Prognostic predictions in psychosis: exploring the complementary role of machine learning models

Prognostic predictions in psychosis: exploring the complementary role of machine learning models | medRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-P4HH5NV'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search Prognostic predictions in psychosis: exploring the complementary role of machine learning models View ORCID Profile Violet van Dee , View ORCID Profile Seyed Mostafa Kia , View ORCID Profile Caterina Fregosi , View ORCID Profile Wilma E. Swildens , View ORCID Profile Anne Alkema , View ORCID Profile Albert Batalla , Coen van den Berg , View ORCID Profile Danko Coric , View ORCID Profile Edwin van Dellen , View ORCID Profile Lotte G. Dijkstra , Arthur van den Doel , View ORCID Profile Livia S. Dominicus , View ORCID Profile John Enterman , View ORCID Profile Frank L. Gerritse , View ORCID Profile Marte Z. van der Horst , View ORCID Profile Fedor van Houwelingen , Charlotte S. Koch , View ORCID Profile Lisanne E.M. Koomen , Marjan Kromkamp , View ORCID Profile Michelle Lancee , View ORCID Profile Brian E. Mouthaan , View ORCID Profile Diane F. van Rappard , View ORCID Profile Eline J. Regeer , View ORCID Profile Raymond W.J. Salet , View ORCID Profile Metten Somers , Jorgen Straalman , Marjolein H.T. de Vette , View ORCID Profile Judith Voogt , View ORCID Profile Inge Winter-van Rossum , View ORCID Profile René S. Kahn , View ORCID Profile Wiepke Cahn , View ORCID Profile Hugo G. Schnack doi: https://doi.org/10.1101/2025.01.30.25321382 Violet van Dee 1 Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center , Utrecht, the Netherlands 2 Department of Psychiatry, St Antonius Hospital , Utrecht, The Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Violet van Dee For correspondence: v.vandee{at}umcutrecht.nl Seyed Mostafa Kia 1 Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center , Utrecht, the Netherlands 3 Donders Institute for Brain, Cognition, and Behaviour, Radboud University , Nijmegen, the Netherlands 4 Department of Cognitive Science and Artificial Intelligence, Tilburg University , Tilburg, the Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Seyed Mostafa Kia Caterina Fregosi 5 Department of Informatics, System and Communication, University of Milano-Bicocca , Milan, Italy Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Caterina Fregosi Wilma E. Swildens 6 Altrecht Institute for Mental Health Care , Utrecht, The Netherlands 7 Inholland University of Applied Sciences , Amsterdam, The Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Wilma E. Swildens Anne Alkema 1 Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center , Utrecht, the Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Anne Alkema Albert Batalla 1 Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center , Utrecht, the Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Albert Batalla Coen van den Berg 1 Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center , Utrecht, the Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site Danko Coric 8 GGZ Delfland , Delft, The Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Danko Coric Edwin van Dellen 1 Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center , Utrecht, the Netherlands 9 Department of Neurology, UZ Brussel and Vrije Universiteit Brussel , Brussels, Belgium Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Edwin van Dellen Lotte G. Dijkstra 1 Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center , Utrecht, the Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Lotte G. Dijkstra Arthur van den Doel 6 Altrecht Institute for Mental Health Care , Utrecht, The Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site Livia S. Dominicus 1 Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center , Utrecht, the Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Livia S. Dominicus John Enterman 10 Parnassia Group for Mental Health Care , The Hague, The Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for John Enterman Frank L. Gerritse 11 Department of Psychiatry, Tergooi MC , Hilversum, The Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Frank L. Gerritse Marte Z. van der Horst 1 Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center , Utrecht, the Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Marte Z. van der Horst Fedor van Houwelingen 1 Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center , Utrecht, the Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Fedor van Houwelingen Charlotte S. Koch 1 Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center , Utrecht, the Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site Lisanne E.M. Koomen 1 Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center , Utrecht, the Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Lisanne E.M. Koomen Marjan Kromkamp 12 Mediant Mental Health Care , Enschede, The Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site Michelle Lancee 1 Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center , Utrecht, the Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Michelle Lancee Brian E. Mouthaan 1 Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center , Utrecht, the Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Brian E. Mouthaan Diane F. van Rappard 1 Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center , Utrecht, the Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Diane F. van Rappard Eline J. Regeer 6 Altrecht Institute for Mental Health Care , Utrecht, The Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Eline J. Regeer Raymond W.J. Salet 1 Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center , Utrecht, the Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Raymond W.J. Salet Metten Somers 1 Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center , Utrecht, the Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Metten Somers Jorgen Straalman 6 Altrecht Institute for Mental Health Care , Utrecht, The Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site Marjolein H.T. de Vette 13 BuurtzorgT Mental Health Care , Utrecht, the Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site Judith Voogt 6 Altrecht Institute for Mental Health Care , Utrecht, The Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Judith Voogt Inge Winter-van Rossum 1 Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center , Utrecht, the Netherlands 14 Department of Psychiatry, Icahn School of Medicine at Mount Sinai , New York City, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Inge Winter-van Rossum René S. Kahn 14 Department of Psychiatry, Icahn School of Medicine at Mount Sinai , New York City, USA Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for René S. Kahn Wiepke Cahn 1 Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center , Utrecht, the Netherlands 6 Altrecht Institute for Mental Health Care , Utrecht, The Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Wiepke Cahn Hugo G. Schnack 1 Department of Psychiatry, UMC Utrecht Brain Center, University Medical Center , Utrecht, the Netherlands 15 Institute for Language Sciences, Utrecht University , Utrecht, the Netherlands Find this author on Google Scholar Find this author on PubMed Search for this author on this site ORCID record for Hugo G. Schnack Abstract Full Text Info/History Metrics Supplementary material Preview PDF ABSTRACT BACKGROUND Predicting outcomes in schizophrenia spectrum disorders is challenging due to the variability of individual trajectories. While machine learning (ML) shows promise in outcome prediction, is has not yet been integrated into clinical practice. Understanding how ML models (MLMs) can complement psychiatrists’ predictions and bridge the gap between MLM capabilities and practical use is key. OBJECTIVE This study aims to compare the performance of psychiatrists and MLMs in predicting short-term symptomatic and functional remission in patients with first-episode psychosis and explore whether MLMs can improve psychiatrists’ prognostic accuracy. METHOD Twenty-four psychiatrists predicted symptomatic and functional remission probabilities based on written baseline information from 66 patients in the OPTiMiSE trial. ML-generated predictions were then shared with psychiatrists, allowing them to adjust their estimates. A questionnaire assessed trust in MLMs, perceived information gaps, and psychiatrists’ self-assessed predictive accuracy, which was compared to actual accuracy. FINDINGS The predictive accuracy of the MLM was comparable to that of psychiatrists for symptomatic remission (MLM: 0.50, psychiatrists: 0.52) and functional remission (MLM: 0.72, psychiatrists: 0.79). Interrater agreement was low but comparable for psychiatrists and the MLM. Although the MLM did not improve overall predictive accuracy, it showed potential in aiding psychiatrists with difficult-to-predict cases. However, psychiatrists struggled to recognize when to rely on the model’s output and we were unable to determine a clear pattern in these cases based on their characteristics. Psychiatrists could not reliably estimate their predictive accuracy. Psychiatrists expressed moderate to high trust in MLMs for prognostic prediction, but highlighted concerns about the lack of transparency and interpretability of model outputs. CONCLUSIONS MLMs are a promising tool for supporting psychiatric decision-making, particularly in challenging cases. However, their potential remains underutilized due to limitations in predictive accuracy and a lack of clarity in how predictions are generated. Addressing these issues is essential to build trust and foster integration into clinical practice. CLINICAL IMPLICATIONS MLMs are best suited as supplementary tools, providing a second opinion while psychiatrists retain decision-making autonomy. Integrating predictions from both sources may help reduce individual biases and improve accuracy. This approach leverages the strengths of MLMs without compromising clinical responsibility. What is already known on this topic While machine learning models (MLMs) show promise in predicting outcomes in psychotic disorders, they have yet to be integrated into clinical practice. Evidence on the predictive accuracy of psychiatrists for these disorders is limited, with only two small studies published before 1990 suggesting moderate accuracy. Comparisons of MLMs and psychiatrists in this context have not been previously conducted. What this study adds This is the first study to compare the predictive accuracy of psychiatrists with that of an MLM for psychotic disorders and to assess whether an MLM can enhance psychiatrists’ performance. It highlights that while MLMs do not improve overall accuracy, they may support psychiatrists in difficult cases. Insights into psychiatrists’ trust in MLMs and the challenges of implementing these models are also provided. How this study might affect research, practice, or policy The findings emphasize the need for advancements in MLM accuracy, interpretability, and strategies to identify cases where MLMs are most beneficial. These improvements could foster effective integration of MLMs as supplementary tools in clinical practice, aiding psychiatrists in decision-making while maintaining their autonomy. BACKGROUND Schizophrenia spectrum disorders have highly variable outcomes, making accurate and individualized outcome prediction - essential to personalized care - a challenging task. The integration of machine learning models (MLMs) into clinical psychiatry presents a promising opportunity to enhance precision in predicting these outcomes. MLMs can detect patterns and statistical dependencies in large datasets, which can be used to support data-driven clinical decision-making. Over the past two decades, numerous studies utilizing MLMs for outcome prediction in SSD have emerged. 1 - 3 However, many of these studies are limited by small sample sizes and a lack external validation, weakening the robustness of their findings. 1 , 4 , 5 Despite these limitations, ongoing advances in algorithmic predictive power signal that the time has come to bridge the gap between MLM capabilities and clinical practice. Currently, no MLMs for psychiatric disorders, including SSDs, have been integrated into clinical practice yet. 6 , 7 Establishing their clinical value requires comparison to clinicians’ predictions, particularly regarding accuracy, interrater reliability, and potential to assist or augment traditional assessments. Current studies mainly assess MLMs by comparing their predictive accuracy to random chance. For clinical relevance, however, MLM performance should be compared to clinicians’ accuracy. If an MLM performs at least as well as the psychiatrist, it could potentially assist or replace the psychiatrist for this task. Little data exists on the accuracy and interrater reliability of psychiatrists’ outcome predictions. 8 Psychiatrists predictions rely on demographic and clinical information, DSM 5 classifications, personal clinical experience and ‘skilled’ intuition. 9 , 10 The multifactorial etiology of psychiatric symptoms, the lack of objective biomarkers, incomplete information, questionable validity of psychiatric classifications, and susceptibility of clinical impression to bias may contribute to low accuracy and low inter-rater agreement of psychiatrists in clinical predictions. 9 , 11 , 12 Regarding schizophrenia spectrum disorders, only two small studies published before 1990 have investigated the prognostic accuracy of outcome prediction by clinicians. 13 , 14 In the first study, psychiatrists’ predictions on both clinical and functional outcome parameters after 1 year scarcely outperformed chance statements. 13 The second study reported ‘acceptable agreement’ between predictions and true outcome, particularly for clinical outcomes (‘length of psychotic episode’ and ‘time spent in hospital’) compared to functional outcomes (‘occupational capacity’ and ‘functioning in family’). 14 Discrepancies in predictive accuracy between studies might stem from differences in defining the criteria for correct predictions between studies. Interrater reliability of psychiatrists in SSD remains unexamined. OBJECTIVE In our study, psychiatrists and MLM predicted individual chance of symptomatic and functional remission at 10 weeks for 66 patients with first episode psychosis, solely based on the collected baseline information. With the generated predictions, we aimed to 1) compare the predictive performance of psychiatrists versus MLMs, 2) explore whether MLM helped psychiatrists to enhance the accuracy of their prognosis predictions, 3) assess the accuracy of psychiatrists in estimating their own predictions, and 4) evaluate the trust of psychiatrists in MLMs. METHODS STUDY DESIGN This study employed a longitudinal cohort design to investigate the predictive accuracy of both psychiatrists and a MLM in predicting symptomatic and functional remission in people with first episode psychosis. This study was preregistered on AsPredicted.org (protocol #132306). DATA COLLECTION OPTiMiSE dataset In this study, data from the OPTiMiSE trial were utilized (trial identifier number NCT01248195 ). 15 The primary objective of the OPTiMiSE trial was to establish a treatment algorithm for individuals with first episode schizophrenia. The first phase involved 446 patients undergoing amisulpride treatment for four weeks. This phase was completed by 371 patients. Patients meeting the symptom severity component of the consensus criteria for symptomatic remission of the Remission in Schizophrenia Working Group (RSWG), automatically completed the study. 16 Subsequently, the 93 patients not in remission proceeded to the second phase, where they were randomly assigned to either continue amisulpride or switch to olanzapine for an additional six weeks in a double-blind fashion. For the current study, data from the 66 patients that completed the second phase with complete Positive And Negative Symptom Scale (PANSS) records were used. Characteristics of these patients are available in Supplement 1. For all included patients the ‘true outcome’ at 10 weeks follow-up was established. Symptomatic remission was defined according to the symptom severity component of the RSWG-criteria 16 . Functional remission was defined as a score of 71 points and higher on the Personal and Social Performance scale (PSP), because a score between 71-100 refers to only mild difficulties. 17 Machine learning model The psychosis prognosis prediction model proposed by van Opstal et al. 18 was employed in this study. This model, based on a recurrent neural network architecture, accommodates multi-model data from diverse sources and can simultaneously predict multiple outcome measures. The model was trained using the OPTiMiSE dataset and used to predict symptomatic and functional outcomes at 10 weeks for those patients not used for training. 18 Available patient information collected at baseline consisted of demographic, lifestyle, somatic and diagnostic information and data of several questionnaires used for symptom screening and scoring ( Table 1 ). View this table: View inline View popup Table 1 Patient information and measurements Participants Twelve psychiatrists and twelve residents in psychiatry, each with at least 1 year of clinical experience with severe mental illness, participated in the study. In the remainder of this paper, for the purpose of clarity and conciseness, all participants will be referred to as psychiatrists, unless stated otherwise. Written informed consent was obtained from all participants. A pilot study showed that reaching a prognosis based on patient information was time-consuming for psychiatrists. To prevent participants from dropping out due to time constraints or the accuracy of predictions being affected by fatigue, we decided to divide the psychiatrists into three groups. Each participant group consisted of four psychiatrists and four residents in psychiatry. We aimed to distribute the level of experience in working with patients with psychotic disorders equally among the groups. The cases were also divided into three groups (group 1-3), with each group of psychiatrists being assigned one set of 22 cases. Instruments All participants completed a questionnaire (Supplement 2) in the Castor Electronic Data Capture online secure survey software program. 19 All participants’ confidentiality agreements regarding the presented (pseudo-anonymized) patient data were obtained before participants commenced the real survey. In the first part of the survey, general information about the participants (age, sex, country of birth, occupation (psychiatrist/resident in psychiatry) and years of experience working with patients with psychotic disorders) was collected. For the second part of the survey, each participant group received baseline information on 22 (of the 66) randomly assigned patient cases from the OPTiMiSE dataset ( Table 1 , Supplement 1). For each case, participants predicted the chance of symptomatic and functional remission at 10 weeks on a scale of 0-100%. This first prediction is referred to as pre-MLM. Participants were also queried on the importance of specific patient information for their predictions. Subsequently, the MLM prognosis and its level of certainty (uncertain, certain, definite) about that prognosis for the same cases, 18 based on identical information, was presented to the participants, allowing them to adjust their predictions and provide reasoning. Whether adjusted or not, this second psychiatrists’ predictions are referred to as post-MLM. The third part of the survey explored participants’ trust in artificial intelligence for prognosis prediction (measured on a 5-point scale), the extent to which they were inclined to consider the MLMs’ prediction (open-ended question), and whether they perceived any crucial information gaps in the patient data (open-ended question). Finally, psychiatrists were asked to estimate their predictive accuracy in this research on a scale between 0-100%. DATA-ANALYSIS In a pre-analysis, we assessed whether the data of psychiatrists and residents in psychiatry could be pooled for further analysis. The means between the groups were compared with an independent two-sample t-test. 1. Who have better predictive performances; psychiatrists or MLM? 1a How accurate are the prognosis predictions of psychiatrists and MLM? For each rater (participants and MLM), by comparing their predictions with the true outcomes, the following performance metrics were calculated: area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, balanced accuracy and Brier score. Mean prognostic performances (across cases) of participants and MLM were compared using the non-parametric Mann-Whitney-U tests, because the participants’ data was not distributed normally for all prognostic metrics. 1b How comparable are the prognosis predictions among participants and between participants and the MLM? Predictive agreements at group level (all participants + MLM) and pairwise interrater agreements were calculated with intraclass correlation coefficients (ICCs), based on a single-rating, absolute-agreement, 2-way random-effects model. In addition, the pairwise predictive agreement between participants and MLM was compared with the pairwise predictive agreement among participants by a Mann-Whitney-U test. To visualize the relationships and distribution of accuracy of prognostic predictions of participants and MLM (for participants both pre-MLM and post-MLM), we calculated the Euclidean distances between the vectors of predictions for each group. Then we performed classical multidimensional scaling to construct rater representations in a 2-dimensional space, the coordinates of which were used for locating each rater in a scatter plot. 2. Can the MLM help participants to enhance the accuracy of their prognosis predictions? 2a Does the MLM help participants to enhance the accuracy of their predictions in general? With a one-sample t-test we calculated whether the mean difference between accuracy of psychiatrists pre- and post-MLM significantly differed from 0. 2b Does the MLM help participants to enhance the accuracy of their predictions in cases with specific characteristics? To effectively utilize the MLM, we aimed to identify cases where the psychiatrist’s prognosis predictions were poor, but the MLM performed well. All cases were grouped based on the number of correct predictions made by psychiatrists, creating nine difficulty groups ranging from 0 – 8 correct predictions. For each group, we calculated the percentage of correct predictions (accuracy) for both the psychiatrists and the MLM. We then focused on the groups where psychiatrists made ≤ 3 correct predictions, defined as the ‘hard cases’. For the ‘hard case’ groups, we calculated the expected number of correct predictions by both the psychiatrists and the MLM. The potential benefit of identifying cases in these groups was visualized. Then, we investigated whether cases in these ‘hard case’ groups shared similar characteristics that could potentially be recognized by psychiatrists (or by computers). To do this, we created a t-distributed Stochastic Neighbor Embedding (t-SNE) plot to visualize the similarity of case characteristics. Data points representing ‘hard cases’ were highlighted with a distinct color to assess whether these cases shared more similar characteristics than others. More information about the characteristics and data-preparation used in this analysis is provided in Supplement 9. 3. How accurately do psychiatrists estimate their own predictive accuracy? For both symptomatic and functional remission, the means of the psychiatrists’ estimated predictive accuracies were compared to their true predictive accuracies using a paired t-test. A Pearson correlation test was then performed to assess the strength and direction of the linear relationship between the estimated and true accuracies. 4. Do psychiatrists trust machine learning algorithms, and what influences their inclination to listen to the algorithm’s predictions? The information participants provided on their level of trust in machine learning algorithms, and on what influenced their inclination to adhere to or deviate from the MLM predictions was summarized. For the questionnaire, see Supplement 2. FINDINGS The mean age of the participants was 37.8 years (SD 9.1 years) and 50% were male. The mean number of years of working experience as a medical doctor in psychiatry was 15.1 (SD 8.5) year for psychiatrists and 4.0 (SD 1.4) years for residents. 0. Pre-analysis: Prognostic accuracy of psychiatrists and residents A pooled variance t-test (Student’s t-test) showed no significant difference in mean accuracy between the resident and psychiatrist groups (symptomatic remission t(22) = -1.15, p = 0.26, and functional remission (t(22) = 0.18, p = 0.86). 1. Predictive performances of psychiatrists and MLM Predictive performances of psychiatrists and the MLM are displayed in Table 2 and Supplement 3. View this table: View inline View popup Download powerpoint Table 2 Predictive performances of psychiatrists and the MLM 1a Comparison of predictive performances of psychiatrists versus MLM The mean prognostic performances – AUC, accuracy, sensitivity, specificity, balanced accuracy and Brier score – of psychiatrists and MLM, for both symptomatic and functional remission, showed no significant differences (all p-values > 0.5). 1b Comparison of predictive agreement among psychiatrists and between psychiatrists and MLM Predictive agreement (ICC) of all raters (psychiatrist and MLM) at group level was in general poor with all group ICCs <0.5 for both symptomatic and functional remission ( Table 3 , Supplement 4 – Pairwise ICC matrices). 20 For both symptomatic and functional remission the pairwise psychiatrist-MLM ICCs were significantly lower than psychiatrist-psychiatrist ICCs for group 3 (symptomatic remission p = 0.02, functional remission p = 0.001), but not for group 1 and 2. View this table: View inline View popup Download powerpoint Table 3 Predictive agreement by intraclass correlation coefficients (ICC) Multidimensional Scaling (MDS) of the relationships and distributions of predictions on case-level for psychiatrists and MLM showed that the MLM’s predictions were largely similar to those of psychiatrists. In all plots, the MLM’s datapoint was positioned at the outer edge of the cloud of data points. However, in most plots there were psychiatrists who deviated from the general cloud as well. Psychiatrists’ predictions post-MLM were more similar (closer) to the MLM than those pre-MLM (Supplement 5). 2. Influence of the MLM on prognostic accuracy of psychiatrists Predictive performances of psychiatrists pre-MLM and post-MLM are displayed in Table 2 . About 25% of all predictions of psychiatrists were changed post-MLM. Some psychiatrists made no changes while others made changes in up to 77% of their predictions. The amount of percent change per prediction was highly variable as well, with the mean absolute amount of change per prediction ranging from 0 to 47%. Approximately the same amount of predictions was changed in the correct direction as in the incorrect direction post-MLM, both for symptomatic and functional remission (Supplement 6 and 7). 2a Influence of the MLM on mean prognostic accuracy of psychiatrists The mean difference of accuracies of psychiatrists pre-MLM and post-MLM did not differ significantly from 0 for both symptomatic (t(23)= -0.84, p = 0.41) and functional remission (t(23)= -3.61e-16, p = 1) predictions. 2b Potential value of the MLM in cases that are difficult to predict for psychiatrists The categorization of cases in groups based on how many psychiatrists predicted them correctly, showed that for functional remission many cases were easy to predict by the psychiatrists (nearly 50% of cases correct by ≥7 psychiatrists) while predicting symptomatic remission appeared more difficult for them (18% of cases correct by ≥7 psychiatrists). For symptomatic remission, the ‘hard case’ groups (correct by ≤3 psychiatrists, 41% of cases) appeared to be more difficult to predict by the MLM as well ( Figure 1 ). For functional remission, for the ‘hard cases’ groups (18% of cases) the accuracy of the MLM remained 50% or higher ( Figure 1 ). Download figure Open in new tab Figure 1 Potential benefit of using the MLM for the prediction of ‘hard cases’. The 66 cases are grouped based on the number of psychiatrists who correctly predicted them. The number of cases in each group is represented by the dark blue bars (‘occurrence’). Psychiatrist accuracy, calculated as the number of correct predictions divided by the total predictions for that group (group number/8), is shown by the blue line (‘psychiatrists’). The accuracy of the MLM for each group is represented by the orange line. For both symptomatic and functional remission, MLM accuracy (orange line) is higher than psychiatrist accuracy (blue line) for cases correctly predicted by < 4 psychiatrists, referred to as the ‘hard cases.’ Example: For the group of cases where 2 psychiatrists correctly predicted symptomatic remission, 10 cases (dark blue bar) were correctly predicted. Assuming psychiatrist performance reflects general practice, the expected number of correct predictions for this group is 2.5 (light blue bar), while the MLM correctly predicts 4 cases (orange bar). Thus, using the MLM instead of psychiatrists could result in 1.5 additional correct predictions, representing a 15% improvement. For the ‘hard cases’, the MLM’s mean prognostic accuracy exceeded that of psychiatrists for both symptomatic and functional remission. The potential benefit of using the MLM for the prediction of these ‘hard cases’ is visualized in Figure 1 . The t-SNE plots showed that the case characteristics of the ‘hard cases’ were not distinct from the other cases (i.e., no outliers), making it impossible to identify the hard cases based on case characteristics (Supplement 8). 3. Psychiatrists’ estimation of their predictive accuracy For symptomatic remission, psychiatrists overestimated their predictive accuracy (mean estimated accuracy = 0.62, mean true accuracy 0.52; t(19) = 2.79, p = 0.01). There was no significant correlation between the estimated and true accuracies (r = -0.19, p = 0.42). For functional remission, psychiatrists underestimated their predictive accuracy (mean estimated accuracy 0.55; mean true accuracy 0.72, t(19)= - 3.03, p < 0.01). Again, no significant correlation was observed (r= -0.23, p = 0.34). 4. The trust of psychiatrists in MLMs Psychiatrists reported medium to high trust in the MLM, especially when it expressed higher certainty. When the MLM’s prediction differed substantially from their own, psychiatrists adjusted their predictions slightly in the direction suggested by the model. In response to open questions, three psychiatrists highlighted the MLM’s strength in processing large data sets and rendering impartial decisions. Six expressed reluctance to fully trust the model due to its lack of transparency. Seven noted the need for direct patient interaction for contextualization and interpretation of information. Quotes like, ‘You’re trying to turn a patient into statistics’ and ‘I treat patients, not questionnaires’, stressed the value of subjective elements like patient biography, motivation for treatment and the therapeutic alliance. Two psychiatrists viewed the MLM’s predictions as a second opinion for re-evaluating their decisions. DISCUSSION SUMMARY AND INTERPRETATION OF KEY FINDINGS Accurate, personalized outcome prediction is essential for tailoring care to individual patients. This study explored whether machine learning models (MLMs) could improve prognosis prediction in first-episode psychosis using data from the OPTiMiSE study. We found that the mean predictive performance of the MLM was similar to that of psychiatrists for both symptomatic and functional remission. Interrater agreement among psychiatrists and between psychiatrists and the MLM was low but similar across groups. Approximately 25% of psychiatrists’ predictions were altered after exposure to the MLM’s output. The interindividual variation in both the frequency and magnitude of these changes was substantial, with roughly equal proportions of corrections in the right and wrong direction. The MLM did not help to improve predictive accuracy of psychiatrists in general. However, Multidimensional Scaling (MDS) of the predictions revealed that although the MLM’s outputs were largely similar to those of psychiatrists, they often exhibited subtle deviations, as indicated by its position at the outer edge of the cloud. This slightly different predictive approach may help explain why the MLM occasionally outperformed psychiatrists in hard cases. A critical issue was psychiatrists’ difficulty in recognizing when to rely on the model and we could not identify a clear pattern to identify ‘hard cases’ based on their characteristics. Psychiatrists tended to overestimate their accuracy for predicting symptomatic remission, while underestimating accuracy for functional remission. These biases may reflect the routine assessment of symptom severity and less frequent evaluation of functional recovery in clinical practice. This discrepancy in self-perceptions likely influenced how psychiatrists engaged with the MLM’s outputs, shaping their reliance on and interpretation of the model’s predictions. While psychiatrists expressed moderate to high trust in MLMs for prognosis prediction, they voiced concerns about the lack of transparency regarding how the model made predictions. For MLMs to be useful, they must be interpretable. 21 Explainable artificial intelligence (XAI) techniques, such as feature importance measures, 22 , 23 can highlight the variables most crucial for predictions, could improve data-driven explanations and aid in shared-decision making with patients. 24 - 26 Surprisingly, we found no difference in accuracy between psychiatrists and psychiatry residents, which contrasts with previous studies suggesting that clinical experience improves decision-making accuracy. 27 One explanation for this discrepancy could be the absence of direct patient interaction in this study, which prevented participants from forming a ‘clinical impression’. The clinical impression might get better with increasing experience, improving predictive accuracy in real-world setting. LIMITATIONS First, the OPTiMiSE dataset lacked key clinical predictors such as trauma history 28 , motivation to and adherence to treatment 29 - 31 and pre-morbid functioning, 32 which may have affected prediction accuracy. Second, psychiatrists in this study relied solely on case data, lacking direct patient interaction, which could have influenced their predictions. Third, the small sample size and low interrater agreement may have limited the detection of small effects. Due to the random assignment, group 1 contained only one patient that achieved functional remission, affecting the reliability of calculations for this group. Fourth, the 10-week follow-up may have been too short to achieve functional remission, which typically takes more time. 33 Finally, outcome measures (PANSS and PSP scores) were selected and rated by researchers and clinicians, not patients. RECOMMENDATIONS FOR FUTURE RESEARCH Incorporating a broader range of prognostic factors may improve the predictive accuracy of both psychiatrists and MLMs. Machine learning techniques from fields such as weather forecasting, 34 which successfully intergrate complex data, may offer strategies to improve psychiatric predictions. Future MLMs incorporating multimodal data such as neuroimaging, EEG, genetics and wearable data may enhance accuracy. Longitudinal data could also refine predictions. 18 However, increasing the data collection burden might limit the model’s feasibility for routine clinical use. Future research should also explore how direct clinical contact influences prognostic accuracy by comparing predictions based on case data with those made after in-person evaluations. If ‘clinical impression’ is found to predict outcomes, methods for integrating this into MLMs should be developed. One approach is to provide MLMs with descriptive diagnoses, which offer a more comprehensive and holistic understanding of the patient’s condition, aligning more closely with clinicians’ impressions than standard disease classifications. In such a descriptive diagnosis, it would be important to also incorporate the perspective of patients and their informal caregivers, because previous research showed that the perspectives of patients, informal caregivers and healthcare professionals on barriers and facilitators of recovery can diverge and possibly complement each other. Larger sample sizes and more balanced case distributions will also be important for detecting subtle effects, particularly in outcomes like functional remission. Extending the follow-up period beyond 10 weeks is crucial to capture long-term recovery, especially for functional outcomes. For personalized treatment, future research should prioritize patient-relevant outcomes measures. 35 , 36 Finally, comparing the decision-making processes of psychiatrists and MLMs could shed light on how they can complement each other. CLINICAL IMPLICATIONS Despite the comparable predictive performance, replacing psychiatrists with MLMs in prognosis prediction tasks is unlikely, due to the clinical responsibility and liability associated with medical decisions. A more realistic approach would involve using MLM predictions as a second opinion, as we did in this study. In this scenario, psychiatrists would retain full autonomy over their final decision, integrating the MLM output as one piece of supplementary information. Since we do not know in which cases psychiatrists should rely on the MLM, a potential strategy could be to consistently average the predictions of the psychiatrist and the MLM. This approach leverages the strengths of multiple sources to mitigate individual biases and errors. 37 - 39 In the current study, this would have resulted in a mean post-MLM accuracy of 0.54 for symptomatic remission (actual values pre-MLM 0.52, post-MLM 0.51, MLM 0.50) and 0.79 for functional remission (actual values pre-MLM 0.72, post-MLM 0.72, MLM 0.79). For MLMs to add value in clinical practice, several improvements are necessary. First, increasing the accuracy of the MLM would enhance its utility in refining psychiatrists’ predictions. Second, identifying when psychiatrists should rely on the MLM is critical. This could be achieved by improving the model’s ability to estimate the certainty of its predictions or by identifying cases that are hard to predict for psychiatrists. Additionally, explaining model decisions would allow psychiatrists to make more informed judgments about when to trust the model’s output. Ethics statement The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2013. All procedures involving human subjects were approved by the Ethics Committee of the University Medical Centre Utrecht (protocol number 20/053). Conflicts of Interest None declared Data availability statement The data that support the findings of this study and the analytic code are available from the corresponding author, [VvD], upon reasonable request. Funding This work was supported by ZonMw (project ID 63631 0011) and by a grant from the working group AI for Health of the Alliance TU/eWUR-UU-UMCU for the project. Authors contribution VD, SK, CF, WS, WC and HS contributed to the conception and design of the study. AA, AB, CB, DC, ED, LDij, AD, LDo, JE, FG, MH, FH, CK, LK, ML, MK, BM, DR, ER, RS, MS, JS, MV, JV participated as participants (residents/ psychiatrists) in the study. VD, SK, WS, WC and HS drafted the manuscript. RK and IW-vR provided the data and organized the database. VD, HS and SK carried out the modeling and statistical analysis. All authors critically reviewed the article and approved the submitted version. Acknowledgements ChatGPT (GPT-4, October 2024 version) was used to review and improve the grammar, language, and flow of the manuscript. References 1. ↵ Del Fabro L , Bondi E , Serio F , et al. Machine learning methods to predict outcomes of pharmacological treatment in psychosis . Transl Psychiatry 2023 ; 13 ( 1 ): 75 . doi: 10.1038/s41398-023-02371-z [published Online First: 20230302] OpenUrl CrossRef PubMed 2. Leighton SP , Upthegrove R , Krishnadas R , et al. Development and validation of multivariable prediction models of remission, recovery, and quality of life outcomes in people with first episode psychosis: a machine learning approach . Lancet Digit Health 2019 ; 1 ( 6 ): e261 – e70 . doi: 10.1016/s2589-7500(19)30121-9 [published Online First: 20190912] OpenUrl CrossRef 3. ↵ de Nijs J , Burger TJ , Janssen RJ , et al. Individualized prediction of three-and six-year outcomes of psychosis in a longitudinal multicenter study: a machine learning approach . NPJ Schizophr 2021 ; 7 ( 1 ): 34 . doi: 10.1038/s41537-021-00162-3 [published Online First: 20210702] OpenUrl CrossRef PubMed 4. ↵ Dwyer DB , Falkai P , Koutsouleris N. Machine Learning Approaches for Clinical Psychology and Psychiatry . Annu Rev Clin Psychol 2018 ; 14 : 91 – 118 . doi: 10.1146/annurev-clinpsy-032816-045037 [published Online First: 20180129] OpenUrl CrossRef PubMed 5. ↵ Schnack HG , Kahn RS . Detecting Neuroimaging Biomarkers for Psychiatric Disorders: Sample Size Matters . Front Psychiatry 2016 ; 7 : 50 . doi: 10.3389/fpsyt.2016.00050 [published Online First: 20160331] OpenUrl CrossRef 6. ↵ Salazar de Pablo G , Studerus E , Vaquerizo-Serrano J , et al. Implementing Precision Psychiatry: A Systematic Review of Individualized Prediction Models for Clinical Practice . Schizophr Bull 2021 ; 47 ( 2 ): 284 – 97 . doi: 10.1093/schbul/sbaa120 OpenUrl CrossRef 7. ↵ Meehan AJ , Lewis SJ , Fazel S , et al. Clinical prediction models in psychiatry: a systematic review of two decades of progress and challenges . Mol Psychiatry 2022 ; 27 ( 6 ): 2700 – 08 . doi: 10.1038/s41380-022-01528-4 [published Online First: 20220401] OpenUrl CrossRef PubMed 8. ↵ Schulz P , Berney P. Clinicians’ predictions of patient response to psychotropic medications . Dialogues Clin Neurosci 2004 ; 6 ( 1 ): 105 – 11 . doi: 10.31887/DCNS.2004.6.1/pschulz OpenUrl CrossRef PubMed 9. ↵ Lennon MJ , Harmer C. Machine learning prediction will be part of future treatment of depression . Australian & New Zealand Journal of Psychiatry 2023 ; 57 ( 10 ): 1316 – 23 . doi: 10.1177/00048674231158267 OpenUrl CrossRef PubMed 10. ↵ Van den Brink N , Holbrechts B , Brand PLP , et al. Role of intuitive knowledge in the diagnostic reasoning of hospital specialists: a focus group study . BMJ Open 2019 ; 9 ( 1 ): e022724 . doi: 10.1136/bmjopen-2018-022724 [published Online First: 20190128] OpenUrl Abstract / FREE Full Text 11. ↵ Parnas J , Parnas AU . Refining the Diagnostic Criteria for Schizophrenia: An Infinite Task . Schizophr Bull 2024 ; 50 ( 1 ): 12 – 13 . doi: 10.1093/schbul/sbad154 OpenUrl CrossRef PubMed 12. ↵ Regier DA , Narrow WE , Clarke DE , et al. DSM-5 field trials in the United States and Canada, Part II: test-retest reliability of selected categorical diagnoses . Am J Psychiatry 2013 ; 170 ( 1 ): 59 – 70 . doi: 10.1176/appi.ajp.2012.12070999 OpenUrl CrossRef PubMed Web of Science 13. ↵ Giel R , Wiersma D , de Jong PA , Slooff C. Prognosis and outcome in a cohort of patients with non-affective functional psychosis . Eur Arch Psychiatry Neurol Sci 1984 ; 234 ( 2 ): 97 – 101 . doi: 10.1007/bf00381214 OpenUrl CrossRef PubMed 14. ↵ Atakan Z , Cliff G , Cooper JE . Can psychiatrists predict the one-year outcome of schizophrenia? Soc Psychiatry Psychiatr Epidemiol 1990 ; 25 ( 3 ): 117 – 24 . doi: 10.1007/bf00782738 OpenUrl CrossRef PubMed 15. ↵ Kahn RS , Winter van Rossum I , Leucht S , et al. Amisulpride and olanzapine followed by open-label treatment with clozapine in first-episode schizophrenia and schizophreniform disorder (OPTiMiSE): a three-phase switching study . Lancet Psychiatry 2018 ; 5 ( 10 ): 797 – 807 . doi: 10.1016/s2215-0366(18)30252-9 [published Online First: 20180813] OpenUrl CrossRef PubMed 16. ↵ Andreasen NC , Carpenter WT , Jr. , Kane JM , et al. Remission in schizophrenia: proposed criteria and rationale for consensus . Am J Psychiatry 2005 ; 162 ( 3 ): 441 – 9 . doi: 10.1176/appi.ajp.162.3.441 OpenUrl CrossRef PubMed Web of Science 17. ↵ Morosini PL , Magliano L , Brambilla L , et al. Development, reliability and acceptability of a new version of the DSM-IV Social and Occupational Functioning Assessment Scale (SOFAS) to assess routine social functioning . Acta Psychiatr Scand 2000 ; 101 ( 4 ): 323 – 9 . OpenUrl CrossRef PubMed Web of Science 18. ↵ van Opstal DPJ , Kia SM , Jakob L , et al. Psychosis Prognosis Predictor: A continuous and uncertainty-aware prediction of treatment outcome in first-episode psychosis . Acta Psychiatr Scand 2024 doi: 10.1111/acps.13754 [published Online First: 20240918] OpenUrl CrossRef 19. ↵ Castor EDC . Castor Electronic Data Capture 2019 [ 27 Aug . 2019 ]. Available from: https://castoredc.com xaccessed August 28, 2019 . 20. ↵ Koo TK , Li MY . A Guideline of Selecting and Reporting Intraclass Correlation Coefficients for Reliability Research . J Chiropr Med 2016 ; 15 ( 2 ): 155 – 63 . doi: 10.1016/j.jcm.2016.02.012 [published Online First: 20160331] OpenUrl CrossRef PubMed 21. ↵ Giddings R , Joseph A , Callender T , et al. Factors influencing clinician and patient interaction with machine learning-based risk prediction models: a systematic review . The Lancet Digital Health 2024 ; 6 ( 2 ): e131 – e44 . doi: 10.1016/S2589-7500(23)00241-8 OpenUrl CrossRef 22. ↵ Smucny J , Davidson I , Carter CS . Comparing machine and deep learning-based algorithms for prediction of clinical improvement in psychosis with functional magnetic resonance imaging . Hum Brain Mapp 2021 ; 42 ( 4 ): 1197 – 205 . doi: 10.1002/hbm.25286 [published Online First: 20201113] OpenUrl CrossRef PubMed 23. ↵ van Dee V , Kia SM , Winter-van Rossum I , et al. Revealing the impact of psychiatric comorbidities on treatment outcome in early psychosis using counterfactual model explanation . Front Psychiatry 2023 ; 14 : 1237490 . doi: 10.3389/fpsyt.2023.1237490 [published Online First: 20231012] OpenUrl CrossRef PubMed 24. ↵ Joyce DW , Kormilitzin A , Smith KA , Cipriani A. Explainable artificial intelligence for mental health through transparency and interpretability for understandability . NPJ Digit Med 2023 ; 6 ( 1 ): 6 . doi: 10.1038/s41746-023-00751-9 [published Online First: 20230118] OpenUrl CrossRef PubMed 25. Barredo Arrieta A , Díaz-Rodríguez N , Del Ser J , et al. Explainable Artificial Intelligence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI . Information Fusion 2020 ; 58 : 82 – 115 . doi: 10.1016/j.inffus.2019.12.012 OpenUrl CrossRef 26. ↵ Wichmann RM , Fagundes TP , de Oliveira TA , et al. Physician preference for receiving machine learning predictive results: A cross-sectional multicentric study . PLoS One 2022 ; 17 ( 12 ): e0278397 . doi: 10.1371/journal.pone.0278397 [published Online First: 20221214] OpenUrl CrossRef PubMed 27. ↵ Spengler PM , Pilipis LA . A comprehensive meta-reanalysis of the robustness of the experience-accuracy effect in clinical judgment . J Couns Psychol 2015 ; 62 ( 3 ): 360 – 78 . doi: 10.1037/cou0000065 [published Online First: 20150323] OpenUrl CrossRef PubMed 28. ↵ Vila-Badia R , Butjosa A , Del Cacho N , et al. Types, prevalence and gender differences of childhood trauma in first-episode psychosis. What is the evidence that childhood trauma is related to symptoms and functional outcomes in first episode psychosis? A systematic review . Schizophr Res 2021 ; 228 : 159 – 79 . doi: 10.1016/j.schres.2020.11.047 [published Online First: 20210110] OpenUrl CrossRef PubMed 29. ↵ Le TP , Ventura J , Subotnik KL , Nuechterlein KH . Intrinsic motivation predicts cognitive and functional gains during coordinated specialty care for first-episode schizophrenia . Schizophr Res 2024 ; 266 : 249 – 55 . doi: 10.1016/j.schres.2024.02.037 [published Online First: 20240304] OpenUrl CrossRef PubMed 30. Taub S , Krivoy A , Whiskey E , Shergill SS . New approaches to antipsychotic medication adherence - safety, tolerability and acceptability . Expert Opin Drug Saf 2022 ; 21 ( 4 ): 517 – 24 . doi: 10.1080/14740338.2021.1983540 [published Online First: 20210929] OpenUrl CrossRef PubMed 31. ↵ Browne J , Wright AC , Berry K , et al. The alliance-outcome relationship in individual psychosocial treatment for schizophrenia and early psychosis: A meta-analysis . Schizophr Res 2021 ; 231 : 154 – 63 . doi: 10.1016/j.schres.2021.04.002 [published Online First: 20210415] OpenUrl CrossRef PubMed 32. ↵ van Dee V , Schnack HG , Cahn W. Systematic review and meta-analysis on predictors of prognosis in patients with schizophrenia spectrum disorders: An overview of current evidence and a call for prospective research and open access to datasets . Schizophr Res 2023 ; 254 : 133 – 42 . doi: 10.1016/j.schres.2023.02.024 [published Online First: 2023/03/03] OpenUrl CrossRef PubMed 33. ↵ Santesteban-Echarri O , Paino M , Rice S , et al. Predictors of functional recovery in first-episode psychosis: A systematic review and meta-analysis of longitudinal studies . Clinical Psychology Review 2017 ; 58 : 59 – 75 . doi: 10.1016/j.cpr.2017.09.007 OpenUrl CrossRef PubMed 34. ↵ Topol EJ . Medical forecasting . Science 2024 ; 384 ( 6698 ): eadp7977 . doi: 10.1126/science.adp7977 [published Online First: 20240523] OpenUrl CrossRef PubMed 35. ↵ Karow A , Naber D , Lambert M , Moritz S. Remission as perceived by people with schizophrenia, family members and psychiatrists . Eur Psychiatry 2012 ; 27 ( 6 ): 426 – 31 . doi: 10.1016/j.eurpsy.2011.01.013 [published Online First: 2011/05/17] OpenUrl CrossRef PubMed Web of Science 36. ↵ Dee V , Swildens , W. , Schnack , H.G. , Cahn , W. In Pursuit of Recovery; A Comparative Study of Stakeholder Perspectives on Outcomes of People with Psychosis . Community Mental Health Journal 2024 37. ↵ Signorini G , Dagani J , Bulgari V , et al. Moderate efficiency of clinicians’ predictions decreased for blurred clinical conditions and benefits from the use of BRASS index. A longitudinal study on geriatric patients’ outcomes . Journal of Clinical Epidemiology 2016 ; 69 : 51 – 60 . doi: 10.1016/j.jclinepi.2015.08.017 OpenUrl CrossRef PubMed 38. Prince V , Bellile EL , Sun Y , et al. Individualized risk prediction of outcomes for oral cavity cancer patients . Oral Oncol 2016 ; 63 : 66 – 73 . doi: 10.1016/j.oraloncology.2016.11.005 [published Online First: 20161123] OpenUrl CrossRef PubMed 39. ↵ Kattan MW , O’Rourke C , Yu C , Chagin K. The Wisdom of Crowds of Doctors: Their Average Predictions Outperform Their Individual Ones . Med Decis Making 2016 ; 36 ( 4 ): 536 – 40 . doi: 10.1177/0272989×15581615 [published Online First: 20150415] OpenUrl CrossRef PubMed View the discussion thread. Back to top Previous Next Posted February 02, 2025. Download PDF Supplementary Material Email Thank you for your interest in spreading the word about medRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Prognostic predictions in psychosis: exploring the complementary role of machine learning models Message Subject (Your Name) has forwarded a page to you from medRxiv Message Body (Your Name) thought you would like to see this page from the medRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Prognostic predictions in psychosis: exploring the complementary role of machine learning models Violet van Dee , Seyed Mostafa Kia , Caterina Fregosi , Wilma E. Swildens , Anne Alkema , Albert Batalla , Coen van den Berg , Danko Coric , Edwin van Dellen , Lotte G. Dijkstra , Arthur van den Doel , Livia S. Dominicus , John Enterman , Frank L. Gerritse , Marte Z. van der Horst , Fedor van Houwelingen , Charlotte S. Koch , Lisanne E.M. Koomen , Marjan Kromkamp , Michelle Lancee , Brian E. Mouthaan , Diane F. van Rappard , Eline J. Regeer , Raymond W.J. Salet , Metten Somers , Jorgen Straalman , Marjolein H.T. de Vette , Judith Voogt , Inge Winter-van Rossum , René S. Kahn , Wiepke Cahn , Hugo G. Schnack medRxiv 2025.01.30.25321382; doi: https://doi.org/10.1101/2025.01.30.25321382 Share This Article: Copy Citation Tools Prognostic predictions in psychosis: exploring the complementary role of machine learning models Violet van Dee , Seyed Mostafa Kia , Caterina Fregosi , Wilma E. Swildens , Anne Alkema , Albert Batalla , Coen van den Berg , Danko Coric , Edwin van Dellen , Lotte G. Dijkstra , Arthur van den Doel , Livia S. Dominicus , John Enterman , Frank L. Gerritse , Marte Z. van der Horst , Fedor van Houwelingen , Charlotte S. Koch , Lisanne E.M. Koomen , Marjan Kromkamp , Michelle Lancee , Brian E. Mouthaan , Diane F. van Rappard , Eline J. Regeer , Raymond W.J. Salet , Metten Somers , Jorgen Straalman , Marjolein H.T. de Vette , Judith Voogt , Inge Winter-van Rossum , René S. Kahn , Wiepke Cahn , Hugo G. Schnack medRxiv 2025.01.30.25321382; doi: https://doi.org/10.1101/2025.01.30.25321382 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Psychiatry and Clinical Psychology Subject Areas All Articles Addiction Medicine (568) Allergy and Immunology (863) Anesthesia (300) Cardiovascular Medicine (4440) Dentistry and Oral Medicine (444) Dermatology (383) Emergency Medicine (608) Endocrinology (including Diabetes Mellitus and Metabolic Disease) (1510) Epidemiology (15229) Forensic Medicine (30) Gastroenterology (1126) Genetic and Genomic Medicine (6605) Geriatric Medicine (668) Health Economics (998) Health Informatics (4541) Health Policy (1369) Health Systems and Quality Improvement (1613) Hematology (543) HIV/AIDS (1265) Infectious Diseases (except HIV/AIDS) (15923) Intensive Care and Critical Care Medicine (1103) Medical Education (623) Medical Ethics (147) Nephrology (668) Neurology (6604) Nursing (346) Nutrition (998) Obstetrics and Gynecology (1145) Occupational and Environmental Health (957) Oncology (3334) Ophthalmology (974) Orthopedics (369) Otolaryngology (420) Pain Medicine (436) Palliative Medicine (130) Pathology (663) Pediatrics (1693) Pharmacology and Therapeutics (692) Primary Care Research (711) Psychiatry and Clinical Psychology (5448) Public and Global Health (9235) Radiology and Imaging (2199) Rehabilitation Medicine and Physical Therapy (1370) Respiratory Medicine (1196) Rheumatology (594) Sexual and Reproductive Health (712) Sports Medicine (530) Surgery (712) Toxicology (99) Transplantation (289) Urology (265) (function(){function c(){var b=a.contentDocument||a.contentWindow.document;if(b){var d=b.createElement('script');d.innerHTML="window.__CF$cv$params={r:'a016da0eacb7c13d',t:'MTc3OTczNjM0Nw=='};var a=document.createElement('script');a.src='/cdn-cgi/challenge-platform/scripts/jsd/main.js';document.getElementsByTagName('head')[0].appendChild(a);";b.getElementsByTagName('head')[0].appendChild(d)}}if(document.body){var a=document.createElement('iframe');a.height=1;a.width=1;a.style.position='absolute';a.style.top=0;a.style.left=0;a.style.border='none';a.style.visibility='hidden';document.body.appendChild(a);if('loading'!==document.readyState)c();else if(window.addEventListener)document.addEventListener('DOMContentLoaded',c);else{var e=document.onreadystatechange||function(){};document.onreadystatechange=function(b){e(b);'loading'!==document.readyState&&(document.onreadystatechange=e,c())}}}})();