A systematic review of prediction models for prevalent pulmonary tuberculosis in adults

A systematic review was conducted to describe the quality and characteristics of prediction models for prevalent pulmonary tuberculosis (PTB) in adults at routine TB care settings. A prediction model was defined as the combination of two or more clinical predictors designed to estimate the probability of having TB. Studies using culture-confirmed PTB as reference standard were included. Models for in-patients, children or specific patient populations were excluded. PubMed, Scopus and the Cochrane Library and abstracts from the International Union Against Tuberculosis and Lung Disease, American Thoracic Society and European Respiratory Society conferences were searched. The CHARMS (Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modelling Studies) checklist was used for data extraction and quality assessment. From 13 671 identified records, six were included for data extraction; three assessed smear-negative, culture-positive PTB as outcome and three focused on human immunodeficiency virus infected individuals only. Reporting of model development, performance and evaluation was poor. In four studies, predictive performance was evaluated using the development data set (apparent performance), one study did an internal validation and one study did an external validation. Results were not pooled due to heterogeneity. Existing prediction models for estimating prevalent PTB in adults at primary care level are poorly reported and validated and are not useful for TB screening. The World Health Organization symptom screen is recommended.