Transdiagnostic behavioral and genetic contributors to repetitive negative thinking: A machine learning approach

Katherine L Forthman; Rayus Kuplicki; Hung-Wen Yeh; Sahib S Khalsa; Martin P Paulus; Salvador M Guinjoan

doi:10.1016/j.jpsychires.2023.05.039

Transdiagnostic behavioral and genetic contributors to repetitive negative thinking: A machine learning approach

J Psychiatr Res. 2023 Jun:162:207-213. doi: 10.1016/j.jpsychires.2023.05.039. Epub 2023 May 5.

Authors

Katherine L Forthman¹, Rayus Kuplicki¹, Hung-Wen Yeh², Sahib S Khalsa³, Martin P Paulus³, Salvador M Guinjoan⁴

Affiliations

¹ Laureate Institute for Brain Research, 6655 South Yale Avenue, Tulsa, OK, 74136, USA.
² Laureate Institute for Brain Research, 6655 South Yale Avenue, Tulsa, OK, 74136, USA; Health Services & Outcomes Research, Children's Mercy Research Institute, 2401 Gilham Road, Kansas City, MO, 64108, USA; School of Medicine, University of Missouri-Kansas City, 2411 Holmes St, Kansas City, MO, 64108, USA.
³ Laureate Institute for Brain Research, 6655 South Yale Avenue, Tulsa, OK, 74136, USA; Oxley College of Health Sciences, University of Tulsa, 1215 South Boulder Ave W, Tulsa, OK, 74119, USA.
⁴ Laureate Institute for Brain Research, 6655 South Yale Avenue, Tulsa, OK, 74136, USA; Department of Psychiatry, Oklahoma University Health Sciences Center, The University of Oklahoma-Tulsa, Schusterman Center, 4502 E. 41st Street, Tulsa, OK, 74135, USA. Electronic address: sguinjoan@laureateinstitute.org.

PMID: 37178517
DOI: 10.1016/j.jpsychires.2023.05.039

Abstract

Background: Repetitive negative thinking (RNT) is a symptom that can negatively impact the treatment and course of common psychiatric disorders such as depression and anxiety. We aimed to characterize behavioral and genetic correlates of RNT to infer potential contributors to its genesis and maintenance.

Methods: We applied a machine learning (ML) ensemble method to define the contribution of fear, interoceptive, reward, and cognitive variables to RNT, along with polygenic risk scores (PRS) for neuroticism, obsessive compulsive disorder (OCD), worry, insomnia, and headaches. We used the PRS and 20 principal components of the behavioral and cognitive variables to predict intensity of RNT. We employed the Tulsa-1000 study, a large database of deeply phenotyped individuals recruited between 2015 and 2018.

Results: PRS for neuroticism was the main predictor of RNT intensity (R²=0.027,p<0.001). Behavioral variables indicative of faulty fear learning and processing, as well as aberrant interoceptive aversiveness, were significant contributors to RNT severity. Unexpectedly, we observed no contribution of reward behavior and diverse cognitive function variables.

Limitations: This study is an exploratory approach that must be validated with a second, independent cohort. Furthermore, this is an association study, limiting causal inference.

Conclusions: RNT is highly determined by genetic risk for neuroticism, a behavioral construct that confers risk to a variety of internalizing disorders, and by emotional processing and learning features, including interoceptive aversiveness. These results suggest that targeting emotional and interoceptive processing areas, which involve central autonomic network structures, could be useful in the modulation of RNT intensity.

Trial registration: ClinicalTrials.gov NCT02450240.

Keywords: Machine learning ensemble method; Neuroticism; Polygenic risk score; Principal component analysis; Repetitive negative thinking.

Publication types

Research Support, N.I.H., Extramural

MeSH terms

Anxiety / psychology
Anxiety Disorders / psychology
Humans
Pessimism* / psychology
Surveys and Questionnaires
Thinking

Associated data

ClinicalTrials.gov/NCT02450240