Evidence for the Use of Triage, Respiratory Isolation, and Effective Treatment to Reduce the Transmission of Mycobacterium Tuberculosis in Healthcare Settings: A Systematic Review

Abstract Evidence is limited for infection prevention and control (IPC) measures reducing Mycobacterium tuberculosis (MTB) transmission in health facilities. This systematic review, 1 of 7 commissioned by the World Health Organization to inform the 2019 update of global tuberculosis (TB) IPC guidelines, asked: do triage and/or isolation and/or effective treatment of TB disease reduce MTB transmission in healthcare settings? Of 25 included articles, 19 reported latent TB infection (LTBI) incidence in healthcare workers (HCWs; absolute risk reductions 1%–21%); 5 reported TB disease incidence in HCWs (no/slight [high TB burden] or moderate [low burden] reduction) and 2 in human immunodeficiency virus-positive in-patients (6%–29% reduction). In total, 23/25 studies implemented multiple IPC measures; effects of individual measures could not be disaggregated. Packages of IPC measures appeared to reduce MTB transmission, but evidence for effectiveness of triage, isolation, or effective treatment, alone or in combination, was indirect and low quality. Harmonizing study designs and reporting frameworks will permit formal data syntheses and facilitate policy making.

Tuberculosis (TB) is the leading infectious cause of death worldwide [1,2]. Healthcare workers (HCWs) are at higher risk of TB than the general population, likely because of exposure in health facilities [3][4][5][6][7]. Infection prevention and control (IPC) measures to reduce Mycobacterium tuberculosis (MTB) transmission in healthcare settings are considered under 3 categories: environmental controls (eg, mechanical ventilation), personal protection (eg, using respirators), and administrative controls (eg, coordinating efforts between governmental health departments) [8]. Evidence is limited, however, for the effectiveness of individual IPC measures in reducing MTB transmission, and guidelines have been written based heavily on expert opinion [9][10][11].
This systematic review was 1 of 7 complementary reviews commissioned by the World Health Organization (WHO) to inform the update of the 2009 TB IPC guidelines [12]. It aimed to answer the question: do (1) triage of people with TB signs, symptoms or with confirmed TB disease; and/or (2) respiratory isolation of presumed or demonstrated infectious TB cases; and/or (3) effective treatment of TB disease reduce the transmission of MTB to HCWs or other populations (including patients and visitors) in healthcare settings, when compared with transmission to the same populations in settings without, or with different, IPC interventions? The primary findings of this review were presented to the WHO guideline development group (GDG) and collated in an online appendix to the 2019 guidelines [13]. The guidelines contain recommendations for practice based on consideration of a wide range of evidence and should be the primary resource for implementation; this article looks more closely at how these interventions have been studied and discusses the implications for future TB IPC research.

METHODS
The review protocol was registered on 12 February 2018 on the International Prospective Register of Systematic Reviews (ref. 1. HCWs (including CHWs) working in health care settings; or 2. Any study that did not report any of the above-stated outcomes of interest.
2. Other staff working in a health care setting; or 3. Any study reporting solely on primary outcomes of interest without a control or comparator group. 3. Persons of all ages (patients and visitors) attending health care settings, anywhere in the world. 4. Any systematic review superseded by an updated systematic review.
Types of intervention At least 1 of the following: 5. Narrative reviews not adding new data or new analysis of data to existing knowledge.
1. Triage of people with TB signs or TB symptoms or confirmed TB; 6. Commentaries and mathematical modelling studies.
2. Respiratory isolation (spatial separation) of presumed infectious TB cases; or 7. Studies with fewer than 10 participants per comparator arm.
3. Effective treatment of TB based on bacteriologic susceptibility. 8. Any study not written in English, Japanese, Chinese, Russian, French, Spanish or Portuguese.
Types of comparator Studies reporting data (for outcomes of interest) from a control or comparator group of HCWs (including CHWs) working in health care settings, or other staff or persons of all ages (patients and visitors) attending health care settings, with no or different administrative infection control interventions. 9. Any study published before 1946.
Types of outcome measures Studies reporting data on at least 1 of the outcome measures of interest (incidence/prevalence of LTBI or TB disease).
Types of study Any consecutive case series, case control study, cohort study, randomized controlled study, systematic review, or metaanalysis.
reports (Table 2) and 19 systematic reviews (Supplementary Table 3). Six TB IPC guidelines were also reviewed for possible primary research articles (Supplementary 3). Of the 25 studies, 17 (68%) were conducted in North America, 3 (12%) in sub-Saharan Africa, 2 (8%) each in Europe and Latin America, and 1 (4%) in East Asia; 19 (76%) were conducted in low TB burden (all high-income countries) and 6 (24%) in high TB burden settings (5 upper middle-and 1 low-income country); and 24 (96%) were conducted in hospitals and 1 (4%) in primary care facilities. Only 2 (8%) studies reported outcomes in non-HCWs attending healthcare facilities; in both cases these were human immunodeficiency virus (HIV)-positive in-patients. Nineteen (76%) studies described LTBI incidence, and 7 (28%) described TB disease incidence (1 described LTBI and TB disease incidence). Sixteen (64%) of 25 studies implemented interventions of interest in combination: 11 triage and isolation; 2 isolation and effective treatment; and 3 triage, isolation, and effective treatment ( Figure 2A). Of the remainder, 8 (32%) studies assessed isolation alone, and 1 (4%) assessed triage alone. An obstacle to the evaluation of the 3 IPC interventions of interest was the paucity of studies that introduced only these interventions: all studies, except 2 [21,22], implemented any or all of the 3 interventions as part of a wider suite of measures, including personal protective equipment (PPE) for HCWs; changes to ventilation and other environmental controls; and broader administrative controls (Table 3; Figure 2B). Disaggregation of the effects of individual measures was not possible, and it was therefore not feasible to attribute the entire reported effect on outcomes to a single intervention or to estimate the proportion of a demonstrated effect that could be attributed to the intervention (whether 1, 2, or all 3 elements of interest).

Studies Implementing Triage of People With TB Signs and/or Symptoms
Fifteen studies implemented triage: 11 (73%) in low burden settings, all in secondary or tertiary health facilities. Definitions of triage varied widely, from screening of patients "with pneumonia or evidence of TB" [35] to an "expanded respiratory isolation policy" [37,41]. The only study to use triage alone used "routine chest x-ray screening for all new admissions" [21]. Among 10 studies reporting changes in LTBI incidence (all implemented composite interventions), estimates of effect ranged from an absolute reduction of 2.3% (n = 21 197) [41] to 20.5% (n = 65; Table 4) [48]. Of the 4 studies reporting incidence rates (IRs) [28,32,37,38], IR ratios ranged from 0.18 to 0.9 (unadjusted; some calculated).

Studies Implementing Effective Treatment Based on Drug Susceptibility
Five studies [31,[45][46][47][48] used effective treatment with other IPC measures. "Effective treatment" was defined variably, from a change in regimen from 3 to 4 drugs [48], to the use of "radiometric susceptibility testing, " [31] which, it was assumed, would have led to appropriate treatment, though this is not stated.
Two studies did not report outcomes for all participants/sites [31,45]. All studies showed absolute reductions in TST conversion after implementation of IPC measures, ranging from 2.1% [46] to 20.5% [48] (crude; calculated), although all studies had small numbers of outcomes (range 10-104) and 2 had small sample sizes (n ≤ 650).
Only 1 study [47] used effective treatment and measured TB disease incidence, employing an "expanded anti-TB regimen" (a change from median 1.5 [range 0-4] to 2.0 [range 0-4] drugs) as part of a composite intervention that included triage, isolation, and changes to diagnostic processes. They found a change in TB disease risk (or "attack rate") among HIV-positive individuals admitted to the ward, from 8.8% before, to 2.6% after intervention (P = .01).

Quality Assessment and GRADE
All 18 retrospective studies scored poorly (median 10/27 [interquartile range 8.3-12.0; range 6-13]; Table 5). The 7 prospective studies also scored poorly ( Table 6): 1 study was marked down for incomplete outcome reporting and 3 for selective outcome reporting. The overall low study quality was reflected in the GRADE assessment (Appendix 3; Supplementary Tables 4-9), where the strength of evidence was consistently downgraded due to serious risk of bias and very serious indirectness (the latter often due to the concurrent use of multiple IPC measures).

DISCUSSION
This review found 25 studies, published from 1957 to 2017, that reported the effects of triage, isolation, and effective treatment on the incidence of LTBI or TB disease in HCWs and others attending health facilities. Most studies were conducted in the 1990s in US hospitals, several in response to outbreaks of TB [22,34,[45][46][47][48]. Almost all studies showed reduced LTBI or TB disease incidence after implementation of a package of IPC measures, but because of heterogeneity in study design and reporting of results, meta-analysis was not conducted. Studies were generally of low quality. All studies, except 2, tested composite interventions, including other administrative measures, PPE, and environmental measures; it was therefore not possible to disaggregate the effects of specific interventions from those of the others described.
It is important that these findings should not be interpreted to suggest a lack of efficacy of the TB IPC measures examined; effective treatment, in particular, is supported by studies outside health care settings [50] and studies from healthcare settings using guinea pigs as infection endpoints [13]. Indeed, the WHO 2019 TB IPC guidelines recommend all 3 examined measures as first-line controls to be used as part of broad suite of interventions [12]. Statements, in the guidelines, around "low certainty" and "indirectness" reflect the overall poor study quality, heterogeneity in study design, implementation of multiple interventions at 1 time, predominance of studies from a particular type of setting, and deficiencies in the reporting of results. These issues are discussed below.

Gaps in the Literature
Most studies were from high-income, low TB burden settings, predominantly the United States. Conspicuously absent were countries with very high TB burdens, such as India and China, and countries in sub-Saharan Africa and South or Central America (other than South Africa, Malawi, and Brazil), where the LTBI burden among HCWs is known to be very high [4,7]. Data from these countries are essential if global policy is to address successfully the broad range of environments in which IPC measures must be implemented. Only 1 study was conducted in a primary care setting [25]. Although many people with TB in low burden countries may receive treatment in hospitals, most in high burden countries are cared for as out-patients and may not visit a hospital at any point in their illness [51]. WHO widely recommends the decentralization of TB care [52,53], although for DR-TB this policy is variably effected [2]. As shown in South Africa [54][55][56][57], HCWs in clinics and the community are also at high risk of TB infection and disease. Evidence is still needed for the effectiveness of IPC measures in these environments, which present different challenges for implementing interventions and measuring outcomes [58,59].
Many studies provided detailed descriptions of interventions used, but often did not describe, in any depth, fidelity to these interventions. Cross-sectional studies [24,25,40,42] were the weakest in this regard, as they were able only to assess whether an intervention or policy had been instated and not if it was being applied as intended. (Additional methodological shortcomings in some cross-sectional studies further reduced confidence in their findings; for example, the study by Claassens et al [25], where IPC coverage was estimated after the period during which outcomes were enumerated.) Some reporting of fidelity is essential to strengthen what is already very indirect evidence for the effectiveness of these interventions.
A consistent finding was the lack of reporting of secular changes in TB incidence or prevalence among people attending the facility over the course of the study. This is particularly relevant given the high number of before-after or during-after  [19] and should be a standard reporting requirement for future studies.
Incidence of LTBI or TB disease in HCWs are useful ways to estimate MTB transmission from patients in health care settings. Transmission between patients and from HCWs to patients does, of course, occur, although this was measured by only 2 studies, both in low TB burden, high-income countries [34,47]. Choice of at-risk population, outcomes, and outcome measurement are critical when studying MTB transmission, but can also make study design more complex [60]. In high TB burden countries, a high proportion of HCWs already have LTBI, limiting the size of the at-risk population. Using TST to measure LTBI incidence (as in several of the included studies) can also be problematic, as reactions can vary based on host factors. The development of TB disease, though easier to measure, is also dependent on a number of interconnected host factors and, in the absence of complementary molecular epidemiological data, is more difficult to reliably attribute to a congregate setting transmission event. More detailed descriptions of at-risk HCW populations would allow for better extrapolation of findings to other key populations, particularly HIV-positive individuals, and provide better guidance on how to prevent TB in HCWs. As discussed by Harries et al [29], robust occupational health programs are critical to the well-being of frontline HCWs; embedding TB IPC studies within existing occupational health frameworks may allow for better reporting of individual HCW risk profiles and improve long-term fidelity to interventions.

Future Research
This review, like others [4,6,7,11,61], found limited and low quality evidence for the effectiveness of administrative IPC measures in reducing MTB transmission, with overrepresentation of data from hospitals in high-income, low TB burden countries. Like previous reviewers, we call for better designed and implemented studies from a wider variety of settings, although we acknowledge the difficulties of doing this in what are often unpredictable environments, and recognize the shortcomings in the methods available to measure MTB transmission in these settings [62,63].
Despite the weaknesses in the data presented here, the weight of evidence to support the use of established TB IPC measures is sufficient that it would be unethical to conduct randomized trials involving a true "control" arm, although trials comparing "best practice" IPC interventions with an established basic standard of care should still be considered, as should the use of pragmatic trial designs, such as stepped wedge cluster randomized trials [64]. We would suggest a change in expectations and an acceptance of the limitations inherent in conducting these complex interventional studies in challenging clinical settings. Standardization of study designs, outcome measurement, and reporting formats, with replication of clusters or sites would facilitate the generation of more robust data syntheses to guide policy making, as would efforts by investigators to provide more precise and comprehensive data in the areas discussed above. We suggest that greater numbers of imperfect but comparable data from studies conducted in a wide range of settings that adhere to a set of standardized rules around design, and reporting would be more useful to decision making than a few perfectly designed studies conducted in places unrepresentative of those where effective interventions are most needed. Additionally, quasi-experimental techniques, such as interrupted timeseries analysis [65][66][67], with or without controls [68], or difference of differences approaches have been employed with success in evaluating complex public health policy interventions in rapidly changing environments [69], and should be considered seriously for future real-world estimations of the effectiveness of TB IPC measures. To this end, given the difficulties outlined above around measurement of outcomes, confounding, bundling of interventions, and valid comparator groups, it would be beneficial to have additional specific guidance, developed by relevant experts, to help investigators plan, conduct, and report studies examining the efficacy of measures to reduce MTB transmission.

Limitations and Strengths
"Prompt initiation of effective treatment" is widely considered a reliable way to reduce MTB transmission, and is the wording used in the WHO 2019 TB IPC guidelines [12]. The 5 studies included in this review that used effective treatment did not report time to treatment, and because "prompt initiation of effective treatment" was not one of the defined interventions of interest, studies examining its efficacy in reducing transmission were not included for analysis.
Heterogeneity of the data and weaknesses in study design prevented meaningful quantitative synthesis, which may have provided a clearer guide for policy makers. Studies may have been overlooked during sifting or published in nonspecified languages. Strengths include the application of a robust search strategy by a professional librarian across a wide range of repositories, all sifting and data extraction being done in duplicate (per PRISMA recommendations; Appendix 4) [70], and the use of GRADE to assess quality. This review found 25 studies implementing triage, isolation, or effective treatment, and measuring the incidence of LTBI or TB disease or both. Overall, packages of IPC measures appeared to reduce MTB transmission, but studies were of low quality and evidence for the effectiveness of individual or combined measures was indirect and of limited utility; heterogeneity of the data prevented meta-analysis. More data are needed from highburden, lower-income, primary care settings. Harmonization of study designs and reporting frameworks will allow for more formal data syntheses, creating a better platform for policy making. The development of specific guidance around conducting and reporting studies to determine the efficacy of TB IPC measures should be prioritized by governing and stakeholder bodies.

Supplementary Data
Supplementary materials are available at Clinical Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.