Factors Influencing Adherence to All-Oral Short-Course Treatment for DR-TB and Establishment of a Predictive Model

To explore factors influencing adherence to all-oral short-course treatment for drug-resistant tuberculosis (DR-TB) and establish a predictive model.

The study is retrospective and single-center. 241 TB patients treated at our hospital from January 2022 to December 2024 were retrospectively selected. Using a random number table method in a 7:3 ratio, patients were divided into a modeling group (n=169) and a validation group (n=72). The modeling group patients were categorized into groups of good adherence (n=89) and poor adherence (n=80). Univariate and binary Logistics regression analyses were used to identify influencing factors. A predictive model was constructed using SPSS, and R language was utilized to assess the model's application value through receiver operating characteristic (ROC) curves, calibration curves, and decision curve analysis (DCA).

Binary logistic regression analysis showed that family monthly income, level of education, history of previous confirmed TB, and family-supervised drug administration were influencing factors for adherence to short-course treatment for DR-TB (P<0.05). The combined predictive model expression was Logit(P) = -1.394 + (0.421*family monthly income) + (0.344*level of education) + (0.310*history of previous confirmed TB) + (0.452*family-supervised drug administration). The model demonstrated good consistency between predicted risks and actual risks, with calibration curves showing a slope close to 1 in both the modeling group and validation group. ROC analysis indicated an area under the curve of 0.88 in the modeling group. For the validation group, the AUC was 0.85. The DCA curve illustrated a clear net benefit, indicating good clinical utility of the model.

Family monthly income, level of education, history of previous confirmed TB, and family-supervised drug administration are factors influencing adherence to short-course treatment for DR-TB. The model established based on these factors holds significant value in predictive applications.