Exploring factors influencing risky behaviors in patients with bipolar disorders using logistic regression model

Abstract Purpose To build a predictive model for predicting risky behaviors of the bipolar disorders (BD) patients in the Guangdong during years 2010–2020; and to explore the specific influencing factors of risky behaviors. Methods Data of BD patients was acquired from the Guangdong Mental Health Center Network Medical System (GDMHS). The data was randomly divided into a primary cohort and a validation cohort. Social-demographic and clinic factors in the primary cohort were screened using univariate logistic regression and multivariate logistic regression model to identify factors associated with risky behaviors. And then, the statistically significant factors were utilized to develop a predictive model. The data of the validation cohort was utilized to the above predictive model for verifying the degree of fit of this model. Results The prevalence of risky behaviors of BD was 2%. Multivariate logistic regression revealed that related factors for risky behaviors of BD mainly contained poverty, younger age, terrible sleeping status, worse treatment effect, lacking of insight of BD, poor medication compliance, moodiness and history of adverse drug reaction. Conclusion The model, based on clinical and socio-demographic factors, provides some usability as a prediction scheme. In addition, the model can also be combined with other diagnostic evidences, such as the brain imaging, genetics and the blood-based biomarker to develop more complete and multi-omics models in future.