Construction Of A Diagnostic Model For Psychiatric Diseases In Basis Of Clinical Blood Test Data

As estimated by the World Health Organization,about a quarter of worldwide population may suffer from a mental or neurological disease in their entire lives.The latest China’s mental health survey(CMHS)illustrated that the weighted annual prevalence of six major mental disorders was 9.3%,excluding dementia.Early diagnosis and early intervention can prevent or delay the onset of first-episode psychosis,it can also delay the functional decline of psychosis.Current clinical practices usually adopt the ICD-10 or DSM5 as the diagnostic criteria for psychosis in basis of the patient’s symptoms,blood test,brain image,and personal experience particularly,which is subjective,time-costing,and sometimes imprecise.Therefore in this study,we aim to utilize extensive clinical blood test data to construct a robust model for diagnosis of mental disorders using the machine-learning methods.The analyses were carried out on 114 distinct blood test data of 65,394 clinical samples derived from the XianYue hospital and the Xiamen No.1 hospital,as well as the clinical records.We analyzed the data distribution and the impact on the model performance.We also adopted three machine learning algorithms for model construction,they were support vector machine(SVM),logistic regression(LR),and random forest(RF).The model performances were evaluated and compared.At last,we also extracted the key features for diagnosis of mental disorders using the information gain theory.After evaluation and comparison,the SVM model was selected for diagnosis of mental disorders,which achieved an AUC of 0.868 and 0.877 for the testing dataset and the first-episode psychosis dataset,respectively.This indicated that the model was highly sensitive and robust in risk assessment of early onset psychosis.Furthermore,we determined that the contributions of 33 blood testing features were weighted between 0.1 and 0.25.Although each of these features had small contributions to the model alone,they together guaranteed the good performance of the diagnostic model.This finding partially supported the mental disorders were multi-system diseases.In summary,this study provides an objective and convenient reference method for clinical diagnosing for mental disorders.It enables early diagnosis of mental disorders in a basis of clinical laboratory evidences,and thus allows early intervention.