Remarks

• Presumptive TB refers to a person who presents with symptoms and/or signs suggestive of TB.24
• Bacteriological confirmation should be sought as part of the integrated treatment decision algorithms whenever possible, using WHO recommended rapid diagnostic tests and appropriate paediatric specimens (including stool, nasopharyngeal aspirate, induced or expectorated sputum or gastric aspirate).
• In follow-up to the GDG meeting, new integrated treatment decision algorithms for specific populations and settings, have been developed and internally validated. These algorithms are detailed with practical guidance on their use in the operational handbook on the management of tuberculosis in children and adolescents.25 National TB and other health programmes are encouraged to use these evidence-based algorithms.
• This interim recommendation will remain valid for a period of 24 months after the publication of these guidelines, after which new evidence will be reviewed.

Subgroup considerations For HIV-infected children under the age of 10 years, the pooled sensitivity of the algorithms reviewed ranged from 24% (Marais et al. criteria) to 92% (Uganda NTLP algorithm), and the pooled specificity from 15% (Uganda NTLP algorithm) to 87% (Stegen-Toledo score, cut-off 5). For children with SAM the pooled sensitivity varied between 33% (Marais et al. criteria) and 93% (Uganda NTLP algorithm and Keith Edward score), while the pooled specificity varied between 10% (Keith Edward score) and 88% (Stegen-Toledo score, cut-off 5). For infants aged below 1 year the pooled sensitivity ranged from 17% (Marais et al. criteria) to 93% Gunasekera et al. algorithm) and the pooled specificity from 13% (Gunasekera et al. algorithm) to 86% (Stegen-Toledo score, cut-off 5). The GDG members highlighted the need for the development of specific evidence-based treatment decision algorithms for specific subgroups at high risk of rapid progression of TB disease, including children living with HIV, children with SAM and infants, if possible.

Implementation considerations Algorithms included in the operational handbook: In the follow-up to the GDG meeting, new integrated treatment decision algorithms for specific populations and settings have been developed and internally validated, using regression modelling with pre-determined cut-off values for sensitivity and specificity against the reference standard (using updated clinical case definitions to define pulmonary TB, outlined in Graham S et al. (31)), based on the individual patient data set used for the evidence review conducted to answer this PICO question. The algorithms are described in the operational handbook and cover the diagnosis of PTB among children under the age of 10 years, including intrathoracic lymphadenopathy. The algorithms are not suitable for the diagnosis of EPTB. Implementation at peripheral levels of the health system: Integrated treatment decision algorithms allow treatment decisions to be made at more decentralized levels of care, where children generally present earlier, with less severe disease and lower bacteriological confirmation rates. Algorithms integrating clinical criteria have an important role to play at these levels of the health system. The decision to start treatment is linked to other recommendations in these guidelines, such as shortening of the treatment duration for children with non-severe forms of TB and on decentralization of TB services. Once a decision to start TB treatment has been made, the severity of disease needs to be assessed to inform the duration of treatment. Detailed criteria for assessing severity of disease are described in the operational handbook. Referral: Defining the criteria for referral of children evaluated for PTB at peripheral levels of the health care system using the algorithms is important. Examples of subgroups in need of referral include infants, children with presumptive severe forms of EPTB (such as TBM, disseminated TB and osteoarticular TB) and children with presumptive DR-TB in regions with a high prevalence of DR-TB. Children presenting with severe acute pneumonia need referral to the appropriate level of care for oxygen supplementation, while children with SAM need to be provided with appropriate nutritional support. A high index of suspicion is important among infants with acute symptoms who are contacts of people with bacteriologically confirmed TB, to make a treatment decision as soon as possible rather than wait for symptoms to persist. This is due to the potential for rapid deterioration in the clinical condition of infants and development of severe TB disease. Clinical monitoring of children started on TB treatment: It is important to acknowledge that the preference for sufficient sensitivity of the algorithms to detect and treat children with TB will mean that some children who do not have TB will be treated with TB treatment. The risk of severe drugrelated toxicity in children is very low, and shorter regimens for non-severe TB (see chapter 5) will further reduce the risks related to treatment. However, it will be critical to monitor children started on TB treatment and to refer them for evaluation for other diseases and appropriate treatment if they fail to respond to TB treatment within 1 month. Implementation in high DR-TB burden settings: Integrated treatment decision algorithms may be implemented in settings with a high burden of DR-TB. Seeking bacteriological confirmation using appropriate paediatric samples and WHO recommended rapid diagnostic tests (such as Xpert MTB/ RIF or Ultra) is critical among children who have a history of contact with a source case with confirmed or highly likely DR-TB (including a TB patient not responding to treatment, or a source case who died of TB while on treatment). Once a decision to treat a child without bacteriological confirmation for TB has been made based on the algorithm, risk factors for the child having DR-TB need to be assessed. Clinicians need to keep a high index of suspicion for DR-TB in these children and ensure they are tested and managed for DR-TB as appropriate (see chapter 5).