Derivation and validation of a pulmonary tuberculosis prediction model

Area of Science:

• Infectious Diseases
• Public Health
• Clinical Prediction Modeling

Background:

• Respiratory isolation is crucial for preventing tuberculosis (TB) transmission.
• Early discontinuation of unnecessary isolation is desirable to optimize resource allocation and patient experience.
• Predictive models can aid clinical decision-making in infectious disease management.

Purpose of the Study:

• To derive and validate a prediction model for pulmonary tuberculosis (TB).
• To enable early and safe discontinuation of respiratory isolation for patients with suspected pulmonary TB.
• To assess the model's performance in a real-world clinical setting.

Main Methods:

• Retrospective analysis of 296 isolation episodes to identify independent predictors of pulmonary TB.
• Development of a weighted prediction model based on identified risk factors.
• Prospective validation of the model on 220 isolation episodes in a university-affiliated hospital.
• Evaluation of model predictability using receiver operating characteristic (ROC) curve analysis and area under the curve (AUC).

Main Results:

• Four independent predictors of pulmonary TB were identified: positive acid-fast sputum smear, localized chest radiograph findings, correctional facility residence, and weight loss history.
• The model demonstrated strong and consistent predictability in both derivation (AUC = 0.86) and validation (AUC = 0.86) cohorts.
• Utilization of the prediction model contributed to a significant decrease in the mean duration of respiratory isolation.

Conclusions:

• A validated pulmonary TB prediction model can effectively quantify individual patient risk.
• The model aids healthcare professionals in making informed decisions regarding the discontinuation of respiratory isolation.
• This approach demonstrates the utility of prediction models in managing clinical challenges like TB and optimizing infection control practices.