JMIR Publications

ABSTRACT

Background:

In Clinical Diagnostic Interviews, mental health professionals (MHPs) implement a care practice that involves open questions (e.g., What do you want from your life? What have you tried before to bring change in your life?) and listening to a patient. Further, MHPs need to gather critical insights from an interview with a patient concerning the patient’s condition. However, partially due to the social stigma associated with mental disorders, the hidden signals and noisy content of the discourse hinder the diagnosis and treatment process. Hence, a focused, well-formed, and elaborative summaries of clinical interviews are critical to MHPs for making informed decisions by enabling a more profound exploration of a patient’s behavior, especially when it endangers life.

Objective:

We propose an unsupervised Knowledge-infused Abstractive Summarization (KiAS) approach that generates summaries to enable MHPs to perform a well-informed follow-up with patients. This aim is to improve the existing summarization methods built on frequency heuristics by creating more informative summaries.

Methods:

Our approach incorporates domain knowledge from the PHQ-9 lexicon into an Integer Linear Programming (ILP) framework that optimizes linguistic quality and informativeness. We utilize three baseline approaches: Extractive summarization (ES) using SumBasic algorithm, Abstractive summarization (AS) using ILP without the fusion of knowledge, and Abstraction over ES to evaluate the performance of KiAS. We demonstrate the capability of KiAS on the Distress Analysis Interview Corpus - Wizard of Oz (DAIC-WoZ) dataset through interpretable qualitative and quantitative evaluations.

Results:

KiAS generates summaries (7 sentences on an average) that capture informative questions and responses exchanged during long (58 sentences on an average), ambiguous, and sparse clinical diagnostic interviews. The summaries generated using KiAS improved upon the three baselines by 23.3%, 4.4%, 2.5%, and 2.2% for thematic overlap, Flesch Reading Ease, contextual similarity, and Jensen Shannon Divergence, respectively. Through visual inspection and substantial inter-rater agreement from MHPs, we validated the quality of generated summaries.

Conclusions:

Our collaborator MHPs observed the potential utility and significant impact of KiAS in reducing follow-up time with patients, in a future real-world clinical setting. Our work shows promise in generating semantically relevant summaries that will help MHPs make informed decisions about patient status.