This study is a single-center, retrospective, cross-sectional study. We plan to work with our network information center to analysis the related indicators of oxidative stress injury in patients with bipolar disorder based on oxidative stress data. During the study, machine learning was used as a data analysis method to screen out the biomarker risk factors with sensitivity and specificity for early recognition of bipolar disorder from major depression disorder with oxidative stress injury as the core. And then build up effective clinical predictive models for early identification of bipolar disorder, which can predict the early quantitative probabilistic of the onset of bipolar disorder.

Inclusion Criteria:

• age is not limited
• gender is not limited
• meets the diagnostic criteria for bipolar disorder of ICD-10 F31,F32 and its sub-categories
• has relevant HIS system data that can be utilized.

Exclusion Criteria:

• patients who did not meet the appeal diagnosis after three-level rounds of ward
• patients who met the above three diagnoses but had severe data loss (missing value ≥ estimated data value of 30%)

Subjects are hospitalized patients in SMHC from 2009 to 2018; meet ICD-10 diagnosis of F31 bipolar disorder, F32 depressive disorder criteria and its subtype (mental examination was conducted by three levels of doctors including at least one attending physician and one chief physician in psychiatric); available relevant HIS system biochemical data; hospitalized patients need to the first admission; age and gender is not limited