Translate this page into:
A clinical prediction rule to identify patients with tuberculosis at high risk for HIV co-infection
Reprint requests: Prof. S.K. Sharma, Chief, Division of Pulmonary, Critical Care & Sleep Medicine, Professor & Chair Department of Medicine, All India Institute of Medical Sciences, New Delhi 110 029, India e-mail: sksharma@aiims.ac.in
-
Received: ,
Abstract
Background & objective:
Many patients presenting with tuberculosis (TB) have underlying human immunodeficiency virus (HIV) co-infection. Routine HIV testing, however, is not a component of the national TB control programme in India. We sought to derive and validate a clinical prediction rule, based on clinical and laboratory parameters, to identify patients at high risk for HIV co-infection among those treated for active TB.
Methods:
Case records of adult patients with active TB treated between 1997 and 2003 at the All India Institute of Medical Sciences hospital, New Delhi were retrospectively reviewed. The data set was randomly split into a training set and a testing set. First a clinical prediction rule was derived by multivariable logistic regression on the training set and was subsequently validated on the testing set.
Results:
The study group comprised 1074 patients [training set 711 (66%), HIV co-infected 66 (9%); testing set 363 (34%), HIV co-infected 30 (8%)]. In the training set, male gender [odds ratio (95% CI) 5.31(1.52- 18.61)], axillary lymphadenopathy [9.71 (3.24-29.10)], anaemia [7.56 (2.48-23.05)], hypoalbuminaemia [3.67(1.31-10.26)], and reduced triceps skinfold thickness [2.91(0.95-8.89)] were independently associated with HIV co-infection. In the testing set, presence of any two of these five features was 94 per cent (95% CI 84-100%) sensitive and 54 per cent (49-60%) specific for predicting HIV co-infection; negative predictive value was 99 per cent (98-100%). Area under the receiver-operating characteristic curve was 0.93 (0.86-1.0) in the testing set.
Interpretation & conclusions:
A simple clinical prediction rule based on clinical and laboratory parameters could be used to identify a subgroup of patients, among those treated for active TB in a hospital setting, for targeted HIV testing.
