Wearables: “Wearable Technology and Artificial Intelligence for Early Detection and Risk Assessment of Unacceptable Toxicities in Paediatric Oncology”

Introduction: Each year 400,000 children develop cancer globally with the International Childhood Cancer Survivor study revealing that 27.5% of survivors have severe or life-threatening health issues as a direct result of cancer treatment. These toxicities range from clearly defined conditions such as heart failure to more subjective disorders like cognitive dysfunction. Leveraging existing wearable technology to monitor cancer therapy-related toxicities could scale routine screening strategies across large populations, facilitating ambulatory care. WEARABLES will use wearable technology and artificial intelligence to develop prediction models to allow earlier detection of toxicities and reduce the suffering of children with cancer. We will start with a toxicity prediction model for sepsis/infection before expanding to address the 21 unacceptable toxicities defined by Andres-Jensen et al 2021.

Hypothesis: The development of a sepsis/infection toxicity prediction model for a paediatric oncology population can assist clinicians in early identification and intervention, leading to better outcomes.

Study Design: Two-arm crossover design, prospective, silent trial collecting physiological data from a wearable device to develop a sepsis/infection prediction model.

Intervention: The intervention arm involves wearing a wearable device for 3 months to collect physiological data.

Objectives: The primary objective of this study is to establish an AI predictive model for sepsis/infection observed during childhood cancer therapy. The secondary objectives are to determine the accuracy of the prediction model and determine the practitioner and consumer acceptability of using a wearable device to detect sepsis/infection.

Inclusion criteria: Eligible patients must meet all the following criteria to be enrolled in this trial:

• Paediatric, adolescent or young adult diagnosis of cancer AND receiving therapy placing them at risk of febrile neutropenia

• Patient at The Royal Children's Hospital Melbourne

• Aged 5-18 years at the time of eligibility screening

• Parent or guardian able to provide consent if patient aged < 16 years

• iPhone 8 or later

• iOS 17 or later (iOS must be up to date/updated at time of enrolment)

• Adequate iPhone storage for WEARABLES app

• Willing to wear a wearable device for a period of 3 months (during waking hours)

• Consent to HealthKit data being shared to the WEARABLES app (owned by the research team)

Methods: Participants will be randomised to arm A (No intervention) or arm B (wearable device) for 3 months, before swapping to the alternative arm for another 3 months. Arm A will be sent weekly surveys to check for symptoms and/or hospital admissions for sepsis/infection. No further involvement will be asked of participants when allocated to arm A. Arm B will receive a wearable device that will collect a range of health metrics (i.e. heart rate, temperature, respiratory rate, blood oxygen) at regular intervals for the duration of the study. Arm B will also receive weekly surveys to check for symptoms and/or hospital admissions for sepsis/infection. At the conclusion of the study, a final survey will be sent to patients to assess the feasibility of using a wearable device to detect sepsis/infection.

All data collected in arm B will be utilised to develop the deep learning model for sepsis/infection. Participants will be considered controls 24 hours after hospital discharge for sepsis/infection and cases 24 hours prior to fever onset. Following the prediction model development, the model will be prospectively validated in a second trial. During this period the clinician will continue to assess and manage patients as per standard of care. Concurrently, the research team will record model predictions based on silent alerts with the aim of identifying false positive, false negative and positive predictive values.

Sample size: 150 episodes of fever and 150 episodes of no fever.

Expected Outcomes:

• A deep-learning model to identify sepsis/infection in paediatric cancer patients with or without the presence of a fever.

• Feasibility and acceptability of using a wearable device to collect health data in paediatric oncology patients.

• Validation of the prediction model through a prospective intervention trial, with the aim to subsequently register the digital health solution as a Medical Device for standard care practice in Australia.

Collier:RMIT/Apple: Other: Scholarship for the Apple Foundation Program at RMIT University; Apple: Other: Apple Watches as part of Apple Investigator Support Program. Elliott:Novo Nordisk: Research Funding; Medical Research Future Fund: Research Funding. Wickramasinghe:Optus: Research Funding. Conyers:Apple: Other: Apple Watches as part of Apple Investigator Support Program; Medical Research Future Fund: Research Funding.