BINARY CLASSIFICATIONANALYSIS FOR PREDICTING MEDICATION ADHERENCE AMONG US ADULTS CURRENTLY TAKING PRESCRIBED MEDICATION FOR DEPRESSION
Author(s)
Kyla Finlayson, MS1, Vicky W. Li, MPH2, Jason N. Kennedy, MS3.
1Biostatistician, Real-World Evidence, Oracle Life Science, Providence, RI, USA, 2Oracle Life Sciences, Austin, TX, USA, 3Oracle Life Sciences, Pittsburgh, PA, USA.
1Biostatistician, Real-World Evidence, Oracle Life Science, Providence, RI, USA, 2Oracle Life Sciences, Austin, TX, USA, 3Oracle Life Sciences, Pittsburgh, PA, USA.
OBJECTIVES: This study examined patient characteristics associated with perfect medication adherence among US adults treated for depression.
METHODS: Data on 7,285 adults who were diagnosed and treated with a prescription medication for depression were identified and analyzed from the 2025 US National Health and Wellness Survey, an online survey of the general US adult population. Medication adherence was measured using self-reported scores of the Adherence to Refills and Medications Scale (ARMS), where a score of 12 indicates perfect medication adherence and scores greater than 12 indicate varying levels of non-adherence. XGBoost binary classification with 10-fold cross-validation was used to predict medication non-adherence and to extract feature importance. Variable selection and feature importance were based on SHAP values. Twenty-four variables including demographic characteristics, depression symptoms and severity, participation in talk therapy, smoking status, alcohol use, body mass index (BMI), exercise habits, and factors related to prescription access such as financial barriers and average monthly medication costs were used in the final model. Parameters were tuned using area under the receiver operating characteristic curve (AUC). Performance metrics included AUC, accuracy, sensitivity, specificity, and Brier score.
RESULTS: Among adults currently taking prescribed medication for depression, 2,700 (37.1%) scored as perfectly adherent while 4,585 (62.9%) were at least somewhat nonadherent. The XGBoost binary classification model had an AUC of 0.73, accuracy of 65.6%, sensitivity of 58.6%, specificity of 76.7%, and Brier score of 0.22. Key predictors features associated with non-adherence included younger age, greater depression severity, cost-related medication barriers, being a drinker and smoker, more recent diagnosis, being male, and higher Charlson Comorbidity Index score.
CONCLUSIONS: In a broadly representative US adult population, XGBoost binary classification identified a selection of factors associated with medication adherence among adults diagnosed with depression.
METHODS: Data on 7,285 adults who were diagnosed and treated with a prescription medication for depression were identified and analyzed from the 2025 US National Health and Wellness Survey, an online survey of the general US adult population. Medication adherence was measured using self-reported scores of the Adherence to Refills and Medications Scale (ARMS), where a score of 12 indicates perfect medication adherence and scores greater than 12 indicate varying levels of non-adherence. XGBoost binary classification with 10-fold cross-validation was used to predict medication non-adherence and to extract feature importance. Variable selection and feature importance were based on SHAP values. Twenty-four variables including demographic characteristics, depression symptoms and severity, participation in talk therapy, smoking status, alcohol use, body mass index (BMI), exercise habits, and factors related to prescription access such as financial barriers and average monthly medication costs were used in the final model. Parameters were tuned using area under the receiver operating characteristic curve (AUC). Performance metrics included AUC, accuracy, sensitivity, specificity, and Brier score.
RESULTS: Among adults currently taking prescribed medication for depression, 2,700 (37.1%) scored as perfectly adherent while 4,585 (62.9%) were at least somewhat nonadherent. The XGBoost binary classification model had an AUC of 0.73, accuracy of 65.6%, sensitivity of 58.6%, specificity of 76.7%, and Brier score of 0.22. Key predictors features associated with non-adherence included younger age, greater depression severity, cost-related medication barriers, being a drinker and smoker, more recent diagnosis, being male, and higher Charlson Comorbidity Index score.
CONCLUSIONS: In a broadly representative US adult population, XGBoost binary classification identified a selection of factors associated with medication adherence among adults diagnosed with depression.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
MSR158
Topic
Methodological & Statistical Research
Topic Subcategory
Artificial Intelligence, Machine Learning, Predictive Analytics