Speech recognition can help evaluate shared decision making and predict medication adherence in primary care setting.

Area of Science:

• Health Informatics
• Artificial Intelligence in Healthcare
• Patient Adherence Research

Background:

• Asthma affects 19 million US adults, with poor adherence to inhaled corticosteroids despite their effectiveness.
• Shared decision-making (SDM) is a strategy to enhance patient activation and improve adherence.

Purpose of the Study:

• To evaluate patient-perceived SDM during primary care encounters using audio recordings.
• To predict inhaled corticosteroid adherence levels based on these encounters.

Main Methods:

• Utilized speech-to-text algorithms to transcribe 80 audio-recorded patient-provider encounters.
• Applied machine learning models (Naive Bayes, SVM, Decision Tree) to assess SDM and predict adherence.
• Assessed SDM with SDM Questionnaire-9 and adherence with MARS-A.

Main Results:

• Automated speech-to-text achieved high accuracy (ROUGE F-score = .9).
• Machine learning models demonstrated strong predictive performance for SDM (F-score = .88) and adherence (F-score = .87).

Conclusions:

• This study is the first to use ML on audio recordings to evaluate SDM and predict adherence.
• ML can identify patients at risk for poor adherence and assess care quality, with potential for broader application.