A retrospective study was conducted including 547 adult patients diagnosed with lymph node fistula TBTB between January 2017 and December 2023. Diagnoses were confirmed by chest CT, bronchoscopy, and combined microbiological or pathological examinations (Figure 1). Inclusion criteria were: (1) adulthood (≥18 years); and (2) a verified diagnosis of bronchial tuberculosis with lymphatic fistula subtype determination based on the WS288-2017 Diagnostic Criteria for Pulmonary Tuberculosis (10) and the Guidelines for the Diagnosis and Treatment of Tracheobronchial Tuberculosis (Trial) (11); and (3) availability of complete demographic, clinical, radiological, and laboratory records. Exclusion criteria included: (1) coexisting non-tuberculous respiratory infections or thoracic malignancies; (2) missing or poor-quality contrast-enhanced chest CT scans; and (3) incomplete medical records precluding data extraction.

Demographic information (age, sex, and occupation), presenting symptoms (cough, sputum production, fever, and hemoptysis), and sputum test results (smear and culture) were extracted from the hospital’s electronic medical record system. Contrast-enhanced chest CT scans were independently reviewed by two experienced thoracic radiologists, blinded to clinical data. The reviewers recorded the presence of cavity formation, the lobe or segment affected by atelectasis, bronchial involvement site, enhancement pattern of enlarged mediastinal or hilar lymph nodes, lymph node calcification, and the visibility of any fistulous opening. Discrepancies between reviewers were resolved through consensus discussion with a senior thoracic radiologist. Atelectasis was defined radiologically as partial or complete collapse of a lung lobe or segment with associated loss of normal aeration and expansion.

Continuous variables that satisfied normality are reported as mean ± SD and contrasted through independent t-testing. Variables lacking normal distribution are given as median with interquartile range (Q₁–Q₃) and analyzed using the Kruskal–Wallis test. Categorical outcomes are summarized as frequency and proportion, with between-group differences evaluated by the Chi-square test or, when expected frequencies were <10, Fisher’s exact test.

Participants were randomly split in a 7: 3 ratio into development and validation cohorts (Figure 1). Univariate logistic regression isolated variables associated with atelectasis (P < 0.05). These candidates were subsequently entered into a multivariable logistic model to determine independent determinants, all of which were retained for construction of a predictive nomogram.

Model discrimination was quantified via the area under the receiver operator characteristic (ROC) curve. Internal validity and predictive consistency were tested through 500 bootstrap replications. Calibration performance was illustrated by overlaying predicted probabilities on actual event rates, and clinical applicability was judged with decision-curve analysis across a spectrum of thresholds. All statistical procedures employed EmpowerStats and R version 4.1.2, with two-sided P < 0.05 indicating significance.

This study included 301 adult patients with lymph node fistula TBTB, consisting of 130 males and 171 females, with ages ranging from 18 to 85 years and a median age of 64 (49, 69) years. The most common symptom was cough, reported in 87.04% (262/301) of patients. At the time of admission, 60.13% (181/301) of patients presented with concurrent atelectasis. Among these cases, 72.93% (132/181) involved the right lung, and 50.28% (91/181) specifically affected the right middle lobe. Significant differences were observed between the atelectasis and non-atelectasis groups in several clinical variables, including age, proportion of patients working as farmers, incidence of night sweats, presence of lymphadenopathy with ring-like enhancement, involvement of the right main bronchus and right middle lobe bronchus, proportion of cases with involvement of two or more bronchial sites, and incidence of visible fistulae (P < 0.05). Demographic and clinical profiles are detailed in Table 1.

To uncover atelectasis-related risks, a univariate logistic regression was performed on the entire sample. As shown in Table 2, eight variables were significantly associated with atelectasis: age, occupation as a farmer, presence of night sweats, lymphadenopathy with ring-like enhancement, involvement of the right main bronchus, involvement of the right middle lobe bronchus, involvement of two or more bronchial sites, and the presence of visible fistulae. These variables were subsequently included in a multivariable logistic regression analysis. The results demonstrated that increasing age, occupation as a farmer, lymphadenopathy with ring-like enhancement, and involvement of the right middle lobe bronchus were independent risk factors for atelectasis.

Based on the independent predictive variables identified by multivariable logistic regression analysis (age, occupation as a farmer, lymphadenopathy with ring-like enhancement, and involvement of the right middle lobe bronchus), a nomogram model was developed to predict the risk of atelectasis (Figure 2). Stepwise selection and correction for multicollinearity were used to ensure the statistical significance and clinical relevance of the selected variables. Lymph node enlargement with annular enhancement represented the dominant risk determinant. The model achieved favorable discrimination with training AUC of 0.824, improving to 0.857 upon validation (Table 3 and Figure 3), thus confirming predictive reliability. Calibration assessment via 500 bootstrap samples indicated optimal correspondence between predicted probabilities and observed events in both groups (Figure 4). Decision curve evaluation (Figure 5) substantiated the model’s clinical advantage compared to blanket treatment scenarios, with demonstrable net benefit observed at threshold probabilities of 0.2–0.9 in the derivation cohort and 0.15–0.85 in the validation cohort.

The clinical importance of TBTB complicated by atelectasis is well recognized (2, 3). To overcome the limitations of relying solely on clinical judgment for risk stratification, this study developed a nomogram based on multivariable analysis to predict the risk of atelectasis in patients with lymph node fistula TBTB. The model incorporates easily obtainable variables, including demographic factors (age and occupation), clinical indicators (symptoms, sputum smear, and culture), and CT findings (bronchial lesions, enhancement patterns of mediastinal, and/or hilar lymph nodes), allowing for a multidimensional risk assessment without significantly increasing the burden of routine diagnosis and treatment. By organizing and quantifying scattered clinical data, the nomogram provides a visual tool to support personalized clinical decisions.

This model may help clinicians identify high-risk patients early, prioritize bronchoscopic evaluations, and adjust anti-tuberculosis regimens, while reducing unnecessary diagnostic procedures in low-risk patients. It may be particularly useful in primary care settings and areas with limited medical resources. Such multidimensional, evidence-based tools could support improved clinical management of TBTB, promote more efficient use of healthcare resources, and help reduce the public health burden of tuberculosis.

Four independent predictors of atelectasis in TBTB were identified: increasing age, farming occupation, involvement of the right middle lobe bronchus, and mediastinal lymphadenopathy with ring-like enhancement. These factors span host characteristics, occupational exposures, local anatomical features, and radiological signs. The findings are consistent with previous studies (7, 12–14) and provide a more systematic understanding of the mechanisms underlying pulmonary atelectasis in lymph node fistula TBTB.

Within 24 h of admission, the proposed model can stratify patient risk using standard enhanced CT scans and electronic medical records. The hospital information system can then automatically assign appropriate follow-up intervals. For surgical candidates identified as high risk, preoperative management—such as mediastinal lymph node debulking or bronchial stenting—may help reduce postoperative atelectasis. In primary care settings, occupational history and CT findings can be used to screen for high-risk individuals and expedite referral, enabling coordinated care across different levels of the healthcare system.

Several important limitations must be acknowledged, each of which also indicates a path for future work. First, the model was developed and internally validated using retrospective data from a single tertiary-care center, which may restrict its generalizability to other institutions, populations, and geographic regions. Although bootstrap validation showed good performance, the lack of external validation remains a critical shortcoming. Future studies should therefore perform prospective external validation in geographically diverse multicenter cohorts and, if necessary, recalibrate the model to ensure robustness and applicability.

Second, variability in CT scanner settings and inter-reader interpretation could compromise the model’s transportability. Subsequent studies should endeavor to employ standardized imaging protocols and explore AI-assisted analysis to minimize instrument- and observer-related variation.

Third, the predictor “occupation (farmer)” is likely a surrogate for unmeasured socioeconomic factors such as income, education, living conditions, nutritional status, healthcare access, and treatment delays, introducing residual confounding. Future research should systematically collect granular socioeconomic variables—including income, education, residential environment, and health-insurance status—to quantify and adjust for these potential confounders more accurately.