Objective: To develop a scoring system for screening children for tuberculosis (TB) and for selecting suspects for further investigation in tuberculosis control programmes. Application of the score model, which would not require sophisticated or expensive technology, would be directed towards resource-poor countries with high prevalences of tuberculosis, where health care workers have to deal with diagnostic problems away from district hospitals or diagnostic facilities.
Design: Based on contributions from members of an IUATLD task group from 10 countries on the use of diagnostic criteria in childhood tuberculosis, criteria were selected to be used as elements in a score model. Data were collected by standardised questionnaire on 879 subjects aged under 15 years. Of these, 794 were considered probable or confirmed cases of tuberculosis by the diagnosing doctors. From each record, the criteria/procedures used in the diagnosis of probable/confirmed TB and regarded by the doctors as relevant criteria were selected. Bacteriology, histology and chest radiography were used either singly or collectively as the definitive reference (gold standard) against which the more subjective criteria (symptoms, clinical signs, skin test) would be evaluated. The latter criteria cited as relevant were then ranked and further explored for inclusion in the score model. The relative importance of each criterion to every other criterion on the list was expressed as weights, determined by employing a logarithmic least squares method to solve the ratio scale estimation problem which underlies decision-making involving more than one criterion. The resultant values were then assigned to each criterion in the final score model.
Results: The five clinical criteria thought to be most relevant as predictors of disease in children were history of contact with a case of tuberculosis, positive skin test, persistent cough, low weight for age, and unexplained/prolonged fever. In selecting the optimal cut-off points for the model at which tuberculosis would be suspected, low sensitivity and specificity (below 70%) but reasonably good positive predictive values (60%-77%) were obtained, depending on age group and epidemiological setting. In low tuberculosis prevalence settings, heavy reliance is placed by the model on a history of contact with a household case of tuberculosis and on a positive skin test, both of which have to be true. For high prevalence settings, more or less equal weighting is assigned to all five elements. Case contact and skin tests are less important, with low body weight, prolonged fever and cough being more indicative of tuberculosis.
Conclusion: The model provides for epidemiological differences between target populations and should prove successful as a screening tool to select children for further investigation by radiography and bacteriology.