Cost-effectiveness analysis of interventions to improve diagnosis and preventive therapy for paediatric tuberculosis in 9 sub-Saharan African countries: A modelling study

Background Over 1 million children aged 0 to 14 years were estimated to develop tuberculosis in 2021, resulting in over 200,000 deaths. Practical interventions are urgently needed to improve diagnosis and antituberculosis treatment (ATT) initiation in children aged 0 to 14 years and to increase coverage of tuberculosis preventive therapy (TPT) in children at high risk of developing tuberculosis disease. The multicountry CaP-TB intervention scaled up facility-based intensified case finding and strengthened household contact management and TPT provision at HIV clinics. To add to the limited health-economic evidence on interventions to improve ATT and TPT in children, we evaluated the cost-effectiveness of the CaP-TB intervention. Methods and findings We analysed clinic-level pre/post data to quantify the impact of the CaP-TB intervention on ATT and TPT initiation across 9 sub-Saharan African countries. Data on tuberculosis diagnosis and ATT/TPT initiation counts with corresponding follow-up time were available for 146 sites across the 9 countries prior to and post project implementation, stratified by 0 to 4 and 5 to 14 year age-groups. Preintervention data were retrospectively collected from facility registers for a 12-month period, and intervention data were prospectively collected from December 2018 to June 2021 using project-specific forms. Bayesian generalised linear mixed-effects models were used to estimate country-level rate ratios for tuberculosis diagnosis and ATT/TPT initiation. We analysed project expenditure and cascade data to determine unit costs of intervention components and used mathematical modelling to project health impact, health system costs, and cost-effectiveness. Overall, ATT and TPT initiation increased, with country-level incidence rate ratios varying between 0.8 (95% uncertainty interval [UI], 0.7 to 1.0) and 2.9 (95% UI, 2.3 to 3.6) for ATT and between 1.6 (95% UI, 1.5 to 1.8) and 9.8 (95% UI, 8.1 to 11.8) for TPT. We projected that for every 100 children starting either ATT or TPT at baseline, the intervention package translated to between 1 (95% UI, −1 to 3) and 38 (95% UI, 24 to 58) deaths averted, with a median incremental cost-effectiveness ratio (ICER) of US$634 per disability-adjusted life year (DALY) averted. ICERs ranged between US$135/DALY averted in Democratic of the Congo and US$6,804/DALY averted in Cameroon. The main limitation of our study is that the impact is based on pre/post comparisons, which could be confounded. Conclusions In most countries, the CaP-TB intervention package improved tuberculosis treatment and prevention services for children aged under 15 years, but large variation in estimated impact and ICERs highlights the importance of local context. Trial registration This evaluation is part of the TIPPI study, registered with ClinicalTrials.gov (NCT03948698).


Analysis of intervention effects
The effect of the intervention as an incidence rate ratio (IRR) was quantified by statistical analysis of before and after data at the site level.This focussed on changes in 2 quantities, stratified by age group (<5 years and 5-14 years): 1) the rate of initiating children on anti-TB treatment (ATT); 2) the rate of initiating children on TPT.These quantities and intervention impacts were modelled with a Bayesian generalised linear mixed-effects model, with checks against a country-wise random effects meta-analysis.For all quantities, the counts were modelled as Poisson distributed counts with means equal to the product of rates and observation-time.

Statistical model specification
We used a Bayesian generalised linear model with site-level random effects.
Let be the event count for record , and (in the notation of Gelman & Hill [1] ) let be the index for the country of the i-th record ( ) and be the site of the i-th record, and be the corresponding observation-time.
We use the model where with priors .Country-level IRRs were computed as Inference was performed using rstanarm [2] using the default priors noted above, and 2 chains of length 6,000 (the first 1,000 iterations of each discarded as burn in).A posterior sample of 10,000 was used as the basis for modelling.

Inference diagnostics
Gelman-Rubin R-hat statistics and effective sample sizes (ESS) suggested good convergence (see Table A2 ).Costing and care cascades

Overview
Costs for the standard of care for TB diagnosis, ATT and TPT were derived from published literature sources.Country-specific costs were used where possible otherwise costs were transferred from other countries by applying relevant purchasing power parity conversion factors.Table A4 shows the assumptions used in deriving unit costs for the standard of care.All historical costs were exchanged to US$ using the World Bank-based period average exchange rates at the time of the costing, [3] and then adjusted to 2020 prices using US inflation rates.[4] We estimated incremental costs of the intervention by considering all CaP-TB-supported costs as incremental to the standard of care unit costs in the cascade of care for childhood TB management.
We defined incremental costs in this context as the additional costs that a public health system would incur in order to improve utilisation (through increased access) of paediatric tuberculosis services, beyond what the system currently provides.We used an activity-based, top-down approach to costing, using budget and expenditure data, together with input from field teams on expenditure breakdowns to estimate intervention costs.Bottom-up estimates were calculated and applied for one activity in one country that was implemented late on during the intervention resulting in very low volumes and unreasonably high unit costs.
CaP TB budget data were organised in cost categories and sub-categories, excluding costs not directly relevant to project implementation, and then the fractional contribution of each subcategory to the overall cost category was calculated, assuming that this remained constant over time.These fractions were applied to actual expenditure data from EGPAF financial records to disaggregate expenditure to the sub-category level.However, for TB medicines and diagnostics, exact costs incurred by each country for each item purchased were taken from expenditure data directly.For both budget and expenditure data, EGPAF overhead costs, staff costs for time spent on donor report/award management (EGPAF related) and CaP TB project specific monitoring and evaluation (M & E) infrastructure costs were excluded because these were assumed not relevant to MoH-led M&E routine implementation.The tool used to categorise and allocate costs is shown in Table A3 .
Costs were then assigned to individual activities across the full range of interventions using the following groups: project set-up and demand generation; community-based household contact tracing; facility-based household contact tracing; screening of non-contacts in HIV entry points; screening of non-contacts in non-HIV entry points; TB evaluation and diagnosis; TB treatment; TB preventive therapy; evaluation (including M & E and research to allow exclusion); and finally, program management.Country teams were asked to assess the proportion of costs for category and sub-category that was associated with each activity.Costs disaggregated across categories and activities were then used to develop unit costs for each direct patient care activity.All research related costs were identified and excluded from the final analysis in order to represent real world implementation.All costs were available in US$ and were adjusted to 2020 prices using US inflation rates.
Costs for intervention and standard of care were therefore calculated as sums over activities: Cost, SoC = ∑(activity, SoC) ⨉ (SoC activity unit cost)

Unit costs and care cascades
The data in Table A4 show the unit costs for each country and activity under the standard of care, and the incremental unit costs for the intervention (ie the unit costs under intervention are the sum of standard of care unit costs and incremental unit costs).We were unable to identify an appropriate value for household contact tracing in the literature, and so considered this in sensitivity analysis.The cascade of care for ATT is shown in Table 1 in the main article.The number of children per child started on ATT that were screened, identified as presumptive TB, and tested with Xpert all rose under the intervention, as did the cost of all activities (including ATT) per child treated (see Figure A8 ).The bulk of the costs per child treated for TB, particularly under the intervention, were associated with screening activities (see Figure A7 ).A5 6.Most countries had a roughly 1:1 ratio between index case households and children started on TPT, except Uganda where around 2 households were traced per child started on TPT.Very few household contacts 5 years or older were initiated on TPT, consistent with current priority targeting; conversely, the majority of children starting TPT at the HIV entry-point were 5 years or older.In all countries, more children started TPT as household contacts than via the HIV entry-point, but this varied from all, to a little over half.The cascade associated with screening of household contacts is shown in Table A6 7, aggregated across countries.Overall, around 2.5% of child contacts were started on ATT.

Overview
Outcomes on mortality and incident cases averted through TPT are lacking from the data and were based on modelling.Models were stratified by age (0-4 years and 5-14 years), and HIV/ART status (assuming all children were on ART).
Changes in mortality were modelled by deriving a change in the fraction of children receiving ATT from the intervention effect analysis and applying estimates of case-fatality ratios from systematic review to the treated and untreated groups (stratified by age and HIV/ART status).The cascade of care, and associated resource use, leading up to a treatment was based on intervention data; the ratios of presumptive TB identified to ATT, and Xpert tests to ATT were assumed to be lower under standard of care by a factor based on comparing with baseline data (see Figure A8 ).The number of children screened was assumed proportional to presumptive TB identified; the ratio was based on data from the intervention in all countries, and data from some countries on how this ratio changed from before the intervention.
The change in children receiving TPT from the intervention effect analysis was disaggregated by entry-point (household contact or HIV entry-point).Different baseline risks of developing incident TB disease were assumed in these groups (ie household exposure vs background annual risk of infection, followed by progression).Hazard ratios for TPT effect from systematic review were applied to calculate the reduction in incidence.The probability of receiving ATT for incident TB was based on WHO age-and country-specific estimates of case detection ratio, scaled-up to account for likely superior detection among household contacts and HIV clinic attendees.Mortality was again modelled using relevant case-fatality ratios from literature for each group.To model the resources required, we used data on the number of households investigated per household contact starting TPT, and assumed that children starting PT via the HIV entry-point would receive screening comparable with that to identify presumptive TB.
Schematically, the health impact is based on the outcomes for those treated (ATT or PT) or not as and are thus calculated relative to each child treated at baseline.
Disability-adjusted life-years were calculated on the basis of life-years lost (i.e.neglecting decrement to quality of life during disease), discounted at 3%.Country-year and age specific UN Population Division World Population Prospects life tables were used to calculate a simple mean across ages within each age group (0-4 years and 5-14 years).
To quantify uncertainty we used probabilistic sensitivity analysis (PSA).Country-level effect estimates and baseline rates were drawn from the Markov chain Monte Carlo (MCMC) samples outputted from inference, and merged against samples of parameters used in the outcome modelling.10,000 samples were used.Unit costs were applied to the activities driving cost included in the TIPPI data.We used gamma distributions to represent uncertainty in all SoC cost parameters used in the model.Uncertainty estimates for INT cost parameters were not available and not represented.The economic evaluation was performed from the public health system perspective, with outputs at country-level including: incremental cost effectiveness ratio (US$ per DALY averted), and plots of the sampled results in the cost-effectiveness plane and cost-effectiveness acceptability curves.Results are presented for each country for the intervention as a whole, as well separately for preventive therapy and improved case detection.

Reproducibility and pre-registration
All code and data to reproduce this analysis are publicly available on GitHub https://github.com/petedodd/tippi .A health economic analysis plan (HEAP) was not produced for this study as it was first undertaken to inform a WHO guideline development plan under time constraints because the randomised studies within CaP-TB were delayed by the impact of the COVID-19 pandemic.

Modelling resource use
For ATT, we considered: screening, testing using Xpert, starting TB treatment, and TB treatment success.Data on numbers screened was only available during the intervention.Comparable baseline data were not available for all quantities and countries.We therefore assumed that the ratio of children screened to presumptive TB identified was the same under the intervention as at baseline, and used data on the increased number of children with presumptive TB identified per child diagnosed with TB under intervention compared to baseline to reduce the baseline number of children screened per child started on ATT (see Figure A8 ).For countries lacking this data, we applied an average of this ratio across other countries.Similarly, we reduced the number of Xpert tests per child treated at baseline compared to under intervention using before/after estimates of its change, and averages across countries were lacking.Data on treatment success were not used in resource modelling: costs were assumed to apply for every child start on ATT.
For TPT, we used data in Table A5 6 and Table A6 7 to disaggregate activities by entry-point and to determine the number of households visited from the number of contacts starting TPT.As with ATT, costs for TPT were assumed to accrue regardless of completion of the course.To include household case-finding, resources were modelled for all countries using the aggregate data in Table A6 7.

Modelling outcomes
Outcomes of TB with and without ATT were modelled using a case fatality ratio (CFR) approach, as followed in Dodd et al. [5] CFRs with and without ATT in each age group were based on the systematic review and meta-analysis of Jenkins et al. [6] The effect of HIV and ART on treatment outcomes was based on a analysis of Jenkins et al. [6] data presented in Dodd et al. [5] and untreated TB outcomes for children with HIV were based on the expert elicitation presented in Dodd et al. [5] All children with HIV were assumed to be on ART.Expected life-years and discounted life-years were based on simple means of interpolations of United Nations (World Population Prospects 2019 revision) country-and year-specific life tables described in Dodd et al. [7] using 2020 as the reference year (implemented using the discly R package available at https://github.com/petedodd/discly).The life-expectancy of children with HIV was not assumed to be the same as those without HIV-infection.
Because there are no child-specific tariffs for untreated TB disease, we used the adult tariff of 0.331 to capture the contribution of morbidity to DALYs.[8] While there are no data to directly inform the duration of TB disease with or without treatment in children, they are likely to be substantially shorter than the global average duration of disease for adults (~ 1 year).We therefore assumed that ultimately untreated TB in children had a mean duration of 6 months, and that children with ultimately treated TB had untreated TB for 3 months prior to treatment.As a sensitivity analysis to explore the importance of these assumptions, we also ran our analyses without including the morbidity contribution to DALYs.
For TPT, entry-point data shown in Table 6 was used to disaggregate those starting TPT by age and entry-point.All those starting PT through the HIV entry-point were assumed to be HIV-infected; HIV-infection prevalence in household contacts was based on data presented in Martinez et al.
(2017).[9] All children with HIV were assumed to be on ART.Incident TB risks for household contacts were base on combining systematic review data on LTBI prevalence from Fox et al, [10] with progression risks in those with LTBI from Martinez et al. (2020).[11] For children at the HIV entry-point, TB risk was based on a background community annual risk of TB infection (ARI), combined with progression risks base on Martinez et al. (2020).[11] Children without LTBI were assumed to have zero risk of incident TB.For children with HIV (ie all those at HIV entry-point and some household contacts), age-specific risks of incident TB had an incidence rate ratios from the systematic review of Dodd et al [12] applied to model the increased risk of TB due to HIV-infection, and the protection due to ART.The TPT hazard ratio from Martinez et al. (2020) [11] was applied to children modelled as having LTBI among household contacts, and an HIV-specific hazard ratio from Zunza et al [13] applied to children at the HIV entry-point.In the base case analysis, all children starting PT were assumed to receive full protection from review meta-analysis.In the 'TPT completion' sensitivity analysis, the proportion of children completing TPT under SoC and the intervention ( Table A1 ) were assumed to receive protection, whereas the rest had no protection from TB incidence.PT was not assumed to have protective effects beyond the immediate risk of TB incidence.
The chances of incident TB being detected were based on WHO country-and age-specific estimates of case detection ratios (CDRs).Development of incident TB was assumed to happen over a 1 to 2 year timescale and discounting of health outcomes therefore neglected.TB outcomes with and without ATT were then modelled as described above.Implicitly a life-time horizon is used, with the assumption that differences in health outcomes and costs occur in the present.We neglected morbidity during TB.
Modelling and analysis was performed using R version 4.2.0.[14]

Parameter distributions
Parameter described in the previous sections were all represented in a probabilistic sensitivity analysis, sampling from the distributions shown in Table A7 8, which also shows the mean and interquartile range associated with each parameter.

Figure
Figure A1 Estimates of the Cap TB intervention effects on TB treatment among children aged 0-4 years.Individual sites are joined with lines.CaP-TB=Catalyzing Pediatric TB Innovations, DRC=Democratic Republic of the Congo, TB=tuberculosis.

Figure
Figure A2 Estimates of the Cap TB intervention effects on TB treatment among children aged 5-14 years.Individual sites are joined with lines.CaP-TB=Catalyzing Pediatric TB Innovations, DRC=Democratic Republic of the Congo, TB=tuberculosis.

Figure
Figure A3 Estimates of the Cap TB intervention effects on TB treatment among children aged 0-14 years.Individual sites are joined with lines.CaP-TB=Catalyzing Pediatric TB Innovations, DRC=Democratic Republic of the Congo, TB=tuberculosis.

Figure
Figure A4 Estimates of the Cap TB intervention effects on TB preventive therapy among children aged 0-4 years.Individual sites are joined with lines.CaP-TB=Catalyzing Pediatric TB Innovations, DRC=Democratic Republic of the Congo, TB=tuberculosis.

Figure
Figure A5 Estimates of the Cap TB intervention effects on TB preventive therapyamong children aged 5-14 years.Individual sites are joined with lines.CaP-TB=Catalyzing Pediatric TB Innovations, DRC=Democratic Republic of the Congo, TB=tuberculosis.

Figure
Figure A6 Estimates of the Cap TB intervention effects on TB preventive therapy among children aged 0-14 years.Individual sites are joined with lines.CaP-TB=Catalyzing Pediatric TB Innovations, DRC=Democratic Republic of the Congo, TB=tuberculosis.

Figure
Figure A7 Composition of intervention costs for diagnosing and treating TB.DRC=Democratic Republic of the Congo, TB=tuberculosis.

Figure
Figure A8 Change in resource use under intervention in comparison to the pre-intervention period (baseline).DRC=Democratic Republic of the Congo, TB=tuberculosis.

Figure
Figure A10 Cost-effectiveness acceptability curves for the Cap TB package of interventions in comparison to standard of care (age 0-4 years).

Figure A1 2
Figure A1 2 Cost-effectiveness acceptability curves for the combined Cap TB intervention package in comparison to standard of care with a discount rate of 0% per year.The figure shows the probability that an intervention is cost-effective in each country, based on the proportion of simulations in which the comparison of the intervention to the standard of care falls below the cost-effectiveness threshold shown on the horizontal axis.CaP-TB=Catalyzing Pediatric TB Innovations, DALY=disability-adjusted life years, DRC=Democratic Republic of the Congo, USD=United States dollar.

Figure A1 3
Figure A1 3 Cost-effectiveness acceptability curves for the combined Cap TB intervention package in comparison to standard of care with a discount rate of 5% per year.The figure shows the probability that an intervention is cost-effective in each country, based on the proportion of simulations in which the comparison of the intervention to the standard of care falls below the cost-effectiveness threshold shown on the horizontal axis.CaP-TB=Catalyzing Pediatric TB Innovations, DALY=disability-adjusted life years, DRC=Democratic Republic of the Congo, USD=United States dollar.

Figure A1 4
Figure A1 4 Cost-effectiveness acceptability curves for the combined Cap TB intervention package in comparison to standard of care with improvements in ATT success and TPT completion rates included.The figure shows the probability that an intervention is cost-effective in each country, based on the proportion of simulations in which the comparison of the intervention to the standard of care falls below the cost-effectiveness threshold shown on the horizontal axis.CaP-TB=Catalyzing Pediatric TB Innovations, DALY=disability-adjusted life years, DRC=Democratic Republic of the Congo, USD=United States dollar.

Table A2
Effective sample size and Rhat convergence statistics across parameters for each model.ATT=anti-tuberculosis treatment, ESS=effective sample sizes, TPT=tuberculosis preventive therapy.

Table A3
Cost data allocation to TB activities with illustrative line cost items.Cost data allocation tool used to allocate Cap TB project expenditures to TB activities with illustrative line cost items.CXR=chest x-ray, MoH=ministry of health, M&E=monitoring and evaluation, TB=tuberculosis, TPT=tuberculosis preventive therapy.

Cost parameter Description Unit cost parameter estimation assumptions Country SoC mean unit cost (SD) Intervention Incremental Cost
Chest X-ray The cost of chest X-ray Applied a mean chest x-ray cost from 2014 in Tanzania, Uganda and CMR 7.35 (0.

Table A5
The distribution of children initiated on TPT by age and entry point per country.DRC=Democratic Republic of the Congo, HIV=human immunodeficiency virus, TPT=tuberculosis preventive therapy.
Table A6 Cascade for child case-finding during household contact screening
Table A8 Healthcare resource use, health outcomes, costs & cost-effectiveness of the Intensified case-finding intervention in comparison to standard of care(baseline) stratified by age.All outcomes are presented per 100 children initiating anti-tuberculosis treatment (ATT) or tuberculosis preventive therapy (TPT) at baseline.Data are presented as n (95% uncertainty interval) unless otherwise stated.All costs are presented in 2020 United States dollars ($USD). CaP-TB=Catalyzing Pediatric TB Innovations, DALY=disability-adjusted life years, DRC=Democratic Republic of the Congo, HIV=human immunodeficiency virus, SoC=standard of care, TB=tuberculosis, USD=United States dollar.Table A9 Healthcare resource use, health outcomes, costs & cost-effectiveness of the Household contact management & HIV clinic preventive therapy intervention in in comparison to standard of care (baseline) stratified by age.All outcomes are presented per 100 children initiating anti-tuberculosis treatment (ATT) or tuberculosis preventive therapy (TPT) at baseline.Data are presented as n (95% uncertainty interval) unless otherwise stated.All costs are presented in 2020 United States dollars ($USD). CaP-TB=Catalyzing Pediatric TB Innovations, DALY=disability-adjusted life years, DRC=Democratic Republic of the Congo, HIV=human immunodeficiency virus, SoC=standard of care, TB=tuberculosis, USD=United States dollar.

Table A10
Incremental cost-effectiveness ratios (ICERs) for different sensitivity analyses per country.ICERs are presented as cost in $US per discounted disability-adjusted life year (DALY) averted.ATT=anti-tuberculosis treatment, DRC=Democratic Republic of the Congo, TPT=tuberculosis preventive therapy.* ie DALYs without a contribution from morbidity Table A11 Cost-effectiveness threshold values where interventions first exceed 50% probability of being cost-effective (CET50) per country.The table does not show Cameroon and Lesotho because intervention does not exceed 50% probability of being cost-effective at all threshold values.DALY=disability-adjusted life year, DRC=Democratic Republic of the Congo, US$=United States dollar.