Development and evaluation of a predictive biomarker model based on the niacin flush pathway for differentiating...

Our previous study demonstrated that the niacin flush response and pathway biomarkers could differentiate schizophrenia patients from healthy controls. Whether these biomarkers also distinguish schizophrenia from bipolar disorder remains unclear. This study aimed to develop and validate predictive models to differentiate schizophrenia, bipolar disorder, and healthy controls. Independent training and validation sets were established for schizophrenia and healthy controls, and model performance was further evaluated in distinguishing schizophrenia from bipolar disorder. We recruited 223 schizophrenia patients, 62 bipolar disorder patients, and 162 healthy controls in Taiwan. Logistic regression was used to construct prediction models. The prediction model demonstrated high discriminative performance in differentiating individuals with schizophrenia from healthy controls, achieving an AUC of 0.97 in the training set and 0.92 in the validation set. Furthermore, the model effectively distinguished schizophrenia from bipolar disorder, with an AUC of 0.96, and 0.91 for cross-validated results. Niacin flush response and GPR109A were the most significant for differentiating schizophrenia patients from healthy controls, while EP2, PGE2, and GPR109A were key for schizophrenia and bipolar disorder classification. Principal component and partial least squares-discriminant analyses confirmed separation among different groups. Shapley Additive Explanation value analysis demonstrated divergent feature importance of biomarkers, indicating differential regulation of the niacin flush pathway in schizophrenia and bipolar patients. We externally validated that biomarkers from the niacin flush pathway were good candidates for distinguishing schizophrenia patients from healthy controls, while EP2, PGE2, and GPR109A may further aid in differentiating schizophrenia from bipolar disorder, potentially reflecting their distinct etiological underpinnings.