4.2.1. Justification and evidence

TB diagnosis in children relies on a thorough assessment of all evidence derived from a careful history of exposure, clinical examination and relevant investigations. While various algorithms and scoring systems for TB diagnosis in children exist, these have not been systematically evaluated. There is a need for evidence-based, practical treatment decision algorithm(s), ideally for different settings with varying access to diagnostic tests and CXR.

PICO question: In children aged below 10 years with presumptive pulmonary TB attending health care facilities, should integrated treatment-decision algorithms be used to diagnose pulmonary TB, compared to a microbiological or composite reference standard?

Evidence: To address the need for evaluating and developing evidence-based, integrated treatment decision algorithm(s), ideally for different settings with varying access to diagnostic tests and CXR, an individual participant dataset (IPD) meta-analysis was conducted. For this analysis, an IPD meta-cohort was developed, consisting of diagnostic evaluations data from children aged below 10 years, to infer the sensitivity and specificity of treatment-decision algorithms (or scores) in identifying pulmonary TB, using the updated clinical case definitions for the classification of intrathoracic TB among children (31). Data were requested from investigators in February 2021. Data for the IPD was sourced from several studies carried out within a geographically diverse set of TB high-burden countries. Algorithm sensitivity and specificity in classifying TB was evaluated against the Union Desk Guide Algorithm (32), which was developed in an attempt to operationalize the 2014 WHO guidance (8) by outlining the steps that a health care worker should take in evaluating a child with presumptive PTB.

To make full use of available data for comparing the performance of these algorithms in different settings, missing variables were imputed, heterogeneous definitions of variables were collapsed, and slight modifications were made to the algorithms or scores to enable the use of the variables available in the IPD. A total of 14 studies, comprising 5494 records were included, of which 4811 records were included in this analysis (26, 33-47). Studies from 13 countries (including 12 high TB, TB/HIV and/or MDR-TB burden countries) within 5 of the 6 WHO regions were included. The cohort in the meta-analysis used to inform this recommendation had a median age of 26 months (interquartile range 13.4-58.3); 38% of the children had TB, of which 30% was bacteriologically confirmed; 20% of the children were living with HIV infection; and 14% had SAM.

Seven algorithms or scoring systems were identified for evaluation, comprising the Uganda National TB/Leprosy Control Program (NTLP) algorithm (48), the Brazilian Ministry of Health Child PTB Scoring System (49), the Gunasekera et al. 2021 algorithm (50), the Keith Edward score (51), the Marcy et al. 2019 algorithm (52), the Stegen-Toledo score (53), and the Marais et al. 2006 criteria (54). The pooled estimates of the sensitivity and specificity of each algorithm/score were compared to the standard of care algorithm (i.e. The Union Desk Guide Algorithm, which constituted the reference standard (32)) for children aged below 10 years, for children living with HIV, for children with SAM and for children aged below 1 year.

For the overall population of children under the age of 10 years, the pooled sensitivity of the seven algorithms or scoring systems ranged from 16% (Marais et al. criteria) to 95% (Gunasekera et al. algorithm), while the pooled specificity ranged from 9% (Gunasekera et al. algorithm) to 89% (Marais et al. criteria) (see web annex 3).

GDG considerations: The GDG felt that algorithms with clinical criteria have an important role to play in making decisions on starting children on TB treatment, particularly at peripheral levels of the health care system. There was strong consensus among the GDG members about the need and importance of working on treatment decision algorithms to improve the gaps in TB case detection in children. An important advantage of evidence-based algorithms (as it allocates a modelled weight to features of clinical evaluation), is that this modelling process allows for specification of the weight of certain clinical features, rather than being based solely on expert opinion. The panel highlighted that data in the IPD were mainly from tertiary settings where the proportion of children with confirmed TB is higher than at district hospital or PHC level. It was acknowledged that the IPD had a high level of heterogeneity. Conducting a meta-regression analysis by level of the health care system was not possible because of the limited number of studies.

The GDG concluded that none of the evaluated algorithms were optimal in terms of either sensitivity or specificity, combined with the very low certainty of evidence. The GDG also noted that algorithms with a high sensitivity (i.e. low number of false negatives) generally have a low specificity (i.e. a high number of false positives) and vice versa. The panel reflected on the consequences of false negative and false positive conclusions based on integrated treatment decision algorithms and agreed that it was most important to avoid missing a TB diagnosis in a child who has TB, considering the large case detection gap and the consequences of a missed diagnosis of TB.

During the GDG deliberations, the following options that emerged from the evidence review were discussed: (i) choose one of the algorithms reviewed for a possible recommendation; (ii) make a generic recommendation on the use of integrated treatment decisions algorithms and present new evidence-based algorithms in the operational handbook; (iii) make a statement about the need for further research on treatment decision algorithms, affirming the need for such algorithms.

The GDG judged the available evidence as inappropriate to support a recommendation for any specific algorithm; and instead decided to make a generic recommendation on the use of treatment decision algorithms, and to describe newly developed algorithms for relevant subgroups and/or settings in the operational handbook. The decision for a generic recommendation considers the current practice of HCWs making decisions on starting TB treatment in children based on a combination of clinical signs and symptoms, history of TB contact and investigations, and the need to develop evidence-based approaches for this practice. In addition, an interim conditional recommendation was deemed most appropriate, considering the need to reduce the TB case detection gap in children and the need for additional evidence on the use of integrated treatment decision algorithms in programmatic settings.

The GDG also prioritized the need to reduce false negative results, while accepting a certain degree of over-diagnosis, as well as limiting unnecessary referrals and tests for children. The GDG members judged that a general recommendation with operational guidance on the use of evidence-based algorithms integrating the use of rapid diagnostics with clinical features would empower HCWs, including those in settings with limited access to diagnostic tools, to make decisions on starting TB treatment in children.

The GDG concluded that a period of 24 months for the interim conditional recommendation would be needed to: set up and conduct studies to generate new data, including studies to externally validate the algorithms, implementation/operational research, modelling studies to determine the potential impact of the treatment decision algorithms and qualitative research into feasibility and acceptability for health care workers and families (refer to chapter 8 on detailed research priorities).

²⁴ Definitions and reporting framework for tuberculosis – 2013 revision (updated December 2014 and January 2020). Geneva: World Health Organization; 2013

²⁵ WHO operational handbook on tuberculosis. Module 5: management of tuberculosis in children and adolescents (https://apps.who.int/iris/bitstream/handle/10665/352523/9789240046832-eng.pdf).

Book navigation