Evaluating an AI Tool for Detecting Thyrotoxic States

Summary

This observational study aims to evaluate the performance of a software-based medical device, Glandy HYPER, in detecting the thyrotoxic state in patients with hyperthyroidism. The device utilizes heart rate data collected from commercially available wearable devices and compares it with thyroid function test results. The study will enroll patients diagnosed with Graves' disease, monitoring their heart rate during sleep and correlating these measurements with free T4 levels obtained through serial blood testing. No investigational device output will be disclosed to participants, and the study will not alter standard clinical care.

Detailed Description

This is a single-center, prospective observational study designed to validate the performance of Glandy HYPER, a software medical device that analyzes sleep heart rate data from wearable devices in conjunction with thyroid function test (TFT) results to detect thyrotoxicosis. The study targets adults aged 22 or older with newly diagnosed or currently treated Graves' disease. Each participant will wear a smartwatch (Apple or Samsung, depending on their smartphone OS) to measure heart rate during sleep over a 12-week period. Blood samples for TFTs will be collected at four separate visits (baseline and at 4, 8, and 12 weeks). The primary endpoint is the F1 score between the investigational device's output and the diagnosis of thyrotoxicosis based on free T4 values. Secondary endpoints include sensitivity, specificity, and area under the curve (AUC) of the device's performance. Data from the wearable device and TFTs will be used to create multiple evaluation-reference data pairs per patient, enabling within-subject validation across different time points. The study does not involve any investigational treatment or alteration to standard care and is classified as non-significant risk (NSR). The output of the software device will not influence clinical decisions during the trial. The study also aims to assess the generalizability of the software's performance by comparing results from this U.S.-based cohort with prior studies conducted in Korea.

Eligibility Criteria

Interventions