Background: The pathophysiology of hikikomori has not been clarified, and biological traits that objectively characterize it remain unexplored.
Methods: Drug-free patients with hikikomori conditions (n=42) and healthy controls (n=41) were recruited. Psychological assessments for the severity of hikikomori (HQ-25) and depression (PHQ-9 and HAMD-17) were conducted. Blood biochemical tests and plasma metabolome analysis were performed. Based on the integrated information, machine-learning models were created to discriminate cases of hikikomori from healthy controls, predict hikikomori severity, stratify the cases, and identify metabolic signatures that contribute to each model.
Results: Long-chain acylcarnitine levels were remarkably higher in patients with hikikomori; bilirubin, arginine, ornithine, and serum arginase were significantly different in male patients with hikikomori. The discriminative random forest model was highly performant, exhibiting an area under the ROC curve of 0.854. To predict hikikomori severity, a partial least squares PLS-regression model was successfully created with high linearity and practical accuracy. Additionally, blood serum uric acid and plasma cholesterol esters contributed to the stratification of cases.
Conclusions: These findings reveal the blood metabolic signatures of hikikomori, which are key to elucidating the pathophysiology of hikikomori. In addition, we believe that these data will shed new light on considering the biological meanings of social-distance in the COVID-19 era.