Abstract
We successfully developed a nomogram model for predicting tuberculous pericarditis, which can assist clinicians in improving diagnostic accuracy and reducing misdiagnoses and missed diagnoses in primary healthcare settings.
Introduction
Tuberculous Pericarditis (TBP)—an inflammatory condition of the heart's lining caused by Mycobacterium tuberculosis—remains a significant clinical challenge, particularly in regions where tuberculosis is endemic. If left untreated, TBP can progress to constrictive pericarditis or cardiac tamponade, both of which are life-threatening.
Despite its severity, TBP is notoriously difficult to diagnose. The "gold standard" of bacterial culture is often too slow for acute management and suffers from low sensitivity due to the paucibacillary nature of pericardial fluid. While advanced molecular diagnostics like Xpert MTB/RIF offer higher precision, their high cost and requirement for specialized infrastructure limit their utility in primary healthcare settings. Consequently, many clinicians rely on subjective experience, leading to high rates of misdiagnosis. This study aimed to develop a data-driven, accessible prediction tool to bridge this diagnostic gap.
Methodology
Patient Cohort and Data Collection
The researchers conducted a retrospective cohort study involving 304 patients treated at the Hangzhou Red Cross Hospital, a provincial tuberculosis center in Zhejiang, China, between January 2015 and August 2023. The cohort included 144 patients confirmed with TBP and 160 patients with non-specific pericarditis. The dataset was randomly partitioned into a training set (70%) for model development and a validation set (30%) for performance assessment.
Statistical Modeling
To identify the most significant predictors from an initial pool of 39 clinical and laboratory features, the team utilized LASSO (Least Absolute Shrinkage and Selection Operator) logistic regression. LASSO is particularly effective for high-dimensional data as it penalizes the absolute size of regression coefficients, effectively filtering out less relevant variables to prevent overfitting.
Following variable selection, multivariate logistic regression was used to finalize the model. The results were then translated into a nomogram—a user-friendly graphical tool that allows clinicians to calculate a patient's risk score manually.
Key Results and Model Performance
The Seven Predictors
The finalized model identified seven key clinical and laboratory markers that collectively predict the likelihood of TBP:
- Weight Loss: A common systemic symptom of TB.
- Body Mass Index (BMI): Lower BMI was associated with higher TBP risk.
- History of Tuberculosis: Previous infection significantly increases current risk.
- History of Dust Exposure: An environmental factor often linked to pulmonary compromise (OR 4.036).
- Moderate-to-Severe Kidney Disease: Identified as a strong positive predictor (OR 4.121).
- Erythrocyte Sedimentation Rate (ESR): A marker of chronic inflammation.
- B-type Natriuretic Peptide (BNP): Used here as a differential marker.
Diagnostic Accuracy
The model demonstrated robust discriminative power, achieving an Area Under the Curve (AUC) of 0.757 in both the training and validation cohorts. Most significantly, the model outperformed traditional clinical judgment. While physicians' initial assessments missed 52.4% of TBP cases, the model’s low-risk threshold reduced the missed diagnosis rate to just 14.3%. Similarly, the misdiagnosis rate for non-TBP patients dropped from 22.5% to 5.0% when using the model's high-risk threshold.
Discussion
The BNP Paradox
One of the study's most intriguing findings was the negative correlation between BNP levels and TBP. While BNP is typically a marker for heart failure, the researchers observed higher levels in the non-specific pericarditis control group. They hypothesize that non-specific pericarditis may cause more acute cardiac dysfunction compared to the chronic, insidious inflammatory progression of TBP, making lower BNP levels a paradoxical clue for tuberculosis in this specific context.
Clinical Utility
Decision Curve Analysis (DCA) confirmed that the nomogram provides a net clinical benefit across a wide range of threshold probabilities. By using routine medical history and basic blood work, the tool is ideally suited for resource-limited environments where advanced imaging or molecular testing is unavailable.
Conclusion and Limitations
This study provides a practical, evidence-based tool for the early diagnosis of TBP. However, the authors note several limitations: the study was conducted at a single specialized center, and the retrospective design may introduce selection bias. Future work should focus on multi-center external validation to ensure the model's generalizability across different geographic regions and patient populations with varying tuberculosis prevalence.
Summary updated