Structured approaches for the screening and diagnosis of childhood tuberculosis in a high prevalence region of South Africa

Objective To measure agreement between nine structured approaches for diagnosing childhood tuberculosis; to quantify differences in the number of tuberculosis cases diagnosed with the different approaches, and to determine the distribution of cases in different categories of diagnostic certainty. Methods We investigated 1445 children aged <2 years during a vaccine trial (2001-2006) in a rural South African community. Clinical, radiological and microbiological data were collected prospectively. Tuberculosis case status was determined using each of the nine diagnostic approaches. We calculated differences in case frequency and categorical agreement for binary (tuberculosis/not tuberculosis) outcomes using McNemar's test (with 95% confidence intervals, CIs) and Cohen's kappa coefficient (K). Findings Tuberculosis case frequency ranged from 6.9% to 89.2% (median: 41.7). Significant differences in case frequency (P < 0.05) occurred in 34 of the 36 pair-wise comparisons between structured diagnostic approaches (range of absolute differences: 1.5-82.3%). Kappa ranged from 0.02 to 0.71 (median: 0.18). The two systems that yielded the highest case frequencies (89.2% and 70.0%) showed fair agreement (K: 0.33); the two that yielded the lowest case frequencies (6.9% and 10.0%) showed slight agreement (K: 0.18). Conclusion There is only slight agreement between structured approaches for the screening and diagnosis of childhood tuberculosis and high variability between them in terms of case yield. Diagnostic systems that yield similarly low case frequencies may be identifying different subpopulations of children. The study findings do not support the routine clinical use of structured approaches for the definitive diagnosis of childhood tuberculosis, although high-yielding systems may be useful screening tools.