Abstract
This study investigates the factors influencing healthcare utilization among diabetic patients in Gansu Province, Northwest China. Utilizing a hybrid statistical approach that combines Random Forest machine learning with traditional Logistic Regression, the research identifies significant disparities in access. While outpatient services show a "pro-poor" distribution, inpatient services remain "pro-rich," skewed toward patients with higher incomes and better insurance coverage. Key predictors include medication frequency, comorbidities, and the presence of a household doctor. The findings underscore the need for targeted policy improvements to meet the "Health China 2030" goals for chronic disease management.
Introduction
China currently manages the world’s largest diabetic population, yet healthcare access remains uneven, particularly in economically underdeveloped regions like Gansu Province. While broad population studies exist, there is a lack of granular data on how patients with specific chronic conditions navigate the healthcare system. This study seeks to understand the social, economic, and medical factors that predict the use of outpatient clinics versus inpatient hospital stays, a critical step toward achieving equitable chronic disease management.
Methodology
The researchers employed a multi-stage stratified sampling method to survey 282 diabetic patients from the 2018 Gansu Provincial Health Services Household Survey. The analytical framework is built on Anderson’s Behavioral Model, which categorizes health-seeking drivers into predisposing, enabling, and need factors.
To enhance the robustness of the analysis, the authors utilized a Random Forest (RF) algorithm to rank variable importance using the Gini index reduction method. This was followed by Binary Logistic Regression to calculate Odds Ratios for the top-ranked factors. Additionally, Wagstaff’s Concentration Index (CI) was calculated to quantify socioeconomic inequality in service use.
Key Insights and Results
The study revealed a significant "Equity Paradox." Outpatient services demonstrated a pro-poor concentration (CI = -0.241), whereas inpatient services showed a pro-rich concentration (CI = 0.107). This suggests that while lower-income patients utilize local clinics, they face significant barriers to more intensive hospital care.
The Random Forest analysis (Figure 6) highlighted several critical predictors:
- Medication as a Behavioral Proxy: The frequency of taking diabetic medication was the most important predictor for outpatient care, suggesting that treatment engagement correlates with better system navigation.
- The Comorbidity Multiplier: Patients suffering from three or more chronic conditions were 2.538 times more likely to require inpatient services than those with diabetes alone.
- The "Household Doctor" Effect: Having a contracted household doctor was a top-three predictor for outpatient use, validating policies aimed at strengthening primary care through family physician contracts.
Utilization rates were high for outpatient services (92.91%) but significantly lower for inpatient care (24.82%). Furthermore, an urban-rural divide was evident, with urban patients spending an average of 5.03 days in the hospital compared to 2.51 days for rural patients. Both the Random Forest and Logistic Regression models showed good predictive power, with Area Under the Curve (AUC) values of 0.706 and 0.748, respectively.
Discussion and Future Directions
While the study provides valuable insights, it is limited by a relatively small sample size and its cross-sectional design, which identifies correlations rather than causation. Future research should focus on longitudinal studies and the unique barriers faced by ethnic minority populations, who exhibited lower healthcare utilization rates. Improving insurance coverage and promoting the rational use of health services are essential steps toward achieving the chronic disease prevention goals of "Health China 2030."
Summary updated