Multimodal prediction of psychotic-like experiences using elastic net modeling: External validation in a clinical sample

Seda Arslan; Merve Kaşıkçı; Osman Dağ; Didenur Şahin-Çevik; Işık Batuhan Çakmak; Evangelos Vassos; Martijn van den Heuvel; Timothea Toulopoulou

doi:10.1017/S0033291725102201

Multimodal prediction of psychotic-like experiences using elastic net modeling: External validation in a clinical sample

Seda Arslan, Merve Kaşıkçı, Osman Dağ, Didenur Şahin-Çevik, Işık Batuhan Çakmak, Evangelos Vassos, Martijn van den Heuvel, Timothea Toulopoulou

Research output: Contribution to Journal › Article › Academic › peer-review

Abstract

Background

Psychotic-like experiences (PLEs) are considered a subclinical component of psychosis continuum. Studies indicate that PLEs arise from multimodal factors, yet research comprehensively examining these factors together remains scarce. Using a large youth sample, we present the first model that simultaneously examines multimodal factors related to PLEs. As a secondary aim, we evaluate the model's ability to explain psychosis in an external validation cohort that included individuals experiencing psychosis.

Methods

After applying variable selection including generalized estimating equations, correlation filtering, Least Absolute Shrinkage and Selection Operator model to 741 variables (i.e., environmental factors, cognitive appraisals, clinical variables, cognitive functioning, and structural brain connectome measures), obtained PLEs predictors (N = 27) and covariates (i.e., age, sex, IQ) were included in the classification model based on Elastic Net algorithm for predicting high/low PLEs in 396 healthy participants aged 14-24 (Mage = 19.72 ± 2.5). We externally validated PLE-related predictors in a clinical sample comprising first-episode psychosis patients (n = 19), their siblings (n = 20), and healthy controls (n = 19).

Results

Eleven factors, including environmental and cognitive appraisals, along with 16 structural network properties spanning frontal, temporal, occipital, and parietal regions, were identified as important predictors of PLEs. The model's performance was moderate in predicting low versus high PLEs (accuracy = 75%, AUC = 0.750). Specificity was high (84.2%) in distinguishing siblings from patients.

Conclusions

Multimodal features, including environmental burden, cognitive schemas, and brain network alterations, predict PLEs and partially generalize to clinical psychosis. These variables may reflect intermediate phenotypes across the psychosis spectrum, offering insights into both vulnerability and resilience.

Original language	English
Article number	e346
Pages (from-to)	1-10
Number of pages	10
Journal	Psychological Medicine
Volume	55
Early online date	14 Nov 2025
DOIs	https://doi.org/10.1017/S0033291725102201
Publication status	Published - 2025

Funding

We would like to thank Timucin Baş, Rabia Sen, Kübra Çelikbaş, Tuba Şahin İlikoğlu, İlayda Aydogan, Eda Gül Karaca, Su Karakelle, Hande Ezgi Atmaca Turan, and Kader Kubat for their contribution to collecting the data for this study.We acknowledge the use of Grammarly and ChatGPT to identify improvements in the writing style. This research was partly funded by the Scientific and Technological Research Institution of Turkiye, Project Number: 119 k410, to Timothea Toulopoulou.

Funders	Funder number
Hande Ezgi Atmaca Turan
Scientific and Technological Research Institution of Turkiye	119 k410

Keywords

elastic net modeling
machine learning
psychosis first episode
psychotic-like experiences
structural connectome

Access to Document

10.1017/S0033291725102201

Persistent URL (handle)

Persistent link

Cite this

@article{b8f7fff78ad24136af7e5da86758a030,

title = "Multimodal prediction of psychotic-like experiences using elastic net modeling: External validation in a clinical sample",

abstract = "Background Psychotic-like experiences (PLEs) are considered a subclinical component of psychosis continuum. Studies indicate that PLEs arise from multimodal factors, yet research comprehensively examining these factors together remains scarce. Using a large youth sample, we present the first model that simultaneously examines multimodal factors related to PLEs. As a secondary aim, we evaluate the model's ability to explain psychosis in an external validation cohort that included individuals experiencing psychosis. Methods After applying variable selection including generalized estimating equations, correlation filtering, Least Absolute Shrinkage and Selection Operator model to 741 variables (i.e., environmental factors, cognitive appraisals, clinical variables, cognitive functioning, and structural brain connectome measures), obtained PLEs predictors (N = 27) and covariates (i.e., age, sex, IQ) were included in the classification model based on Elastic Net algorithm for predicting high/low PLEs in 396 healthy participants aged 14-24 (Mage = 19.72 ± 2.5). We externally validated PLE-related predictors in a clinical sample comprising first-episode psychosis patients (n = 19), their siblings (n = 20), and healthy controls (n = 19). ResultsEleven factors, including environmental and cognitive appraisals, along with 16 structural network properties spanning frontal, temporal, occipital, and parietal regions, were identified as important predictors of PLEs. The model's performance was moderate in predicting low versus high PLEs (accuracy = 75\%, AUC = 0.750). Specificity was high (84.2\%) in distinguishing siblings from patients. Conclusions Multimodal features, including environmental burden, cognitive schemas, and brain network alterations, predict PLEs and partially generalize to clinical psychosis. These variables may reflect intermediate phenotypes across the psychosis spectrum, offering insights into both vulnerability and resilience.",

keywords = "elastic net modeling, machine learning, psychosis first episode, psychotic-like experiences, structural connectome",

author = "Seda Arslan and Merve Ka{\c s}ık{\c c}ı and Osman Dağ and Didenur {\c S}ahin-{\c C}evik and {\c C}akmak, \{I{\c s}ık Batuhan\} and Evangelos Vassos and \{van den Heuvel\}, Martijn and Timothea Toulopoulou",

year = "2025",

doi = "10.1017/S0033291725102201",

language = "English",

volume = "55",

pages = "1--10",

journal = "Psychological Medicine",

issn = "0033-2917",

publisher = "Cambridge University Press",

}

TY - JOUR

T1 - Multimodal prediction of psychotic-like experiences using elastic net modeling

T2 - External validation in a clinical sample

AU - Arslan, Seda

AU - Kaşıkçı, Merve

AU - Dağ, Osman

AU - Şahin-Çevik, Didenur

AU - Çakmak, Işık Batuhan

AU - Vassos, Evangelos

AU - van den Heuvel, Martijn

AU - Toulopoulou, Timothea

PY - 2025

Y1 - 2025

N2 - Background Psychotic-like experiences (PLEs) are considered a subclinical component of psychosis continuum. Studies indicate that PLEs arise from multimodal factors, yet research comprehensively examining these factors together remains scarce. Using a large youth sample, we present the first model that simultaneously examines multimodal factors related to PLEs. As a secondary aim, we evaluate the model's ability to explain psychosis in an external validation cohort that included individuals experiencing psychosis. Methods After applying variable selection including generalized estimating equations, correlation filtering, Least Absolute Shrinkage and Selection Operator model to 741 variables (i.e., environmental factors, cognitive appraisals, clinical variables, cognitive functioning, and structural brain connectome measures), obtained PLEs predictors (N = 27) and covariates (i.e., age, sex, IQ) were included in the classification model based on Elastic Net algorithm for predicting high/low PLEs in 396 healthy participants aged 14-24 (Mage = 19.72 ± 2.5). We externally validated PLE-related predictors in a clinical sample comprising first-episode psychosis patients (n = 19), their siblings (n = 20), and healthy controls (n = 19). ResultsEleven factors, including environmental and cognitive appraisals, along with 16 structural network properties spanning frontal, temporal, occipital, and parietal regions, were identified as important predictors of PLEs. The model's performance was moderate in predicting low versus high PLEs (accuracy = 75%, AUC = 0.750). Specificity was high (84.2%) in distinguishing siblings from patients. Conclusions Multimodal features, including environmental burden, cognitive schemas, and brain network alterations, predict PLEs and partially generalize to clinical psychosis. These variables may reflect intermediate phenotypes across the psychosis spectrum, offering insights into both vulnerability and resilience.

AB - Background Psychotic-like experiences (PLEs) are considered a subclinical component of psychosis continuum. Studies indicate that PLEs arise from multimodal factors, yet research comprehensively examining these factors together remains scarce. Using a large youth sample, we present the first model that simultaneously examines multimodal factors related to PLEs. As a secondary aim, we evaluate the model's ability to explain psychosis in an external validation cohort that included individuals experiencing psychosis. Methods After applying variable selection including generalized estimating equations, correlation filtering, Least Absolute Shrinkage and Selection Operator model to 741 variables (i.e., environmental factors, cognitive appraisals, clinical variables, cognitive functioning, and structural brain connectome measures), obtained PLEs predictors (N = 27) and covariates (i.e., age, sex, IQ) were included in the classification model based on Elastic Net algorithm for predicting high/low PLEs in 396 healthy participants aged 14-24 (Mage = 19.72 ± 2.5). We externally validated PLE-related predictors in a clinical sample comprising first-episode psychosis patients (n = 19), their siblings (n = 20), and healthy controls (n = 19). ResultsEleven factors, including environmental and cognitive appraisals, along with 16 structural network properties spanning frontal, temporal, occipital, and parietal regions, were identified as important predictors of PLEs. The model's performance was moderate in predicting low versus high PLEs (accuracy = 75%, AUC = 0.750). Specificity was high (84.2%) in distinguishing siblings from patients. Conclusions Multimodal features, including environmental burden, cognitive schemas, and brain network alterations, predict PLEs and partially generalize to clinical psychosis. These variables may reflect intermediate phenotypes across the psychosis spectrum, offering insights into both vulnerability and resilience.

KW - elastic net modeling

KW - machine learning

KW - psychosis first episode

KW - psychotic-like experiences

KW - structural connectome

UR - https://www.scopus.com/pages/publications/105021718870

UR - https://www.scopus.com/inward/citedby.url?scp=105021718870&partnerID=8YFLogxK

U2 - 10.1017/S0033291725102201

DO - 10.1017/S0033291725102201

M3 - Article

C2 - 41234002

AN - SCOPUS:105021718870

SN - 0033-2917

VL - 55

SP - 1

EP - 10

JO - Psychological Medicine

JF - Psychological Medicine

M1 - e346

ER -

Multimodal prediction of psychotic-like experiences using elastic net modeling: External validation in a clinical sample

Abstract

Funding

Keywords

Access to Document

Persistent URL (handle)

Other files and links

Fingerprint

Cite this