Attrition in Conversational Agent-Delivered Mental Health Interventions: Systematic Review and Meta-Analysis

BACKGROUND

Conversational agents (CAs) or chatbots are computer programs that mimic human conversation. They have the potential to improve access to mental health interventions through automated, scalable, and personalized delivery of psychotherapeutic content. However, digital health interventions, including those delivered by CAs, often have high attrition rates. Identifying the factors associated with attrition is critical to improving future clinical trials.

OBJECTIVE

This review aims to estimate the overall and differential rates of attrition in CA-delivered mental health interventions (CA interventions), evaluate the impact of study design and intervention-related aspects on attrition, and describe study design features aimed at reducing or mitigating study attrition.

METHODS

We searched PubMed, Embase (Ovid), PsycINFO (Ovid), Cochrane Central Register of Controlled Trials, and Web of Science, and conducted a gray literature search on Google Scholar in June 2022. We included randomized controlled trials that compared CA interventions against control groups and excluded studies that lasted for 1 session only and used Wizard of Oz interventions. We also assessed the risk of bias in the included studies using the Cochrane Risk of Bias Tool 2.0. Random-effects proportional meta-analysis was applied to calculate the pooled dropout rates in the intervention groups. Random-effects meta-analysis was used to compare the attrition rate in the intervention groups with that in the control groups. We used a narrative review to summarize the findings.

RESULTS

The systematic search retrieved 4566 records from peer-reviewed databases and citation searches, of which 41 (0.90%) randomized controlled trials met the inclusion criteria. The meta-analytic overall attrition rate in the intervention group was 21.84% (95% CI 16.74%-27.36%; I2=94%). Short-term studies that lasted ≤8 weeks showed a lower attrition rate (18.05%, 95% CI 9.91%- 27.76%; I2=94.6%) than long-term studies that lasted >8 weeks (26.59%, 95% CI 20.09%-33.63%; I2=93.89%). Intervention group participants were more likely to attrit than control group participants for short-term (log odds ratio 1.22, 95% CI 0.99-1.50; I2=21.89%) and long-term studies (log odds ratio 1.33, 95% CI 1.08-1.65; I2=49.43%). Intervention-related characteristics associated with higher attrition include stand-alone CA interventions without human support, not having a symptom tracker feature, no visual representation of the CA, and comparing CA interventions with waitlist controls. No participant-level factor reliably predicted attrition.

CONCLUSIONS

Our results indicated that approximately one-fifth of the participants will drop out from CA interventions in short-term studies. High heterogeneities made it difficult to generalize the findings. Our results suggested that future CA interventions should adopt a blended design with human support, use symptom tracking, compare CA intervention groups against active controls rather than waitlist controls, and include a visual representation of the CA to reduce the attrition rate.

TRIAL REGISTRATION

PROSPERO International Prospective Register of Systematic Reviews CRD42022341415; https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD42022341415.

Systematic review and meta-analysis of AI-based conversational agents for promoting mental health and well-being

Conversational artificial intelligence (AI), particularly AI-based conversational agents (CAs), is gaining traction in mental health care. Despite their growing usage, there is a scarcity of comprehensive evaluations of their impact on mental health and well-being. This systematic review and meta-analysis aims to fill this gap by synthesizing evidence on the effectiveness of AI-based CAs in improving mental health and factors influencing their effectiveness and user experience. Twelve databases were searched for experimental studies of AI-based CAs' effects on mental illnesses and psychological well-being published before May 26, 2023. Out of 7834 records, 35 eligible studies were identified for systematic review, out of which 15 randomized controlled trials were included for meta-analysis. The meta-analysis revealed that AI-based CAs significantly reduce symptoms of depression (Hedge's g 0.64 [95% CI 0.17-1.12]) and distress (Hedge's g 0.7 [95% CI 0.18-1.22]). These effects were more pronounced in CAs that are multimodal, generative AI-based, integrated with mobile/instant messaging apps, and targeting clinical/subclinical and elderly populations. However, CA-based interventions showed no significant improvement in overall psychological well-being (Hedge's g 0.32 [95% CI -0.13 to 0.78]). User experience with AI-based CAs was largely shaped by the quality of human-AI therapeutic relationships, content engagement, and effective communication. These findings underscore the potential of AI-based CAs in addressing mental health issues. Future research should investigate the underlying mechanisms of their effectiveness, assess long-term effects across various mental health outcomes, and evaluate the safe integration of large language models (LLMs) in mental health care.

DSE-NN: Deeply Supervised Efficient Neural Network for Real-Time Remote Photoplethysmography

Non-contact remote photoplethysmography can be used in a variety of medical and healthcare fields by measuring vital signs continuously and unobtrusively. Recently, end-to-end deep learning methods have been proposed to replace the existing handcrafted features. However, since the existing deep learning methods are known as black box models, the problem of interpretability has been raised, and the same problem exists in the remote photoplethysmography (rPPG) network. In this study, we propose a method to visualize temporal and spectral representations for hidden layers, deeply supervise the spectral representation of intermediate layers through the depth of networks and optimize it for a lightweight model. The optimized network improves performance and enables fast training and inference times. The proposed spectral deep supervision helps to achieve not only high performance but also fast convergence speed through the regularization of the intermediate layers. The effect of the proposed methods was confirmed through a thorough ablation study on public datasets. As a result, similar or outperforming results were obtained in comparison to state-of-the-art models. In particular, our model achieved an RMSE of 1 bpm on the PURE dataset, demonstrating its high accuracy. Moreover, it excelled on the V4V dataset with an impressive RMSE of 6.65 bpm, outperforming other methods. We observe that our model began converging from the very first epoch, a significant improvement over other models in terms of learning efficiency. Our approach is expected to be generally applicable to models that learn spectral domain information as well as to the applications of regression that require the representations of periodicity.

A Survey on Databases for Multimodal Emotion Recognition and an Introduction to the VIRI (Visible and InfraRed Image) Database

Multimodal human–computer interaction (HCI) systems pledge a more human–human-like interaction between machines and humans. Their prowess in emanating an unambiguous information exchange between the two makes these systems more reliable, efficient, less error prone, and capable of solving complex tasks. Emotion recognition is a realm of HCI that follows multimodality to achieve accurate and natural results. The prodigious use of affective identification in e-learning, marketing, security, health sciences, etc., has increased demand for high-precision emotion recognition systems. Machine learning (ML) is getting its feet wet to ameliorate the process by tweaking the architectures or wielding high-quality databases (DB). This paper presents a survey of such DBs that are being used to develop multimodal emotion recognition (MER) systems. The survey illustrates the DBs that contain multi-channel data, such as facial expressions, speech, physiological signals, body movements, gestures, and lexical features. Few unimodal DBs are also discussed that work in conjunction with other DBs for affect recognition. Further, VIRI, a new DB of visible and infrared (IR) images of subjects expressing five emotions in an uncontrolled, real-world environment, is presented. A rationale for the superiority of the presented corpus over the existing ones is instituted.