Scholarlink

Objective <p>Drug-resistant pulmonary tuberculosis (DR-TB) remains a major public health challenge with suboptimal treatment outcomes. This study aimed to identify pre-treatment predictors of unfavorable outcomes in DR-TB patients and develop an interpretable predictive model using routinely available admission variables.</p> Methods <p>We retrospectively enrolled 184 DR-TB patients from the Second People’s Hospital of Fuyang City between January 2021 and December 2023. Patients were categorized into favorable outcome group (cured or treatment completed, <i>n</i> = 143) and unfavorable outcome group (treatment failure, death, loss to follow-up, and not evaluated, <i>n</i> = 41). LASSO regression was employed for variable selection from 30 candidate variables, followed by multivariate logistic regression to identify independent predictors. A nomogram was constructed to facilitate individualized risk estimation. SHAP (SHapley Additive exPlanations) analysis was performed to interpret model predictions at global and individual levels. Model performance was assessed using ROC curve analysis, calibration curves, decision curve analysis, and bootstrap resampling. Restricted cubic spline (RCS) regression was used to examine linearity assumptions for continuous predictors.</p> Results <p>LASSO regression selected three admission variables: age, BMI, and albumin. Multivariate logistic regression confirmed these as independent predictors: age (OR = 1.047, 95% CI: 1.018–1.077, <i>P</i> = 0.001), BMI (OR = 0.728, 95% CI: 0.599–0.885, <i>P</i> = 0.001), and albumin (OR = 0.894, 95% CI: 0.826–0.967, <i>P</i> = 0.005). A nomogram was developed based on these three predictors for bedside risk estimation. The model demonstrated good discrimination with AUC of 0.833 (95% CI: 0.758–0.908) and bias-corrected C-index of 0.819 after bootstrap validation. Calibration was satisfactory (Hosmer-Lemeshow <i>P</i> = 0.564), and decision curve analysis confirmed clinical utility across threshold probabilities of 0.1–0.8. SHAP analysis identified age as the most influential predictor, with higher BMI and albumin contributing protectively. RCS analysis confirmed linear associations for all three predictors (P for nonlinear &gt; 0.05 for all), supporting model simplicity.</p> Conclusion <p>This simple, pre-treatment nomogram, using three routinely available variables, enables early identification of high-risk patients at the time of diagnosis and may support targeted interventions to improve treatment outcomes in drug-resistant pulmonary tuberculosis.</p> Clinical trial <p>Not applicable.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Development and validation of a LASSO-derived nomogram for predicting unfavorable treatment outcomes in drug-resistant pulmonary tuberculosis

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月3日

查看全文
  • Wei Zhang,
  • Tuantuan Li,
  • Haiqing Liu,
  • Manman Liu,
  • Fangfang Song,
  • Lili Zhou

摘要

Objective

Drug-resistant pulmonary tuberculosis (DR-TB) remains a major public health challenge with suboptimal treatment outcomes. This study aimed to identify pre-treatment predictors of unfavorable outcomes in DR-TB patients and develop an interpretable predictive model using routinely available admission variables.

Methods

We retrospectively enrolled 184 DR-TB patients from the Second People’s Hospital of Fuyang City between January 2021 and December 2023. Patients were categorized into favorable outcome group (cured or treatment completed, n = 143) and unfavorable outcome group (treatment failure, death, loss to follow-up, and not evaluated, n = 41). LASSO regression was employed for variable selection from 30 candidate variables, followed by multivariate logistic regression to identify independent predictors. A nomogram was constructed to facilitate individualized risk estimation. SHAP (SHapley Additive exPlanations) analysis was performed to interpret model predictions at global and individual levels. Model performance was assessed using ROC curve analysis, calibration curves, decision curve analysis, and bootstrap resampling. Restricted cubic spline (RCS) regression was used to examine linearity assumptions for continuous predictors.

Results

LASSO regression selected three admission variables: age, BMI, and albumin. Multivariate logistic regression confirmed these as independent predictors: age (OR = 1.047, 95% CI: 1.018–1.077, P = 0.001), BMI (OR = 0.728, 95% CI: 0.599–0.885, P = 0.001), and albumin (OR = 0.894, 95% CI: 0.826–0.967, P = 0.005). A nomogram was developed based on these three predictors for bedside risk estimation. The model demonstrated good discrimination with AUC of 0.833 (95% CI: 0.758–0.908) and bias-corrected C-index of 0.819 after bootstrap validation. Calibration was satisfactory (Hosmer-Lemeshow P = 0.564), and decision curve analysis confirmed clinical utility across threshold probabilities of 0.1–0.8. SHAP analysis identified age as the most influential predictor, with higher BMI and albumin contributing protectively. RCS analysis confirmed linear associations for all three predictors (P for nonlinear > 0.05 for all), supporting model simplicity.

Conclusion

This simple, pre-treatment nomogram, using three routinely available variables, enables early identification of high-risk patients at the time of diagnosis and may support targeted interventions to improve treatment outcomes in drug-resistant pulmonary tuberculosis.

Clinical trial

Not applicable.