Association between nutritional-inflammation status and treatment outcomes in elderly patients with rifampicin-susceptible pulmonary tuberculosis: a single-center retrospective cohort study

In this study, we aimed to identify independent nutritional–inflammatory risk factors for adverse treatment outcomes, and develop a multivariate prediction model. A single-center retrospective cohort analysis was conducted on 526 elderly inpatients aged 60–90 years with rifampicin-susceptible PTB admitted between July 2021 and December 2024. Baseline demographic and nutritional data (Nutritional Risk Screening 2002 [NRS-2002]), as well as routine peripheral inflammatory indicator data, were collected. Therapeutic outcomes were dichotomized into treatment success (cured/completed treatment, n = 472) and adverse outcomes (treatment failure/all-cause death, n = 54). Univariate and multivariate binary logistic regression were performed to determine independent predictors of adverse treatment outcomes. Furthermore, receiver operating characteristic (ROC) curves and DeLong pairwise tests facilitated the comparison of the predictive performance of the combined model. In addition, double internal validation was performed using bootstrap self-calibration and stratified 10-fold cross-validation. Multivariate logistic regression confirmed that NRS-2002 nutritional risk (odds ratio [ OR ] = 0.147, 95% confidence interval [ CI ]: 0.043–0.500, P = 0.002) and elevated CRP ( OR = 1.015, 95% CI : 1.009–1.021, P < 0.001) were independent predictors of adverse outcomes. ROC analysis showed areas under the curve (AUCs) of 0.695 for NRS-2002, 0.785 for CRP, and 0.820 for the combined model (sensitivity = 83.33%, specificity = 72.88%, negative predictive value = 97.45%). The DeLong test indicated that the combined model was significantly superior to NRS-2002 alone ( P < 0.001). Bootstrap internal validation yielded an original full-sample AUC of 0.8307, optimism of 0.0062, a corrected AUC of 0.8245, a Hosmer–Lemeshow χ² statistic of 8.6177 ( P = 0.3756), and a Brier score of 0.082. Stratified 10-fold cross-validation yielded a mean AUC of 0.8030 (standard deviation = 0.0793), signifying minimal overfitting and stable model performance across data subsets. Concurrent high nutritional risk (NRS-2002 ≥ 3) and elevated CRP (> 35.8 mg/L) acted as independent predictors of poor anti-TB outcomes in elderly patients with rifampicin-susceptible PTB. The combined nutritional–inflammatory prediction model demonstrated favorable discrimination, calibration, and internal stability after dual validation.