A critical review of the XTEND trial comparing Xpert MTB/RIF with smear microscopy for the diagnosis of tuberculosis under routine settings: why a highly sensitive test failed to improve patient mortality

Abstract

Beginning in 2011, a cartridge-based PCR assay with a two-hour turn around time known as Xpert MTB/RIF (hereafter ‘Xpert’) was rolled out as a replacement for smear microscopy for tuberculosis diagnosis in South Africa. The staged roll-out made it possible to conduct a pragmatic cluster-randomised trial (Xpert for TB – evaluating a new diagnostic, or ‘XTEND’) to determine the impact of Xpert on patient and programme outcomes. XTEND results indicated that Xpert identified one and a half times more tuberculosis cases. However, there was no difference in loss to follow up, the proportion of participants started on tuberculosis treatment or mortality rates across intervention (Xpert) and control (smear microscopy) arms. This PhD aimed to investigate reasons for the absence of a reduction in mortality by exploring contextual factors that may have impacted Xpert performance during the XTEND trial. A literature review of impact evaluations of Xpert conducted between 2012-2019 and done as a part of this PhD suggests that the more systematically and rigorously tuberculosis diagnostic tests are conducted at patient enrolment and follow-up as part of study procedures, the less likely a study will find a mortality difference between intervention (Xpert) and control (smear) arms. Findings from the literature review suggested that a careful evaluation should be done of the extent of adherence to routine tuberculosis diagnostic algorithms in the XTEND study. In a secondary data analysis of XTEND data, we showed poor levels of adherence to tuberculosis diagnostic algorithms. We identified greater levels of adherence to algorithms for tuberculosis diagnosis in the smear microscopy arm and amongst persons who were more ill, male or visited clinic repeatedly. In addition, we observed large between-clinic variation in adherence rates. In mapping the pathways to care for XTEND participants who were initiated empirically on tuberculosis treatment, we showed that empiric tuberculosis treatment was infrequently prescribed in the real-world setting of XTEND, where participating clinics were left to implement tuberculosis diagnostic algorithms without study intervention. Furthermore, we observed that a majority of empiric tuberculosis treatment initiations took place in the smear microscopy arm. In a qualitative evaluation at eight representative XTEND clinics, we identified large clinic-to-clinic variation with regard to the extent of Xpert implementation and embedding, and district health system strengths and weaknesses. We showed the impact of co-ordination of resources, creation and sustaining of referral networks and ownership of the tuberculosis programme by managers in supporting full implementation of and adherence to the algorithm for TB diagnosis using Xpert. Within clinics, we identified that motivated and well-trained facility managers, and a willing primary health clinic staff complement have the resources and motivation required to co-ordinate tuberculosis diagnosis, maintain referral networks, interpret Xpert results and translate tuberculosis testing into improved outcomes. Drawing these findings together, we recognised that the XTEND trial did not measure certain factors that impacted on Xpert performance, including the degree of adherence to the algorithms for tuberculosis diagnosis, nor the rates of empiric tuberculosis treatment. The XTEND trial did not assess how tuberculosis diagnosis was implemented by health care workers. We showed that the provision of additional investigations to XTEND participants who were more ill decreased the mortality rates in both study arms. This may have led to a reduction in statistical power to detect a mortality difference between study arms. Further, we

viii

demonstrated higher rates of empiric treatment amongst persons in the smear microscopy arm, which may have decreased the mortality difference between study arms. Our qualitative work revealed high clinic-toclinic variation in the degree of integration of Xpert, which is unlikely to have introduced a systematic bias in mortality rates across study arm. In conclusion, we found that the Xpert roll-out did not impact patient outcomes in the XTEND trial because increased sensitivity of Xpert and increased yield was offset by additional efforts to diagnose tuberculosis in participants who were at risk of death, and the more frequent provision of empiric TB treatment in the XTEND smear microscopy study arm. While Xpert may not have affected patient outcomes, a plethora of new diagnostic tests are under evaluation. The findings from this PhD highlight important methodological considerations in the design and conduct of impact evaluations of new tuberculosis diagnostic tests.