Table 1. Demonstrates the key characteristics of the included studies: AI tools, study design, location, findings, and clinical significance.

SI – suicidal ideation; NLP – natural language processing; EMR – electronic medical records; EHR – electronic health record; AI – artificial intelligence; ML – machine learning; SAIPH – Suicide Artificial Intelligence Prediction Heuristic; DSM – Diagnostic and Statistical Manual of Mental Disorders; ANN – artificial neural network; MMI – major mental illness; PPV – positive predictive value; LR – logistic regression.

S. No	Author	Subjects	AI Tools	Study design	Location	Findings	Clinical significance	References
1.	Carson et al. (2019) [9]	Identification of suicidal behavior among hospitalized adolescents admitted in the psychiatric department.	NLP of EHR	Survey	USA	NLP model showed specificity, sensitivity, accuracy, and PPV of 0.22, 0.83, 0.47, and 0.42, respectively. The model also identified key terms related to suicidal attempts that were closely associated to family members, suicidal thoughts, psychotropic medications, and psychiatric disorders.	Use of NLP and ML showed modest success in determining suicide attempts in the psychiatric department.	8
2.	Cook et al. (2016) [11]	Assessment of psychiatric symptoms and SI among adult psychiatric inpatients discharged from the hospital	NLP and ML	Qualitative study	Madrid	The NLP-based model demonstrated a better performance for assessing SI including specificity, PPV, and sensitivity with values of 0.57, 0.61, and 0.56, respectively, compared to the traditional models where the values were 0.62, 0.73, and 0.76, respectively. The outcome was overall positive for NLP-based models in assessing heightened psychiatric symptoms.	NLP-based models demonstrated better outcomes in assessing general mood. Have a higher predictive value in determining suicide risk and psychological distress simply based on general questions.	10
3.	Glaz et al. (2021) [12]	Understanding the significance of studies that used AI techniques for determining mental status.	NLP and ML models	Qualitative study	-	Assessed 327 articles. This review highlighted that language-specific features should be incorporated to enhance the usefulness and performance of different NLP methods.	ML and NLP proved a significant tool in determining the state of mental health. These tools should be used to gain perspective in clinical practice.	11
4.	Cohen et al. (2020) [13]	Use of AI to assess the risk of suicide by use of a mobile application by patients undergoing therapy sessions. Parameters used: Scores of suicidality and depression standardized scale, language samples, and impression of client’s mental status according to the therapist.	NLP and ML models	Qualitative study	USA	Interviews were collected and the therapist evaluated patients’ risk for self-harm or suicide. Findings demonstrate better outcome with logistic regression and vector machines with an accuracy of 0.76 and 0.75, respectively. These models have good discriminative ability allowing improved overall performance.	ML methods can help improve mental health outcomes during therapy sessions. They are effective in identifying mental health based on language samples and voice collections.	12
5.	Fernandes et al. (2018) [14]	Categorize the severity of suicide ideation and investigate the number of suicide attempts reported in a clinical psychiatric database.	NLP and hybrid ML	Scientific report	London	A hybrid model was successful in identifying the number of suicide attempts, with a precision of 82.8%, and a sensitivity of 98.2%. The precision for correct identification of suicidal ideation was 91.7%.	Algorithm based on a dual model of NLP and hybrid ML was able to successfully extract suicide behavior factors, and clinical data from a psychiatric database.	21
6.	Oh et al. (2020) [15]	Determine the different factors related to suicidal ideation and create AI models to predict risks.	ML	Cross sectional	Korea	ML models demonstrate better prediction performance than conventional LR model. AI models outperform traditional models.	ML approach had better performance in predicting suicidal ideation than other models.	20
7.	Graham et al. (2019) [16]	Categorizing mental illness based on: EHR, social media platforms, brain imaging data, novel monitoring systems (e.g., mobile apps), mood rating scales	ML algorithms	Narrative synthesis	USA	ML algorithms useful in classifying mental health disorders (schizophrenia, depression, and suicidal ideation and attempts). Potential to help mental health clinicians differentiate mental health disorders more objectively than relying on traditional DSM-5 scores.	AI techniques can determine mental health outcomes effectively, helping mental health practitioners better understand mental disorders.	14
8.	Corke et al. (2021) [17]	Relationship between number of suicide risk factors included in an algorithm and strengthening of predictions in suicide prediction models.	ML	Review article	England	Results found a higher odds ratio when numerous risk factors for suicide were included in the ML-based studies. (P = 0.02)	ML can enhance the performance of predicting the risk of suicide by increasing the number of suicide risk factors evaluated. Superiority over other methods is yet to be confirmed.	18
9.	Kumar et al. (2020) [18]	Use of AI for the identification of self-harm (uncoded) in severe mental disorders (major depression, schizophrenia, schizoaffective disorder, and bipolar disorder).	ML	Narrative synthesis	USA	AI learning models had an agreement of 83.5% with traditional models. Also found that patients with MMI had the highest overall incidence of encountering coded self-harm.	ML displayed significant success in identifying self-harm visits.	19
10.	Roy et al. (2020) [19]	Based on Twitter data (publicly available) predicting the risk of suicidal thoughts in the future	Algorithm SAIPH	Qualitative study	Canada	A neural network model assessed tweets based on SI cases. AI-based random forest model enabled the prediction of suicidal ideation status. The model demonstrated an accuracy of 0.88 in differentiating suicidal ideation from controls.	SAIPH, as an algorithm has the potential to predict future SI. Can enable clinicians to use this tool for suicide screening and risk assessment.	13
11.	Yang et al. (2021) [20]	Investigating the risk of suicide among users of the website “Zou Fan Tree Hole” and conducting the suicide crisis intervention with high suicide risk (level 6-10).	Tree Hole Intelligent Agent (Artificial Intelligence Program)	Qualitative study	China	The “Tree Hole Action” served as a significant online tool to prevent 3629 potential suicides.	Tree Hole Action showed improved outcomes in monitoring suicide risk and interventions for online users. Example of coordination of AI, social forces, and mental health services to provide the required support to individuals at high suicide risk.	17
12.	Gong et al. (2019) [21]	Use of AI models to determine patterns and progression of depression according to fitted depression trajectories.	ANN		Southern California, Colorado & Washington	ANN model had a strong tendency to demarcate depression severity.	AI models proved helpful in the detection of patterns of depression, its progression, and changes in symptoms over time. Will help design a trajectory-based method for depression patients.	16
13.	Zhong et al. (2019) [22]	Classifying AI algorithm that helps predict the risk of suicidal behavior in pregnant women.	NLP in EHR	Review	Boston	NLP models were successful in improving the identification of suicidal ideation and suicidal attempts. The model was able to correctly predict suicide cases that increased from 125 to 1423, even though the PPV was similar to diagnostic codes of suicidal ideation.	NLP helped predict suicidal behavior by mining unorganized clinical notes. NLP was effective in predicting suicidal behavior in pregnant women by 11-fold.	15