Comparing Predictive Models: Traditional Versus Modern Statistical Approaches for Economic Outcomes in Major Depressive Disorder

OBJECTIVES: This study aims to compare conventional and contemporary statistical techniques using negative binomial regression (NB2) and XGBoost to predict healthcare expenses and its drivers in patients diagnosed with Major Depressive Disorder (MDD).

METHODS: A longitudinal retrospective study was conducted using the household component of the Medical Expenditure Panel Survey (MEPS) database (2017-2021). Patients with a primary diagnosis of MDD were included; however, those with a diagnosis of bipolar disorder, anxiety disorder, unspecified mental disorder, and schizophrenia were excluded. The index event was defined as the first diagnosis of MDD. NB2 and XGBoost were used to estimate the healthcare costs for patients within a year of the index event.

RESULTS: A total of 2,596 patients met the inclusion criteria with a mean (SD) age of 52.5 (17.7) years consisting of 69% women. About 65% of patients were in the high- and middle-income category. In the 12-month baseline period, the most prevalent comorbidities were anxiety (47%), hypertension (36%), and sleep disorders (23%). NB2 identified individual income, depressive medications, psychoactive substance disorder, and reaction to severe stress as the major predictors of healthcare cost. Age, gender, anxiety, hypertension, and selective serotonin reuptake inhibitors medications were determined as additional significant covariates using XGBoost. Root mean squared error (RMSE) was computed to compare the performance metrics, and XGBoost yielded a 19% lower RMSE when compared to NB2 indicating a better prediction accuracy. To achieve the best performance of the model, the parameters were optimized using hyperparameter tuning.

CONCLUSIONS: The XGBoost model outperformed the traditional regression-based model in terms of predictive accuracy. Both the models observed similar predictors, but XGBoost proved to be more effective in highlighting additional key drivers of costs.