Development of a clinical prediction rule for tuberculous meningitis in adults in Lima, Peru.

Abstract

Diagnosis of tuberculous meningitis (TM) is a challenge in countries with a high burden of the disease and constrained resources and clinical prediction rules (CPRs) could be of assistance. We aimed at developing a CPR for diagnosis of TM in a Latin American setting with high tuberculosis incidence and a concentrated HIV epidemic. We enrolled adult patients with clinical suspicion of TM attending two hospitals in Lima, Peru. We obtained information on potential anamnestic, clinical and laboratory predictive findings that are easy to collect and promptly available. We independently diagnosed TM according to a composite reference standard that included a series of microbiological tests. We performed bivariate analysis and constructed a logistic regression model to select the predictive findings associated with TM. With the selected predictors included in the model, we developed a score-based CPR. We assessed its internal validity and diagnostic performance. Of 155 analysed patients, 59 (38%) had TM. The CPR we derived includes three predictors: cough for 14 days or more, 10-500 cells in CSF and adenosine deaminase ≥ 6 U/l in CSF. It classifies patients into high-, moderate- or low-score groups and has an overall area under the ROC curve of 0.87. 59% of patients were assigned to either the high- or the low-score group, permitting prompt decision-making. In patients in the high-score group, it attains a positive likelihood ratio for TM of 10.6 and in patients with low scores, a negative likelihood ratio of 0.10. Bootstrap analysis indicated high internal validity. This CPR could support decision-making in patients with clinical suspicion of TM. External validation and further assessment of its clinical impact are necessary before application in other settings.