PROJECT SUMMARY/ABSTRACT There are approximately 500,000 childhood cancer survivors in the U.S. today. Childhood cancer survivors are vulnerable to late effects of therapy including chronic health conditions and premature death. Predicting survivor-specific risk of late effects, discussing how to manage these risks, and offering early preventions and interventions are critical components of survivorship care. Over 75% of childhood cancer survivors have prevalent symptoms, and constantly poor or worsening symptoms are associated with onset of medical late effects. However, regular symptom monitoring is uncommon in survivorship or primary care. The core concept of this R01 grant proposal is to enable regular monitoring of patient-generated health data (PGHD), including symptoms, physical activity, energy expenditure, sleep behavior and heart rate variability, and utilize these data in predicting survivor-specific risk of late effects to improve survivorship care and outcomes. The proposed application will enroll 620 adult survivors of childhood cancer from the St. Jude Lifetime Cohort Study who are ≥5 years post diagnosis and currently ≥18 years of age at enrollment to achieve the following 3 specific aims: Aim 1) use a mobile health platform to collect dynamic PGHD data over 3 months and use them to develop and validate risk prediction models for future quality-of-life (QOL); Aim 2) develop/validate risk prediction models and establish personalized risk prediction scores for other outcomes (unplanned care utilization, physical performance deficits, onset of chronic health conditions) using the same approach as Aim 1; and Aim 3) create a web-based tool to calculate and report personalized outcome-specific risks, and facilitate integration of risk scores into the survivor’s patient portal and hospital’s Electronic Health Record (EHR). We have a series of preliminary data to support this R01 grant proposal: a) in a pilot study assessing 20 common symptoms with a mobile health platform, childhood cancer survivors completed 90% of all required evaluations over 3 months; and b) in a prediction analysis from ongoing cohort of childhood cancer survivors, the inclusion of longitudinal symptom data generated a superior model performance in predicting future QOL (prediction measure, AUC=0.85) compared to the use of only age, sex, and childhood cancer type (AUC=0.63). Linking through a mobile health platform, we will use a smartphone to collect symptom data, a wrist-worn accelerometer to collect momentary activity/behavioral data, and a finger sensor to collect heart rate variability data. We will predict patient-reported outcomes (poor QOL, unplanned healthcare utilization) and clinically- assessed outcomes (physical performance deficits, onset of chronic health conditions) on the 12th and 24th months after collecting risk factors. We will apply state-of-the-art machine/statistical learning techniques to capture features of dynamic changes in PGHD to predict these outcomes. We will build a Central Cancer Survivorship Platform to integrate predicted risks presented with interpretable scores into a patient portal and EHR, and to inform clinicians and survivors about potential adverse-event risks for risk management/intervention.

PROJECT NARRATIVE We propose to develop survivor-specific risk models to predict adverse outcomes of childhood cancer through collecting daily symptoms and other patient-generated health data (physical activity, sleep behavior, heart rate variability) using mobile health devices. We will apply machine/statistical learning methods to develop and validate risk prediction models, and use qualitative methods to receive input from clinicians and survivors for developing a user-friendly dashboard and decision support tools. Our risk prediction tool will facilitate informing individualized levels of specific adverse outcomes, and providing key information for clinicians and survivors to discuss in view of preventive interventions.