Comparing Tuberculosis Diagnostic Yield in Smear/Culture and Xpert(®) MTB/RIF-Based Algorithms Using a Non-Randomised Stepped-Wedge Design

SETTING: Primary health services in Cape Town, South Africa. STUDY AIM: To compare tuberculosis (TB) diagnostic yield in an existing smear/culture-based and a newly introduced Xpert(®) MTB/RIF-based algorithm. METHODS: TB diagnostic yield (the proportion of presumptive TB cases with a laboratory diagnosis of TB) was assessed using a non-randomised stepped-wedge design as sites transitioned to the Xpert(®) based algorithm. We identified the full sequence of sputum tests recorded in the electronic laboratory database for presumptive TB cases from 60 primary health sites during seven one-month time-points, six months apart. Differences in TB yield and temporal trends were estimated using a binomial regression model. RESULTS: TB yield was 20.9% (95% CI 19.9% to 22.0%) in the smear/culture-based algorithm compared to 17.9% (95%CI 16.4% to 19.5%) in the Xpert(®) based algorithm. There was a decline in TB yield over time with a mean risk difference of -0.9% (95% CI -1.2% to -0.6%) (p<0.001) per time-point. When estimates were adjusted for the temporal trend, TB yield was 19.1% (95% CI 17.6% to 20.5%) in the smear/culture-based algorithm compared to 19.3% (95% CI 17.7% to 20.9%) in the Xpert(®) based algorithm with a risk difference of 0.3% (95% CI -1.8% to 2.3%) (p = 0.796). Culture tests were undertaken for 35.5% of smear-negative compared to 17.9% of Xpert(®) negative low MDR-TB risk cases and for 82.6% of smear-negative compared to 40.5% of Xpert(®) negative high MDR-TB risk cases in respective algorithms. CONCLUSION: Introduction of an Xpert(®) based algorithm did not produce the expected increase in TB diagnostic yield. Studies are required to assess whether improving adherence to the Xpert(®) negative algorithm for HIV-infected individuals will increase yield. In light of the high cost of Xpert(®), a review of its role as a screening test for all presumptive TB cases may be warranted.