Adaptive immunity, including T-cell-mediated pathogenic immune responses, is known to play an important role in the inflammatory process of tuberculosis[8]. Nearly all patients with PTM have a history of ipsilateral tuberculous pleurisy[9]. Some patients initially present with unilateral pleurisy, which later progresses to bilateral PTM. This progression usually occurs gradually and subtly after the complete absorption of pleural fluid during anti-tuberculosis therapy[3]. However, the exact mechanism by which tuberculous pleurisy transforms into tuberculous pleuroma, as well as the risk factors for the development of PTM, remain unclear. In this retrospective study, we analyzed the clinical features, and laboratory examinations of 79 PTM and compared them to a control group of 60 TPE patients who were effectively treated without developing PTM. The aim was to identify the risk factors associated with the occurrence of PTM.
This study initially examined 13 risk factors that could potentially be associated with PTM. These factors included 1 binary variable(gender), and 12 continuous variables such as age, Adenosine deaminase (ADA), lactate Dehydrogenase (LDH), protein, CD3, CD4, CD8, CD4/CD8, HLA-DR+CD4+T, HLA-DR+CD8+T, Ki-67+CD4+T, Ki-67+CD8+T. We first compared the consistency of demographic data and clinical features between the PTM group and the TPE group by T-test or chi-square test(Table 1). The results showed that the occurrence of PTM was not related to gender differences (P = 0.960) and age characteristics (P = 0.548). However, it was observed that PTM patients were mostly young, with an average age of 30.75 ± 7.48 years. The potential risk factors for PTM identified in this study were ADA, CD3, CD4, Ki-67+CD4+T, and Ki-67+CD8+T. These findings were consistent with the results of the univariate logistic regression analysis. Besides, the average onset time of PTM in this study was found to be 4.67 ± 2.41 months after the TPE diagnosis. This finding aligns with previous studies that have shown the onset of PTM typically occurs within 6 months after diagnosis[10].
We developed two models using logistic regression, lasso regression, and best subset selection. Model 1 included Ki-67+CD4+T and Ki-67+CD8+T, while Model 2 included Ki-67+CD4+T, Ki-67+CD8+T, and ADA. Notably, three clinical features (ADA, CD3, CD4) that showed a significant impact on PTM occurrence in univariate logistic regression were excluded in multivariate logistic regression. Only Ki-67+CD4+T and Ki-67+CD8+T were identified as risk factors for PTM occurrence. Indeed, a strong correlation existed among CD3, CD4, and CD8 (Fig. 2), potentially leading to biased model estimation or challenges in accurate estimation when variables were included. This could partly explain the discrepancy in significance tests for CD3 and CD4 between univariate logistic regression and multivariate logistic regression.
Pathogen-specific T-cell immunity is a crucial mechanism for host control of Mycobacterium tuberculosis infection[11]. Research in both human and animal models has demonstrated that acquired immunity against M. tuberculosis necessitates the participation of various T cell subsets, particularly CD4+ T cells and CD8+ T cells[12, 13]. However, a patient's immune status should not only be assessed by enumerating immune cells but also by evaluating activated T cells[14, 15]. Therefore, our study not only analyzed immune cells like CD3, CD4, and CD8 but also included activated T cells like HLA-DR+CD4+T and Ki-67+CD4+T, which have cell activation markers. Our research revealed that while the absolute count of MTB-specific CD3 and CD4 cells, and the ratio of CD4/CD8 did not have a statistically significant impact on the development of PTM (P > 0.05), the presence of the activation marker Ki-67 in MTB-specific CD4 and CD8 cells was identified as a significant risk factor for PTM development. The impairment of T lymphocyte function plays a crucial role in the progression of tuberculosis, with Ki-67 serving as an early indicator of intracellular proliferation. Several studies have indicated a positive correlation between the presence of Ki-67 on MTB-specific CD4 T + cells and the load of mycobacterium[16, 17]. Besides, it has been observed that the expression of Ki-67 would decrease following treatment[18, 19]. The heightened Ki-67 expression in CD4 + and CD8 + T cell subpopulations following in vitro restimulation with MTB antigen in PTM patients suggested a greater burden of mycobacterium antigen in slow responders, hinting at the possibility of PTM being a chronic infectious disease involving both innate and acquired immunity in response to TB antigen[20]. The disparity in CD4 + and CD8 + T cell subset activation implied that immune dysregulation might be a characteristic of PTM. Therefore, further exploration of cell activation could lead to significant advancements in our comprehension of PTM pathogenesis.
For the diagnosis of tuberculous pleurisy, the most widely used clinical indicators are ADA, LDH, and total protein[21, 22]. ADA, an enzyme found throughout human tissues and body fluids, is secreted by activated lymphocytes, macrophages, and neutrophils. It serves as a general indicator of inflammation, being present in lymphocyte-based exudates and recognized as a sign of cell-mediated immune reactions[23]. A high ADA index is regarded as a key element in the diagnostic criteria for tuberculous pleurisy[24]. In our research, despite the similar AUC values of Model 1 and Model 2(0.987 (95%CI: 0.969-1.000) vs 0.986 (95%CI: 0.969-1.000)), the Hosmer-Lemeshow goodness of fit test indicated that Model 2 had better prediction accuracy, most likely due to the introduction of ADA as a clinical feature in Model 2 compared to Model 1, which increased the prediction accuracy for conditions with a lower risk of PTM occurrence(mostly tuberculous pleurisy). Anyway, Model 1 displayed a larger net benefit area when assessing the clinical significance of both models in predicting PTM (Fig. 4.e). Particularly, Model 1 demonstrated better clinical utility when the risk of PTM occurrence ranged from 0.7 to 0.85. In addition, we assessed whether the incorporation of ADA as a clinical feature enhanced the predictive capability of the model by computing the Integrated Discrimination Improvement(IDI). The findings revealed that ADA made a minimal contribution to the predictive model of PTM occurrence (IDI: 0.004 (95%CI: -0.009-0.018); P-value: 0.53). Considering the aforementioned factors, we ultimately selected Model 1 over Model 2 and achieved a satisfactory performance evaluation during internal validation (Accuracy: 0.96; Kappa: 0.93; AUC: 0.98; Sensitivity: 0.95; Specificity: 0.96). LDH, the primary enzyme responsible for sugar metabolism, is extensively present in the cytoplasm of tissue cells and holds significant clinical value in identifying leaky or exudative pleural effusion. While LDH serves as a crucial marker for inflammatory responses, it may not provide specific information for determining the cause of the condition[25]. As a result, it is suggested to utilize the LDH/ADA ratio for diagnosing TB effusions instead of relying solely on LDH levels[26].
There are several limitations in this study. Firstly, it should be noted that this study is retrospective, which may introduce bias in patient selection. Additionally, the study was conducted at two centers and included a relatively small number of PTM patients (79 cases), so caution must be exercised when applying the findings to a larger population of PTM patients. Secondly, there is a scarcity of reports on the clinical risk prediction of PTM, making it impossible to externally validate our model and assess its generalizability. Lastly, we believe that the limited calibration of model predictions may be attributed to the incomplete inclusion of risk factors for PTM occurrence, particularly the potential presence of risk factors in T cells with cell activation markers that were not considered in this study. Further investigation is warranted in this regard.