A model to rule out smear-negative tuberculosis among symptomatic HIV patients using C-reactive protein

Abstract

Improved diagnostic algorithms for sputum smear-negative tuberculosis (SNTB) are needed to address the dramatic increase in SNTB in regions with high human immunodeficiency virus (HIV) prevalence. To determine whether the addition of C-reactive protein (CRP) to a prediction model using simple clinical criteria improves the diagnosis of SNTB among mostly antiretroviral-naïve adult HIV TB suspects in an out-patient setting. A multiple logistic regression model was derived from a database of 228 HIV patients to predict the risk of SNTB using data from a previous prospective study. The derived model demonstrated that male sex, night sweats, fever, low body mass index and anemia increased the probability of having SNTB. CRP improved the accuracy of the model (without CRP, area under the curve [AUC] 0.75, 95%CI 0.68-0.81 vs. model with CRP, AUC 0.81, 95%CI 0.76-0.87, P = 0.0014) to predict SNTB. Using reclassification tables, CRP correctly reclassified 27.9% of the patients (net reclassification improvement, P = 0.0005) into higher or lower risk categories. The strongest effect was seen in the reclassification improvement among patients with no TB, which was 20.6% (P = 0.0023). CRP improved the performance of the prediction model in the diagnosis of SNTB in HIV patients, and may play a role in ruling out SNTB in this population. Prospective validation of this model is needed.