Background:  In developing nations, incident tuberculosis (TB) is often the index event leading to a diagnosis of underlying HIV infection. It is recommended that all patients with active TB be screened for HIV co-infection. However, in practice HIV-testing rates are often poor. The objective of this study was to develop and validate a clinical prediction rule to identify patients at high risk for HIV co-infection (HIV/TB) among those presenting with TB.

Methods:  HIV serostatus and clinical and laboratory data were prospectively collected at the time of diagnosis in 1074 consecutive patients (HIV/TB 96 [9%]) treated for active TB at a tertiary care hospital in north India, during the period January 1997 to December 2003. This group was randomly split a posteriori, into a training set (711 [66%]; HIV/TB 66 [9%]) and a testing set (363 [34%]; HIV/TB 30 [8%]). Potentially useful variables significantly (p < 0.1) associated with HIV/TB on univariate analyses in the training set were entered as covariates to develop a multivariate logistic regression equation to obtain adjusted risk estimates. A summary variable was derived by combining the β-coefficients of variables in the equation and its performance was evaluated in the testing set by plotting receiver operating characteristic (ROC) curve.

Results:  Mean age of the study group was 32 ± 14 years (females 436 [41%]). No significant differences were found between training and testing sets. On univariate analysis, 10 variables were identified as potentially useful. Adjusted risk estimates (OR; 95% CI) for variables in the equation were: Male gender (5.3; 1.5 to 18.6), hemoglobin <11g/dL (7.6, 2.5 to 23), albumin < 3.5 g/dL (3.7; 1.3 to 10.3), triceps skinfold thickness < 8 mm (2.9 [1 to 8.9]) and axillary lymphadenopathy (9.7; 3.2 to 29.1). Summary variable was calculated as:  (1.7×male gender) + (2 × hemoglobin < 11 g/dL) + (1.3 × albumin < 3.5 g/dL) + (1.1 × triceps skinfold thickness < 8 mm) + (2.3×axillary adenopathy); score 1 if present and 0 if not, for all variables. In the testing set, at a cut-off of ≥ 4.1, summary variable had a sensitivity of 94% and specificity of 85% for predicting HIV seropositivity. Corresponding positive and negative predictive values were 29% and 99% respectively. Area under the ROC curve was 0.93 (0.86 to 1).

Conclusions:  A prediction rule based on simple clinical and laboratory parameters could be useful in identifying patients at high-risk for HIV co-infection, in those presenting with TB and would facilitate targeted testing in resource-limited settings.

Keywords: clinical prediction rule; tuberculosis; HIV testing