Number Needed to Screen for Tuberculosis Disease Among Children: A Systematic Review Free

This review was part of a larger systematic review conducted on behalf of the WHO Guidelines Development Group on systematic screening for TB, which described the number needed to screen (NNS) to detect 1 active TB case for several different populations and risk groups and served as an update to a similar review conducted in 2013.^14–16  The goal of this review was to examine the evidence for screening children in these and in other pediatric populations to inform the recent guideline update for TB screening in children.¹⁶  This systematic review was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analysis guidelines.¹⁷ 

We searched 4 databases (PubMed, Embase, Scopus, and the Cochrane Library) for articles published from November 2010 to February 2020. Search terms and subject indexing vocabulary included a combination of “tuberculosis,” “mass screening,” and other terms related to mass chest x-ray (CXR) or radiography screening, prevalence surveys, and case finding; the search strategy for each database is described in more detail in Supplemental Table 3. Titles and abstracts were entered into Covidence. Duplicates were removed, and 2 reviewers independently screened the remaining entries, with adjudication performed by a third reviewer as needed. For the initial title and abstract review, the eligibility criteria were only that the publication be original research (ie, not a review, commentary, or author reply letter), suggest that ACF for TB (testing for TB among individuals not seeking clinical care for TB symptoms) had taken place, and be written in English, French, or Spanish.

Full texts were then assessed for eligibility independently by 2 reviewers, with adjudication by a third reviewer if the first 2 reviewers disagreed. Studies were excluded if (1) the paper did not report original data, (2) no ACF was conducted, (3) the number of persons screened was not reported, (4) outcomes of active and passive case finding were not disaggregated, (5) only an abstract was published, (6) the paper was published in a language other than English, French, or Spanish, (7) the paper presented duplicate data from another publication, or 8) the paper was unavailable.

For this analysis, articles were also excluded if the study population did not include children aged <15 years or the data could not be stratified by age. Studies that included children up to 18 years were included if the data could not be disaggregated to exclude children aged 16 to less than 18 years. The age range for each study is listed in Supplemental Table 4. We excluded papers if the study population was prescreened for symptoms or other indicators suggestive of TB unless the source population size was also reported. The primary screening strategies of interest (set by the WHO Guidelines Development Group) were symptom-based screening (including screening for cough alone, the adult WHO 4-symptom screen, or other combinations of symptom screening), CXR, Xpert, or nutritional status. We excluded any paper conducting ACF that did not use any of these strategies.

Two reviewers independently abstracted the following from each included paper, following a standard protocol: study characteristics and results, including study location, demographic information about the study population, screening strategy, case definition for TB, number screened, and number of TB cases identified by screening strategy. If multiple screening strategies were reported, each was abstracted separately. Discrepancies in full-text reviews and data abstraction were resolved by consensus and/or consultation with a third reviewer. We assessed risk of bias using a modified version of the AXIS tool for cross-sectional studies.¹⁸ 

The definitions and flow for the possible screening strategies are shown in Fig 1. We considered the study population to be the total number of eligible children (<15 years old or <18 if unable to disaggregate) screened for TB according to individual study criteria. If available, we used the total number screened, regardless of the availability of results. If only the number of people screened for whom results were available was reported, then that number was used as the study population. We defined primary screening strategies as those used for the entire study population. Because children aged <15 years with TB may have trouble producing sputum and tend to have low bacterial loads,¹⁹  which limits the use of sputa-based diagnostics in children, active TB was defined as any diagnosis of TB disease according to individual study criteria, including clinical diagnoses.

The primary outcome was the NNS to detect 1 case of active TB. We first calculated the prevalence of TB detected for a given screening strategy for each study. We then weighted the prevalence from each study by the population size (to account for substantial heterogeneity in the size of the population screened between studies) and estimated the weighted mean prevalence for each screening strategy (Fig 1); a study could contribute to multiple NNS calculations if it reported multiple screening strategies. We then indirectly estimated the weighted mean NNS for a given strategy and population using the inverse of the weighted prevalence. This method allowed for studies that detected 0 TB cases to contribute to the weighted NNS; these studies have an undefined NNS and would have been excluded if the weighted NNS was calculated directly, rather than by using the inverse of the prevalence. We reported the upper bound of the range for weighted mean NNS estimates including studies that detected no TB cases as undefined.

We estimated NNS for children stratified by screening modality, HIV status, location of screening, and WHO estimated country-level TB incidence (low/moderate incidence: <100 per 100 000; medium/high incidence: >100 per 100 000).¹ 

We screened 27 221 titles and abstracts, reviewed 1146 full texts, and identified 428 articles eligible for inclusion in the parent ACF review (Fig 2). Of these, we identified 31 studies of ACF in children^20–50  (Supplemental Table 4) for inclusion in this analysis.

The majority of studies were conducted in settings with medium or high TB incidence (28/31 studies of children^{20–35,37,39–49} ). Eight studies among children^{26,28,31,33,37,47,48}  had results stratified by HIV status or were conducted solely among people with HIV; these were all conducted in medium or high TB incidence settings. The eligible age range varied across studies; 12 studies enrolled children <15, whereas 3 included participants up to aged 18 years.^29,33,36  Other studies focused on young children^20,46,49  or adolescents^22,35,40 ; a few studies did not define the eligible age range.^24,26,47  Only 8 studies required microbiologic confirmation of TB^{21,25,27,37,38,43,44,47} ; the others allowed for clinical diagnosis. More details about the setting, population, and screening procedures for each study are shown in Supplemental Table 4.

Studies that screened by symptoms only primarily screened for cough alone (n = 5) or used the WHO adult 4-symptom screen (cough >2 weeks, fever, night sweats, and/or unexplained weight loss) or a modified version with additional symptoms (n = 9). For children aged <15 years in medium- or high-incidence countries living with HIV, the weighted mean NNS using a positive symptom screen only was 78 (range, 5-undefined), suggesting that 78 children living with HIV would need to be screened for symptoms to detect a single active case of TB (Table 1, Fig 3). Using the same symptom screen among children without HIV or with unknown status would require screening 259 children to detect a single active case of TB. For children <5 years old (not known to be HIV-positive), the weighted mean NNS using a positive symptom screen was 24 (range 18–58).

For children aged <15 years living with HIV in medium- or high-incidence countries, the weighted mean NNS was 17 (range, 17–17; n = 2) using an abnormal CXR only, 21 using either symptoms and/or abnormal CXR (n = 1), 19 (range, 11–21; n = 2) using underweight status only, and 23 (n = 1) using Xpert (there was only 1 study, which was among children living with HIV, which used Xpert as the primary screening modality). The weighted mean NNS among children not known to have HIV was 415 (range, 343–585) using a positive symptom screen and/or positive tuberculin skin test, 295 (range, 6–388; n = 4) using an abnormal CXR only, 65 using either symptoms and/or abnormal CXR (n = 1), and 426 (range, 25–1551) using underweight status only.

For children in low- or moderate-burden countries, the weighted mean NNS using abnormal CXR alone was 30 891 (range, 30 860–30 922) and the NNS using abnormal CXR and/or positive symptom screen was 19 069 (n = 1). We found no papers using other screening modalities in low- or moderate-burden settings and no studies in low/moderate incidence settings among children with HIV.

In studies conducted in health care settings, the weighted mean NNS to detect a case of active TB (regardless of screening modality) was 48 (range, 5-undefined; n = 8) among children with HIV and 109 (range, 9–241; n = 8) among children with negative or unknown HIV status (Table 2). One study of inpatient children with HIV found a low weighted mean NNS (23, n = 1); the weighted mean NNS was higher among inpatient children with negative or unknown HIV status (216; range, 18–241; n = 3). The weighted mean NNS in outpatient settings for children with HIV (53; range, 5-undefined; n = 6) was similar to that for children with negative or unknown HIV status (68; range, 50–154; n = 4). Most studies conducted in outpatient settings were in HIV clinics (n = 7/10). The overall weighted mean NNS was 50 (range, 5-undefined; n = 7) in outpatient HIV clinics and 75 (range, 75–154; n = 3) in all other outpatient settings including general and pediatric clinics. In non-clinical settings, the weighted mean NNS among children in school settings was 464 (range, 118–665; n = 2) and 1117 (range, 28–5146; n = 10) in community settings.

In low- and moderate-burden settings, there were 2 studies conducted using immigration records and 1 study conducted in the general community. The weighted mean NNS among studies of migrant populations was 30 891 (range, 2676–33 167). In the 1 study of ACF in the community of a low/moderate burden country, the NNS was 19 069.

The overall risk of bias, based on select criteria from the AXIS tool, was low (Supplemental Table 3). The primary source of potential bias was lack of detailed data describing non-responders.

In this systematic review, we used the current literature to estimate the NNS to detect a case of active TB among children using different screening modalities and among different populations and locations. Screening strategies, locations, and yield varied widely, reflecting the global landscape of ACF. We found that screening for TB among children in outpatient settings, including both HIV clinics and general clinical settings, as well as screening in children younger than 5 years in both health care and community settings, may be efficient strategies for detecting undiagnosed pediatric TB. We also provide additional evidence supporting screening for TB among children with HIV for all screening modalities. The importance of ACF for reducing TB burden and transmission is well-recognized,⁵¹  but ACF is also critical for linking individuals in hard-to-reach or hard-to-diagnose populations, including children,⁵²  to potentially lifesaving treatment. The United Nations high-level meeting on TB in 2018 set targets of initiating 2.5 million children on TB treatment by 2022. However, by the end of 2020, only 41% of that target had been met.¹  It is therefore important to identify high-yield screening strategies for detecting TB in children.

Overall, we found that symptom-based screening was most commonly used for ACF but requires screening more individuals to detect a case of TB compared with screening algorithms that consider either symptoms and/or an abnormal CXR as a positive screen; combining symptom screening with tuberculin skin test did not appear to provide an added benefit. The impact of CXR on NNS is not surprising given that CXR has a higher sensitivity for detecting TB compared with symptom screening.⁵³  Even so, given its simplicity and low cost, symptom screening may be the only option in settings where the costs and infrastructure required for facility-based CXR may be prohibitive. As portable CXR- and computer-aided detection are being expanded, this may improve the efficiency of pediatric ACF. Although we did not include studies of novel diagnostics or the use of nonsputum specimens in this review, they will likely become important components of TB ACF in children.

Notably, for all screening strategies the NNS was lower for children with HIV; a previous review showed that screening adults with HIV for TB is similarly efficient.¹⁵  As children co-infected with TB and HIV are at increased risk for death,⁵⁴  screening for TB among children with HIV and linkage to comprehensive TB and HIV treatment is critical. In addition to detecting active TB, ACF interventions also represent an opportunity to link children with HIV to TB preventive treatment (TPT), once TB has been excluded.⁵⁵  The UN high level meeting targets include providing 30 million PLHIV with TPT from 2019-2022 by 2020. This target had been reached but children living with HIV lag behind.¹  Therefore, ACF can be used to increase not only the number of children diagnosed with TB, but also the number of children accessing TPT.

Health care settings, including both inpatient wards and outpatient clinics, had a lower NNS than school- or community-based settings. Targeted inpatient screening strategies for TB in children have been suggested in malnutrition wards and general pediatric hospital wards.^11,56  Although children with TB symptoms are often inadequately evaluated for TB in inpatient settings,²⁴  case finding in these settings remains important to reduce TB-associated morbidity and mortality. Additionally, the low NNS of ACF conducted in outpatient settings, including HIV clinics and general outpatient departments, represents an opportunity for more TB screening among children seeking care at maternal and child health clinics, outpatient pediatric clinics, antiretroviral clinics, nutrition services, and private provider clinics.¹¹  The low NNS for children younger than age 5 years in health care settings suggests that screening young children for TB may be particularly efficient in these settings. Young children are also at increased risk for severe or disseminated disease, and earlier diagnosis of TB may improve health outcomes among this population. Strategies for screening in health care settings may help reach the large proportion of children who are exposed to TB outside of the household, and thus not identified during contact investigation.⁵⁷  Additional studies are needed to assess the efficiency of ACF strategies in various outpatient and inpatient settings where children access healthcare.⁵⁸ 

The heterogeneity of populations, settings, and case finding approaches limit our ability to draw conclusions regarding the best methods for case finding in children. This work represents the existing body of evidence and identifies strategies that warrant additional research. Specifically, there were few studies in low-burden settings, so we limit our main conclusions to moderate- and high-burden settings. There were also few studies representing different pediatric outpatient settings, limiting our conclusions on which settings may be optimal to integrate with TB services. Additionally, we did not require microbiologic confirmation to diagnose TB in this study because of the difficulties in confirming TB diagnoses in children. Including clinical TB diagnoses may inflate the number of cases detected (and thus reduce the NNS) if many of those cases do not have TB or introduce bias; the results of certain screening modalities may influence a clinician’s diagnosis. However, clinical diagnoses currently make up the majority of pediatric TB diagnoses in the world. Thus, these strategies represent practical approaches to pediatric diagnosis, and this is unavoidable given the lack of a gold standard for TB diagnosis in children. Finally, this review covered a limited set of diagnostic criteria and did not include the pediatric TB screen (a tool used for screening child household contacts)⁵⁹  or newer diagnostics such as TB-LAM, which are of growing importance. Future reviews should include these new diagnostics.

We found substantial heterogeneity in approaches to screening children for TB. Case finding interventions should be designed based on the evidence as well as local context. There is a particular opportunity to screen children for TB when they are in contact with the health care system for other reasons. Improving case detection for children with TB, with appropriate linkage to TB treatment or TB-preventive treatment, is critical to improving outcomes among children in high-burden TB settings.

We thank Lori Rosman from the Johns Hopkins University Welch Library and the systematic review team: Ethan Valinetz, Niccolo Dosto, Matthew Murrill, Samuel Ayeh, Fatima Qamar, and Rabia Jalalzai. In addition, we thank Cecily Miller and the Wold Health Organization Guideline Development Group on systematic screening for tuberculosis.

ACF

active case finding

CXR

chest x-ray

NNS

number needed to screen

TB

tuberculosis

WHO

World Health Organization