Evaluation of clinical prediction rules for respiratory isolation of inpatients with suspected pulmonary tuberculosis.

Background: In the framework of hospital infection control, various clinical prediction rules (CPRs) for respiratory isolation of patients with suspected pulmonary tuberculosis (PTB) have been developed. Our aim was to evaluate their performance in an emergency department setting with a high prevalence of PTB.

Methods: We searched the MEDLINE and OVID databases to identify CPRs to predict PTB. We used a previously collected database containing clinical, radiographical, and microbiological information on patients attending an emergency department with respiratory complaints, and we applied each CPR to every patient and compared the result with culture for Mycobacterium tuberculosis as the reference standard. We also simulated the proportion of isolated suspects and missed cases for PTB prevalences of 5% and 30%.

Results: We withheld 13 CPRs for evaluation. We had complete data on 345 patients. Most CPRs achieved a high sensitivity but very low specificity and very low positive predictive value. Mylotte's score, which includes results of sputum smear as a predictive finding, was the best-performing CPR. It attained a sensitivity of 88.9% and a specificity of 63.9%. However, at a 30% PTB prevalence, 498 of 1000 individuals with suspected PTB would have to be isolated; 267 of these cases would be true PTB cases, and 33 cases would be missed. Two consecutive sputum smears had a sensitivity of 75.6% and a specificity of 99.7%.

Conclusions: In a setting with a high prevalence of PTB, only 1 of the 13 assessed CPRs demonstrated high sensitivity combined with satisfactory specificity. Our results highlight the need for local validation of CPRs before their application.