Treatment decisions and mortality in HIV-positive presumptive smear-negative TB in the Xpert® MTB/RIF era: a cohort study

Tuberculosis (TB) remains the main cause of death among HIV-infected patients in sub-Saharan Africa. Sputum smear microscopy has a low sensitivity for detecting TB particularly among people living with HIV. Because of its low cost, it is the mainstay of diagnosis in many high burden countries. In December 2010 the World Health Organization (WHO) endorsed Xpert® MTB/RIF (Cepheid, Sunnyvale, California, USA) (XP) [1], the first fully automated, real-time nucleic acid amplification technology for the rapid detection of TB [1, 2] which can be performed with only 1 day of training by most health care workers. A WHO policy statement came out in 2011 recommending its use as the initial diagnostic test among people suspected of multi-drug resistant (MDR) or HIV-associated TB [2]. WHO policy updates in October 2013 recommending broad use of Xpert for TB diagnosis [3] and concessionary pricing for high-burden countries have resulted in widespread roll-out in routine care settings [4].

Because of a higher sensitivity than sputum smear microscopy and a running time of two hours, routine use of XP was predicted to increase case finding and improve TB treatment outcomes [5–7]. However, this impact has not been realized in trials to date [8–11]. Some have argued that the impact may been blunted by high rates of empirical treatment in sub-Saharan Africa: up to 40% of TB cases are treated empirically due to diagnostic uncertainties and the risk of severe morbidity and mortality if TB treatment is delayed [12]. It is therefore still unclear what impact XP will have on case finding and TB treatment outcomes in routine care settings [13]. To date no studies have investigated the impact of XP on health care workers’ pre- and post-test management [13].

In an integrated TB-HIV clinic with adequate staffing and training in Kampala, Uganda, we implemented XP for use as an add-on test in sputum smear negative presumptive TB patients. We investigated if and how clinicians changed their diagnostic decision-making on the basis of the additional test available to them. Considering the high level of experience of the clinicians, we hypothesized that this would be limited. We analyzed XP usage and yield compared to empirical diagnosis without access to XP, and examined treatment decisions subsequent to an XP result. We also compared mortality in the first year after being investigated for TB between patients who were treated empirically, treated on the basis of microbiological confirmation, or not treated at all.

This study investigated the impact of XP usage on the diagnosis, treatment and mortality of HIV- positive smear-negative presumptive TB patients. Despite current WHO guidelines recommending the use of XP as the first diagnostic test rather than as an add-on to a negative sputum smear [3], this study offers insights into clinicians’ diagnostic and treatment decisions with the availability of XP and the associated treatment outcomes. We found that XP was only used in 43% of eligible patients despite free access to the test. The higher likelihood of an X-ray being ordered in this same group of patients suggests clinicians had a higher level of clinical suspicion compared to those not sent for XP. High levels of empirical treatment prevailed although XP testing did change who was ultimately treated for TB. A pre-XP empirical diagnosis by an experienced clinician did not predict XP positivity.

Although XP availability led to little change in the total proportion of presumptive TB patients actually treated for TB, the use of XP led to treatment of those who would not have been treated pre-XP and vice versa. Despite the clinicians’ experience, XP positivity was not more likely among patients with than without a pre-XP empirical diagnosis. This corroborates previous work on the poor accuracy of empirical diagnoses (using sputum culture as gold standard) in this clinic and in Tanzania [16, 17]. Despite having access to all clinical data, we were not able to determine which clinical factors most influenced clinicians in their decision-making process pre-XP. Among hospitalized patients, a recent study in Kampala identified productive cough, fever and tachycardia as predictors of empirical treatment [18]. The lack of sputum culture in our clinic precludes the determination whether empirical treatment was true- or false-positive (both pre- and post-XP).

Our data suggests that clinicians used the XP test as a “rule-in” test and only partly as a “rule-out” test. This is justifiable as the test is known to have low sensitivity as an add-on test to a negative sputum smear (61% in a HIV-positive population) [19, 20]. Nevertheless, the clinicians decided against treatment in just over half of the patients whom they would have treated without access to an XP, which may indicate an overestimated confidence in a negative XP. This potential false sense of security has also been described during roll-out of routine XP as the first-line diagnostic in South Africa [21]. This is likely to be exaggerated in rural areas where more mid-level health care workers are present with more limited TB diagnostic training [22]. In future, this risk might be mitigated by use of the next generation Xpert cartridge, the Xpert MTB/RIF Ultra, which has been shown to have a higher sensitivity [23].

Low XP utilisation has also been described in a study evaluating its use in 18 health facilities throughout Uganda (21%) [24], as well as in studies from the Democratic Republic of Congo (37%) and Swaziland (51%) [25, 26]. It has been suggested that the varying degrees of utilisation may be associated with the level of training and support provided to health care workers [24]. It may also take time for clinicians to get used to the new test. This has been shown in Swaziland where the utilisation increased to 73% in the second and third years after implementation, and in Cape Town, where the impact on empirical treatment reached its maximum in the third year after implementation [26, 27].

The lack of association between post-XP empirical treatment or microbiologically confirmed treatment and the risk of death was also found by three recent studies among both ambulant and hospitalized ART naïve patients [18, 28–30]. However, a recent systematic review on the impact of tuberculosis nucleic amplification tests, such as Xpert® MTB/RIF, concluded that the evidence has been of limited scientific rigour and from a relatively small number of settings, limiting its validity and generalisability [13]. We found that CD4 count confounded the negative association between empirical treatment and mortality. It is well known that a low CD4 count is a strong risk factor for mortality, and is likely to be associated with being treated empirically for TB [12]. Our study design was also not optimal to investigate this association, however, and we may not have captured potential confounding factors of the relationship between empirical treatment and mortality, including anemia, co-morbidities such as malignancies, and others.

The main limitation of this study was the risk of misclassification bias if clinicians failed to answer the clinical decision question truthfully due to misunderstanding of its purpose (whether it might affect the procedure in the laboratory or be used to review their performance). This was minimized with continued training and regular checks of their understanding. More than half of eligible patients were not sent for an XP; their higher probability of an EPTB diagnosis suggests a higher clinical suspicion of extrapulmonary involvement on the basis of additional information not captured in our data.

XP usage to aid the diagnosis of smear-negative TB by experienced clinicians was lower than expected. In those tested, experienced clinicians mainly used the XP as a “rule-in” test and only partly as a “rule-out” test. The lower sensitivity of XP in smear-negative HIV-positive patients led to the majority of patients still being treated empirically.