A nutritional–inflammatory composite score to support nursing risk stratification in suspected multidrug-resistant tuberculosis patients: implications for identifying nontuberculous mycobacterial pulmonary disease

OBJECTIVE: Nontuberculous mycobacterial pulmonary disease (NTM-PD) and multidrug-resistant pulmonary tuberculosis (MDR-PTB) share overlapping clinical and radiological features, leading to frequent misdiagnosis and inappropriate treatment. Simple, low-cost risk stratification tools are needed, particularly for nursing staff in resource-limited settings. This study developed and evaluated a nutritional-inflammatory composite score (NICS) for differentiating NTM-PD from MDR-PTB. METHODS: We retrospectively enrolled 500 patients with suspected MDR-PTB who underwent confirmatory microbiological testing. Final diagnoses were MDR-PTB (n = 338, 67.6%) and NTM-PD (n = 162, 32.4%). Three logistic regression models were compared: clinical (age, BMI, smoking, diabetes, COPD, bronchiectasis), laboratory (CRP, albumin, ferritin, lymphocytes), and NICS-based (age, BMI, diabetes, COPD, bronchiectasis, and NICS). Model discrimination was assessed by area under the ROC curve (AUC). RESULTS: NICS was the strongest independent predictor of NTM-PD (OR = 1.14 per unit increase, 95% CI: 1.08-1.20, P < 0.001). The NICS-based model achieved the highest AUC (0.66, 95% CI: 0.61-0.71) compared with the clinical model (AUC = 0.57, 95% CI: 0.52-0.62) and the laboratory model (AUC = 0.60, 95% CI: 0.54-0.65). Using the Youden-derived optimal cut-off, sensitivity was 35.19%, specificity 73.96%, PPV 39.31%, and NPV 70.42% (overall correct classification 61.40%). CONCLUSIONS: NICS improves NTM-PD versus MDR-PTB discrimination modestly but significantly (ΔAUC = 0.09 vs. clinical model). The nomogram represents a preliminary research instrument for risk estimation; however, the low sensitivity at the optimal threshold precludes its use as a diagnostic tool, limiting its role to hypothesis generation pending prospective validation. Prospective usability testing and external validation remain necessary.