Currently submitted to: JMIR Formative Research
Date Submitted: Nov 18, 2026
Open Peer Review Period: Nov 19, 2026 - Jan 14, 2027
(currently open for review)
Warning: This is an author submission that is not peer-reviewed or edited. Preprints - unless they show as "accepted" - should not be relied on to guide clinical practice or health-related behavior and should not be reported in news media as established information.
A Mobile App for Coronary Artery Disease Risk Assessment Using Federated Learning and a Wearable Band in Adults in Lima, Peru: Development and Preliminary Evaluation Study
ABSTRACT
Background:
Cardiovascular diseases are the leading cause of death worldwide. In Peru, the burden is aggravated by a high prevalence of risk factors among adults and by limited access to monitoring technologies. Centralized artificial intelligence approaches to cardiovascular diagnosis require transferring clinical data to external servers, which raises privacy concerns.
Objective:
This study aimed to develop a mobile app that combines federated learning and a commercial wearable band to assess coronary artery disease (CAD) risk in adults in Lima, Peru, and to perform a preliminary evaluation of its classification performance, user satisfaction, and usability.
Methods:
The app was developed in 5 phases: technology selection, dataset definition, model training, metric selection, and app development. A neural network was trained with Federated Averaging (FedAvg) in a simulation with 5 virtual clients, using 32 noninvasive predictors derived from the Z-Alizadeh Sani dataset (303 records). Inference runs on the phone with TensorFlow Lite, combining questionnaire data with heart rate read from a Xiaomi Smart Band 10 through Health Connect. The model was evaluated on an internal test set (n=61) and on external data from 100 Peruvian adults older than 35 years, whose reference labels were assigned by a cardiologist. Wilson and bootstrap 95% CIs were estimated. Participants completed a 10-item satisfaction questionnaire (5-point Likert scale), and a cardiologist completed the System Usability Scale (SUS).
Results:
In the internal test set, the model achieved an area under the receiver operating characteristic curve (AUC-ROC) of 0.818 (95% CI 0.646-0.956), recall of 0.954 (95% CI 0.845-0.987), and specificity of 0.667 (95% CI 0.438-0.837). In the external validation (30 CAD and 70 Normal reference labels), the model identified all 30 CAD cases (recall 1.000, 95% CI 0.887-1.000) but classified 37 of 70 Normal participants as CAD (specificity 0.471, 95% CI 0.359-0.587; precision 0.448, 95% CI 0.335-0.566); the AUC-ROC was 0.878 (95% CI 0.804-0.940) and the F1-score was 0.619 (95% CI 0.495-0.722). Mean satisfaction was 3.48 of 5, and the SUS score given by the cardiologist was 80 of 100.
Conclusions:
The prototype showed high sensitivity with respect to the reference labels, accompanied by a large number of false alerts at the 0.5 threshold. These results provide preliminary evidence on the feasibility and acceptability of the proposed mobile health solution but do not establish its accuracy for confirmed CAD or its clinical utility. Multicenter evaluation with a confirmatory reference standard, calibration analysis, and federated training with real clients are required.
Citation
Request queued. Please wait while the file is being generated. It may take some time.
Copyright
© The authors. All rights reserved. This is a privileged document currently under peer-review/community review (or an accepted/rejected manuscript). Authors have provided JMIR Publications with an exclusive license to publish this preprint on it's website for review and ahead-of-print citation purposes only. While the final peer-reviewed paper may be licensed under a cc-by license on publication, at this stage authors and publisher expressively prohibit redistribution of this draft paper other than for review purposes.