Abstract
Background: Managing chronic illness effectively depends not only on treatment availability but also on patients’ ability to adhere to prescribed medications.
Objectives: This study examined the factors influencing medication adherence among Jordanian adults with long-term conditions, using both traditional regression and machine learning methods.
Method: In this cross-sectional study, patients with chronic conditions completed an online survey that assessed demographic, clinical, and behavioral variables, including health literacy (HLQ-12) and adherence (MARS-5). Quantile regression and machine learning models were applied.
Results: A total of 981 patients (63.1% females) were enrolled in the study. Quantile regression showed that higher health literacy, a diagnosis of diabetes or cardiovascular disease, and fewer prescribed medications were positively associated with adherence. In contrast, being married or having public, military, or no insurance was linked to lower adherence scores. The Random Forest model achieved the highest predictive accuracy (R² = 0.38), and SHAP analysis identified health literacy, disease duration, and age as the most influential features.
Conclusion: These findings highlight the need for targeted interventions that address both individual understanding and structural challenges such as insurance type and treatment complexity. Improving health literacy, simplifying medication regimens, and ensuring equitable healthcare access may help support better adherence in this population. The use of explainable machine learning, alongside conventional statistical approaches, offers new opportunities to improve the understanding and prediction of adherence behaviors in resource-constrained settings.
Objectives: This study examined the factors influencing medication adherence among Jordanian adults with long-term conditions, using both traditional regression and machine learning methods.
Method: In this cross-sectional study, patients with chronic conditions completed an online survey that assessed demographic, clinical, and behavioral variables, including health literacy (HLQ-12) and adherence (MARS-5). Quantile regression and machine learning models were applied.
Results: A total of 981 patients (63.1% females) were enrolled in the study. Quantile regression showed that higher health literacy, a diagnosis of diabetes or cardiovascular disease, and fewer prescribed medications were positively associated with adherence. In contrast, being married or having public, military, or no insurance was linked to lower adherence scores. The Random Forest model achieved the highest predictive accuracy (R² = 0.38), and SHAP analysis identified health literacy, disease duration, and age as the most influential features.
Conclusion: These findings highlight the need for targeted interventions that address both individual understanding and structural challenges such as insurance type and treatment complexity. Improving health literacy, simplifying medication regimens, and ensuring equitable healthcare access may help support better adherence in this population. The use of explainable machine learning, alongside conventional statistical approaches, offers new opportunities to improve the understanding and prediction of adherence behaviors in resource-constrained settings.
| Original language | English |
|---|---|
| Article number | 2548979 |
| Number of pages | 11 |
| Journal | Annals of Medicine |
| Volume | 57 |
| Issue number | 1 |
| DOIs | |
| Publication status | Published - 20 Aug 2025 |