Ethical approval was obtained from the Eswatini National Health Research Board (formerly Scientific and Ethics Committee [Approval reference: MH/599C]), CDC Institutional Review Board (IRB) (Reference: CGH-HSR Tracking #: 2015-196), and University Research Co. LLC IRB (02 March 2015). Participants signed informed consent written in their preferred language (SiSwati or English). Participants received no incentives but were reimbursed transport costs for additional visits to read TST. Diagnostics tests and treatment, if required, were free of out-of-pocket charges.

We conducted a cross-sectional study enrolling pregnant and postpartum women, aged 18 years and older, attending antenatal and postnatal care clinics, from 01 April to 30 November 2015 at three public health facilities in three of the four regions of Eswatini. Sociodemographic and clinical data, including past tuberculosis screening results where applicable, were collected. Eswatini is categorised by WHO as a high tuberculosis/HIV burdened country with a co-infection rate of 70% and a tuberculosis incidence rate of 398 per 100 000 population.²⁷

Participants were pregnant and postpartum women who were not on anti-tuberculosis treatment at enrolment or who had not taken anti-tuberculosis medicines, including isoniazid for tuberculosis preventive therapy, within the 2 months preceding enrolment, based on documented evidence from patient clinical records, and who provided informed consent. Four groups of women were enrolled: HIV-positive pregnant, HIV-negative pregnant, HIV-positive postpartum, and HIV-negative postpartum.

A sample size of 183 in each group was determined and a full narrative of sample size calculation is fully described in our protocol paper titled ‘Screening in Maternity to Ascertain Tuberculosis Status (SMATS) study’.⁴ Participants were consecutively enrolled until the sample size was reached.

Of the 990 women who were enrolled, 516 (52%) were pregnant: 101 (10%) in first trimester, 219 (22%) in second trimester and 196 (20%) in third trimester (Figure 2). The remaining 474 (48%) were all postpartum women. Among the participants, 470 (47%) were HIV-positive and, among them, 434 (92%) were on antiretroviral therapy.

Most women (790, 80%) had secondary or tertiary education, and 300 (30%) were in formal employment (Table 1). The median household density (number of people per room within household) was 1.5 (interquartile range: 1–2 people per room). The median age was 26 (interquartile range: 22–31) years.

Of the 990 participants screened, 48 (5%) screened positive for tuberculosis using the NTBSS tool. There were 181 (18%) with at least one tuberculosis symptom (Table 1). Participants who screened positive for specific tuberculosis symptoms included 103 (10%) with a cough of any duration, 38 (4%) with night sweats, 29 (3%) with fever and 48 (5%) with weight loss. Among 516 pregnant women, 433 (84%) were screened for tuberculosis at their last clinical encounter compared with 283 (60%) of the 474 postpartum women. In 531 (54%) patients, sputum collection was spontaneous, 158 (16%) by induction and 87 (9%) as early morning samples. However, 214 (22%) participants failed to produce sputum using any of the three methods. In total, 776 (78%) participants produced sputum samples and 758 (98%) had samples available for culture testing and 704 (93%) had valid culture results. However, only 361 (47%) had sputum samples available for Xpert^® MTB/RIF testing and Xpert^® MTB/RIF results were available for 361 (100%) participants, although 18 (5%) of those with Xpert^® MTB/RIF results had no culture results for comparison.

Tuberculosis status was either bacteriologically confirmed or ruled out by culture testing in 704 (93%) participants who had sputum samples available for culture. The overall prevalence of tuberculosis in this cohort of pregnant and postpartum women was 2% (95% CI: 1–3) (Table 2). M. tuberculosis was found in 15 (2%) participants and 12 (80%) were pregnant. The prevalence of tuberculosis among pregnant women was 3% (95% CI: 1–5) compared to 1% (95% CI: 0–2) in postpartum women. The prevalence of tuberculosis among those who were HIV-negative was 1% (95% CI: 0–2) compared to a prevalence of 3% (95% CI: 1–5) among those who were HIV-positive (Table 2).

A total of 450 participants chose between TST and IGRA as their preferred test method. Of this number, 277 (62%) chose TST as the preferred method compared to 173 (38%) for IGRA (data not shown in tables). Among the 540 with no decision on preferred method, 512 (95%) were not sure, and 28 (5%) chose not to respond to the question. Tuberculin skin tests were done on 961 (97%) participants, of which 659 (69%) came back for reading within 48 h – 78 h, whereas 302 (31%) were lost to follow-up. Of the 659 participants who had TST results, 237 (36%) were TST-positive. IGRAs were done on 465 HIV-positive participants. Of the 465 IGRA results, 153 (33%) were positive, 273 (59%) negative, 23 (5%) indeterminate and 16 (3%) were missing.

The standard NTBSS tool failed to identify women with tuberculosis disease with a sensitivity of 0% (95% CI: 0–29) among HIV-positive and 0% (90% CI: 0–60) among HIV-negative participants (Table 3). The enhanced screening tool, by including history of contact with a person with tuberculosis or presence of tuberculosis symptoms within the household of the participant, did not improve the sensitivity of the standard NTBSS tool. Only the inclusion of a tuberculosis contact history significantly improved the sensitivity of the algorithm to 18% among HIV-positive women, although specificity decreased from 94% to 62%.

M. tuberculosis was detected in two (1%) of the 361 Xpert MTB/RIF results, and no rifampicin resistance was detected in either one. No M. tuberculosis was detected in the remaining 359 (99%) samples. Sputum smear microscopy was done on 724 (73%) participants, and 4 (1%) had acid-alcohol-fast bacilli. LF-LAM was done on 411 (87%) participants and 327 (80%) had culture results. When compared to culture, the sensitivity of both the tuberculosis screening and the tuberculosis diagnostic tests were less than 50% in both HIV-positive and HIV-negative women (Table 3).

Specific algorithms could not be analysed due to the missing data of tests used to construct the different algorithms, as well as low prevalence of tuberculosis symptoms among our study participants. The study showed that those who were HIV-positive had a threefold risk of culture-positive tuberculosis compared to HIV-negative individuals (odds ratio = 3.23; 95% CI: 1.00–10.40, p = 0.05) (Table 4). Those residing in the Central region (odds ratio = 0.08; 95% CI: 0.01–0.61, p = 0.015) and Southern region (odds radio = 0.19; 95% CI: 0.04–0.89, p = 0.035) were independently less likely to have culture-positive tuberculosis compared to the northern region (reference region) and (Table 4).

According to our review, this is a unique study in this setting to determine the burden of tuberculosis among pregnant and postpartum women regardless of the presence of tuberculosis symptoms. We observed that higher proportions of pregnant women (84%) were previously screened for tuberculosis during their last clinic visit prior to enrolment compared to postpartum women (60%). However, these were lower than the universal screening (99%) reported among people living with HIV attending antiretroviral therapy clinics in Eswatini.³⁵

Even though 80% of participants who were confirmed to have tuberculosis disease were pregnant, there were no statistical differences between the prevalence of tuberculosis in pregnant and postpartum women. The highest prevalence of tuberculosis was among HIV-positive pregnant women (5%), which is comparable to the 3.3% observed in neighbouring South Africa.¹¹ Although 93% of participants who were found to have active tuberculosis reported no tuberculosis symptoms, there were no differences in tuberculosis prevalence between those reporting symptoms and those with no symptoms. A study conducted in South Africa found a higher tuberculosis prevalence among patients who did not report symptoms of tuberculosis.¹⁸ A study from Ethiopia did not find any person with active tuberculosis disease among pregnant women but did not test for tuberculosis among those who did not have tuberculosis symptoms.³⁶ This has significant public health implications for tuberculosis control, considering previous reports that asymptomatic patients with culture-positive tuberculosis can transmit tuberculosis.¹⁸

The WHO four-symptom NTBSS screening tool failed to identify women with active tuberculosis disease, as the majority of women with confirmed tuberculosis disease did not have symptoms of tuberculosis. Almost a third of participants who had a TST done did not have results, because participants did not come back within the stipulated time for reading, despite the provision of transport imbursements for additional visits for TST reading.

Pregnancy is known to suppress the T1-helper pro-inflammatory response, resulting in masking of tuberculosis symptoms and increased susceptibility to M. tuberculosis reactivation⁸ and primary infection with M. tuberculosis.³⁷ Women are unlikely to show typical symptoms like sweating at night and fever, and these are further masked by pregnancy.^8,17 Weight loss can be masked by physiological changes during pregnancy.¹⁸ The low sensitivity of the tuberculosis screening tool has been reported in many other studies.^18,19,38,39 Adapting the screening tool to include a history of tuberculosis contact as an independent indicator of positive tuberculosis screen improved sensitivity by 18% but only in HIV-positive women. This sensitivity (18%) is still too low for effective tuberculosis case finding and ruling out tuberculosis among pregnant and postpartum women. LaCourse et al.¹⁹ also showed poor performance of the four-symptom screening tool in HIV-positive pregnant women in Kenya. Tuberculosis diagnosis in pregnancy is often delayed due to atypical symptoms.^11,39,40 Culture prevailed as the reliable gold standard to diagnose tuberculosis; some authors recommend that it should be mandatory for establishing tuberculosis diagnosis in this group.^11,39

A false-negative symptom screening, which is the initial screening method for triaging for tuberculosis diagnostic testing, will often lead to a delay in diagnosis and treatment of active tuberculosis and consequently poor foetal and maternal outcomes.¹⁷ A false-positive screening result leads to inconvenient and costly laboratory procedures.¹⁷ However, the routine symptom screening tool had high negative predictive values (97% – 99%), which supports its utility in identifying people who are unlikely to have tuberculosis, especially people living with HIV. Therefore, those who screen negative with the standard NTBSS tool and are at high risk of progressing from latent tuberculosis to active tuberculosis can be given tuberculosis preventive therapy in this setting of high HIV-tuberculosis burden.^39,41

We demonstrated that TST had poor performance when used as a screening method for exposure to tuberculosis and its feasibility is further challenged by a third of participants who were lost to follow-up for a reading of skin reaction. However, alternative follow-up methods, including home visits, should be included to complete the TST readings. IGRA was a less preferred screening method compared to TST, possibly due to the need for blood draw. In addition, IGRA demands laboratory infrastructure⁴² compared to TST and is currently available only in private laboratories in this setting.

We considered it a strength that our methodology included participants who did not have the usual symptoms of tuberculosis, minimising the risk of under-reporting of true tuberculosis prevalence in this study population.^39,43 We also induced sputum in women unable to produce sputum spontaneously, thus maximising the tuberculosis diagnostic yield.¹⁹ In addition to symptom screening, we attempted five tuberculosis tests for all participants.

We had planned to test 10 different algorithms using different testing combinations and ordering of individual diagnostic tests. However, given the low agreement, the low number of positives, and the poor performance of the WHO symptom screening tool (several of the algorithms began with the WHO symptom screening), the analysis was of no utility, and we did not include these results in this report. Although unintended by the study design, more than half of the participants did not have an Xpert^® MTB/RIF assay done, which is a near point-of-care test that allows results to be quickly available (as early as 2 h) to clients and service providers, leading to a quick clinical management decision, unlike culture which may take several weeks.

The four-symptom screening tool appears likely to miss women with active tuberculosis. Without sensitive, symptom-based tuberculosis screening in this subpopulation, a high index of suspicion of tuberculosis is necessary and factors such as a history of tuberculosis contact should prompt clinicians to consider tuberculosis in their differential diagnosis, especially in a setting of high HIV- tuberculosis burden. Bold, deliberate decisions to invest in laboratory-based, quality-assured culture testing are required to maximise detection of people who have active tuberculosis disease in countries with a high HIV- tuberculosis burden in order to end tuberculosis by 2035 as envisaged by the WHO’s End TB Strategy.⁴⁴ However, the feasibility of increasing access to culture in this setting is confronted by costs related to culture testing, transportation, specimen storage and lengthy waiting periods for culture results by clinicians and patients. Therefore, low-income and middle-income countries should strike the right balance to ensure access to culture for those who could benefit from culture and availability of the newer, more sensitive Xpert^® MTB/RIF Ultra platform for rapid tuberculosis diagnosis.

Prevalence of tuberculosis was particularly high (4.5%) among HIV-positive pregnant women and low (0.6%) among HIV-negative postpartum women. Although we were unable to test different tuberculosis screening algorithms, the poor performance of the standard NTBSS tool that serves as the entry to tuberculosis services highlights the challenge of diagnosing tuberculosis in pregnant and postpartum women.