People living with HIV (PLWH) are at increased risk of developing and dying from tuberculosis (TB). In areas where HIV-associated TB is prevalent, routine screening for TB among PLWH is recommended in order to detect cases earlier, and to exclude TB prior to initiating isoniazid preventive therapy (IPT). The latter two objectives are anticipated to result in a decrease of TB-related morbidity and mortality. The WHO recommends using a symptom-based questionnaire for TB screening among PLWH in resource-constrained settings [1]. The questionnaire categorizes patients into two groups: asymptomatic individuals in whom IPT can be initiated, and symptomatic individuals (i.e., TB suspects) requiring further evaluation to exclude TB. The recommended questionnaire was selected based on a meta-analysis of screening studies in adult PLWH [2].

In high TB and HIV burden settings, HIV-associated TB remains a significant cause of morbidity and mortality even among persons on antiretroviral therapy (ART) [3,4]. However, few data from adults on ART were included in the meta-analysis which motivated the selection of the recommended screening questionnaire [2]. Since the meta-analysis was published, two reports from South Africa have described the performance of symptom-based screening among adults on ART and arrived at disparate conclusions [5,6]. In the larger of the two studies (N = 1429), diagnostic performance was inferior among participants on ART compared with those not taking ART [5], whereas in the smaller study (N = 422), symptom-based screening was highly sensitive regardless of ART status [6]. Importantly, neither study evaluated screening strategies that incorporated chest radiography (CXR) in the screening process.

In South Africa, where both HIV and TB are highly prevalent, the number of PLWH receiving ART has increased substantially over the past decade (75% in the last 2 years) [7]. Increases in ART uptake have also been achieved elsewhere [7]. As more PLWH are receiving ART, it is critically important to define the TB screening questionnaire's performance among PLWH on ART.

Our primary objective was to compare the performance of the WHO-recommended symptom-based screening questionnaire for the diagnosis of pulmonary TB (PTB) in adult PLWH on ART and those not taking ART at two HIV clinics in South Africa. As a secondary objective, we sought to evaluate diagnostic performance of a screening strategy utilizing CXR and the questionnaire.

We enrolled 825 participants. While enrollment was ongoing and prior to analysis, data from the first 73 enrolled at one of the sites were removed because of concerns about data integrity. Fifteen participants were excluded because sputa were submitted more than 14 days after enrollment. Supplement Figure S1, http://links.lww.com/QAD/A511 describes recruitment, enrollment, and exclusion.

In this study of TB screening among adult PLWH in South Africa, the diagnostic performance of the WHO-recommended symptom-based questionnaire was limited and varied by ART status. Among participants on ART, the questionnaire had low sensitivity and was not useful for differentiating those with and without PTB. Among participants not taking ART, the questionnaire was highly sensitive and diagnostically useful. Sensitivity was improved, particularly for those on ART, with incorporation of CXR in the screening strategy. An association between ART status and diagnostic performance was also observed after controlling for potential confounders, and in post-hoc analyses in which only culture-confirmed cases were categorized as PTB.

Our findings have major implications for the utility of the WHO-recommended TB screening questionnaire. In the meta-analysis used to derive the questionnaire, in which nearly all data came from PLWH not on ART, the authors reported important between-study heterogeneity in diagnostic performance [2]. The questionnaire's diagnostic performance among PLWH not on ART in our study is consistent with that reported in the meta-analysis [2], but conflicts with recent reports of poor sensitivity in the same population [10] and in antenatal clinics [11]. Between-study heterogeneity is now also evident for the questionnaire's performance among PLWH on ART. For culture-confirmed TB, the questionnaire's sensitivity was lowest in the study by Rangaka et al. (23.8%), followed by our study (60%), and highest in the study by Kufa et al. (100%). The converse was true for specificity, which was highest in the study by Rangaka et al. (94.4%), followed by our study (55.1%), and the study by Kufa et al. (18.1%). This diagnostic heterogeneity was seen despite similar settings and reported participant characteristics, raising the question of what factors (such as viral load, opportunistic infections, or language) could explain these differences, and underscoring the need for further research.

There are a number of possible explanations for the observed association between ART use and the questionnaire's diagnostic performance. First, it is possible that a greater proportion of TB cases on ART were in an early, and hence asymptomatic, phase of TB, compared with cases not on ART. Second, ART use may have affected the diagnostic performance by promoting immune reconstitution, thereby preventing other opportunistic illnesses (including extrapulmonary TB) that could cause a positive screen. The lower sensitivity and greater specificity of the questionnaire among those on ART support this hypothesis. Finally, sensitivity may have been lower among participants on ART if they were more likely to have been previously screened for TB than those not on ART [2,12,13]. However, such differential TB screening is unlikely, as all PLWH should have been screened for TB at enrollment into HIV care and at each routine clinic visit as per national guidelines.

It is possible the questionnaire's sensitivity is higher during the initial months of ART when immune reconstitution ‘unmasks’ TB that was prevalent, but asymptomatic prior to ART [14]. There were too few participants on ART for less than 3 months in our study (N = 43, 8.2%) to permit meaningful exploration of this hypothesis. Our findings are thus most generalizable to PLWH who have been on ART for more than 3 months. Further studies are needed to establish the questionnaire's diagnostic performance during the early phase of ART use.

To our knowledge, ours is the first study to report performance of CXR in TB screening among PLWH on ART. Because prevalence of abnormal findings on CXR was high irrespective of PTB diagnosis, incorporation of CXR into the screening strategy increased sensitivity and decreased specificity of TB screening for participants on ART. Similar results have been reported among ART-naive populations [10,12,15,16], with sensitivity reaching over 90% in three studies [10,15,16]. Further research is warranted to verify our findings, as we were unable to assess inter-reader variability.

In high HIV and TB burden settings, PLWH remain at increased risk of TB even after several years of ART [3]. In such settings, routine TB screening among PLWH is expected to lower morbidity and mortality through earlier case detection, and by expediting the exclusion of TB prior to IPT [1]. The questionnaire's poor sensitivity among participants on ART is concerning – it suggests the questionnaire is not useful for active case finding in this group. However, the questionnaire mostly failed to identify TB cases that were smear-negative culture-positive, and smear-positive culture-negative, the majority of cases. Some have questioned the clinical relevance of identifying these cases prior to symptom onset, arguing that a false-negative screen will result in poor treatment outcomes only if TB becomes substantially more extensive before it is diagnosed [17], a situation that could arise with infrequent follow-up or screening. However, even if initiation of PTB treatment in the early stages of disease as opposed to once disease is symptomatic does not improve treatment outcomes, it may be important for preventing transmission in household, community, and healthcare settings [18–20].

Fear of generating isoniazid-resistant TB is often cited as a reason for not prescribing IPT [21,22], although proponents of IPT note that a meta-analysis and recent cluster randomized trials have shown the risk of developing isoniazid-resistant TB following IPT is minimal [23–25]. However, none of the studies in the meta-analysis, nor the cluster randomized trials, relied solely on symptom-based screening to exclude TB prior to IPT initiation. In the meta-analysis, which pooled results from six studies of IPT in PLWH, five studies used both symptom-based screening and CXR to exclude TB prior to IPT, and two studies also used sputum smear and culture [23]. In the two cluster randomized trials, participants were screened with both symptoms and CXR prior to IPT [24,25]. With the use of a less sensitive tool such as the screening questionnaire, the risk of developing isoniazid-resistant TB may be higher in populations in whom TB is common (as a greater proportion of those starting IPT will already have TB).

In our study, if only the symptom-based questionnaire was used to exclude TB prior to IPT, then among persons receiving isoniazid, 3.0% on ART and 1.6% not on ART would be cases of asymptomatic TB exposed to isoniazid monotherapy, roughly the same proportions expected to benefit from nontuberculin skin test (TST)-targeted IPT [26,27]. Screening with tests more sensitive than the questionnaire alone, and targeting IPT to TST-positive patients [25,28], will help ensure the probability of benefit from IPT outweighs the risk of inadvertent isoniazid monotherapy among PLWH, both on and off ART.

Our study has a number of strengths. First, the study included a large number of participants, both on and off ART. Second, use of sputum induction is likely to have increased the proportion of participants with TB who were identified by smear status. Third, our study was conducted under programmatic conditions in a resource-constrained area and our results are generalizable to settings in which the questionnaire is used.

Our study also has limitations. First, if smear-positive culture-negative specimens were due to laboratory error rather than PTB, then classification of participants with these results as having PTB spuriously lowered the sensitivity among those taking ART. Second, 17.1% of participants did not have culture results, which could have resulted in missed cases of PTB. However, results in post-hoc analysis addressing both limitations were similar to the primary analysis. Third, the routinely collected data used for some participant characteristics were incomplete or may not have reflected values at the time of study enrollment. Fourth, we did not assess inter-reader reliability for CXR interpretation. Fifth, each hospital used a different method to read smears; however, the proportion of positive smears was not significantly different between the hospitals (data not shown) and all multivariable models adjusted for site. A final limitation is that we were unable to determine whether using the questionnaire as the sole method of TB screening increases the risk of poor outcomes.

We found the WHO-recommended TB screening questionnaire performs adequately among PLWH not taking ART, and poorly among those on ART. This has important implications as the population of PLWH on ART is growing rapidly and their TB risk remains elevated in high TB burden areas. Research is needed to better understand the determinants of diagnostic heterogeneity of symptom-based TB screening in PLWH, both within and between populations on and off ART. There is also a need for prospective implementation studies, which follow PLWH screened with the WHO-recommended questionnaire to determine subsequent TB incidence, drug resistance, and treatment outcomes. Finally, there remains a need to identify inexpensive TB screening strategies for PLWH in resource-limited settings with higher sensitivity and less diagnostic heterogeneity. The increasing availability of the Xpert MTB/RIF assay provides an opportunity to operationally explore its role in routine TB screening in resource-limited settings.

Our study also has implications for screening policies in settings in which TB is common among PLWH. To lower risks of exposing PLWH on ART with unrecognized TB to isoniazid monotherapy, it may be prudent to combine the screening questionnaire with CXR or perform sputum cultures regardless of symptom status. It may be safer to use one of these more sensitive screening strategies even among PLWH who are not on ART, particularly if IPT is not targeted to TST-positive persons. The resources required to implement such screening are unavailable in many places where TB is prevalent among PLWH. The limited effectiveness of symptom-based screening in our study, and results of an analysis demonstrating the cost–effectiveness of TB screening with GeneXpert or sputum culture regardless of symptoms in South Africa [29], support scale-up of TB diagnostic capacity in areas burdened by TB-HIV coinfection.

In settings where the questionnaire remains the sole method of TB screening prior to IPT initiation, implementing the WHO recommendation that PLWH be screened for TB during every follow-up visit regardless of IPT status [1], and performing drug susceptibility testing in all PLWH who develop TB on IPT, may help avoid important delays in TB treatment initiation and detection of drug-resistant TB among those with an initial negative screen.