Symptomatology differences of major depression in psychiatric versus general hospitals: A machine learning approach

Symptomatology differences of major depressive disorder (MDD) in psychiatric and general hospitals in China leads to possible misdiagnosis. Looking at the symptomatology of first-visit patients with MDD in different mental health services, and identifying predictors of health-seeking behavior using machine learning may help to improve diagnostic accuracy.

1500 patients first diagnosed with MDD were recruited from 16 psychiatric hospitals and 16 general hospitals across China. Socio-demographic characteristics, causal attribution, symptoms of depression within and outside Diagnostic and Statistical Manual of Mental Disorders (DSM) framework were collected using a self-made questionnaire. A predictive model of 62 variables was established using Random forest, symptom frequencies of patients in general hospitals and psychiatric hospitals were compared.

The machine learning approach revealed that symptoms were strong predictors of health-seeking behavior among patients with MDD. General hospitals patients had higher frequencies of suicidal ideation (χ²=15.230, p<0.001), psychosis (χ²=14.264, p<0.001), weight change (all p<0.001), hypersomnia (χ²=25.940, p<0.001), and a tendency of denying emotional/cognitive symptoms compared with psychiatric hospitals patients.

Stigma and preference bias were not measured. Severity of current depressive episodes was not assessed. Data of previous episode(s) was not presented.

Symptom evaluation targeting specific patient population in different hospitals is crucial for diagnostic accuracy. Suicide prevention reliant on collaboration between general hospitals and psychiatric hospitals is required in the future construction of Chinese mental health system.