intelligence to Estimate Personal Symptom Network for Cancer Survivorship Care

Childhood cancer survivors frequently experience a high burden of persistent, co-occurring symptoms. These symptoms often interact in complex ways, forming interconnected symptom networks that reflect underlying health vulnerabilities. Approximately half of all childhood cancer survivors report moderate to high symptom burden, which is strongly linked to incident chronic health conditions, reduced quality of life, and premature death. However, current methods of symptom assessment fall short in capturing these intricate interdependencies, especially at the individual level. Most existing methods impose restrictive statistical assumptions, estimate population- rather than individual-level symptom networks, and lack flexibility to accommodate high-dimensional covariate and ordinal symptom data, all of which are essential for personalized survivorship care. This R21 application seeks to advance symptom assessment and its translational impact for long-term survivors of childhood cancer by developing personal symptom networks using an interpretable machine learning (ML) framework. Building on our foundational work, we propose estimating personal symptom networks that incorporate a broad range of risk factors, including cancer therapy, genetic profile, personal demographics/socioeconomic status (SES), lifestyle, and neighborhood-level social determinants of health (SDOH). These networks will identify critical symptoms and interactions that contribute to all-cause and cause- specific mortality outcomes. We will leverage secondary data from an NCI-funded R01 study (R01CA238368; PI: Huang), which collected 141 ordinal symptom items derived from a modified version of the Patient-Reported Outcomes Common Terminology Criteria for Adverse Events (PRO-CTCAE). The dataset comprises 1,600 survivors recruited from the St. Jude Lifetime Cohort Study (SJLIFE; U01CA195547; MPIs: Hudson/Ness) or the Childhood Cancer Survivor Study (CCSS; U24CA055727; PI: Armstrong), providing a robust foundation for large- scale personal network estimation. The project will pursue three Specific Aims. Aim 1. Symptom Network Architecture: To develop an interpretable ML framework to estimate personal symptom networks for ordinal symptom data and evaluate this framework using simulation techniques. Aim 2. Clinical Validation: To apply the ML framework from Aim 1 to real data for estimating personal symptom networks and examining their associations with all-cause and cause-specific mortality. Aim 3. Clinical Application: To develop an interactive online platform to visualize personal symptom networks and facilitate potential personalized, targeted symptom management. By estimating personal symptom networks and developing a visualization tool, this project will facilitate clinical teams to move symptom management from a reactive to a proactive approach. The integration of these personal networks into electronic health records will enable early detection of health deterioration and guide precision interventions. Ultimately, this work will lay the foundation for a future multicenter randomized clinical trial to evaluate the impact of network-informed survivorship care on long-term outcomes. Project Number: 1R21CA313586-01 | Fiscal Year: 2026 | NIH Institute/Center: National Cancer Institute (NCI) | Principal Investigator: Yiwang Zhou (+1 co-PI) | Institution: ST. JUDE CHILDREN'S RESEARCH HOSPITAL, MEMPHIS, TN | Award Amount: $491,925 | Activity Code: R21 | Study Section: Special Emphasis Panel[ZRG1 HSS-S (90)] View on NIH RePORTER: https://reporter.nih.gov/project-details/11356463