Trust and acceptance of a virtual psychiatric interview between embodied conversational agents and outpatients

Mental health disorders necessitate frequent consultations, creating a strain on resources. Virtual medical agents (VMAs), particularly embodied conversational agents (ECAs), offer a promising solution for patient follow-up and physician assistance. ECAs, through verbal and nonverbal interaction, foster empathy and encourage disclosure of sensitive information. They also offer potential benefits such as time savings for physicians, reduced variability in interventions, 24/7 accessibility, and efficient data management. However, patient engagement is crucial for long-term success. This study focuses on two key dimensions influencing engagement: acceptance and trust, drawing on the HCI literature. Acceptance is related to perceived ease of use and perceived usefulness, while trust involves perceived expertise and benevolence of the agent. While previous research has explored these concepts, standardized measurement tools are lacking, particularly in medical contexts. This study investigated the impact of user characteristics (age, gender, education, and health conditions) on engagement, acceptance, and trust, along with the influence of the medical domain (depression or addiction screening).

The introduction section extensively reviews existing literature on VMAs in healthcare, highlighting their potential benefits and the importance of patient engagement, acceptance, and trust. It discusses relevant theoretical frameworks from HCI, emphasizing the concepts of perceived ease of use, perceived usefulness, credibility, and benevolence as key factors influencing technology acceptance and trust. The literature review notes a lack of standardized scales for measuring these aspects, particularly in medical contexts, and underscores the need to consider user characteristics in VMA design and evaluation. Previous work by the authors on using ECAs for clinical interviews and psychiatric diagnoses is also mentioned.

This quantitative study analyzed data from two previously published protocols involving a VMA for diagnosing major depressive disorder (MDD) and another for screening alcohol and tobacco use disorders. A total of 318 outpatients from a sleep clinic participated. The Acceptability E-scale (AES) measured system acceptance (with subscales for usability and satisfaction), and the ECA Trust Questionnaire (ETQ) measured trust in the VMA (with subscales for credibility and benevolence). Engagement was assessed via a single question about willingness to interact with the VMA in the future. Participants’ characteristics (age, gender, education level, type of sleep disorder) were also collected. Univariate and multivariate analyses (linear and logistic regression) were conducted to examine the relationships between patient characteristics, VMA type, and acceptance, trust, and engagement. Receiver operating characteristic (ROC) analyses determined thresholds for acceptance and trust scores associated with engagement.

The VMA was highly accepted and trusted by participants. 68.2% of patients were “very satisfied” with usability, and 78.1% rated satisfaction above three out of five. Regarding trust, 68.2% “totally agreed” with the VMA's benevolence, and 72.9% rated credibility above two out of three. Over half (57.23%) of patients were willing to interact with the VMA again. Age and education significantly influenced acceptance: older and less-educated patients demonstrated higher acceptance. The type of interview (MDD vs. addiction screening) affected usability and engagement, with the addiction interview perceived as easier and depression screening linked to lower engagement. ROC analysis identified cutoff scores for acceptance and trust measures predicting future engagement; credibility showed the best classification performance. Table 1 details sample characteristics. Figures 1 and 2 illustrate acceptance/trust and engagement distributions, respectively. Supplementary Tables provide further statistical details.

The high acceptance and trust levels suggest the potential of VMAs in various settings, highlighting the importance of considering user characteristics during design. The findings contradict the stereotype of older adults' reluctance to adopt technology; older patients showed greater acceptance, possibly due to the VMA serving as a companion in managing health. Less-educated patients also exhibited greater satisfaction, potentially explained by the widespread availability of free healthcare apps. Lower engagement with MDD interviews might be due to the less-structured nature of the interview compared to the addiction screening. The identified acceptance and trust thresholds are valuable for clinical settings, especially in long-term autonomous use. Credibility is crucial for VMA success, particularly in chronic disease management. The study contributes to the growing body of work on the therapeutic alliance in digital mental health.

This study demonstrates high acceptance and trust in VMAs for psychiatric interviews. Age and education levels significantly impact acceptance, while interview type influences engagement. The identified thresholds for acceptance and trust can inform VMA design and clinical implementation. Future research should investigate the impact of VMA gender, interview design, and long-term use in various settings.

The study's sample was drawn from a sleep clinic, potentially limiting generalizability. The cross-sectional design prevents causal inferences. The use of self-reported measures could introduce biases. Further research is needed to compare the validity of the used scales with other evaluation tools and to investigate the impact of repeated VMA usage.