Systematic review of artificial intelligence enabled psychological interventions for depression and anxiety: A comprehensive analysis

Abstract

Psychiatric disorders like depression and anxiety represent significant global health challenges, affecting individuals of all ages and contributing to a substantial disability burden worldwide. Despite advancements in mental health care, barriers such as cost, geographical limits, and social stigma may prevent individuals from receiving early psychological interventions. In recent years, artificial intelligence (AI) has emerged as a prominent tool to address these issues by facilitating early detection, personalized treatment, and intervention delivery for individuals experiencing depression and anxiety. Hence, the present systematic review focused on AI-enabled conversational chatbots for the identification and management of depression and anxiety. The current systematic review yielded a total of ten studies after a thorough analysis. The majority of studies were randomized controlled trials. The most frequently utilized AI method was conversational AI agents which are chatbots available through online software accessible via computers or smartphones. The investigations revealed significant outcomes by using AI for the enhancement of psychotherapy. The majority of studies showed a low risk (71.67%), indicating their reliability, while unclear studies (15%) exhibited some ambiguity without invalidating results. Conversely, studies classified as high risk (13.33%) indicated significant bias and potential errors. Chatbots have emerged as an effective medium for self-help depression and anxiety management. Studies have revealed significant positive outcomes, showing the potential of AI augmentation in psychotherapy to reduce clinical symptomatology. Notably, chatbot-delivered therapies have proven to be more successful than limited bibliotherapy, demonstrating their ability to effectively reduce the symptoms of depression and anxiety while encouraging a stronger therapeutic alliance among participants.

Anxiety and depressive disorders are the most prevalent mental health issues worldwide.[1,2] Studies suggest that approximately 3.8% of the global population experiences depression, while 4% are affected by anxiety disorders.[3,4] According to a recent WHO report, within India the prevalence of depression and anxiety were found to be 4.5% and 3.0% respectively.[5] To prevent these concerning issues, interventions are required,[6] in which psychiatry plays a crucial role in addressing mental illnesses.[7] Cognitive Behavior Therapy (CBT) is a form of psychotherapeutic treatment aiming to address maladaptive thought patterns and behaviors associated with depressive and anxiety disorders.[8]

In recent years, artificial intelligence (AI) has emerged as a vital tool in mental health care. Its advanced algorithms and data analysis techniques offer promising potential to improve the identification and treatment of mental disorders. Integrating AI into psychological interventions opens new pathways to enhance treatment outcomes and increase the accessibility of effective therapies.[9] The rapid growth of machine learning techniques within the AI field has raised hopes that algorithms could overcome the trial-and-error approach prevalent in mental health care.[10] These techniques support precise diagnosis, prognosis, and therapeutic choice.[11] Methods based on deep learning within the machine learning field have proven effective in improving the current state of psychotherapy.[1,12]

Moreover, chatbots which are computer programs designed to engage in text-based or voice-activated conversations, play a significant role in mental health care.[13] Research by Ho et al.,[14] suggests that chatbot conversations can provide emotional, relational, and psychological benefits similar to those experienced in human interactions. Chatbots targeting mental health include Woebot, Shim, Kokobot, Wysa, Vivibot, Pocket Skills, and Tess.[15]

Woebot, an artificial conversational agent, is designed to detect mood.[16] Similarly, Tess is another automated mental health chatbot powered by AI. These chatbots represent innovative approaches to enhance mental health care accessibility and support.[16] Thus, the integration of AI into psychological interventions holds significant promise for addressing conditions like depression and anxiety. Psychological AI delivering CBT has demonstrated feasibility, engagement, and effectiveness in alleviating signs of depression and anxiety in college students. However, additional research is necessary to evaluate the overall effectiveness of psychological AI in delivering comprehensive mental health care, including CBT.[17]

In Psychiatry, CBT remains a highly effective method for managing anxiety and depression; however, evidence suggests that different therapeutic approaches can yield similarly successful outcomes.[18] It has also been evident that the CBT had outperformed other psychotherapeutic measures particularly psychodynamic therapy at post-treatment.[19] This literature review aims to provide a systematic analysis of the use of AI-enabled chatbots/applications for individuals suffering from depression and anxiety and to assess their impact in improving depression and anxiety. Thus, this emphasizes the necessity of such research in advancing mental health care.

MATERIALS AND METHODS

The study framework was developed in compliance with the authorized reporting standards for systematic review.

Search strategy

We conducted a comprehensive literature search across multiple electronic databases, such as MEDLINE, Google Scholar, PubMed, Cochrane Library, Web of Science, Scopus, and other digital resources. Articles were searched using relevant keywords such as “psychological interventions,” “diagnosis of mental disorders,” “deep learning,” “depression, anxiety in patients, detection using AI,” and “use of AI in psychological disorders,” with a restriction to publications in English. The studies were selected as per the inclusion and exclusion criteria mentioned below.

Inclusion criteria

This study included articles published between 2014 and 2024, specifically focusing on human clinical trials that implement an AI technique. The target population includes individuals with emotional problems such as depression or anxiety. The main focus must involve psychological treatment with the use of AI tools to improve ongoing intervention.

Exclusion criteria

The exclusion criteria for this study include incomplete or irrelevant studies, studies exclusively focused on AI for the detection of depression and anxiety without intervention, articles not written in English, and duplicate articles.

Data analysis

The data from articles were input into an Excel spreadsheet and duplicates were removed after extraction from the databases. The abstracts of each article were then independently evaluated, and papers were selected according to the established criteria. The final choice of pertinent studies was made by carefully reviewing the full content of the selected publications.

Evaluation of study quality

Using RevMan software, we assessed the risk of bias according to the Cochrane methodology. The risk was categorized as high, unclear, or low. In this review, 10 studies were included. RevMan software version 5.4 and Microsoft Office Excel 2013 (Microsoft Corporation, USA) were used for piloting the data. The included trials were rated as high risk (-), unclear risk (?), or low risk (+) using the Risk of Bias Tool for RCT.

RESULTS

The present systematic review initially comprised 884 articles obtained from the initial search. Among these, 23 duplicate articles were removed. An additional 214 articles were excluded based on the literature review. Subsequently 647 articles underwent screening. During the detailed screening, an additional 296 articles were excluded due to missing parameters required for this study, leaving 351 published studies to be assessed for eligibility.

Subsequent review of titles and abstracts identified 135 articles as irrelevant or incomplete. An additional 104 articles were excluded for other reasons. Furthermore, 73 articles categorized as case reports/studies and 29 letters to the editor were also excluded. Following these exclusions, full-text articles were evaluated based on the established inclusion criteria. Ultimately, 10 articles were selected for inclusion in the systematic review [Figure 1].

Figure 1.

Illustrates the PRISMA flow diagram and the associated databases mentioned in the article

As seen in Table 1, of the ten studies that were selected, six were RCTs,[16,17,20,21,22,23] one was a nonclinical trial,[24] one was a pre-pilot study,[25] one was a pilot trial[26] and one study was a controlled trial.[27]

Table 1.

Characteristics of the selected studies

Author, year, reference number	Country	Study design	Sample size	Emotional issues evaluated (with scales used)	Type of AI or automated conversational agent used	Mental health intervention/procedure used	Findings
Fitzpatrick KK et al., 2017,[16]	USA	RCT	70 (Woebot group=34 and control group=36)	Depression (PHQ-9) and anxiety (GAD-7)	Woebot	Web based CBT	Individuals from Woebot group showed significant reduction in depressive symptoms compared to the control group (P=0.01) as assessed by PHQ-9. While, in terms of anxiety, both groups showed significant reduction as assessed by GAD-7.
Fulmer R et al., 2018,[17]	USA	RCT	74 (2 test groups with total 50 participants who were randomized to receive Tess access unlimitedly for 2 weeks [group 1 with n=24] and 4 weeks [group 2 with n=26], and the control group [n=24])	Depression (PHQ-9) and Anxiety (GAD-7)	Tess	Web based survey that included GAD-7, PHQ-9 and PANAS	Compared to the control group, test group 1 showed significant reduction of depressive symptoms which was assessed by PHQ-9 (P=0.03). Compared to the control group, test group 1 (P=0.045) and test group 2 (P=0.02) showed significant reduction in anxiety symptoms as assessed by GAD-7. Statistically significant difference was seen on PANAS between test group 1 and control group (P=0.03) which was suggestive of significant impact of Tess.
Sadeh-Sharvit S, et al., 2023,[20]	USA	RCT	47 (23 participants were randomized to an AI platform group and 24 were randomized to treatment as usual [TAU] group)	Depression (PHQ-9) and Anxiety (GAD-7)	AI platform developed by Eleos Health	CBT	Participants from AI platform group had 34% reduction in depressive symptoms and 29% reduction in anxiety symptoms compared to only 20% and 8% reductions in depressive and anxiety symptoms respectively in TAU group.
Kaywan P et al., 2023,[24]	Australia	Nonclinical trial	50	Depression (SIGH-D and IDS-C)	DEPRA chatbot	Structured and authoritative early detection depression interview guide containing 27 combined questions from SIGH-D and IDS-C	DEPRA chatbot uses 2 scientific scoring systems namely IDS-SR and QIDS-SR for early detection of depression. The satisfaction rate of using DEPRA chatbot was found to be 79%.
He Y et al., 2022,[21]	China	RCT	148 (3 arm RCT with 1:1:1 randomization of participants with depressive symptoms to mental health chatbot [XiaoE, n=49], e-book [n=49], and general chatbot group [Xiaoai, n=50])	Depression (PHQ-9)	XiaoE chatbot	CBT	An intention-to-treat analysis showed lower scores on PHQ-9 amongst the study participants from XiaoE group compared to Xiaoai and e-book groups at both T1 (1 week later, P=<0.001) and at T2 (1 month later, P=0.005).
Liu H et al., 2022,[22]	China	RCT	83 (41 study participants were randomly assigned to chatbot test group and 42 were randomly assigned to bibliotherapy control group)	Depression (PHQ-9) and Anxiety (GAD-7)	XiaoNan chatbot (can be accessed by using the smartphone app “WeChat”)	CBT	Intention-to-treat analysis showed that the university students from the XiaoNan chatbot test group had significant reduction of PHQ-9 scores (P=<0.01) and GAD-7 scores (P=0.02). During follow up, it was evident that the reduction in anxiety was significant only during first 4 weeks period of the study out of total 16 weeks period.
Green EP et al., 2020,[25]	Kenya	Single case experimental design pre-pilot study	41	Perinatal depression (PHQ-9)	Tess (Zuri in Kenya)	Healthy Moms perinatal depression intervention curriculum – a CBT based intervention	Most of the study participants who used Zuri had reported trust in Zuri and positive attitude towards its use. They also inferred positive life changes to Zuri with 7% improvement in their mood.
Bird T et al., 2018,[23]	UK	A web-based RCT	213 participants completed baseline questionnaire, of which 171 completed the process of randomization, of which 67 (35 ELIZA and 32 MYLO) provided the data at all 3 points i.e., at baseline, post-intervention and 2-weeks follow-up	Problem distress (11-points scale), Depression and Anxiety (DASS-21)	MYLO (AI based program)	After completion of baseline questionnaire, participants were randomized to MYLO program or to control group with ELIZA program	Both MYLO and ELIZA programs were associated with improvements in depression, anxiety, and problem distress post-intervention and again after 2 weeks period. Although MYLO was rated more beneficial than ELIZA, it was not found to be more effective than later.
Wahle F et al., 2016,[26]	Switzerland	Pilot trial	126 (64 participants uninstalled MOSS app during first 2 weeks and another 26 uninstalled it during next 2 weeks, thus only 36 participants used MOSS app for the 2 weeks or more.)	Depression (PHQ-9)	MOSS	CBT delivered via MOSS	A significant reduction in PHQ-9 score was evident (P=0.01).
Paredes P et al., 2014,[27]	Ecuador, South America	Controlled trial	95 (Only 20 participants used the mobile app for all the 4 weeks.)	Depression (PHQ-9)	Mobile App on Windows Phone v. 8.1	Response data on PHQ-9 was analyzed from those 20 participants who used mobile app for all the 4 weeks.	Participants showed statistically significant lower level of depression while they use the app.

FOOTNOTE:- GAD-7 : Generalized Anxiety Disorder 7 items questionnaire, PHQ-9: Patient Health Questionnaire 9 itmes, PANAS: Positive and Negative Affect Schedule, SIGH-D: structured interview guide for the hamilton depression rating scale, IDS-C: Inventory of Depressive Symptoms-Clinician rated, CBT: cognitive behavior therapy, MOSS: Mobile Sensing and Support, MYLO: manage your life online

The findings showed that chatbots are a viable tool for treating depression and anxiety. AI chatbots demonstrated a noteworthy reduction in depression as indicated by the PHQ-9. The decrease in anxiety was determined by the GAD-7.

Most articles received a low-risk grade (71.67%) based on overall bias evaluation.[16,17,20,21,22,23,24,25,26,27] A “low-risk” rating indicates that these studies used reliable methods to categorize patients into different treatment groups, ensuring the validity of their results. Some studies (15%) were rated as “unclear”,[16,17,20,22,23,24,25] suggesting potential bias, but not significant enough to invalidate their findings, possibly due to missing data. Figure 2 illustrates that “high-risk” studies (13.33%)[16,22,24,25,26,27] exhibited considerable bias, which could lead to incorrect conclusions, often due to gaps in information or reporting errors. Bias is assessed by the judgments (high [-], unclear [?], or low [+]) for the individual elements from 5 domains such as selection, attrition, reporting, detection, and others.

Figure 2.

Risk of bias summary – showing the risk of bias of every domain in every study

This review employed the Cochrane “Risk of Bias” tool to assess RCTs. This tool developed by the Cochrane Collaboration evaluates bias across five domains (bias due to randomization methods, bias due to deviations from intended interventions, bias due to missing outcome data, bias due to selection of reported results, and other bias) as depicted in Figure 3.

Figure 3.

Risk of bias graph – depicting the overall risk of bias of each domain

Use of Artificial Intelligence (AI) for the detection as well as management of depression and anxiety

AI is currently receiving widespread attention across various fields due to its vast applications. In mental health care, AI shows promise in identifying high-risk individuals, offering interventions, and enhancing the prevention as well as treatment of mental illness.[28] In healthcare, AI is primarily employed for diagnostic and treatment recommendations, enhancing patient engagement as well as adherence and streamlining administrative tasks.

The integration of AI into mental health care is widely recognized by practicing psychiatrists. Doraiswamy et al.,[29] conducted a global survey among the psychiatrists, revealing that 47% participants thought that AI or machine learning can moderately change their jobs over a period of next 25 years. Deep learning and artificial intelligence have great potential to enhance mental health outcomes. The rapid advancement of depression and anxiety detection through machine learning has leveraged the extensive text data available from social media platforms.[13]

Chatbots have been developed to support and improve mental health care. The demand for AI chatbots has increased over the years for their role in therapy for mental disorders with depression being the most prevalent. These chatbots enhance the quality of patient interactions, offering an alternative to in-person therapy and freeing up time for medical professionals. AI chatbots can simulate psychotherapist interactions, assess individual levels of depression, and recommend self-help strategies. Individuals using the chatbot demonstrated a noteworthy reduction in PHQ-9 and GAD-7 scores, with a P value of <0.05. These results are supported by the findings from studies referenced[16,17,22] According to the enrolled study by Kaywan P et al., an AI chatbot – DEPRA showed a 79% satisfaction rate in detecting depression and demonstrated a higher rate of user engagement among participants.[24] This chatbot primarily focused on depression detection. Wahle et al.,[26] observed that the binary classification performance for bi-weekly samples of PHQ-9 with a cut-off of >11 based on random forest and support vector machine resulted in 60.1% and 59.1% accuracy, respectively. Fulmer R et al.,[17] found that 86% of participants were satisfied with using Tess to decrease the symptoms of depression and anxiety. The study by Sadeh-Sharvit S et al.,[20] observed that using an AI platform provided by Eleos Health showed a 34% decrease in depression and a 29% decrease in anxiety. Green EP et al., reported a 7% improvement in the mood by using an alpha version of the AI system Zuri.[25]

Questionnaires most commonly used by the final selected studies for the psychological assessment

Patient Health Questionnaire (PHQ-9)

The PHQ-9 is a commonly used tool in psychological research to assess depression. This 9-item self-assessment questionnaire measures the prevalence and intensity of depressive symptoms using a scale from 0 (not at all) to 3 (nearly every day). It aligns with DSM-IV criteria for major depressive disorder and is recognized for its reliability and validity.[30] The PHQ-9 was frequently employed by studies to select participants and measure depression levels for evaluating outcomes.[16,17,20,21,22,25,26,27]

Generalized Anxiety Disorder-7 (GAD-7) –

The GAD-7 is a consistent self-assessment tool used to gauge the severity of anxiety symptoms.[31] Comprising 7 items, it uses a scoring range from 0 to 3, with a total score ranging from 0 to 21. Scores between 5 to 9 indicate mild anxiety, between 10 to 14 indicate moderate anxiety, while scores above 15 suggest severe anxiety. This criterion is frequently used in studies to select participants and assess anxiety as a primary outcome.[16,17,20,22]

DISCUSSION

Depression and anxiety are leading causes of economic burden and premature mortality, with millions of people estimated to suffer from major depressive disorder.[32] Despite the availability of numerous psychological interventions, such as therapy and medication, access to mental healthcare remains a major issue. This is characterized by significant gaps in both access to and quality of care, as well as a limited availability of mental health professionals, particularly in rural areas.[33] Hence, this study was performed to examine the efficacy of using a conversational AI agent as an adjuvant to engage in early depression and anxiety detection and management.

In this systematic review, we assessed studies using electronic databases. We found that among the included studies, there was considerable variation in sample sizes, study designs, mental health interventions delivered, and AI techniques used. This variability underscores the need for further research to enhance replicability and generalizability, as well as to obtain more robust findings in this field.

In the present systematic review, Fitzpatrick et al.,[16] showed significant reductions in depression symptoms with Woebot, while both groups experienced reduced anxiety. The study suggests that conversational agents are a feasible, engaging, and effective method for delivering CBT. Similarly, Kamita et al.,[34] conducted experiments to evaluate Woebot with college students as participants. Results showed a significant decrease in depressive symptoms among participants. Students reported finding their interactions with Woebot more receptive than therapy sessions with healthcare professionals. This suggests that Woebot may offer a promising alternative or complement to traditional therapy for addressing depression symptoms among college students.

Similarly, studies by Green EP et al.[25] and Fulmer R et al.[17] examined the effectiveness of the AI chatbot Tess for diagnosing depression. The Tess chatbot was a cost-effective integrative psychological solution to traditional treatment, effectively reducing symptoms of depression and anxiety. Similarly, Joerin A et al.,[35] found that Tess provided appropriate support and the coping tips on most of the times in 76% participants.

Chen et al.,[36] used WeChat to screen perinatal depression. Participants were selected based on their Edinburgh Postnatal Depression Scale (EPDS) scores, and Long Short-Term Memory (LSTM) networks were used to evaluate outcomes. In another study, Ackerman et al.,[37] found that interactions with Tess provided valuable emotional support to a small group of healthcare system employees (n = 26). Most participants reported that Tess helped offer relevant support and coping strategies.

A present systematic review revealed that several AI techniques have been used in psychological therapies, and the outcomes showed that AI is useful, effective, and has the potential to improve psychotherapy and lessen clinical symptomatology. Conversational AI agents or chatbots accessible via online software on computers or smartphones were the most commonly used AI approach, with the potential to enhance psychotherapy and reduce clinical symptoms.[38] Eleos Health-supported therapy has shown better results for anxiety and depression as well as higher rates of patient retention.[20]

The mental health services provided through community-based clinics utilizing an AI platform specializing in behavioral treatment have proven more effective in alleviating key symptoms. AI provides a cost-effective and accessible solution for mental health care and has the potential to enhance traditional treatment methods on a broad scale. However, it is crucial to recognize that integrative psychological AI is not intended to substitute for a qualified therapist. Rather, it should be seen as a supportive tool that enhances therapeutic outcomes and improves accessibility to mental health care.

Limitations

There exist a few limitations in the current review. In the current review, a literature search was restricted to the studies that were published in the English language only. Thus, there might be a possibility of missing out on other relevant studies that were published in other languages. Another limitation includes a lack of clear differentiation regarding the types of AI used and their effectiveness. Additionally, the available data were insufficient and too heterogeneous to establish AI as a promising method for diagnosing psychological conditions. Excessive dependence on use of modern AI technology can raise the risk of isolation among the users and they may not receive sufficient help during crisis period to the level that may be gained through human psychiatrist interactions. Another limitation that exists is that the AI technology may not be able to take detailed mental status examination to the level of human psychiatrist and may not be able to handle the psychiatric emergencies like acute psychotic episode involving aggressive behavior as well as suicidal and homicidal thoughts.

CONCLUSION

To conclude, this systematic review has revealed the promising potential of AI chatbots in the early detection and diagnosis of symptoms. AI tools offer valuable support in identifying and monitoring patients’ mental health conditions. By using advanced technology, psychiatrists can enhance their diagnostic accuracy and provide timely interventions, ultimately improving patient outcomes. However, further research and development are needed to fully identify the ability of AI in mental healthcare and ensure its integration into clinical practice effectively and ethically.

Authors’ contributions

Concepts and design: ACJ and ASG. Drafting and revision: ACJ, ASG and PBM. Final editing and approval of the version to be published: ACJ, ASG and PBM.

Data availability statement

Not applicable.

Conflicts of interest

There are no conflicts of interest.

Funding Statement

Nil.