TY - JOUR
T1 - Machine learning-based prediction of anxiety disorders using blood metabolite and social trait data from the UK Biobank
AU - Smith, Annabel
AU - Miller, Jack J.
AU - Anthony, Daniel C.
AU - Radford-Smith, Daniel E.
PY - 2025/7
Y1 - 2025/7
N2 - Anxiety disorders are the most prevalent type of mental health disorders and are characterised by excessive fear and worry. Despite affecting one in four individuals within their lifetime, there remains a gap in our understanding regarding the underlying pathophysiology of anxiety disorders, which limits the development of novel treatment options. Exploring blood-based biomarkers of anxiety disorder offers the potential to predict the risk of clinically significant anxiety in the general population, increase our understanding of anxiety pathophysiology, and to reveal options for preventative treatment. Here, using psychosocial variables in combination with blood and urine biomarkers, reported in the UK Biobank, we sought to predict future anxiety onset. Machine learning accurately predicted (ROC AUC: ∼0.83) ICD-10-coded anxiety diagnoses up to 5 years (mean 3.5 years) after blood sampling, against lifetime anxiety-free controls. Analysis of the blood biochemistry measures indicated that anxious individuals were more anaemic and exhibited higher levels of markers of systemic inflammation than controls. However, blood biomarkers alone were not predictive of resilience or susceptibility to anxiety disorders in a subset of individuals rigorously matched for a wide range of psychosocial covariates (ROC AUC: ∼0.50). Overall, we demonstrate that the integration of biological and psychosocial risk factors is an effective tool to screen for and predict anxiety disorder onset in the general population.
AB - Anxiety disorders are the most prevalent type of mental health disorders and are characterised by excessive fear and worry. Despite affecting one in four individuals within their lifetime, there remains a gap in our understanding regarding the underlying pathophysiology of anxiety disorders, which limits the development of novel treatment options. Exploring blood-based biomarkers of anxiety disorder offers the potential to predict the risk of clinically significant anxiety in the general population, increase our understanding of anxiety pathophysiology, and to reveal options for preventative treatment. Here, using psychosocial variables in combination with blood and urine biomarkers, reported in the UK Biobank, we sought to predict future anxiety onset. Machine learning accurately predicted (ROC AUC: ∼0.83) ICD-10-coded anxiety diagnoses up to 5 years (mean 3.5 years) after blood sampling, against lifetime anxiety-free controls. Analysis of the blood biochemistry measures indicated that anxious individuals were more anaemic and exhibited higher levels of markers of systemic inflammation than controls. However, blood biomarkers alone were not predictive of resilience or susceptibility to anxiety disorders in a subset of individuals rigorously matched for a wide range of psychosocial covariates (ROC AUC: ∼0.50). Overall, we demonstrate that the integration of biological and psychosocial risk factors is an effective tool to screen for and predict anxiety disorder onset in the general population.
KW - Anxiety
KW - Metabolomics
KW - Neuroticism
KW - Random forest
UR - https://www.scopus.com/pages/publications/105004681318
U2 - 10.1016/j.bbih.2025.101010
DO - 10.1016/j.bbih.2025.101010
M3 - Journal article
C2 - 40485662
AN - SCOPUS:105004681318
SN - 2666-3546
VL - 46
JO - Brain, Behavior, and Immunity - Health
JF - Brain, Behavior, and Immunity - Health
M1 - 101010
ER -