Identifying major depressive disorder among US adults living alone using stacked ensemble machine learning algorithms.

Background: It has been increasingly recognized that adults living alone have a higher likelihood of developing Major Depressive Disorder (MDD) than those living with others. However, there is still no prediction model for MDD specifically designed for adults who live alone. Objective: This study aims to investigate the effectiveness of utilizing personal health data in combination with a stacked ensemble machine learning (SEML) technique to detect MDD among adults living alone, seeking to gain insights into the interaction between personal health data and MDD. Methods: Our data originated from the US National Health and Nutrition Examination Survey (NHANES) spanning 2007 to 2018. We finally selected a set of 30 easily accessible variables encompassing demographic profiles, lifestyle factors, and baseline health conditions. We constructed a SEML model for MDD detection, incorporating three conventional machine learning algorithms as base models and a Neural Network (NN) as the meta-model. Furthermore, SHapley Additive exPlanations (SHAP) analysis was used to explain the impact of each predictor on MDD. Results: The study included 2,642 adult participants who lived alone, of whom 10.6% (279 out of 2,642) had a PHQ-9 score of 10 or above, indicating the presence of MDD. The performance of our SEML model was robust, with an area under the curve (AUC) of 0.85. Further analysis using SHAP revealed positive correlations between the occurrence of MDD and factors such as sleep disorders, number of prescription medications, need for specific walking aids, leak urine during nonphysical activities, chronic bronchitis, and Healthy Eating Index (HEI) scores for sodium. Conversely, age, the Family Monthly Poverty Level Index (FMMPI), and HEI scores for added sugar showed negative correlations with MDD occurrence. Additionally, a U-shaped relationship was noted between the occurrence of MDD and both sleep duration and Body Mass Index (BMI), as well as HEI scores for dairy. Conclusion: The study has successfully developed a predictive model for MDD, specifically tailored for adults living alone using a stacked ensemble technique, enhancing the identification of MDD and its risk factors among adults living alone.