Verification Bias in a Diagnostic Accuracy Study of Symptom Screening for Tuberculosis in HIV-Infected Pregnant Women

TO THE EDITOR—We read with interest the article by Gupta et al on symptom screening among pregnant women infected with human immunodeficiency virus (HIV) [1]. This work examines the diagnostic accuracy of the World Health Organization (WHO) symptom screen for active tuberculosis [2] among HIV-infected pregnant women, a population in whom tuberculosis screening is important because tuberculosis increases maternal mortality and HIV perinatal transmission [3, 4]. Diagnosis of maternal tuberculosis is also important in the prevention of infant tuberculosis. In total, 799 women were screened using the WHO symptom screen and tuberculin skin test. Symptomatic women and those with a positive skin test were assessed by chest radiography. Women suspected of active tuberculosis based on symptom screen, positive skin test, suspicious chest radiograph, or physician findings were requested to provide a sputum sample for microscopy and culture. Sputum smear and culture were used to verify tuberculosis disease in 107 of 799 women. This differential verification of tuberculosis disease means there is no reference standard against which the screening test can be evaluated in the entire population, which can lead to verification bias. Unless the WHO symptom screen had perfect sensitivity, misclassification of false-negatives as true-negatives results in underestimated disease prevalence and overestimated sensitivity and specificity of the symptom screen [5]. In the case of this study, where only 11 tuberculosis cases were identified, the impact of misclassification of even a few women could be significant. This bias may be corrected under certain conditions with additional information on the study population [6, 7]. Two other concerns are worth noting. First, the authors emphasize the high negative predictive value of 99.3%, with exact 95% confidence limits estimated at 98.4% and 99.8%. Given the reported prevalence of 1.4%, the negative predictive value of assuming no one has tuberculosis is 98.6% (1 – prevalence). The gain of screening in terms of negative predictive value, therefore, is somewhat uncertain, demonstrating the difficulty in interpreting the negative predictive value where disease prevalence is low. Second, the authors state that the WHO symptom screening test was effective in ruling out tuberculosis. However, 5 of the 11 cases of tuberculosis were asymptomatic. Reliance on the WHO screen alone would therefore have missed nearly half the cases of tuberculosis in this population. In light of these issues, we must be cautious in interpreting the reported diagnostic accuracy for the symptom screen and the reported prevalence of tuberculosis in this study population of HIV-infected pregnant women. C. H. receives support from the National Insitutes of Health (NIH)/National Institute of Allergy and Infectious Diseases (NIAID) (401AI069463). A. V. R. receives support from the NIH/National Institute of Child Health and Human Development (NICHD) (R01-HD053216, NIH/NICHD R01-HD058972, NIH/NICHD R01-HD0691175, and NIH/NIAID 401AI069462). D. W. receives support NIH/NICHD (4R00-HD-06-3961), the NIH/NIAID (2P30-AI-06-4518-06), and the Duke Center for AIDS Research. All authors: No reported conflicts. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.