MoodMon: novel optimization of bipolar disorder monitoring through patient-driven voice parameter submission and AI technology.

Abstract

Purpose: Behavioral data collected from smartphones can assist artificial intelligence (AI) in assessing and predicting fluctuations in mental states in patients with bipolar disorder (BD). In Poland, the MoodMon online system is used to integrate passive and active data, including voice parameters, for analysis and the issue of alerts based on changes in individual's mental state. The study aims to explore whether active engagement of the patient enhances the efficacy of the advanced MoodMon tool. This clinical trial is embedded in a broader research initiative.

Methods: Methodologically, smartphones were used to automatically collect daily activity data from wristbands and phones of 75 BD patients. Clinical evaluations, using the Hamilton Depression and Young Mania Rating Scales were conducted via a web app, regular visits, calls, or system-initiated contacts after alerts. The MoodMon system, trained on patient data, was compared against clinical evaluations, successfully predicting mental states.

Results: Results showed high alert accuracy: true positive ratio (TPR) at 86.6% (sensitivity) and true negative ratio (TNR) at 98.59% (specificity). Active patient voice data submissions notably improved the prediction of changes or stability in mental states.

Conclusions: Active patient participation in data submission enhances MoodMon's effectiveness as an AI-driven monitoring tool for BD. This underscores the potential of behavioral markers and mobile health applications in mental health care.