What question did this study set out to answer?

This research aims to create and validate a clinical prediction model using the pleural fluid ADA-to-LDH ratio to differentiate between tuberculous and malignant pleural effusions.

May 7, 2026BMC Pulmonary Medicine1 citationsOpen Access

Derivation and validation of a clinical prediction model incorporating the pleural fluid ADA-to-LDH ratio for differentiating tuberculous from malignant pleural effusions: a multi-center study

ZPZhiyu Pan STSha Tao HSHaoyu Sheng

Key Points

This research aims to create and validate a clinical prediction model using the pleural fluid ADA-to-LDH ratio to differentiate between tuberculous and malignant pleural effusions.
Multicenter retrospective study involving 455 patients from five hospitals in Anhui Province, China.
Data analyzed using Firth penalized logistic regression to construct the prediction model based on the ADA/LDH ratio and other variables.
Cohorts included training (n=290), internal validation (n=72), and external validation (n=93).
An ADA/LDH ratio ≥ 6.0% was the strongest predictor of malignant effusion (OR = 13.32, 95% CI 6.51-27.28, P < 0.001).
The model showed excellent discriminative ability with AUCs of 0.901, 0.893, and 0.916 across cohorts.
Good calibration was confirmed with Brier scores 0.1235, 0.1249, and 0.1159.

Abstract

BACKGROUND: Accurate discrimination between tuberculous (TPE) and malignant pleural effusion (MPE) is a major clinical challenge. Most existing models rely on non-routine laboratory tests and lack rigorous multicenter external validation. OBJECTIVE: To develop and validate a clinical prediction model integrating the pleural fluid adenosine deaminase to lactate dehydrogenase ratio (ADA/LDH) and routine indicators for TPE vs. MPE differentiation. METHODS: In this multicenter retrospective study conducted between January 2023 and December 2025, patients from five hospitals in Anhui Province, China, were divided into a training cohort (n = 290), an internal validation cohort (n = 72), and an external validation cohort (n = 93). Predictors were screened via univariable analysis and backward stepwise regression based on the Akaike Information Criterion (AIC). The optimal ADA/LDH cutoff was identified as 5.83% using restricted cubic splines (RCS) and simplified to 6.0% for clinical practicability without compromising model performance. A Firth penalized logistic regression model was constructed to mitigate data separation caused by the strong predictive effect of the ADA/LDH ratio. RESULTS: The final model included three statistically significant variables: pleural fluid ADA/LDH ratio (≥ 6.0% vs. < 6.0%), age, and sex. An ADA/LDH ratio ≥ 6.0% was the strongest independent predictor (OR = 13.32, 95% CI 6.51-27.28, P < 0.001). The model demonstrated excellent and stable discriminative ability with AUCs of 0.901 (training cohort), 0.893 (internal validation cohort), and 0.916 (external validation cohort). Calibration was good across all cohorts (Brier scores: 0.1235, 0.1249, 0.1159, respectively). Decision curve analysis demonstrated that the model provided numerically higher net benefit than the "treat all" and "treat none" strategies across the clinically relevant threshold range of 0%-90%. CONCLUSION: This multicenter study developed and validated a robust Firth penalized prediction model centered on the pleural fluid ADA/LDH ratio. The model demonstrates excellent discriminative ability, good calibration, and potential clinical utility for differentiating TPE from MPE in TB-endemic regions of China.

KI fragen

Bookmark

View Full Paper