Early detection of high-risk users in a digital peer-to-peer mental health support platform

Early detection of high-risk users in a digital peer-to-peer mental health support platform | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (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],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Early detection of high-risk users in a digital peer-to-peer mental health support platform Sourav Das, John Naslund, Ben Locke This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-9553769/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Digital peer support platforms have emerged as scalable solutions to mental health problems, but they face critical challenges in identifying users at risk of escalation, a situation that requires urgent clinical or administrative intervention. This study investigates whether user-level characteristics, rather than textual content alone, can be leveraged for early detection of high-risk cases. We analyzed a dataset with N = 178,612 users from Togetherall, a clinically moderated, anonymous online peer-support platform, covering demographics, health background, and engagement patterns (189 predictors). Escalation events (occurring in 1% of users) were modeled as a binary classification outcome. Using machine learning methods such as Logistic Regression, Support Vector Machines (SVM), Random Forest, and XGBoost we compared predictive performance under different feature strategies, including the full feature set, LASSO-based selection (aggressive and moderate), and factor analysis. Results showed that ensemble models, particularly XGBoost (F1 ≈ 0.863) and Random Forest (F1 ≈ 0.855), achieved the highest predictive accuracy with the full feature set. Moderate LASSO preserved predictive performance (XGBoost F1 ≈ 0.862; Random Forest F1 ≈ 0.863), whereas aggressive LASSO substantially reduced accuracy. Factor analysis offered computational efficiency but with some loss in predictive power (XGBoost and Random Forest F1 ≈ 0.811). Most importantly, even with a highly parsimonious feature set derived from aggressive LASSO (time on site, page views, and age) the models retained reasonably strong predictive power (XGBoost F1 ≈ 0.762; Random Forest F1 ≈ 0.769). This underscores the central role of basic engagement metrics and demographics in identifying escalation risk. Our findings demonstrate, for the first time, that user-level sociodemographic, health, and engagement characteristics provide strong predictive signals for escalation risk in online peer-support platforms. In particular, time on site, page views, and age emerged as the most predictive factors. Integrating such models into platform moderation systems could enable earlier detection of risks, enhance user safety, and improve cost-effectiveness in delivering digital peer-to-peer mental health support. Biological sciences/Computational biology and bioinformatics Health sciences/Health care Physical sciences/Mathematics and computing Full Text Additional Declarations No competing interests reported. Supplementary Files CorrespondenceforIRB241479.pdf Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-9553769","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":634301607,"identity":"9edb3ac7-0d00-4fa0-9c74-684d7c745dba","order_by":0,"name":"Sourav Das","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA3ElEQVRIiWNgGAWjYHACAwaGCgYGNiAEARkGhgRitJxBaOEhTgtjG4gmVot8A/PGjz/nHc7jY2BL/Pjjjw0PA3vyAfxWHGArlubddrgY6LDD0rxtaTwMPM/wW2PAwGMgzbjtcGKb/PMGacaGwzwMEjkGBBzGY/zz5xygFgb25p8//vwHasn/gN8zB3jMJHgbQFrYjknwsB0A2YJfh8FhtjJrnmPpIC1p1rxtyTxsPM8IOKy9efPNHzXWifMb2Ixv/vhjJ8fPnvwAvzXM6AJs+NWPglEwCkbBKCAGAACJzTzaBKgdEgAAAABJRU5ErkJggg==","orcid":"","institution":"University of Kassel","correspondingAuthor":true,"prefix":"","firstName":"Sourav","middleName":"","lastName":"Das","suffix":""},{"id":634301608,"identity":"b0c31ebe-ad5d-4514-bc00-11164c7778e6","order_by":1,"name":"John Naslund","email":"","orcid":"","institution":"Harvard University","correspondingAuthor":false,"prefix":"","firstName":"John","middleName":"","lastName":"Naslund","suffix":""},{"id":634301609,"identity":"f7c67fd9-86a8-47ca-81f0-2fd6c2425f10","order_by":2,"name":"Ben Locke","email":"","orcid":"","institution":"Togetherall","correspondingAuthor":false,"prefix":"","firstName":"Ben","middleName":"","lastName":"Locke","suffix":""}],"badges":[],"createdAt":"2026-04-28 11:54:11","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-9553769/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-9553769/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":108493708,"identity":"56f2687f-11c2-4dc1-ac57-efc9ec940d10","added_by":"auto","created_at":"2026-05-05 10:01:22","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":608950,"visible":true,"origin":"","legend":"","description":"","filename":"Togetherallproject20260211.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9553769/v1_covered_09d82478-b2a4-472a-bb49-bf19185bb027.pdf"},{"id":108466261,"identity":"c3fb820f-c2e0-40fe-b692-a3243bf716f6","added_by":"auto","created_at":"2026-05-05 03:41:54","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":110040,"visible":true,"origin":"","legend":"","description":"","filename":"CorrespondenceforIRB241479.pdf","url":"https://assets-eu.researchsquare.com/files/rs-9553769/v1/d4be63ad81985c57d9569401.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Early detection of high-risk users in a digital peer-to-peer mental health support platform","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-9553769/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-9553769/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eDigital peer support platforms have emerged as scalable solutions to mental health problems, but they face critical challenges in identifying users at risk of escalation, a situation that requires urgent clinical or administrative intervention. This study investigates whether user-level characteristics, rather than textual content alone, can be leveraged for early detection of high-risk cases.\u003c/p\u003e\n\u003cp\u003eWe analyzed a dataset with N = 178,612 users from Togetherall, a clinically moderated, anonymous online peer-support platform, covering demographics, health background, and engagement patterns (189 predictors). Escalation events (occurring in 1% of users) were modeled as a binary classification outcome. Using machine learning methods such as Logistic Regression, Support Vector Machines (SVM), Random Forest, and XGBoost we compared predictive performance under different feature strategies, including the full feature set, LASSO-based selection (aggressive and moderate), and factor analysis.\u003c/p\u003e\n\u003cp\u003eResults showed that ensemble models, particularly XGBoost (F1 ≈ 0.863) and Random Forest (F1 ≈ 0.855), achieved the highest predictive accuracy with the full feature set. Moderate LASSO preserved predictive performance (XGBoost F1 ≈ 0.862; Random Forest F1 ≈ 0.863), whereas aggressive LASSO substantially reduced accuracy. Factor analysis offered computational efficiency but with some loss in predictive power (XGBoost and Random Forest F1 ≈ 0.811). Most importantly, even with a highly parsimonious feature set derived from aggressive LASSO (time on site, page views, and age) the models retained reasonably strong predictive power (XGBoost F1 ≈ 0.762; Random Forest F1 ≈ 0.769). This underscores the central role of basic engagement metrics and demographics in identifying escalation risk.\u003c/p\u003e\n\u003cp\u003eOur findings demonstrate, for the first time, that user-level sociodemographic, health, and engagement characteristics provide strong predictive signals for escalation risk in online peer-support platforms. In particular, time on site, page views, and age emerged as the most predictive factors. Integrating such models into platform moderation systems could enable earlier detection of risks, enhance user safety, and improve cost-effectiveness in delivering digital peer-to-peer mental health support.\u003c/p\u003e","manuscriptTitle":"Early detection of high-risk users in a digital peer-to-peer mental health support platform","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-05-05 03:41:50","doi":"10.21203/rs.3.rs-9553769/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"c028845d-d855-4fbc-9b8c-23c10fdeb16f","owner":[],"postedDate":"May 5th, 2026","published":true,"recentEditorialEvents":[{"type":"editorAssigned","content":"","date":"2026-05-04T16:42:04+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-05-01T04:19:50+00:00","index":"","fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":67504535,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":67504536,"name":"Health sciences/Health care"},{"id":67504537,"name":"Physical sciences/Mathematics and computing"}],"tags":[],"updatedAt":"2026-05-05T03:41:50+00:00","versionOfRecord":[],"versionCreatedAt":"2026-05-05 03:41:50","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-9553769","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-9553769","identity":"rs-9553769","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}