Abstract
Background Children with severe acute malnutrition (SAM) face higher risks of underdiagnosis and death from tuberculosis. In 2022, the World Health Organization (WHO) recommended use of treatment decision algorithms (TDAs) for tuberculosis diagnosis in children. There is currently no cost-effectiveness evidence for TDA-based approaches compared to routine practice.
Methods The TB-Speed SAM study developed i) a one-step TDA including Xpert, clinical, radiological and echography features, and ii) a two-step TDA, which also included a screening phase, for children under 5 years hospitalised with SAM at tertiary hospitals in Uganda and Zambia. We assessed the diagnostic accuracy and cost-effectiveness of deploying TB-Speed and WHO TDA-based approaches compared to the standard of care (SOC). Estimated outcomes included children started on tuberculosis treatment, false positive rates, disability-adjusted life years (DALYs) and incremental cost-effectiveness ratios (ICERs).
Findings Per 100 children hospitalised with SAM, averaging 19 children with tuberculosis, the one-step TDA initiated 17 true positive children (95% uncertainty intervals [UI]: 12-23) on tuberculosis treatment, the WHO TDA 16 (95%UI: 13-21), the two-step TDA 16 (95%UI: 10-23), and SOC 4 (95%UI: 1-9). The WHO TDA generated the most false positives (50%, 95%UI: 44-57), followed by the one-step TDA (22%, 95%UI: 17-27), the two-step TDA (17%, 95%UI: 12-22), and SOC (13%, 95%UI: 9-17). All TDA-based approaches had ICERs below plausible country cost-effectiveness thresholds compared to SOC (one-step: $44-51/DALY, two-step: $34-39/DALY, WHO: $43-49/DALY).
Interpretation Our findings show that these TDA-based approaches are highly cost-effective for the vulnerable group of children hospitalised with SAM, compared to current practice.
Funding Unitaid
Evidence before this study In 2022, the WHO conditionally recommended the use of treatment decision algorithms (TDAs) for tuberculosis diagnosis in children aged <10 with presumptive pulmonary tuberculosis. Two TDAs were suggested for use in settings with (TDA A) and without (TDA B) access to chest X-ray. These WHO-suggested TDAs propose a single approach to TB diagnosis in all children. The TB Speed SAM study developed specific algorithms for children <5 hospitalised with severe acute malnutrition. Aiming to identify studies assessing cost-effectiveness of using TDAs for childhood TB, we searched the PubMed database using (“Decision Support Systems, Clinical”[MeSH] OR “clinical decision support” OR “decision support” OR “clinical decision-making”) AND (“Algorithms”[MeSH] OR “algorithm” OR “decision-making” OR “decision model” OR “treatment decision algorithm”) AND (“Tuberculosis”[MeSH] OR “tuberculosis” OR “TB”) AND (“Costs and Cost Analysis”[MeSH] OR “cost-effectiveness” OR “cost analysis” OR “costs”) between January 1st, 2004 and October 18th, 2024, without language restrictions. Of 31 articles identified, 2 articles reported on the cost-effectiveness of interventions aiming to improve clinical decision making for tuberculosis diagnosis. Other articles were excluded because they were not an economic evaluation, not on tuberculosis, or only compared microbiological testing approaches related to tuberculosis care (microscopic observation drug susceptibility test versus Xpert MTB/RIF test, QuantiFERON-TB Gold In-Tube versus tuberculin skin test for tuberculosis diagnosis). Debes et al. assessed the cost-effectiveness of tuberculosis diagnosis using microscopic observation drug susceptibility, Xpert MTB/RIF and empiric treatment for all patients, in addition to current clinical diagnostic practices in Ugandan children. Van’t Hoog et al. explored combinations of sensitivity, specificity and cost at which a hypothetical triage test would improve affordability of the Xpert assay. We found no economic evaluations of a treatment decision algorithm (TDA)-based approach (screening, testing, treatment) for tuberculosis diagnosis.
Added value of this study This is the first study to assess the cost-effectiveness of using treatment decision algorithms in childhood tuberculosis diagnosis, focusing on children <5 years hospitalised with severe acute malnutrition using the TB-Speed SAM one-step and two-step TDAs that includes a screening step before the diagnostic step, and the WHO-suggested TDA A. We also evaluated the accuracy of the WHO-suggested TDA for paediatric tuberculosis in this patient group. This study found that for children hospitalised with SAM all three TDA-based approaches for paediatric tuberculosis diagnosis were cost-effective compared to the standard of care from a health systems perspective in Uganda and Zambia, including in lower tuberculosis prevalence settings. The TB-Speed two-step approach had a smaller resource footprint than the TB-Speed one-step and WHO TDAs because its first step resulted in fewer assessments overall, but also a smaller health impact due to a slightly lower sensitivity. The TB-Speed one-step and WHO TDAs were similar in cost and health impact, but the WHO TDA involved substantial rates of overtreatment.
Implications of all the available evidence The WHO has conditionally recommended incorporating TDAs, pending validation, into existing case detection strategies to support the decentralisation of clinical tools and improve the identification of tuberculosis in children. Our findings show that TDA-based approaches are cost-effective for the vulnerable group of children hospitalised with SAM, compared to current practices, and our sensitivity analysis suggests that these results are robust. While not developed in children hospitalised with SAM, the WHO-suggested TDA for paediatric tuberculosis performs well in this patient group. This analysis contributes valuable evidence to support the interim WHO recommendation on decentralised models of care.
Introduction
Tuberculosis mortality remains high in children globally, with 187,500 deaths estimated by the World Health Organization (WHO) for 1.25 million paediatric cases in 2023.1,2 The vast majority of childhood tuberculosis deaths occur because the disease remains undiagnosed and, therefore, untreated.3 Tuberculosis diagnosis is challenging in children, unlike in adults, due to the low yield of microbiological tests in this age group explained by mostly paucibacillary disease and challenging sample collection.4 It relies mostly on clinical and radiographic features, which lack specificity, notably in children with immunodeficiency due to HIV or other conditions. Recent studies show that children under 5 years of age with severe acute malnutrition (SAM) are an important risk group for the development of tuberculosis, and face a high risk of being undiagnosed for the disease.5–7
SAM is defined by WHO as severe wasting, i.e. low weight for height ratio, or low mid upper arm circumference (<115 mm), or clinical signs of bilateral nutritional pitting oedema.8 It is the most severe manifestation of undernutrition, and is associated with high mortality. Childhood malnutrition is a major global health challenge with 45 million children affected, accounting for almost 50% of the deaths in children under the age of 5 years.9,10 Therefore, improving case detection in children with SAM could contribute to reducing mortality in this vulnerable group and is essential to achieving the global target of zero deaths from tuberculosis in children by 2030.6,7
Treatment decision algorithms (TDAs) are scoring systems combining specific clinical, radiological and microbiological features. Treatment initiation is recommended above a pre-determined score threshold.11 Aiming to standardise and accelerate the identification of tuberculosis in children, TDAs could fill the detection gaps in vulnerable groups, often difficult to diagnose. WHO recently conditionally recommended the use of TDAs for the diagnosis of tuberculosis in children below 10 years old with symptoms suggestive of pulmonary tuberculosis.12,13 In addition, WHO suggested using two TDAs for settings with chest X-ray [CXR] (algorithm A) and for settings without CXR (algorithm B). These two TDAs use a similar diagnostic approach for the general paediatric population and highly vulnerable children such as those with severe acute malnutrition that may require specific approaches.
In 2023, the TB-Speed SAM study was the first to develop TDAs specifically for children under 5 years hospitalised with SAM, enrolled at tertiary hospitals in Uganda and Zambia.14 Two TDAs were developed: i) a one-step TDA including Xpert, clinical, radiological and echography features assessment in all children, and ii) a two-step TDA including a screening phase followed by similar assessment in only those who screened positive. Both TDAs demonstrated satisfactory diagnostic accuracy with a sensitivity (ability to diagnose a child infected with tuberculosis as positive) of 85.2% and specificity (ability to designate a child not infected with tuberculosis as non-TB) of 81.6% for the one-step TDA, and a sensitivity and specificity of 77.2% and 85.7% for the two-step TDA, with a reduction of 30% of tuberculosis assessments needed, due to the screening step.
In this study, we sought to assess the overall diagnostic accuracy and quantify the cost-effectiveness of the two TB-Speed SAM TDAs and the WHO-suggested TDA A (with CXR; WHO TDA thereafter) compared to routine clinical practice, from a health system perspective, for TB screening and diagnosis in children hospitalised with SAM.
Methods
Study design
We conducted a cost-effectiveness analysis associated with the TB-Speed SAM study using patient pathways, costs and cost-effectiveness modelling. We assessed the two TB-Speed SAM TDAs and the WHO TDA compared to routine clinical practice, from a health system perspective. The TB-Speed SAM study design has been described elsewhere.14 In addition, we evaluated retrospectively the diagnostic accuracy of the WHO TDA.
The study was approved by the sponsor’s (Inserm) institutional review committee, the WHO ethical review committee, as well as the national ethics committees and institutional review boards in Uganda and Zambia. The TB-Speed SAM study is registered with clinicaltrials.gov (NCT04240990).
Patient pathways
We developed conceptual models with country experts and the TB-Speed SAM study investigators to represent patient care pathways and resource use for four comparator arms: a standard of care (SOC), the one-step TDA, two-step (screening and diagnosis) TDA, and the WHO TDA (Figure 1). The SOC pathway represented typical care available in high tuberculosis incidence countries at hospital level with a mix of options for assessment (clinical only; clinical and CXR; clinical and Xpert testing; clinical, CXR, and Xpert testing). The one-step diagnosis, two-step screening and diagnosis approach were based on the TB-Speed SAM study protocol and results,14 and the WHO TDA was based on the 2022 WHO operational handbook.12 (Appendix figures 1, 2 and 3). The TB-Speed SAM one-step TDA proposes systematic Xpert testing on two types of samples (stool and gastric aspirate), clinical evaluation, CXR and echography (abdominal ultrasound). Children with a score >=10 during clinical assessment are immediately initiated on TB treatment and do not receive echography. They also receive Xpert testing and CXR to assess whether there is a case of TB drug-resistance and to assess the TB disease severity (not primarily for diagnostic purposes). The two-step TDA presents a screening phase solely based on a clinical examination and HIV testing, then similar features as the one-step TDA. The WHO TDA presents similar features as the one-step TDA, except for the echography which is not included and for the fact that it recommended for use in children with presumptive TB recognized on the basis of chronic symptoms, unlike the TB-Speed SAM one-step TDA that is applied to any child < 5 years hospitalised with SAM. Scores attributed to the different features vary between the three TDAs. We paid particular attention to screening for initial assessment (by TDA or SOC assessment) and reassessment following negative initial assessment, informing their diagnostic accuracy with data from TB-Speed SAM.
Model patient care pathways for the diagnosis and treatment of tuberculosis in children hospitalised with SAM
Screening for tuberculosis
In the SOC, children admitted to hospital with SAM received non-systematic screening for tuberculosis, with a coverage of 80% suggested by clinicians. For the TDA arms, systematic screening for TB was conducted for all. Children in the SOC were considered to screen positive for tuberculosis if they had at least one chronic symptom (>2 weeks) for fever and cough, or a history of contact tuberculosis. In the one-step TDA, all children admitted to hospital with SAM received a TB clinical assessment (‘artificial’ screening sensitivity of 100% and specificity of 0%). Children in the two-step TDA screened positive if they had history of TB contact, or one of the signs/symptoms from clinical exam (cough >3 weeks, Temperature >38 °C, tachycardia, chest-in-drawing, crackles on auscultation, depressed level of consciousness, cervical or supra-clavicular adenopathy), or an HIV-positive test result.14 For the WHO TDA, the screening step was based only on a child’s chronic symptoms (fever, cough, fatigue or loss of playfulness, weight loss, and loss of appetite) being longer than 2 weeks.12 A child presenting with at least one of these chronic symptoms would be considered to screen positive. For all arms, we did not permit tuberculosis rescreening. Thus, if a child initially screened negative for tuberculosis, they were excluded from the model’s care pathway. See Appendix table 1 and 2 for details.
Reassessment following negative initial tuberculosis assessment
Using the TB-Speed SAM cohort data, we defined a two-step reassessment process using i) the clinician’s choice to reassess where the sensitivity is: true TB who are reassessed / true TB not diagnosed at the initial clinical assessment, and specificity: true non-TB not reassessed / true non-TB not diagnosed at the initial clinical assessment; and ii) reassessment exam’s sensitivity and specificity. Reassessment was taken to comprise CXR, Xpert on gastric aspirate, and clinical assessment, and their sensitivity and specificity were conditional on results of (clinical) components used in initial assessment being negative (Appendix section 4). We assumed that children can only be reassessed once.
Evaluation of TDA diagnostic accuracy
A synthetic population of 10,000 children was created matching frequency cross-tabulations of signs and symptoms among children evaluated as true tuberculosis and not tuberculosis by an expert committee within the TB-Speed SAM study14, and scores calculated for each TDA. Symptoms included fatigue, loss of appetite, fever, and weight loss. Signs included results from the clinical exam, radiological testing (CXR), abdominal ultrasound, and microbiological test results (Xpert and HIV testing). Two symptoms (night sweat and haemoptysis) used in the WHO diagnostic score were not collected in the TB-Speed SAM cohort. These signs were simulated in the synthetic cohort based on their tuberculosis-stratified frequency in the individual patient database Gunasekera and al. used to develop the WHO TDAs.13 Diagnostic accuracy of all TDAs for the synthetic cohort was assessed against expert committee tuberculosis status from the TB-Speed SAM study, using the updated NIH Clinical Case Definition.14,15
Costing approach
Cost data collection tools were adapted from the Value TB costing tool suite developed in conjunction with the Global Health Cost Consortium, with reference to the WHO guidance ‘Costing Guidelines for Tuberculosis Interventions’.16 Labour costs were sourced from national pay scales and project accounts, medications from the Stop TB partnership Global Drug Facility catalogue17, consumables (e.g. diagnostic tests), staff training, and equipment from project accounts, and hospitalisation cost of an inpatient bed day from the WHO-CHOICE unit cost estimates for service delivery.18 As this intervention did not change facility infrastructure, we excluded facility-associated overhead costs from the analysis.
To calculate the proportionate use of major equipment such as X-ray or Xpert machines we used the expected lifespan and annual number of uses of each item. These data were obtained through key informants at MSF-Logistique (https://www.msflogistique.org/) for expected lifespan and from laboratory managers for the number of uses between March 2019 and November 2021 in the three tertiary hospitals. To value the contribution of labour we conducted a time and motion study to estimate the length of time that staff spent on each patient care task under the TB-Speed intervention.19
Unit costs were estimated using an ingredient-based costing approach, in which the expected resources required for each child enrolled into the study were listed, costed, and summed to estimate direct health service utilisation and cost varying by patient characteristics and the route taken on the pathway. Services (number and type of diagnostic and treatment procedures, materials, laboratory investigations and medications) provided to patients were valued by multiplying the quantities required by their unit costs. The value of all time spent by staff for each patient was estimated as the product of ‘hours spent’ and ‘hourly labour costs’. Costs were estimated in 2021 US dollars (US$), using a discount rate of 3% for the annualisation of the economic costs of equipment following guidelines.20,21 Costs of conducting a lateral flow urine lipoarabinomannan (LF-LAM) test were extracted from the literature at 2014 US$11.49 (9.15-14.89) per patient tested.22 See Appendix table 5 and 6 for cost parameters.
Modelling approach
A decision-analytic cost-effectiveness model was developed in R software (version 4.3.0) to assess the clinical benefits, cost-effectiveness of each diagnostic approach. A patient-level decision tree model represented clinical pathways shown in Figure 1, with outcomes depending on true tuberculosis status. The probabilities of having tuberculosis, diagnosis, treatment, and death were primarily based on TB-Speed SAM study data. In particular, counts of children among the cohort of stratified by tuberculosis status were used to parametrize beta distributions for branching probabilities, assuming uniform priors. See Appendix table 7 for non-cost parameters.
Outcomes
Country-specific unit costs associated with resource use at each step of care were accumulated to produce total mean costs. Country-specific life expectancy from United Nations estimates was used to calculate the mean life-years lost over a lifetime horizon (with and without 3% discounting). Case fatality rates by tuberculosis status and treatment status were estimated from the TB-Speed SAM cohort. We disregarded the contribution of morbidity to disability adjusted life-years (DALYs) and this has previously been shown to be a good approximation.23 All results were calculated using a probabilistic sensitivity analysis with 1,000 parameter and cohort samples.
For each arm, we report the sensitivity and specificity of the tuberculosis screening step only and of the overall arm (screening, initial assessment and reassessment), false positive and false negative rates, positive predictive value (PPV: probability that children who test positive for TB actually have TB), and negative predictive value (NPV: probability that children who test negative for TB truly do not have TB), TB cascade of care. We also report total and incremental costs, number of deaths and deaths averted, number of (discounted and undiscounted) DALYs and DALYs averted, per 100 children admitted to hospital with SAM, and the incremental cost-effectiveness ratios (ICERs) for all diagnostic approaches in each country. ICERs were compared to various options for cost-effectiveness thresholds (presented as a range) in each country to assess potential cost-effectiveness.24 We also report cost-effectiveness acceptability frontiers (CEAF).25 First, we identify the mean cost and mean effect for each option, to calculate which option is optimal (has the highest expected net benefit) at different thresholds. Second, at given thresholds, we plot the probability of highest net benefit for each arm. We present our findings at the overall cohort prevalence (18.9%) and at lower prevalence levels similar to those reported in other studies on children with SAM (10.0%, 5.0%).5 We complied with the Consolidated Health Economic Evaluation Reporting Standards (CHEERS 2022) reporting guidelines.26
Sensitivity analysis
We conducted sensitivity analyses to assess the impact of all model parameters on estimated ICERs compared to the SOC, using the interquartile range’s lower and upper limits, differentiated by TDA approach and by country.
Role of the funding source
The study funders had no role in the study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication.
Results
Accuracy of the SOC and TDA-based approaches
The two-step TDA had the highest screening sensitivity (88%) followed by the WHO TDA (79%), whereas the SOC had the lowest sensitivity (37%) explained by an assessment solely based on the presence of chronic fever or cough (>2 weeks), and tuberculosis contact history (Table 1). Screening specificity was the highest for SOC (79%), followed by two-step TDA (34%), and WHO TDA (28%). Combined screening and treatment decision sensitivity was the highest for one-step TDA (93%), followed by the WHO TDA (92%), the two-step TDA (86%) and the SOC (25%). The overall specificity was the lowest for the WHO TDA (50%), followed by the one-step TDA (78%), the two-step TDA (83%) and the SOC (87%).
Per 100 children hospitalised with SAM, tuberculosis diagnostic assessment was much higher in the three TDA arms (73-100 children) compared to the SOC (19 children). The WHO TDA initiated the highest number of children on tuberculosis treatment (58 [53 to 63]), followed by the one-step TDA (35 [30 to 41]), the two-step TDA (30 [24 to 36]), and only 15 (11 to 19) children were initiated on treatment in the SOC. However, the WHO TDA also presented the highest rate of false positives (50 [44 to 57]), followed by the one-step TDA (22 [17 to 27]), the two-step TDA (17 [12 to 22]) and only a few cases in the SOC (13 [9 to 17]). Therefore, the number of true positive cases initiated on treatment was similar across the three TDA arms (16 to 17 children) and was three to four times higher than the SOC (5 children). Across all arms, the vast majority of children (>97%) were initiated on tuberculosis treatment following the first assessment. PPV was highest in the two-step TDA (54%) and was lowest in the WHO TDA (29%), whereas NPV was highest in the one-step TDA (97%) and lowest in the SOC (80%). At lower tuberculosis prevalence, the NPV remained high, but the PPV significantly decreased due to the rarity of true positive children detected.
Costs and cost-effectiveness
Costs were slightly higher in Zambia than in Uganda, however overall findings on costs and cost-effectiveness were similar between the two countries (Table 2). We present the results for Uganda, with the full model outputs (both discounted and undiscounted) provided in the appendix at varying levels of tuberculosis prevalence (Appendix table 8). Incremental costs per child going through the patient care pathway were highest for the one-step TDA (75 [70 to 81]) and for the WHO TDA (72 [66 to 78]), followed by the two-step TDA (52 [45 to 60]),compared to the SOC. Compared to the SOC, the three TDA-based approaches had similar impact on child deaths averted (6 [4 to 9]), resulting in DALYs averted of 172 (112 to 237) for the one-step TDA, 155 (96 to 218) for the two-step TDA, and 170 (112 to 232) for the WHO TDA.
Incremental costs and DALYs averted are shown on a cost-effectiveness plane (Figure 2). All three TDA-based approaches compared to the SOC presented ICERs well below estimates of implied country cost-effectiveness thresholds (UG: 150-194 $/DALY, ZM: 364-500 $/DALY, Appendix table 9), and so appear cost-effective, compared to the SOC. The probability of two-step TDA being cost-effective at these thresholds compared to SOC was close to 100%, with ICERs of $34/DALY (UG) or $39/DALY (ZM) (Figure 2). The other TDAs were more effective but costlier. ICERs for the WHO TDA compared to the two-step TDA (UG: $130/DALY, ZM: $148/DALY) are also below the country-specific threshold ranges, making this highly likely to be cost-effective, while ICERs for the one-step TDA compared to the WHO TDA (UG: $159/DALY, ZM: $221/DALY), are below the threshold range for Zambia, but within the range for Uganda, so the cost-effectiveness of one-step TDA as compared to the WHO TDA in Uganda is uncertain (Figure 2). The CEAF indicates that in Zambia, the one-step TDA is the optimal choice, offering the highest mean net benefit and remaining cost-effective, making it the preferred option (Figure 3). In Uganda, the one-step TDA also provides the highest mean net benefit at thresholds of $159 and above, staying within the country’s cost-effectiveness range at this level.
Cost-effectiveness planes for the TB-Speed SAM and WHO treatment decision algorithms, each compared to the standard of care, by country. DALY: disability-adjusted life year. Numerical annotations represent incremental cost-effectiveness ratios corresponding to the dashed line of the convex hull.
Cost-effectiveness acceptability frontiers and probability of highest net benefit for the TB-Speed SAM and WHO treatment decision algorithms, each compared to the standard of care, by country (in US$ per DALY averted). DALY, disability-adjusted life year.
Sensitivity analysis
The sensitivity analysis of the lower and upper quartiles of all model parameters indicates that the key drivers of cost-effectiveness are the case fatality rates for children with tuberculosis, whether treated or untreated, the cohort tuberculosis prevalence, and the screening sensitivity of the SOC (Appendix figure 5).
Discussion
Benchmarked against estimates of country cost-effectiveness thresholds, this study found that for children hospitalised with SAM all three TDA-based approaches for paediatric tuberculosis diagnosis were cost-effective compared to the SOC from a health systems perspective in Uganda and Zambia. The TB-Speed SAM two-step TDA had a smaller resource footprint than the TB-Speed SAM one-step and WHO TDAs because its first step resulted in fewer assessments overall, but also a smaller health impact due to a slightly lower sensitivity. The TB-Speed one-step and WHO TDAs were similar in cost and health impact.
When multiple interventions are cost-effective, the one with the highest expected net benefit is optimal.27 TB-Speed one-step was the most effective algorithm, and was optimal in Zambia for thresholds over $221/DALY, and in Uganda for thresholds over $159/DALY. Ultimately, the choice of cost-effectiveness threshold is for local decision makers.
Tuberculosis prevalence was a key determinant of cost-effectiveness, indicating that at lower prevalence levels, only the TB-Speed two-step approach may be cost-effective. This finding has important implications for the decentralised use of TDAs in secondary and primary healthcare settings, where tuberculosis prevalence is typically lower. The TB-Speed one-step TDA had the highest sensitivity, followed by the WHO TDA. The TDA-based approaches achieved higher sensitivity at the expense of lower specificity, with the WHO TDA having the highest rate of overdiagnosis (50%). Overtreatment has important implications for resource footprint and may impact negatively on children who are inappropriately treated, including their caregivers (household costs, stigma). Notably, if TDAs are to be decentralised to lower healthcare levels the number of false positives is likely to increase in these settings.
The overall sensitivity of the diagnostic approach to detect tuberculosis is the key driver of health impact due to the high case fatality rate for children with untreated tuberculosis (61% in the TB-SAM cohort). From the TB-Speed SAM cohort data, we found the case fatality rate was higher in children with SAM and without tuberculosis (12%) than in children with tuberculosis and receiving tuberculosis treatment (9%). Although the difference is small, it may reflect genuinely better outcomes when correctly identifying a serious but treatable disease.
The overall sensitivity of the diagnostic approach to detect tuberculosis depends not only on the TDA or other assessments, but also on the screening used prior to initial assessment, and reassessment following negative initial assessment. We found that the sensitivity of the screening step significantly impacted the overall sensitivity of each intervention. This was further confirmed by our sensitivity analysis, which identified the sensitivity of the SOC screening as the third most influential factor affecting cost-effectiveness, following case fatality rates and tuberculosis prevalence. The WHO definition of positive tuberculosis screening includes weight loss >2 weeks, whereas failure to respond to nutritional therapy is suggested as more appropriate for children hospitalised with SAM.5 Our sensitivity analyses indicate that weight loss should still be considered for WHO TDA in children with SAM, in order to maintain sensitivity (Appendix figure 6). Reassessment was less important for the TDA-based approaches, which had high overall sensitivity. We represented reassessment as full tuberculosis reassessment, including Xpert testing and CXR, and used TB-Speed SAM data to determine the accuracy of these procedures in children previously assessed as non-TB. If data become available on sign and symptom progression, it may become possible to model the use of repeated TDAs in reassessment.
Many limitations of our study stem from having less robust data to characterise the SOC arm. For example, we used a screening coverage of 80% for the SOC based on expert opinion. For the TDA-based approaches, we assumed complete coverage of systematic tuberculosis screening in children hospitalised with SAM, in line with WHO guidelines and the intended use of the TB-Speed TDAs. Real-life clinical practice may not achieve full screening coverage. Other studies have found screening characteristics are important for cost-effectiveness. Van’t Hoog et al.28 explored combinations of sensitivity, specificity and cost at which a hypothetical triage test would improve affordability of the Xpert assay. They found that a triage test with sensitivity equal to Xpert, 75% specificity, and costs of US$5 per patient tested could reduce total diagnostic costs by 42% in the Uganda setting, and by 34% and 39% respectively in the India and South Africa settings. Our sensitivity analyses showed that the SOC and WHO TDA screening specifications were key determinants of cost-effectiveness, thus providing more comprehensive evidence about the impact of a tuberculosis triage test on the overall cost-effectiveness of the TDA-based strategies compared to SOC. For all groups, due to a lack of data on the progression of signs and symptoms, tuberculosis rescreening was not permitted. Nevertheless, TDA-based approaches demonstrated high screening sensitivity (79% to 100%), indicating that rescreening would likely have minimal impact on these groups. Additional data, however, would be beneficial for the SOC. Lastly, we did not account for a potential negative health impact on children who are inappropriately treated for tuberculosis and costs for their caregivers.
Conversely, our study had major strengths in basing the diagnostic accuracy and clinical outcomes underlying modelling assumptions, and also the costs on primary empirical data collection and analysis conducted in the TB-Speed SAM studies conducted in two high tuberculosis incidence countries. These assumptions were applied within a framework that accounted for the complexity of patient pathways, including tuberculosis screening and follow-up assessments. Additionally, we developed new unit cost parameters for tuberculosis care specifically for children hospitalised with SAM.
In 2017, Debes et al.29 assessed the cost-effectiveness of tuberculosis diagnosis using microscopic observation drug susceptibility, Xpert and empiric treatment for all patients (the decision to treat all children who present for suspected tuberculosis), in addition to current clinical diagnostic practices in Ugandan children. In line with our findings, they found that case fatality rate for untreated tuberculosis, SOC specifications, and tuberculosis prevalence were major determinants of cost-effectiveness, and recommended empiric treatment in settings with high risk of tuberculosis death. We found that the TB-Speed two-step TDA (with a screening step), was also cost-effective in high tuberculosis incidence countries, and even presented a lower ICER and false positives rate, compared to the WHO and one-step TDA.
Further research is needed to assess the total health and budget impact of TDA-based interventions, as well as their cost-effectiveness when adapted to different populations of children and implemented in decentralised settings. Future studies should also explore the combination of improved child tuberculosis diagnosis strategies with tuberculosis disease severity assessment and eligibility for shortened anti-tuberculosis treatments, which have been shown to be highly cost-effective.30 Additionally, the potential of innovative technologies, such as artificial intelligence for reading CXR, should be evaluated to further enhance tuberculosis diagnosis and treatment strategies.
A key aspect of the TB-Speed TDAs is the inclusion of results from an abdominal ultrasound exam, which raises questions about the feasibility of decentralised implementation and scaling at secondary-level care facilities, such as district health hospitals, due to the costs of ultrasound equipment and the availability of trained personnel. However, Chabala et al.14 found no significant added value of abdominal ultrasound in the TB-Speed diagnostic algorithms, suggesting that the TB-Speed TDA could be more cost-effective without it.
In considering the relevance of these results to other contexts, decision-makers will need to take into account relevant cost-effectiveness thresholds in their setting and resources available for new interventions, the applicability of the case fatality rates and tuberculosis prevalence in the population studied in this analysis, and the relative importance to give to the false positive rate. Different TDAs may be preferred in different contexts, but this study provides strong evidence that any of the three TDAs may be preferred to current care.
The WHO has conditionally recommended incorporating TDAs, pending validation, into existing case detection strategies to support the decentralisation of clinical tools and improve the identification of tuberculosis in children. To our knowledge, this is the first multi-country study to assess the cost-effectiveness of using treatment decision algorithms in childhood tuberculosis services. This analysis focuses on the vulnerable group of children hospitalised with SAM, and contributes valuable evidence to support the interim WHO recommendation on decentralised models of care. Our findings show that TDA-based approaches are highly cost-effective for the vulnerable group of children hospitalised with SAM, compared to current practices, and our sensitivity analysis suggests that these results are robust.
Contributors
MDE, MH, NM, PD, OM, MB, EW conceived the health economics analysis plan. PD developed the cost-effectiveness model. MDE, MH, PD and NM had access to and verified the underlying study data. MDE conducted the cost and the cost-effectiveness analyses. MDE and MH coordinated international economic data collection and design of the patient pathways. MDE, PD, MH, NM interpreted economic results. OM, CC, MB, EW, CR provided scientific expertise and guidance. OM, CC, MB, EW, GB, PS, BN, VM provided clinical expertise.
OM, MB, EW conceived and designed the TB-Speed Severe Acute Malnutrition study, and acquired the project financial support. OM, MB, and EW led the study at international level. EW led the study in Uganda, MHTNN conducted the statistical analysis of the TB-Speed Severe Acute Malnutrition study.
MDE wrote the first draft and all authors reviewed, edited and approved the final version of the manuscript. MDE, PD, OM, CC, MB, EW were responsible for the decision to submit the manuscript.
Data sharing
Aggregated data for all analyses will be publicly available with the publication on a GitHub repository under a Creative Commons Attribution (CC-BY 4.0) licence (URL: https://github.com/petedodd/TBSsam).
Declaration of interests
All authors declare no competing interests.
Acknowledgments
We thank all the children and their families who participated in the study and the healthcare workers of the participating hospitals and laboratories. We thank the Ministries of Health and National Tuberculosis Programmes of participating countries for their support. We thank members of the TB-Speed Scientific Advisory Board who gave technical advice on the design of the study and approved the protocol: Anneke Hesseling (Stellenbosch University, Cape Town, South Africa), Luis Cuevas (Liverpool School of Tropical Medicine, Liverpool, UK), Malgorzata Grzemska and Sabine Verkuijl (WHO, Geneva, Switzerland), Philippa Musoke (Makerere University, Kampala, Uganda), and Mark Nicol (University of Western Australia, Perth, WA, Australia).
References
