Clinical-Laboratory Prediction Rule Derivation for Pulmonary Tuberculosis Diagnosis in General Hospitals in a High-Burden Country

Abstract

Background Clinical prediction rules (CPR) for pulmonary tuberculosis (PTB) diagnosis in the hospital setting in countries with high prevalence of PTB are rare. Methods With the purpose to derive a CPR for PTB and assess its accuracy in patients 15 years or older admitted to a general hospital in Rio de Janeiro city, Brazil, a diagnostic accuracy study was conducted with hospitalized patients from 2008 to 2013. Reference standard: positive Mycobacterium tuberculosis culture in respiratory specimens. Logistic regression-adjusted odds ratios, receiver operating characteristic curve of the value predicted by the model and its accuracy parameters were calculated. Results This study assessed 284 patients, 52 with PTB (18.3%). The score of the derived CPR ranged from 0 to 5, 2 points being attributed to fever and 1 to each of the following predictors: weight loss greater than 10%, alcohol consumption and typical/compatible chest imaging. The cutoff point was 3 or greater, with 0.70 area under the receiver operating characteristic curve, 72.1% sensitivity, 64.5% specificity, 30.1% and 91.6% positive and negative predictive values, respectively, and 2.0 and 0.4 positive and negative likelihood ratios, respectively. Conclusions The derived CPR corroborated the result of a recent systematic review of PTB predictors for hospitalized patients, but should be prospectively validated. The high false-positive rate justifies its combination with PTB-specific tests.