China is one of the countries with a high incidence of pulmonary tuberculosis globally, ranking third in the world¹⁴. Given the high clinical similarity between NTM-PD and PTB, including but not limited to symptoms such as coughing, sputum production, intermittent fever, night sweats, and hemoptysis, the differential diagnosis of these diseases is significantly challenging^15,16. With numerous primary healthcare institutions and varying levels of medical resource allocation and technical capabilities, many grassroots medical facilities lack the capacity to efficiently conduct NTM cultures or polymerase chain reaction (PCR) tests¹⁷. These factors collectively increase the difficulty of diagnosing NTM-PD, leading to potential misdiagnosis of NTM-PD patients as PTB cases during the initial stages of medical consultation.

This study shows significant differences between the NTM-PD and PTB groups in factors such as age, BMI, bronchiectasis, and lung cavitation. The average age of NTM-PD patients was significantly higher than that of PTB patients, consistent with multiple studies^18,19, indicating that as people age, their immune system function gradually weakens, making the elderly more susceptible to NTM infections. Elderly patients often have other chronic diseases or are in an immunosuppressed state, which also increases the risk of NTM disease. Therefore, there should be increased attention and early screening for NTM disease in the elderly population. The higher fever rate (62%) observed in our NTM-PD cohort may reflect secondary bacterial infections or advanced disease stages in this hospitalized population, as previous studies have reported fever to be less common in mild NTM-PD cases²⁰. This discrepancy highlights the clinical heterogeneity of NTM-PD presentations. NTM-PD patients had a lower BMI, possibly reflecting malnutrition or weight loss due to chronic illness. The association between low BMI and NTM-PD has been confirmed in research²¹. Studies have indicated that lower BMI is an independent risk factor for distinguishing PTB from NTM-PD²². Low BMI may result from prolonged chronic inflammation consuming large amounts of energy and protein, thereby affecting overall health and immunity. The presence of bronchiectasis and lung cavitation was significantly higher in NTM-PD patients compared to PTB patients. Bronchiectasis provides a suitable environment for NTM growth, while lung cavitation can lead to compromised local immune defenses, both increasing the likelihood of NTM infection. This conclusion aligns with reports by several scholars^23,24,25,26, emphasizing the importance of closer monitoring and management for patients with structural lung diseases, especially those known to have bronchiectasis or lung cavitation. Notably, CRP and SAA indicators did not show significant differences between the two groups in this study but were much higher than normal values, indicating significant inflammatory responses in both groups without specific differentiation between the two diseases. This finding is inconsistent with Cowman SA et al.‘s study²⁷ but largely consistent with reports by Park HJ et al.²⁸ and Zhang W et al.¹³, suggesting variability in research findings on inflammatory markers. Future work will involve expanding the sample size to further validate these results.

The nomogram model constructed in this study demonstrated excellent predictive performance, with an AUC of 0.861, indicating high accuracy in distinguishing between NTM-PD and PTB. The performance in calibration curves and decision curve analysis also confirmed its potential value in clinical practice. This robust performance was achieved through LASSO regression, which was specifically chosen for three key advantages: (1) handling multicollinearity among clinical predictors (e.g., interrelated inflammation markers), (2) automatic selection of discriminative features via L1 regularization while eliminating noise variables, and (3) prevention of overfitting-a critical consideration given our moderate sample size. These properties collectively enhanced model generalizability, as evidenced by the consistent AUC (0.861) and calibration slopes across validation. This suggests that clinicians can use the predictive information provided by the model to more precisely tailor treatment strategies at corresponding probability thresholds, thereby improving patient outcomes. By integrating multiple relevant factors, the model provides an intuitive and quantitative risk assessment tool. Clinically, this model can be used to conduct personalized risk assessments for patients, enabling earlier and more precise treatment planning. Additionally, the model aids in deepening our understanding of the pathogenesis of NTM-PD and PTB, providing new avenues for future research. This nomogram could be directly translated into clinical practice as a screening tool to identify high-risk NTM-PD patients for targeted diagnostic testing. With further validation, it may form the basis for point-of-care prediction tools and guide personalized therapeutic decisions when combined with molecular diagnostics. Future multicenter studies should focus on implementing and validating its utility in real-world clinical settings.

This study also has certain limitations. First, due to the data being sourced from a single center with a relatively limited sample size, this may restrict the generalizability of some conclusions. Second, while cavitation was identified as a significant discriminator, detailed radiological characteristics (e.g., cavity size, wall thickness) were not analyzed due to inconsistent documentation in retrospective imaging reports. Third, symptom duration data were not systematically collected, precluding analysis of this potentially discriminative clinical feature. During the modeling process, we endeavored to control for confounding factors, but it is still possible that unmeasured variables could influence the results. In future studies, we will aim to expand the sample size, implement standardized imaging protocols and symptom tracking, and incorporate additional relevant factors to enhance the precision and applicability of the model.