Abstract
Disseminated histoplasmosis is a common differential diagnosis of tuberculosis in disease-endemic areas. We aimed to find a predictive score to orient clinicians towards disseminated histoplasmosis or tuberculosis when facing a non-specific infectious syndrome in patients with advanced HIV disease. We reanalyzed data from a retrospective study in Cayenne Hospital between January 1997-December 2008 comparing disseminated histoplasmosis and tuberculosis: 100 confirmed disseminated histoplasmosis cases and 88 confirmed tuberculosis cases were included. A simple logit regression model was constructed to predict whether a case was tuberculosis or disseminated histoplasmosis. From this model, a score may be obtained, where the natural logarithm of the probability of disseminated histoplasmosis/tuberculosis = +3.917962 × WHO performance score (1 if >2, 0 if ≤2) -1.624642 × Pulmonary presentation (1 yes, 0 no) +2.245819 × Adenopathies > 2 cm (1 yes, 0 no) -0.015898 × CD4 count - 0.001851 × ASAT - 0.000871 × Neutrophil count - 0.000018 × Platelet count + 6.053793. The area under the curve was 98.55%. The sensitivity of the model to distinguish between disseminated histoplasmosis and tuberculosis was 95% (95% CI = 88.7-98.3%), and the specificity was 93% (95% CI = 85.7.3-97.4%). In conclusion, we here present a clinical-biological predictive score, using simple variables available on admission, that seemed to perform very well to discriminate disseminated histoplasmosis from tuberculosis in French Guiana in well characterized patients.