@article{pmid41457119,

title = {The Use of Artificial Intelligence for Personalized Treatment in Psychiatry},

author = {Sara Jalali and Qiong You and Victoria Xu and Langfan Chen and Jonathan Zini and Tihare Zamorano and Yangyu Luo and Timothy Friesen and Gabriel James Fontana and David Benrimoh},

doi = {10.1007/s11920-025-01656-y},

issn = {1535-1645},

year  = {2025},

date = {2025-12-01},

journal = {Curr Psychiatry Rep},

volume = {28},

number = {1},

pages = {7},

abstract = {PURPOSE OF REVIEW: This review examines the role of artificial intelligence (AI) in psychiatry in the past 5 years across four domains: screening; outcome prediction; risk and relapse prediction; and psychotherapy.nnRECENT FINDINGS: Machine learning models applied to questionnaires, electronic health records, neuroimaging, and digital phenotyping data demonstrate promising results for predicting symptom trajectories, relapse risk and treatment response, but external and clinical validation is rare. Randomized controlled trials provide some evidence for AI-enabled clinical decision support, but only preliminary evidence for chatbot-delivered psychotherapy. Some preliminary evidence for chatbots in screening exists. Ethical risks, including automation bias, model opacity and socioemotional harms, complicate integration into practice. Current evidence only supports AI's role as a complement to clinical expertise. To realize safe integration of AI into clinical practice, future work should focus on prospective, multi-site trials with active comparators, external validation across diverse populations, transparent reporting, and governance frameworks that prioritize explainability, oversight, and equity.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{pmid41432275,

title = {Dynamic Indicators of Adherence and Retention in Adults Using a Digital Mental Health App: Longitudinal Observational Analysis From the Brighten Study},

author = {Dylan Hamitouche and Youcef Barkat and Deven Parekh and Eva Hammer and David Benrimoh},

doi = {10.2196/69464},

issn = {2292-9495},

year  = {2025},

date = {2025-12-01},

journal = {JMIR Hum Factors},

volume = {12},

pages = {e69464},

abstract = {BACKGROUND: Making optimal use of mobile health technologies requires the validation of digital biomarkers, which in turn demands high levels of participant adherence and retention. However, current remote digital health studies have high attrition rates and low participant adherence, which may introduce bias and limit the generalizability of their findings.nnOBJECTIVE: This study aimed to identify longitudinal indicators of participant retention and adherence, which may serve to develop strategies to improve data collection in digital health studies and improve understanding of how study cohorts are shaped by participant withdrawal and nonadherence.nnMETHODS: We performed secondary analyses on the Brighten study, which consisted of 2 remote, smartphone-based randomized controlled trials evaluating mobile apps for depression treatment, enrolling 2193 participants in total. Participants were asked, after baseline assessment, to complete 7 digital questionnaires regularly. We assessed adherence to digital questionnaires, engagement (postbaseline participation), and retention rates (the proportion of participants who continued completing questionnaires over time) as outcomes. We investigated the relationship between these outcomes and both static measures (eg, demographics and average questionnaire scores) and dynamic measures (eg, changes in questionnaire scores over time).nnRESULTS: The study included 2201 participants, of whom 1093 completed at least 1 nonbaseline questionnaire, with a median completion rate of 37.6% (IQR 15.5%-67.9%). We found significantly higher adherence rates in participants who were less depressed on average over the course of the study (t752=-5.63; P<.001) and in those who perceived clinical improvement (t744=3.78; P=.001). There were significant demographic differences in adherence and engagement, including differences by gender, race, education, income, and income satisfaction. Participants who were more depressed at baseline were more likely to withdraw before completing any nonbaseline questionnaire (t1917=-2.53; P=.01). However, participants who showed improvement in depressive symptoms during the study showed better adherence (Mann-Whitney U=127,084; P<.001) and retention (hazard ratio 0.78, 95% CI 0.67-0.91; P=.002), despite showing greater depressive symptoms at baseline.nnCONCLUSIONS: We show that participants' clinical trajectory of depressive symptoms, as well as their perception of improvement, are important indicators of engagement, adherence, and retention. Expanding knowledge regarding these longitudinal indicators may improve interpretation of outcomes and help build strategies to improve retention and adherence in future clinical trials.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{pmid41237332,

title = {Sleep and Activity Patterns as Transdiagnostic Behavioral Biomarkers in Psychiatry: Longitudinal Observational Study From the DeeP-DD Study},

author = {Dylan Hamitouche and Tihare Zamorano and Youcef Barkat and Deven Parekh and Lena Palaniyappan and Sara Jalali and David Benrimoh},

doi = {10.2196/81107},

issn = {2561-326X},

year  = {2025},

date = {2025-11-01},

journal = {JMIR Form Res},

volume = {9},

pages = {e81107},

abstract = {BACKGROUND: Despite widespread use of symptom rating scales in psychiatry, these tools are limited by reliance on self-report, infrequent administration, and lack of predictive power. This constrains clinicians' ability to monitor illness trajectories or anticipate adverse outcomes like relapse. Actigraphy, a passive wearable-based method for measuring sleep and physical activity, offers objective, high-resolution behavioral data that may better reflect symptom fluctuations. Prior research has shown associations between actigraphy features and mood or psychosis symptoms, but most studies have focused on narrow diagnostic groups or fixed time windows, limiting clinical translation.nnOBJECTIVE: This study aims to examine whether actigraphy-derived sleep and activity features correlate with psychiatric symptom severity in a transdiagnostic psychiatric sample, and to identify which features are most clinically relevant across multiple temporal resolutions.nnMETHODS: We present a feasibility case series study analyzing preliminary data from 8 outpatients (ages 18-52 years) enrolled in the Deep Phenotyping and Digitalization at Douglas (DeeP-DD) study, a prospective transdiagnostic study of digital phenotyping. Participants wore wrist-based actigraphy devices (GENEActiv) for up to 5 months. Symptom severity was measured using a variety of self- and clinician-rated scales. We performed intraindividual Spearman correlations and interindividual repeated measures correlations across daily, weekly, monthly, and full-duration averages.nnRESULTS: Intraindividual analyses revealed that later rise times were significantly associated with higher weekly 9-item Patient Health Questionnaire (PHQ-9) scores in participant 7 (ρ=0.74, P<.001) and participant 4 (ρ=0.78, P=.02), as well as higher weekly 7-item General Anxiety Disorder (GAD-7) scores in participant 7 (ρ=0.59, P=.03). While similar trends were observed at daily and monthly timescales, the weekly resolution yielded the most robust significance. Interindividual analyses showed that weeks with later average rise time correlated with higher PHQ-9 (r=0.48, P<.001) and GAD-7 scores (r=0.38, P=.03), with the PHQ-9 association remaining significant after Bonferroni correction (Bonferroni-corrected P=.02). Increased light physical activity was linked to lower PHQ-9 scores weekly (r=-0.44, P=.001) and monthly (r=-0.53, P=.01). Over the whole duration of the study, increased levels of sedentary activity were associated with lower GAD-7 scores (ρ=0.74; P<.001).nnCONCLUSIONS: Our findings highlight actigraphy-derived sleep and activity features, particularly rise time and physical activity, as promising transdiagnostic markers of psychiatric symptom burden. Their consistent associations across temporal scales and diagnostic groups underscore their potential utility for scalable, real-world clinical monitoring. Future work should validate these findings in larger cohorts and explore advanced analytical methods to capture circadian rhythmicity and symptom dynamics more precisely.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{pmid40374762,

title = {Subtyping first-episode psychosis based on longitudinal symptom trajectories using machine learning},

author = {Yanan Liu and Sara Jalali and Ridha Joober and Martin Lepage and Srividya Iyer and Jai Shah and David Benrimoh},

doi = {10.1038/s44184-025-00129-7},

issn = {2731-4251},

year  = {2025},

date = {2025-05-01},

journal = {Npj Ment Health Res},

volume = {4},

number = {1},

pages = {18},

abstract = {Clinical course after first episode psychosis (FEP) is heterogeneous. Subgrouping and predicting longitudinal symptom trajectories after FEP may help develop personalized treatment approaches. We utilized k-means clustering to identify clusters of 411 FEP patients based on longitudinal positive and negative symptoms. Three clusters were identified. Cluster 1 exhibits lower positive and negative symptoms (LS), lower antipsychotic dose, and relatively higher affective psychosis; Cluster 2 shows lower positive symptoms, persistent negative symptoms (LPPN), and intermediate antipsychotic doses; Cluster 3 presents persistently high levels of both positive and negative symptoms (PPNS), and higher antipsychotic doses. We predicted cluster membership (AUC of 0.74) using ridge logistic regression on baseline data. Key predictors included lower levels of apathy, affective flattening, and anhedonia/asociality in the LS cluster, compared to the LPPN cluster. Hallucination severity, positive thought disorder and manic hostility predicted PPNS. These results help parse the FEP trajectory heterogeneity and may facilitate the development of personalized treatments.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{pmid38906883,

title = {Development of a differential treatment selection model for depression on consolidated and transformed clinical trial datasets},

author = {Kelly Perlman and Joseph Mehltretter and David Benrimoh and Caitrin Armstrong and Robert Fratila and Christina Popescu and Jingla-Fri Tunteng and Jerome Williams and Colleen Rollins and Grace Golden and Gustavo Turecki},

doi = {10.1038/s41398-024-02970-4},

issn = {2158-3188},

year  = {2024},

date = {2024-06-01},

journal = {Transl Psychiatry},

volume = {14},

number = {1},

pages = {263},

abstract = {Major depressive disorder (MDD) is the leading cause of disability worldwide, yet treatment selection still proceeds via "trial and error". Given the varied presentation of MDD and heterogeneity of treatment response, the use of machine learning to understand complex, non-linear relationships in data may be key for treatment personalization. Well-organized, structured data from clinical trials with standardized outcome measures is useful for training machine learning models; however, combining data across trials poses numerous challenges. There is also persistent concern that machine learning models can propagate harmful biases. We have created a methodology for organizing and preprocessing depression clinical trial data such that transformed variables harmonized across disparate datasets can be used as input for feature selection. Using Bayesian optimization, we identified an optimal multi-layer dense neural network that used data from 21 clinical and sociodemographic features as input in order to perform differential treatment benefit prediction. With this combined dataset of 5032 individuals and 6 drugs, we created a differential treatment benefit prediction model. Our model generalized well to the held-out test set and produced similar accuracy metrics in the test and validation set with an AUC of 0.7 when predicting binary remission. To address the potential for bias propagation, we used a bias testing performance metric to evaluate the model for harmful biases related to ethnicity, age, or sex. We present a full pipeline from data preprocessing to model validation that was employed to create the first differential treatment benefit prediction model for MDD containing 6 treatment options.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{pmid38302562,

title = {On the proportion of patients who experience a prodrome prior to psychosis onset: A systematic review and meta-analysis},

author = {David Benrimoh and Viktor Dlugunovych and Abigail C Wright and Peter Phalen and Melissa C Funaro and Maria Ferrara and Albert R Powers and Scott W Woods and Sinan Guloksuz and Alison R Yung and Vinod Srihari and Jai Shah},

doi = {10.1038/s41380-024-02415-w},

issn = {1476-5578},

year  = {2024},

date = {2024-05-01},

journal = {Mol Psychiatry},

volume = {29},

number = {5},

pages = {1361--1381},

abstract = {BACKGROUND: Preventing or delaying the onset of psychosis requires identification of those at risk for developing psychosis. For predictive purposes, the prodrome - a constellation of symptoms which may occur before the onset of psychosis - has been increasingly recognized as having utility. However, it is unclear what proportion of patients experience a prodrome or how this varies based on the multiple definitions used.nnMETHODS: We conducted a systematic review and meta-analysis of studies of patients with psychosis with the objective of determining the proportion of patients who experienced a prodrome prior to psychosis onset. Inclusion criteria included a consistent prodrome definition and reporting the proportion of patients who experienced a prodrome. We excluded studies of only patients with a prodrome or solely substance-induced psychosis, qualitative studies without prevalence data, conference abstracts, and case reports/case series. We searched Ovid MEDLINE, Embase (Ovid), APA PsycInfo (Ovid), Web of Science Core Collection (Clarivate), Cochrane Database of Systematic Reviews, Cochrane Central Register of Controlled Trials, APA PsycBooks (Ovid), ProQuest Dissertation & Thesis, on March 3, 2021. Studies were assessed for quality using the Critical Appraisal Checklist for Prevalence Studies. Narrative synthesis and proportion meta-analysis were used to estimate prodrome prevalence. I and predictive interval were used to assess heterogeneity. Subgroup analyses were used to probe sources of heterogeneity. (PROSPERO ID: CRD42021239797).nnRESULTS: Seventy-one articles were included, representing 13,774 patients. Studies varied significantly in terms of methodology and prodrome definition used. The random effects proportion meta-analysis estimate for prodrome prevalence was 78.3% (95% CI = 72.8-83.2); heterogeneity was high (I 97.98% [95% CI = 97.71-98.22]); and the prediction interval was wide (95% PI = 0.411-0.936). There were no meaningful differences in prevalence between grouped prodrome definitions, and subgroup analyses failed to reveal a consistent source of heterogeneity.nnCONCLUSIONS: This is the first meta-analysis on the prevalence of a prodrome prior to the onset of first episode psychosis. The majority of patients (78.3%) were found to have experienced a prodrome prior to psychosis onset. However, findings are highly heterogenous across study and no definitive source of heterogeneity was found despite extensive subgroup analyses. As most studies were retrospective in nature, recall bias likely affects these results. While the large majority of patients with psychosis experience a prodrome in some form, it is unclear if the remainder of patients experience no prodrome, or if ascertainment methods employed in the studies were not sensitive to their experiences. Given widespread investment in indicated prevention of psychosis through prospective identification and intervention during the prodrome, a resolution of this question as well as a consensus definition of the prodrome is much needed in order to effectively direct and organize services, and may be accomplished through novel, densely sampled and phenotyped prospective cohort studies that aim for representative sampling across multiple settings.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{pmid38441079,

title = {Studying Psychosis Using Natural Language Generation: A Review of Emerging Opportunities},

author = {Lena Palaniyappan and David Benrimoh and Alban Voppel and Roberta Rocca},

doi = {10.1016/j.bpsc.2023.04.009},

issn = {2451-9030},

year  = {2023},

date = {2023-10-01},

journal = {Biol Psychiatry Cogn Neurosci Neuroimaging},

volume = {8},

number = {10},

pages = {994--1004},

abstract = {Disrupted language in psychotic disorders, such as schizophrenia, can manifest as false contents and formal deviations, often described as thought disorder. These features play a critical role in the social dysfunction associated with psychosis, but we continue to lack insights regarding how and why these symptoms develop. Natural language generation (NLG) is a field of computer science that focuses on generating human-like language for various applications. The theory that psychosis is related to the evolution of language in humans suggests that NLG systems that are sufficiently evolved to generate human-like language may also exhibit psychosis-like features. In this conceptual review, we propose using NLG systems that are at various stages of development as in silico tools to study linguistic features of psychosis. We argue that a program of in silico experimental research on the network architecture, function, learning rules, and training of NLG systems can help us understand better why thought disorder occurs in patients. This will allow us to gain a better understanding of the relationship between language and psychosis and potentially pave the way for new therapeutic approaches to address this vexing challenge.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{pmid36964175,

title = {Using dimensionality-reduction techniques to understand the organization of psychotic symptoms in persistent psychotic illness and first episode psychosis},

author = {Leah M Fleming and Ann Catherine Lemonde and David Benrimoh and James M Gold and Jane R Taylor and Ashok Malla and Ridha Joober and Srividya N Iyer and Martin Lepage and Jai Shah and Philip R Corlett},

doi = {10.1038/s41598-023-31909-w},

issn = {2045-2322},

year  = {2023},

date = {2023-03-01},

journal = {Sci Rep},

volume = {13},

number = {1},

pages = {4841},

abstract = {Psychotic disorders are highly heterogeneous. Understanding relationships between symptoms will be relevant to their underlying pathophysiology. We apply dimensionality-reduction methods across two unique samples to characterize the patterns of symptom organization. We analyzed publicly-available data from 153 participants diagnosed with schizophrenia or schizoaffective disorder (fBIRN Data Repository and the Consortium for Neuropsychiatric Phenomics), as well as 636 first-episode psychosis (FEP) participants from the Prevention and Early Intervention Program for Psychosis (PEPP-Montreal). In all participants, the Scale for the Assessment of Positive Symptoms (SAPS) and Scale for the Assessment of Negative Symptoms (SANS) were collected. Multidimensional scaling (MDS) combined with cluster analysis was applied to SAPS and SANS scores across these two groups of participants. MDS revealed relationships between items of SAPS and SANS. Our application of cluster analysis to these results identified: 1 cluster of disorganization symptoms, 2 clusters of hallucinations/delusions, and 2 SANS clusters (asocial and apathy, speech and affect). Those reality distortion items which were furthest from auditory hallucinations had very weak to no relationship with hallucination severity. Despite being at an earlier stage of illness, symptoms in FEP presentations were similarly organized. While hallucinations and delusions commonly co-occur, we found that their specific themes and content sometimes travel together and sometimes do not. This has important implications, not only for treatment, but also for research-particularly efforts to understand the neurocomputational and pathophysiological mechanism underlying delusions and hallucinations.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{pmid34953204,

title = {Evaluating the perceived utility of an artificial intelligence-powered clinical decision support system for depression treatment using a simulation center},

author = {Myriam Tanguay-Sela and David Benrimoh and Christina Popescu and Tamara Perez and Colleen Rollins and Emily Snook and Eryn Lundrigan and Caitrin Armstrong and Kelly Perlman and Robert Fratila and Joseph Mehltretter and Sonia Israel and Monique Champagne and Jérôme Williams and Jade Simard and Sagar V Parikh and Jordan F Karp and Katherine Heller and Outi Linnaranta and Liliana Gomez Cardona and Gustavo Turecki and Howard C Margolese},

doi = {10.1016/j.psychres.2021.114336},

issn = {1872-7123},

year  = {2022},

date = {2022-02-01},

journal = {Psychiatry Res},

volume = {308},

pages = {114336},

abstract = {Aifred is a clinical decision support system (CDSS) that uses artificial intelligence to assist physicians in selecting treatments for major depressive disorder (MDD) by providing probabilities of remission for different treatment options based on patient characteristics. We evaluated the utility of the CDSS as perceived by physicians participating in simulated clinical interactions. Twenty physicians who were either staff or residents in psychiatry or family medicine completed a study in which they had three 10-minute clinical interactions with standardized patients portraying mild, moderate, and severe episodes of MDD. During these scenarios, physicians were given access to the CDSS, which they could use in their treatment decisions. The perceived utility of the CDSS was assessed through self-report questionnaires, scenario observations, and interviews. 60% of physicians perceived the CDSS to be a useful tool in their treatment-selection process, with family physicians perceiving the greatest utility. Moreover, 50% of physicians would use the tool for all patients with depression, with an additional 35% noting that they would reserve the tool for more severe or treatment-resistant patients. Furthermore, clinicians found the tool to be useful in discussing treatment options with patients. The efficacy of this CDSS and its potential to improve treatment outcomes must be further evaluated in clinical trials.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{pmid34694234,

title = {Evaluating the Clinical Feasibility of an Artificial Intelligence-Powered, Web-Based Clinical Decision Support System for the Treatment of Depression in Adults: Longitudinal Feasibility Study},

author = {Christina Popescu and Grace Golden and David Benrimoh and Myriam Tanguay-Sela and Dominique Slowey and Eryn Lundrigan and Jérôme Williams and Bennet Desormeau and Divyesh Kardani and Tamara Perez and Colleen Rollins and Sonia Israel and Kelly Perlman and Caitrin Armstrong and Jacob Baxter and Kate Whitmore and Marie-Jeanne Fradette and Kaelan Felcarek-Hope and Ghassen Soufi and Robert Fratila and Joseph Mehltretter and Karl Looper and Warren Steiner and Soham Rej and Jordan F Karp and Katherine Heller and Sagar V Parikh and Rebecca McGuire-Snieckus and Manuela Ferrari and Howard Margolese and Gustavo Turecki},

doi = {10.2196/31862},

issn = {2561-326X},

year  = {2021},

date = {2021-10-01},

journal = {JMIR Form Res},

volume = {5},

number = {10},

pages = {e31862},

abstract = {BACKGROUND: Approximately two-thirds of patients with major depressive disorder do not achieve remission during their first treatment. There has been increasing interest in the use of digital, artificial intelligence-powered clinical decision support systems (CDSSs) to assist physicians in their treatment selection and management, improving the personalization and use of best practices such as measurement-based care. Previous literature shows that for digital mental health tools to be successful, the tool must be easy for patients and physicians to use and feasible within existing clinical workflows.nnOBJECTIVE: This study aims to examine the feasibility of an artificial intelligence-powered CDSS, which combines the operationalized 2016 Canadian Network for Mood and Anxiety Treatments guidelines with a neural network-based individualized treatment remission prediction.nnMETHODS: Owing to the COVID-19 pandemic, the study was adapted to be completed entirely remotely. A total of 7 physicians recruited outpatients diagnosed with major depressive disorder according to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition criteria. Patients completed a minimum of one visit without the CDSS (baseline) and 2 subsequent visits where the CDSS was used by the physician (visits 1 and 2). The primary outcome of interest was change in appointment length after the introduction of the CDSS as a proxy for feasibility. Feasibility and acceptability data were collected through self-report questionnaires and semistructured interviews.nnRESULTS: Data were collected between January and November 2020. A total of 17 patients were enrolled in the study; of the 17 patients, 14 (82%) completed the study. There was no significant difference in appointment length between visits (introduction of the tool did not increase appointment length; F=0.805; mean squared error 58.08; P=.46). In total, 92% (12/13) of patients and 71% (5/7) of physicians felt that the tool was easy to use; 62% (8/13) of patients and 71% (5/7) of physicians rated that they trusted the CDSS. Of the 13 patients, 6 (46%) felt that the patient-clinician relationship significantly or somewhat improved, whereas 7 (54%) felt that it did not change.nnCONCLUSIONS: Our findings confirm that the integration of the tool does not significantly increase appointment length and suggest that the CDSS is easy to use and may have positive effects on the patient-physician relationship for some patients. The CDSS is feasible and ready for effectiveness studies.nnTRIAL REGISTRATION: ClinicalTrials.gov NCT04061642; http://clinicaltrials.gov/ct2/show/NCT04061642.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{pmid34250463,

title = {Identification of Suicidal Ideation in the  Using Deep Learning},

author = {Sneha Desai and Myriam Tanguay-Sela and David Benrimoh and Robert Fratila and Eleanor Brown and Kelly Perlman and Ann John and Marcos DelPozo-Banos and Nancy Low and Sonia Israel and Lisa Palladini and Gustavo Turecki},

doi = {10.3389/frai.2021.561528},

issn = {2624-8212},

year  = {2021},

date = {2021-01-01},

journal = {Front Artif Intell},

volume = {4},

pages = {561528},

abstract = { Suicidal ideation (SI) is prevalent in the general population, and is a risk factor for suicide. Predicting which patients are likely to have SI remains challenging. Deep Learning (DL) may be a useful tool in this context, as it can be used to find patterns in complex, heterogeneous, and incomplete datasets. An automated screening system for SI could help prompt clinicians to be more attentive to patients at risk for suicide.  Using the Canadian Community Health Survey-Mental Health Component, we trained a DL model based on 23,859 survey responses to classify patients with and without SI. Models were created to classify both lifetime SI and SI over the last 12 months. From 582 possible parameters we produced 96- and 21-feature versions of the models. Models were trained using an undersampling procedure that balanced the training set between SI and non-SI; validation was done on held-out data.  For lifetime SI, the 96 feature model had an Area under the receiver operating curve (AUC) of 0.79 and the 21 feature model had an AUC of 0.77. For SI in the last 12 months the 96 feature model had an AUC of 0.71 and the 21 feature model had an AUC of 0.68. In addition, sensitivity analyses demonstrated feature relationships in line with existing literature.  Although further study is required to ensure clinical relevance and sample generalizability, this study is an initial proof of concept for the use of DL to improve identification of SI. Sensitivity analyses can help improve the interpretability of DL models. This kind of model would help start conversations with patients which could lead to improved care and a reduction in suicidal behavior.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{pmid33403948,

title = {Using a simulation centre to evaluate preliminary acceptability and impact of an artificial intelligence-powered clinical decision support system for depression treatment on the physician-patient interaction},

author = {David Benrimoh and Myriam Tanguay-Sela and Kelly Perlman and Sonia Israel and Joseph Mehltretter and Caitrin Armstrong and Robert Fratila and Sagar V Parikh and Jordan F Karp and Katherine Heller and Ipsit V Vahia and Daniel M Blumberger and Sherif Karama and Simone N Vigod and Gail Myhr and Ruben Martins and Colleen Rollins and Christina Popescu and Eryn Lundrigan and Emily Snook and Marina Wakid and Jérôme Williams and Ghassen Soufi and Tamara Perez and Jingla-Fri Tunteng and Katherine Rosenfeld and Marc Miresco and Gustavo Turecki and Liliana Gomez Cardona and Outi Linnaranta and Howard C Margolese},

doi = {10.1192/bjo.2020.127},

issn = {2056-4724},

year  = {2021},

date = {2021-01-01},

journal = {BJPsych Open},

volume = {7},

number = {1},

pages = {e22},

abstract = {BACKGROUND: Recently, artificial intelligence-powered devices have been put forward as potentially powerful tools for the improvement of mental healthcare. An important question is how these devices impact the physician-patient interaction.nnAIMS: Aifred is an artificial intelligence-powered clinical decision support system (CDSS) for the treatment of major depression. Here, we explore the use of a simulation centre environment in evaluating the usability of Aifred, particularly its impact on the physician-patient interaction.nnMETHOD: Twenty psychiatry and family medicine attending staff and residents were recruited to complete a 2.5-h study at a clinical interaction simulation centre with standardised patients. Each physician had the option of using the CDSS to inform their treatment choice in three 10-min clinical scenarios with standardised patients portraying mild, moderate and severe episodes of major depression. Feasibility and acceptability data were collected through self-report questionnaires, scenario observations, interviews and standardised patient feedback.nnRESULTS: All 20 participants completed the study. Initial results indicate that the tool was acceptable to clinicians and feasible for use during clinical encounters. Clinicians indicated a willingness to use the tool in real clinical practice, a significant degree of trust in the system's predictions to assist with treatment selection, and reported that the tool helped increase patient understanding of and trust in treatment. The simulation environment allowed for the evaluation of the tool's impact on the physician-patient interaction.nnCONCLUSIONS: The simulation centre allowed for direct observations of clinician use and impact of the tool on the clinician-patient interaction before clinical studies. It may therefore offer a useful and important environment in the early testing of new technological tools. The present results will inform further tool development and clinician training materials.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{pmid33733120,

title = {Analysis of Features Selected by a Deep Learning Model for Differential Treatment Selection in Depression},

author = {Joseph Mehltretter and Colleen Rollins and David Benrimoh and Robert Fratila and Kelly Perlman and Sonia Israel and Marc Miresco and Marina Wakid and Gustavo Turecki},

doi = {10.3389/frai.2019.00031},

issn = {2624-8212},

year  = {2019},

date = {2019-01-01},

journal = {Front Artif Intell},

volume = {2},

pages = {31},

abstract = { Deep learning has utility in predicting differential antidepressant treatment response among patients with major depressive disorder, yet there remains a paucity of research describing how to interpret deep learning models in a clinically or etiologically meaningful way. In this paper, we describe methods for analyzing deep learning models of clinical and demographic psychiatric data, using our recent work on a deep learning model of STAR*D and CO-MED remission prediction.  Our deep learning analysis with STAR*D and CO-MED yielded four models that predicted response to the four treatments used across the two datasets. Here, we use classical statistics and simple data representations to improve interpretability of the features output by our deep learning model and provide finer grained understanding of their clinical and etiological significance. Specifically, we use representations derived from our model to yield features predicting both treatment non-response and differential treatment response to four standard antidepressants, and use linear regression and -tests to address questions about the contribution of trauma, education, and somatic symptoms to our models.  Traditional statistics were able to probe the input features of our deep learning models, reproducing results from previous research, while providing novel insights into depression causes and treatments. We found that specific features were predictive of treatment response, and were able to break these down by treatment and non-response categories; that specific trauma indices were differentially predictive of baseline depression severity; that somatic symptoms were significantly different between males and females, and that education and low income proved important psycho-social stressors associated with depression.  Traditional statistics can augment interpretation of deep learning models. Such interpretation can lend us new hypotheses about depression and contribute to building causal models of etiology and prognosis. We discuss dataset-specific effects and ideal clinical samples for machine learning analysis aimed at improving tools to assist in optimizing treatment.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

@article{pmid30286415,

title = {A systematic meta-review of predictors of antidepressant treatment outcome in major depressive disorder},

author = {Kelly Perlman and David Benrimoh and Sonia Israel and Colleen Rollins and Eleanor Brown and Jingla-Fri Tunteng and Raymond You and Eunice You and Myriam Tanguay-Sela and Emily Snook and Marc Miresco and Marcelo T Berlim},

doi = {10.1016/j.jad.2018.09.067},

issn = {1573-2517},

year  = {2019},

date = {2019-01-01},

journal = {J Affect Disord},

volume = {243},

pages = {503--515},

abstract = {INTRODUCTION: The heterogeneity of symptoms and complex etiology of depression pose a significant challenge to the personalization of treatment. Meanwhile, the current application of generic treatment approaches to patients with vastly differing biological and clinical profiles is far from optimal. Here, we conduct a meta-review to identify predictors of response to antidepressant therapy in order to select robust input features for machine learning models of treatment response. These machine learning models will allow us to learn associations between patient features and treatment response which have predictive value at the individual patient level; this learning can be optimized by selecting high-quality input features for the model. While current research is difficult to directly apply to the clinic, machine learning models built using knowledge gleaned from current research may become useful clinical tools.nnMETHODS: The EMBASE and MEDLINE/PubMed online databases were searched from January 1996 to August 2017, using a combination of MeSH terms and keywords to identify relevant literature reviews. We identified a total of 1909 articles, wherein 199 articles met our inclusion criteria.nnRESULTS: An array of genetic, immune, endocrine, neuroimaging, sociodemographic, and symptom-based predictors of treatment response were extracted, varying widely in clinical utility.nnLIMITATIONS: Due to heterogeneous sample sizes, effect sizes, publication biases, and methodological disparities across reviews, we could not accurately assess the strength and directionality of every predictor.nnCONCLUSION: Notwithstanding our cautious interpretation of the results, we have identified a multitude of predictors that can be used to formulate a priori hypotheses regarding the input features for a computational model. We highlight the importance of large-scale research initiatives and clinically accessible biomarkers, as well as the need for replication studies of current findings. In addition, we provide recommendations for future improvement and standardization of research efforts in this field.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}