关键词: 结核,肺, 老年人, 共病现象, 模型, 统计学

Abstract:

Objective: Using the Chinese Multimorbidity-weighted Index (CMWI), a model for predicting the prognosis of elderly pulmonary tuberculosis patients with comorbidities was established, to provide a reference basis for formulating prevention and treatment strategies for comorbidities of pulmonary tuberculosis in the elderly. Methods: A retrospective cohort study was conducted on 1423 elderly pulmonary tuberculosis patients with comorbidities admitted to Guangzhou Chest Hospital from January 1, 2016 to December 31, 2023. The study started from the date of diagnosis of pulmonary tuberculosis, and the follow-up observation period was 18 months, taking prognosis of study subjects as final outcomes. Information of the subjects was collected from the “TB Information Management System” of the “China Disease Control and Prevention Information System”, medical records, the “tuberculosis patient treatment record card of Guangzhou”, and the “tuberculosis patient treatment management registration card of Guangzhou”. Characteristics and prognosis of patients were analyzed using descriptive methods, followed by logistic regression analysis to identify influencing factors and establish a predictive model. Results: Among 1423 elderly pulmonary tuberculosis patients with comorbidities, 1205 cases (84.68%) got treatment success, and 218 cases (15.32%) got unfavorable outcomes. Higher age (HR=1.054,95%CI:1.033-1.076), non-local-registered residence (HR=1.655,95%CI:1.058-2.621), clinical symptoms (HR=2.216,95%CI:1.333-3.683), positive etiological examination result (HR=3.802,95%CI:2.512-5.754), severe pulmonary tuberculosis (HR=4.628,95%CI:2.968-7.216), and high CMWI (Chinese multimorbidity-weighted index,CMWI)(HR=1.301,95%CI:1.196-1.415) were risk factors for adverse treatment outcomes; regular treatment (HR=0.285,95%CI: 0.180-0.451) was a protective factor. We established a predictive model as logit P=-8.136+0.053X₁+0.510X₂+0.796X₃+1.335X₆+1.532X₉-1.254X₁₀+0.263X₁₁. The goodness of fit test showed that the model fit well (χ²=8.055, P=0.428). The total accuracy using backtesting and interaction verification methods were 87.70% and 88.19%, respectively, indicating good prediction accuracy. The diagnostic performance of the model was as follows: area under the curve (AUC) was 0.870 (95%CI: 0.844-0.896), optimal critical point was 0.189, sensitivity was 73.53%, missed diagnosis rate was 26.47%, specificity was 86.98%, misdiagnosis rate was 13.02%, and Jordan index was 0.605, showing the model having made good performance. A decision tree model was constructed using machine learning algorithms which consisted of 3 layers, 11 nodes, and 4 explanatory variables, with regular treatment as the root node, followed by severe tuberculosis, etiological examination result, and CMWI as child nodes. The probability of getting treatment success was highest (96.36%) for patients with regular treatment, non-severe pulmonary tuberculosis, and CMWI≤3.2. ROC curve for this model showed: area under the curve (AUC) was 0.843 (95%CI: 0.812-0.875), optimal critical point was 0.134, sensitivity was 77.98%, missed diagnosis rate was 22.02%, specificity was 79.34%, misdiagnosis rate was 20.66%, and Jordan index was 0.573. Brier score was 0.089. Calibration curve indicated good predictive accuracy. Decision curve demonstrated good clinical applicability. Conclusion: The prognostic outcome predictive model for elderly pulmonary tuberculosis patients with comorbidities based on the CMWI has good application value and provides scientific basis for exploring the management of comorbidities of pulmonary tuberculosis in the elderly.

Key words: Tuberculosis, pulmonary, Aged, Comorbidity, Models, statistical