JMIR Publications

ABSTRACT

Background:

The diagnosis of hypothyroidism is often delayed due to its nonspecific symptoms, posing a significant clinical challenge. A non-invasive method to pre-screen individuals and prioritize them for diagnostic testing is needed.

Objective:

This study aimed to develop and validate an artificial intelligence (AI)-assisted system using wearable-derived sleep heart rate (HR) data as a sensitive tool to identify individuals who warrant further thyroid function tests.

Methods:

This retrospective cohort study utilized 1,677 HR-thyroid function test (TFT) data pairs from 413 participants. The AI-assisted system was developed to predict hypothyroidism by learning temporal changes in individual sleep HR patterns. Model performance was robustly evaluated using patient-level leave-one-out cross-validation.

Results:

The system achieved an accuracy of 80.8%, with sensitivity of 82.0%, specificity of 80.8%. In severe overt hypothyroidism with free T4 < 0.5 ng/dL, sensitivity increased to 96.3%. Incorporating multiple reference HR–TFT pairs enhanced performance: using two or three reference datasets improved accuracy to 82.9% (sensitivity 84.0%, specificity 82.8%) and 84.6% (81.8%, 84.8%), respectively, demonstrating that temporal HR patterns strengthened predictive ability.

Conclusions:

This AI-assisted system demonstrates significant potential as a sensitive, non-invasive pre-screening and monitoring tool for hypothyroidism. Its ability to detect a high percentage of hypothyroid cases—particularly the most severe ones—suggests clinical utility in prioritizing at-risk individuals for confirmatory blood tests, thereby enabling earlier diagnosis and intervention.