Development and validation of a risk prediction model for pulmonary tuberculosis among presumptive tuberculosis cases in Ethiopia

Abstract

BackgroundEarly diagnosis and treatment of tuberculosis (TB) is one of the key strategies to achieve the WHO End TB targets. This study aimed to develop and validate a simple, convenient risk score to diagnose pulmonary TB among presumptive TB cases.MethodsThis prediction model used Ethiopian national TB prevalence survey data and included 5459 presumptive TB cases from all regions of Ethiopia. Logistic regression was used to determine which variables are predictive of pulmonary TB. A risk prediction model was developed, incorporating significant variables (p<0.05). The Youden Index method was used to choose the optimal cut-off point to separate the risk score of the patients as high and low. Model performance was assessed using discrimination power and calibration. Internal validation of the model was assessed using Efron’s enhanced bootstrap method, and the clinical utility of the risk score was assessed using decision curve analysis.ResultsOf total participants, 94 (1.7%) were confirmed to have TB. The final prediction model included three factors with different scores: (1) TB contact history, (2) chest X-ray (CXR) abnormality and (3) two or more symptoms of TB. The optimal cut-off point for the risk score was 6 and was found to have a good discrimination accuracy (c-statistic=0.70, 95% CI: 0.65 to 0.75). The risk score has sensitivity of 51.1%, specificity of 79.9%, positive predictive value of 4.3% and negative predictive value of 98.9%. After internal validation, the optimism coefficient was 0.003, which indicates the model is internally valid.ConclusionWe developed a risk score that combines TB contact, number of TB symptoms and CXR abnormality to estimate individual risk of pulmonary TB among presumptive TB cases. Though the score is easy to calculate and internally validated, it needs external validation before widespread implementation in a new setting.