Integrated treatment-decision algorithms for childhood TB: modelling diagnostic performance and costs

Word count: 200 27 Main text word count: 2493 28

23 29 Main text: 3 figures, 2 tables 30 Supplementary Data: 3 tables, 1 figure 31 Key words: pediatrics, tuberculosis, diagnostics, cost-effectiveness, modelling 32 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 20, 2025. ; https://doi.org/10.1101/2025.06.20.25329945doi: medRxiv preprint NOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice. 2

33

To improve childhood tuberculosis (TB) diagnosis, treatment-decision 34 algorithms (TDAs) with and without chest X-ray (CXR) were developed for children under 35 age 10. We aimed to model diagnostic performance and costs of implementing TDAs in 36 primary healthcare (PHC) and district hospital (DH) settings in Uganda. 37

We developed decision-tree models following the TDA pathway from evaluation to 38 treatment-decision. We compared six scenarios with combinations of diagnostic testing 39 (stool and respiratory Xpert, urine lipoarabinomannan, and/or CXR) at PHCs and DHs. 40 Outcomes were diagnostic accuracy and cost per correct treatment-decision for a cohort of 41 10,000 children with presumptive TB using a Monte Carlo simulation from a health system 42 perspective. Costs were reported in 2024 International dollars. 43

In all scenarios, TDA’s had high sensitivity (80.8–91.9%) but low specificity (51.2-44 60.9%). Total diagnostic and treatment costs for the cohort were I$1,768,958–2,458,790; 45 largely driven by overtreatment of false-positive cases. Diagnostic costs were mostly offset 46 by reducing overtreatment. The cost per treatment-decision was lowest using mobile CXR at 47 PHC (I$287.40) and highest with DH referral (I$445.84). 48

The TDAs have high sensitivity and can be implemented at PHCs with lower 49 costs than DHs. Improving specificity and reducing treatment costs would enable affordable, 50 large-scale implementation. 51 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 20, 2025. ; https://doi.org/10.1101/2025.06.20.25329945doi: medRxiv preprint 3

52 The burden of childhood tuberculosis (TB) remains high globally, driven by 53 challenges in diagnosis and subsequent treatment initiation. There were an estimated 1.3 54 million new cases of TB and 191,000 deaths due to TB in children under 15 years of age in 55 2023,1 and it is estimated that 96% of deaths are in children not diagnosed and treated.2 56 In young children, TB often presents with non-specific symptoms and sputum-based 57 testing is rarely feasible due to challenges obtaining samples. 3 Even when sputum can be 58 collected, sensitivity of culture and molecula r testing is reduced due to the paucibacillary 59 nature of childhood disease. 4 Furthermore, chest X-ray (CXR) findings can be 60 heterogeneous and difficult to interpret, especially in children living with HIV (CLHIV) and 61 CXR is typically not available at primary health centers (PHC). 5 Diagnostic capacity 62 combining clinical assessment with clinic al, radiographic, and laboratory information may 63 only be present in district hospitals (DH). Decentralizing childhood TB services has been 64 shown to increase case-finding and improve uptake at PHCs.6 65 To standardize diagnostic approaches and enable more children to initiate treatment 66 earlier, two treatment-decision algorithms (TDAs) were developed for settings with and 67 without CXR, guided by a large individual-patient data meta-analysis. 7 The algorithms were 68 designed with high sensitivity to reduce missed cases, at the expense of low specificity, 69 potentially leading to significant overtreatme nt and associated costs. Previous studies 70 evaluating the TDAs have only validated the clinical scoring system and have not 71 incorporated other aspects such as different combinations of diagnostic tests and settings.8, 9 72 The World Health Organization (WHO) gave a conditional recommendation for the TDAs 73 pending further validation of their accuracy and c onsiderations for implementation, including 74 costs, to guide programmatic adoption and scale-up. 10, 11 The objectives of this analysis 75 were to model the diagnostic performance and costs of implementing TDAs across different 76 implementation scenarios in primary care and hospital-based settings. 77 78 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 20, 2025. ; https://doi.org/10.1101/2025.06.20.25329945doi: medRxiv preprint 4

79 Study population and setting 80 We modelled the WHO TDAs among children under 10 years of age presenting with 81 symptoms suggestive of pulmonary TB, including cough or fever for two weeks or more, 82 poor appetite, weight loss or failure to thrive, and fatigue or reduced playfulness, who do not 83 require urgent care. The setting for the analysis were outpatient clinics in PHCs and DHs in 84 Uganda, which was chosen as a representative high-burden country with a TB incidence 85 rate of 198/100,000 and 37% of TB cases are living with HIV. 1 Of the TB cases reported 86 nationally in 2023, over 12,000 (14%) were among children 0–14 years, 1 but this may be 87 underestimating the true burden. 12 The current standard of care is screening children at 88 health facility entry points, and evaluating those meeting criteria for presumptive TB with HIV 89 testing, clinical examination, CXR, and coll ecting samples for bacteriological confirmation. 12 90 In this setting, most resources for diagnostic testing, including X-ray facilities, on-site 91 laboratories with GeneXpert, and clinical reso urces to perform gastric aspirate and sputum 92 induction are at centralized facilities. Howeve r, approximately 54% of children initially 93 present at PHCs, which have limited access to diagnostics, so children with presumptive TB 94 are often referred to DHs.13 95 96 WHO Treatment Decision Algorithms 97 Each TDA follows a sequence of evaluations. Because the algorithms aim to detect as many 98 cases as possible, a negative result on any step leads to further assessment until a 99 treatment decision is reached. Children at hi gh-risk for rapid disease progression (under 2 100 years of age, living with HIV, or with severe acute malnutrition) are tested for TB 101 immediately, including molecular testing on stool or respiratory samples with GeneXpert, 102 and urine lateral flow lipoarabinomannan (LF-LAM, Determine TB LAM, Abbott, Chicago) for 103 CLHIV. The children not in a high-risk group are assessed for other likely conditions related 104 to their symptoms and return for follow-up in two weeks. Children whose symptoms resolve 105 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 20, 2025. ; https://doi.org/10.1101/2025.06.20.25329945doi: medRxiv preprint 5 by the follow-up visit exit the algorithm and undergo no further testing. Children with 106 persistent symptoms continue for molecular testing. If these tests are negative or not 107 available, the child proceeds to the clinical scoring step, which considers history of TB 108 contact and presence of individual symptoms. Al gorithm A (TDA-A) includes CXR findings in 109 the scoring and can be used where X-ray is available. While this is typically relevant for DHs, 110 mobile vans with portable X-ray machines may be available to expand access to PHCs. 14 111 Algorithm B (TDA-B) is for settings without X-ray, typically PHCs, and only includes clinical 112 signs and symptoms. 113 114 Model structure 115 We developed six scenarios comparing a range of strategies at PHC and the centralized DH 116 approach. We converted the TDAs into a decis ion-tree model with separate arms for each 117 scenario, following the above-described pathway from initial evaluation to treatment decision 118 (Figure 1). Scenario 1 (‘TDA-B’) considered the simplest scenario with only clinical diagnosis 119 using the TDA-B scoring system and no molecular testing. Scenario 2 (‘TDA-A’) explored the 120 impact of adding mobile CXR and use of the TDA-A scoring. Scenario 3 (‘Stool + TDA-B’) 121 and Scenario 4 (‘Stool + TDA-A’) included molecular testing with Xpert Ultra on stool 122 samples and urine LF-LAM for CLHIV, with or without mobile CXR. Scenario 5 (‘Stool + 123 TDA-B + Referral’) began with stool testing at PHC and referred a portion of children with a 124 negative result to DHs for respiratory Xpert Ultra testing and TDA-A with CXR. Children in 125 the high-risk group were more likely to be referred, and those who remained at PHC were 126 evaluated with the TDA-B score. This scenario most closely reflected the current standard of 127 care in Uganda. Scenario 6 (‘Referral’) reflected the centralized approach where all children 128 had CXR and respiratory sample testing at DHs. 129 130 Model parameters and analysis 131 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 20, 2025. ; https://doi.org/10.1101/2025.06.20.25329945doi: medRxiv preprint 6 Estimates for clinical and cost parameters matching the modelled setting were obtained from 132 the literature (Table S1). Several clinical parameter estimates were obtained from a multi-133 site study on decentralization of TB testing and the authors (MB, OM, EW) provided 134 additional data for the Ugandan site. 13 We also consulted expert opinion (see 135 Acknowledgement) for input on clinical parameters, including TB prevalence and resolution 136 of symptoms during follow-up. Due to limited data available on pre-diagnostic loss and the 137 possible impact of new strategies for children in this setting, loss to follow-up before and 138 during the diagnostic process was not included. The costs of clinical examination, sample 139 collection, HIV testing, and TB molecular testing were included as specified for each 140 scenario. The cost of respirat ory sample collection at DH included gastric aspirates for 141 children under five and induced or expectorated sputum for older children. TB treatment 142 costs included direct costs of medications a nd follow-up visits until completion of therapy. 143 Patient costs such as transportation and t he burden of treatment (i.e., giving daily 144 medications and side effects) were not included. Drug-resistant TB was not included as the 145 rates are low among children in Uganda.15 146 The outcomes were diagnostic accuracy and cost per correct treatment-decision 147 (both true-positives and true-negatives). Treatm ent costs were reported separately for true 148 TB cases, overtreatment, and risk group. Outcomes were calculated for a cohort of 10,000 149 children using a Monte Carlo simulation. The analysis adopted a health system perspective, 150 using a time horizon of one year for program implementation and no discounting was 151 applied. Costs were converted to 2024 Internat ional dollars (I$) using World Bank inflation 152 data.16, 17 153 One-way and probabilistic sensitivity analyses (PSA) were conducted to evaluate 154 how uncertainty in model parameters impacted outcomes. TreeAge Pro 2024 was used for 155 analysis. Details of the evaluation are reported following the Consolidated Health Economic 156 Evaluation Reporting Standards (CHEERS) guidance (Table S2).18 Ethical approval was not 157 required as all parameters were obtained from published literature and there was no human 158 subject participation. 159 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 20, 2025. ; https://doi.org/10.1101/2025.06.20.25329945doi: medRxiv preprint 7 160

161 Diagnostic performance 162 Overall, the diagnostic accuracy of the sc enarios was moderate, ranging from 55.0–61.6%, 163 balancing a high sensitivity and low specificity (Table 1, full results in Table S3). The 164 sensitivity was high in all scenarios, ranging fr om 80.8–91.9%, indicating that the TDAs are 165 not missing many TB cases. However, the s pecificity was consistently low (51.2–60.9%), 166 resulting in a low positive predictive value (PPV ) (5.5–6.3%), particularly in Scenarios 1-5 at 167 PHCs with low prevalence of TB disease (3%). The negative predictive value (NPV) was 168 above 98% across all scenarios. 169 When comparing Scenario 1 and 2 (clinical diagnosis with and without CXR 170 respectively), the addition of mobile CXR at PHCs improved both sensitivity (80.8% versus 171 82.7%, respectively) and specificity (56.1% ve rsus 60.9%, respectively). The addition of 172 molecular testing, with or without CXR, in Models 3 and 4 improved sensitivity (86.3% 173 versus 87.0%, respectively) and specificity (54. 0 versus 58.7%, respectively) compared to 174 Scenarios 1 and 2. Referring either some or all children for CXR and respiratory sample 175 testing (Scenarios 5 and 6, respectively) im proved sensitivity (86.3% versus 91.9%, 176 respectively) but lowered specificity (56.6% versus 51.2%, respectively). These relationships 177 were similar when comparing high and low-risk groups. However, low-risk groups had on 178 average 10% lower sensitivity and 30% higher specificity compared to the overall results. 179 180 Costs 181 The total costs of TB testing and treatment for a cohort of 10,000 children ranged from 182 I$1,768,958 in Scenario 2 to I$2,458,790 in Scenario 6 (Table 3). The costs were driven by 183 the overtreatment of false-positive cases, whic h resulted in more than I$1.4 million in every 184 scenario. The lower specificity in the high-ri sk group resulted in a larger proportion of 185 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 20, 2025. ; https://doi.org/10.1101/2025.06.20.25329945doi: medRxiv preprint 8 overtreatment costs than in the low-risk group. Due to the low prevalence of TB at PHC, the 186 cost for treating true TB cases was only 5% of total costs for Scenarios 1-5. 187 The diagnostic costs are lowest in Scenario 1 with only clinical diagnosis (I$139,647) 188 and highest in Scenario 6 with referral testing (I$540,192). When comparing scenarios, the 189 increased cost of diagnostics was mostly offset by the decreased cost of overtreatment. For 190 example, the addition of mobile CXR between Scenarios 3 and 4 increased diagnostic costs 191 by I$101,142 but reduced overtreatment by I$165,259. However, referring all children in 192 Scenario 6 substantially increased diagnostic cost s but did not reduce overtreatment. This is 193 reflected in the cost per correct treatment-dec ision, which was lowest using mobile CXR at 194 PHC (Scenario 2, I$287.40) and highest with DH referral (Scenario 6, I$445.84). 195 196 Sensitivity Analyses 197 One-way sensitivity analyses showed that the parameters with the greatest impact were the 198 specificity of TDA’s, cost of TB treatment , and proportion of children who were high-risk 199 (Figure 2). As TDA specificity increases, reduc tion in overtreatment lowers the cost per 200 case. Decreasing treatment costs and proportions of high-risk children also decreased the 201 cost per case. The PSA indicated Scenario 2 was more cost-effective than other scenarios 202 across a range of willingness-to-pay thresholds (Figure 3). Additional PSAs are in Figure S1. 203 204

205 This is the first analysis to model both the diagnostic accuracy and costs of the new 206 TDAs in a cohort of children across a range of implementation scenarios. The TDAs were 207 developed to support the clinical assessment for childhood TB and reduce the diagnostic 208 gap. We showed that the TDAs have high sensitivity to detect TB cases with high NPV, but 209 their low specificity leads to substantial overtreatment costs. The current standard of 210 referring children to DHs had both the highest sensitivity and highest cost. However, 211 expanding diagnostic capacity at PHCs improved specificity, reduced overtreatment, and 212 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 20, 2025. ; https://doi.org/10.1101/2025.06.20.25329945doi: medRxiv preprint 9 consequently offset costs. These findings suggest that implementation of TDAs at PHCs 213 with and without TB-specific testing would have high sensitivity to detect TB, and lower costs 214 than referral to a DH, but further improvements are needed to reduce costs of overtreatment. 215 The scenarios with DH referral do not substantially improve accuracy or reduce costs 216 compared to scenarios at PHCs. Although children have access to respiratory testing and 217 CXR at DHs, if these results are negative then children could still be initiated on TB 218 treatment based on the less-specific clinical score. At the same time, there are additional 219 costs for the DH assessment and any previous PHC visits before referral. In PHC scenarios, 220 mobile CXR improved specificity and higher diagnostic costs were offset by the reduction in 221 overtreatment. Stool-based testing did not improve accuracy, again as children with negative 222 stool tests would then be assessed with the clinical score. The difference in clinical pathway 223 for high-risk children increased the sensitivity, but with a trade-off of lower specificity and 224 associated higher overtreatment cost. However, it is important to recognize the benefits of 225 microbiological confirmation, including detection of drug resistance.19 CXR also has benefits 226 of classifying disease severity and eligibility for shorter treatment regimens, or identifying 227 alternative diagnoses.11 Additional benefits of decentralization for children and caregivers 228 include reduced costs and burden of visits to referral facilities. 229 However, it is important to recognize the high total costs. Our model estimated the 230 costs for a cohort of 10,000 children, including diagnostics and treatment, were over I$1.7 231 million for all scenarios. The 2023 funding for TB in Uganda was $32 million and 84% came 232 from international sources.1 In a time of decreased global health funding, it may not be 233 feasible for national programs or donors to cover costs of expanding services, and the cost 234 of diagnostics alone may not be affordable for many countries.20 An analysis of 235 decentralization strategies, which included scale-up costs such as training and equipment, 236 found that decentralization to the PHC level would likely not be cost-effective,21 and a cost-237 effectiveness analysis of TB screening in Uganda found similar challenges in an adult 238 population with low TB prevalence.22 Our sensitivity analyses showed that improved TDA 239 specificity, especially among high-risk children, and reduced treatment costs would provide 240 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 20, 2025. ; https://doi.org/10.1101/2025.06.20.25329945doi: medRxiv preprint 10 the greatest impact. Moreover, implementing newly-recommended shorter treatment 241 regimens for children with non-severe disease would also reduce treatment costs. 10 23 242 Strengths of this analysis include using parameters from studies conducted in similar 243 high-burden settings, including a childhood TB the decentralization study in Uganda.12 We 244 also consulted expert opinion to support the limited data available for some clinical 245 parameters. However, there are still limited data available for childhood TB, and the impact 246 of this uncertainty was explored in the sensitivity analyses. These models did not include 247 pre-diagnostic loss to follow-up, so our estimations are likely overoptimistic. We also did not 248 include patient costs which would likely support more patient-centered algorithms at PHCs. 249 Implementation will require additional resources for training and supporting healthcare 250 workers. 251 Clinical studies to validate TDA performance in high-burden settings are ongoing and 252 the results will inform future implementation. When these studies are completed, it will be 253 valuable to conduct formal budget impact analyses. Additional evaluations including patient 254 costs, caregiver preferences regarding location of care, and feedback from healthcare 255 workers on their experience using the algorithms will inform stakeholder decision-making. 256 Modifications to improve algorithm performance, especially increasing specificity (e.g. 257 through more scalable and accessible pathogen-based diagnostics) should be considered. 258 259

260 Increasing children’s access to TB diagnostic tools is important. Our models indicate that the 261 TDAs have high sensitivity and negative predictive value, enabling increased detection of 262 childhood TB, and can be implemented at primary care centers at lower cost than district 263 hospitals. However, the low specificity and subsequent overtreatment costs could reduce the 264 feasibility of implementation in real-world settings, unless there are further efforts to improve 265 specificity and reduce treatment costs. 266 267 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 20, 2025. ; https://doi.org/10.1101/2025.06.20.25329945doi: medRxiv preprint 11 268 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 20, 2025. ; https://doi.org/10.1101/2025.06.20.25329945doi: medRxiv preprint 12 List of abbreviations: 269 Children living with HIV (CLHIV), Chest X-ray (CXR), District hospital (DH), Lateral flow urine 270 lipoarabinomannan assay (LF-LAM), negative predictive value (NPV), Primary health center 271 (PHC), positive predictive value (PPV), probabilistic sensitivity analysis (PSA), Tuberculosis 272 (TB), treatment-decision algorithm (TDA), World Health Organization WHO). 273 274

275 We would like to thank the following experts for providing input on the clinical parameters for 276 childhood TB: Beate Kampmann, Ben Marais, and Steve Graham. We would also like to 277 thank Ken Gunasekera and James Seddon for providing input on how the algorithms were 278 originally developed. 279 This work was supported by the National Institute of Allergy and Infectious Diseases at the 280 National Institutes of Health [U01AI152087 to CMD]; the National Heart, Lung, and Blood 281 Institute at the National Institutes of Health [K23HL153581 to DJ]; and the German Center 282 for Infection Research (DZIF) [TTU.02.813, funding indicator 8029802813 to MG]. The 283 funders had no role in the identification, design, conduct, and reporting of the analysis. 284 The authors have no conflicts of interest to report. 285 This analysis used data from published studies and did not require ethical approval. 286 287 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 20, 2025. ; https://doi.org/10.1101/2025.06.20.25329945doi: medRxiv preprint 13

288 1. World Health Organization. Global tuberculosis report 2024. 2024. 289 2. Dodd PJ, Yuen CM, Sismanidis C, Seddon JA, Jenkins HE. The global burden of 290 tuberculosis mortality in children: a mathematical modelling study. Lancet Glob Health. 291 2017;5(9):e898-e906. 292 3. Ioos V, Cordel H, Bonnet M. Alternative sputum collection methods for diagnosis of 293 childhood intrathoracic tuberculosis: a systematic literature review. Arch Dis Child. 294 2019;104(7):629-35. 295 4. Atherton RR, Cresswell FV, Ellis J, Kitaka SB, Boulware DR. Xpert MTB/RIF Ultra for 296 Tuberculosis Testing in Children: A Mini-Review and Commentary. Frontiers in Pediatrics. 297 2019;7. 298 5. Berteloot L, Marcy O, Nguyen B, Ung V, Tejiokem M, Nacro B, et al. Value of chest 299 X-ray in TB diagnosis in HIV-infected children living in resource-limited countries: the ANRS 300 12229-PAANTHER 01 study. Int J Tuberc Lung Dis. 2018;22(8):844-50. 301 6. Gunasekera KS, Marcy O, Muñoz J, Lopez-Varela E, Sekadde MP, Franke MF, et al. 302 Development of treatment-decision algorithms for children evaluated for pulmonary 303 tuberculosis: an individual participant data meta-analysis. Lancet Child Adolesc Health. 304 2023;7(5):336-46. 305 7. Kitonsa PJ, Kikaire B, Wambi P, Nalutaaya A, Nakafeero J, Nanyonga G, et al. The 306 Accuracy of the Uganda National Tuberculosis and Leprosy Program diagnostic algorithm 307 and the World Health Organisation treatment decision algorithms for childhood tuberculosis: 308 A retrospective analysis. PLOS Global Public Health. 2025;5(4):e0004026. 309 8. Triasih R, Yani FF, Wulandari DA, Nababan BWY, Ardiyamustaqim MB, Meyanti F, 310 et al. Treatment-Decision Algorithm of Child TB: Evaluation of WHO Algorithm and 311 Development of Indonesia Algorithm. Trop Med Infect Dis. 2025;10(4). 312 9. World Health Organization. TDA4Child initiative 2024 [Available from: 313 https://tdr.who.int/activities/TDA4Child-314 initiative#:~:text=TDR%20(the%20Special%20Programme%20for,evaluate%20the%20perfo315 rmance%2C%20feasibility%2C%20acceptability. 316 10. World Health Organization. WHO consolidated guidelines on tuberculosis Module 5: 317 Management of tuberculosis in children and adolescents. 2022. 318 11. Republic of Uganda Ministry of Health NTaLP. National Clinical Guidelines on the 319 Management of Tuberculosis in Children. 2016. 320 12. Wobudeya E, Nanfuka M, Ton Nu Nguyet MH, Taguebue J-V, Moh R, Breton G, et 321 al. Effect of decentralising childhood tuberculosis diagnosis to primary health centre versus 322 district hospital levels on disease detection in children from six high tuberculosis incidence 323 countries: an operational research, pre-post intervention study. eClinicalMedicine. 324 2024:102527. 325 13. Jo Y, Kagujje M, Johnson K, Dowdy D, Hangoma P, Chiliukutu L, et al. Costs and 326 cost-effectiveness of a comprehensive tuberculosis case finding strategy in Zambia. PLOS 327 ONE. 2021;16(9):e0256531. 328 14. Orikiriza P, Tibenderana B, Siedner MJ, Mueller Y, Byarugaba F, Moore CC, et al. 329 Low resistance to first and second line anti-tuberculosis drugs among treatment naive 330 pulmonary tuberculosis patients in southwestern Uganda. PLoS One. 2015;10(2):e0118191. 331 15. Turner HC, Lauer JA, Tran BX, Teerawattananon Y, Jit M. Adjusting for Inflation and 332 Currency Changes Within Health Economic Studies. Value in Health. 2019;22(9):1026-32. 333 16. Group WB. A Global Database of Inflation 2025 [Available from: 334 https://www.worldbank.org/en/research/brief/inflation-database. 335 17. Husereau D, Drummond M, Augustovski F, De Bekker-Grob E, Briggs AH, Carswell 336 C, et al. Consolidated Health Economic Evaluation Reporting Standards 2022 (CHEERS 337 2022) statement: updated reporting guidance for health economic evaluations. BMC 338 Medicine. 2022;20(1). 339 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 20, 2025. ; https://doi.org/10.1101/2025.06.20.25329945doi: medRxiv preprint 14 18. Kay AW, González Fernández L, Takwoingi Y, Eisenhut M, Detjen AK, Steingart KR, 340 et al. Xpert MTB/RIF and Xpert MTB/RIF Ultra assays for active tuberculosis and rifampicin 341 resistance in children. Cochrane Database Syst Rev. 2020;8(8):Cd013359. 342 19. Zawedde-Muyanja S, Nakanwagi A, Dongo JP, Sekadde MP, Nyinoburyo R, 343 Ssentongo G, et al. Decentralisation of child tuberculosis services increases case finding 344 and uptake of preventive therapy in Uganda. Int J Tuberc Lung Dis. 2018;22(11):1314-21. 345 20. Pantoja A, Kik SV, Denkinger CM. Cost s of novel tuberculosis diagnostics--will 346 countries be able to afford it? J Infect Dis. 2015;211 Suppl 2:S67-77. 347 21. d’Elbée M, Harker M, Mafirakureva N, Nanfuka M, Huyen Ton Nu Nguyet M, 348 Taguebue J-V, et al. Cost-effectiveness and budget impact of decentralising childhood 349 tuberculosis diagnosis in six high tuberculosis incidence countries: a mathematical modelling 350 study. eClinicalMedicine. 2024:102528. 351 22. Murray M, Cattamanchi A, Denkinger C, Van't Hoog A, Pai M, Dowdy D. Cost-352 effectiveness of triage testing for facility-based systematic screening of tuberculosis among 353 Ugandan adults. BMJ Glob Health. 2016;1(2):e000064. 354 23. Zwerling A, Gomez GB, Pennington J, Cobelens F, Vassall A, Dowdy DW. A 355 simplified cost-effectiveness model to guide decision-making for shortened anti-tuberculosis 356 treatment regimens. The International Journal of Tuberculosis and Lung Disease. 357 2016;20(2):257-60. 358 359 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 20, 2025. ; https://doi.org/10.1101/2025.06.20.25329945doi: medRxiv preprint 360 361 Figure 1. Decision-tree model and scenario-specific changes 362 15 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 20, 2025. ; https://doi.org/10.1101/2025.06.20.25329945doi: medRxiv preprint 16 Table 1. Diagnostic accuracy of scenarios for a cohort of 10,000 children under 10 years of age with presumptive TB 363 364 365 366 Scenario* Scenario 1: TDA-B Scenario 2: TDA-A Scenario 3: Stool + TDA-B Scenario 4: Stool + TDA-A Scenario 5: Stool + TDA-B + Referral Scenario 6: Referral TDA Algorithm B Algorithm A Algorithm B Algorithm A Algorithm A and B Algorithm A Chest X-ray Available No Yes No Yes Partial Yes TB-specific testing No** No** Yes Yes Yes Yes Accuracy, overall 56.8% 61.6% 55.0% 59.6% 57.6% 55.2% High-risk group 32.9% 40.3% 30.0% 37.3% 34.2% 41.6% Not high-risk group 86.0% 87.4% 85.4% 86.7% 86.1% 86.4% Sensitivity, overall 80.8% 82.7% 86.3% 87.0% 86.3% 91.9% High-risk group 86.7% 89.8% 92.8% 94.0% 93.4% 97.0% Not high-risk group 73.8% 74.5% 78.7% 78.7% 78.0% 79.2% Specificity, overall 56.1% 60.9% 54.0% 58.7% 56.6% 51.2% High-risk group 31.2% 38.8% 28.0% 35.5% 32.3% 35.6% Not high-risk group 86.4% 87.8% 85.6% 86.9% 86.3% 87.1% Positive-predictive value 5.5% 6.3% 5.6% 6.3% 5.9% 16.8% High-risk group 3.8% 4.4% 3.9% 4.3% 4.1% 14.2% Not high-risk group 14.9% 16.5% 15.0% 16.3% 15.6% 38.0% Negative predictive value 98.9% 99.1% 99.2% 99.3% 99.2% 98.3% High-risk group 98.7% 99.2% 99.2% 99.5% 99.4% 99.1% Not high-risk group 99.0% 99.1% 99.2% 99.2% 99.2% 97.7% Evaluated with clinical score 64.8% 64.8% 62.1% 62.1% 61.1% 67.3% High-risk group 100% 100% 96.2% 96.2% 94.8% 88.1% Not high-risk group 21.9% 21.9% 20.4% 20.4% 20.1% 19.3% *The setting for all scenarios is at primary health clinics, unless specified as referral to district hospitals **Scenario 1 and Scenario 2 follows the algorithm without TB-specific testing (Xpert Ultra or LF-LAM) Legend for abbreviations: urine lateral flow lipoarabinomannan (LF-LAM), Tuberculosis (TB), treatment-decision algorithm (TDA) . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 20, 2025. ; https://doi.org/10.1101/2025.06.20.25329945doi: medRxiv preprint 17 367 368 369 370 371 372 373 374 375 376 377 378 Table 2. Costs of diagnostic testing and TB treatment of scenarios for a cohort of 10,000 children under 10 years of age with presumptive TB Scenario Scenario 1: TDA- B Scenario 2: TDA- A Scenario 3: Stool + TDA-B Scenario 4: Stool + TDA-A Scenario 5: Stool + TDA-B + Referral Scenario 6: Referral Total cost $1,772,666.21 $1,768,957.61 $1,995,927.74 $1,932,535.10 $2,001,305.07 $2,458,790.18 Treatment, total $1,633,019.46 $1,466,310.86 $1,712,749.66 $1,548,215.52 $1,618,885.47 $1,918,598.54 Treatment, TB cases $89,877.68 $92,052.14 $96,038.65 $96,763.47 $96,038.65 $321,457.67 High-risk $52,187.04 $53,999.09 $55,811.14 $56,535.96 $56,173.55 $242,814.70 Not high-risk $37,690.64 $38,053.05 $40,227.51 $40,227.51 $39,865.10 $78,642.97 Overtreatment $1,543,141.78 $1,374,258.72 $1,616,711.01 $1,451,452.05 $1,522,846.82 $1,597,140.87 High-risk $1,328,232.65 $1,181,819.01 $1,389,479.94 $1,244,878.35 $1,306,850.46 $1,468,847.73 Not high-risk $214,909.13 $192,439.71 $227,231.07 $206,573.70 $215,996.36 $128,293.14 Non-treatment costs $139,646.75 $302,646.75 $283,178.08 $384,319.58 $382,419.60 $540,191.64 Cases correctly diagnosed 5683 6155 5497 5955 5756 5515 Cost per correct decision $311.92 $287.40 $363.09 $324.52 $347.69 $445.84 Legend for abbreviations: Tuberculosis (TB), treatment-decision algorithm (TDA) . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 20, 2025. ; https://doi.org/10.1101/2025.06.20.25329945doi: medRxiv preprint 379 Figure 2. One-way sensitivity analyses for each scenario 380 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 20, 2025. ; https://doi.org/10.1101/2025.06.20.25329945doi: medRxiv preprint 19 381 Figure 3: Cost-Effectiveness Acceptability Curve 382 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 20, 2025. ; https://doi.org/10.1101/2025.06.20.25329945doi: medRxiv preprint 20 Figure legends: 383 Figure 1. Decision-tree model and scenario-specific changes 384 Legend: Part A shows a simplified version of the clinical pathway converted into a decision-385 tree model for each scenario. Part B shows the diagnostic tests used for each model 386 scenario. 387 388 Figure 2. One-way sensitivity analyses for each scenario 389 Legend: Parameters are shown in order of decreasing impact on the outcome of cost per 390 correct treatment-decision. Blue=low range of parameter, red=high range of parameter 391 392 Figure 3: Cost-Effectiveness Acceptability Curve 393 Legend: Probabilistic sensitivity analysis for the joint uncertainty of all parameters using a 394 Monte-Carlo simulation with 10,000 iterations. The cost-effectiveness acceptability curve 395 shows which scenario was most cost-effective at the given willingness-to-pay threshold. 396 397 398 399 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 20, 2025. ; https://doi.org/10.1101/2025.06.20.25329945doi: medRxiv preprint