Expanding molecular diagnostic coverage for tuberculosis by combining computer-aided chest radiography and sputum specimen pooling: a modeling study from four high burden countries

Expanding molecular diagnostic coverage for tuberculosis by combining computer-aided chest radiography and sputum specimen pooling: a modeling study from four high burden countries | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Expanding molecular diagnostic coverage for tuberculosis by combining computer-aided chest radiography and sputum specimen pooling: a modeling study from four high burden countries Andrew James Codlin, Luan Nguyen Quang Vo, Tushar Garg, Sayera Banu, and 9 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3813705/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 11 You are reading this latest preprint version Abstract Background In 2022, fewer than half of persons with tuberculosis (TB) have access to molecular diagnostic tests for TB due to their high costs. Studies have found that computer-aided detection using artificial intelligence (AI) for chest X-ray (CXR) and sputum specimen pooling can each reduce testing costs. We modeled the combination of both strategies to estimate potential savings in consumables that could be used to expand access to molecular diagnostics. Methods We obtained Xpert testing and positivity data segmented into deciles by AI probability scores for TB from community- and healthcare facility-based active case finding conducted in Bangladesh, Nigeria, Viet Nam and Zambia. AI scores in the model were based on CAD4TB version 7 (Zambia) and qXR (all other countries). We modeled four ordinal screening and testing approaches involving computer-aided CXR to indicate individual and pooled testing. Setting a false negative rate of 5%, for each approach we calculated additional and cumulative savings over the baseline of universal Xpert testing as well as the theoretical expansion in diagnostic coverage. Results In each country, the optimal screening and testing approach was to use AI to rule out testing in deciles with low AI scores and guide pooled and individual testing in persons with moderate and high AI scores, respectively. This approach yielded cumulative savings in Xpert tests over baseline ranging from 50.8% in Zambia to 57.5% in Nigeria and 61.5% in Bangladesh and Viet Nam. Using these savings, diagnostic coverage theoretically could be expanded by 34–160% across the different approaches and countries. Conclusions Using a combination of AI and CXR to inform different pooling strategies may optimize TB diagnostic test use, and could extend molecular tests to more people who need them. The optimal AI thresholds and pooled testing strategy varied across countries, which suggests that bespoke screening and testing approaches may be needed for differing populations and settings. GeneXpert Pooling AI CAD X-ray CXR ACF Active case finding Tuberculosis diagnosis Figures Figure 1 Figure 2 Figure 3 Figure 4 Introduction When the World Health Organization (WHO) recommended the Xpert MTB/RIF assay (Cepheid; Sunnyvale, CA, USA) for diagnosis of tuberculosis (TB) in 2010, it was heralded as a game-changer.[ 1 ] This novel technology of cartridge-based nucleic acid amplification test (NAAT) ushered in a new era of progress in TB diagnostics and was followed shortly thereafter by the second molecular WHO-recommended rapid diagnostic test (mWRD), the Molbio Truenat.[ 2 ] Today, the TB diagnostic pipeline is healthier than ever with at least 35 other NAATs for diagnosing TB in development.[ 3 ] Despite the bright future, only 47% of people newly diagnosed with TB received a mWRD as their initial test in 2022.[ 4 , 5 ] Meanwhile, most persons with TB were still diagnosed by smear microscopy, the same method Robert Koch used to isolate M. tuberculosis as the etiologic agent in 1882.[ 6 ] Among the many causes for this diagnostic coverage gap, a major reason is cost.[ 7 , 8 ] Despite recent price reductions of consumables and reagents[ 9 ], ensuring universal mWRD coverage may cost over $ 1 billion per year.[ 10 , 11 ] To mitigate high mWRD costs, a screening step can be used to rule out people with a low probability of TB disease.[ 12 ] Among various options, chest X-ray (CXR) has become the screening tool of choice in many settings due to the ability to identify the large cohort of asymptomatic people with TB.[ 13 , 14 ] More recently, computer-aided detection (CAD) and artificial intelligence (AI) platforms to support CXR reading such as qXR (Qure.ai; Mumbai, India), CAD4TB (Delft Imaging; Delft, the Netherlands), and other platforms, have gained popularity with value propositions ranging from capacity creation to overcome the lack of trained human readers to workload reduction through triaging of normal CXR images.[ 15 ] For individuals 15 years and above, CAD/AI may be used in place of human reading for screening and triage, and working alongside of human readers to help automating and standardizing interpretation.[ 12 ] CAD/AI offers a key advantage over human readers through its continuous score (0–1 or 1–100) which confers a probability of TB among people screened, and grants greater flexibility to tailor follow-on testing to limited public health budgets.[ 16 ] Another recent process innovation to address high costs of laboratory tests in resource-constrained settings that was effectively employed during the COVID-19 pandemic is sample pooling. This method involves mixing specimen for a two-step hierarchical diagnostic algorithm that foregoes individual testing in the event of a negative pooled sample.[ 17 ] While initial TB pooling studies had found that dilution of the bacterial load can lead to lower levels of detection and potentially missed diagnoses [ 18 ], more recent evaluations have reported sensitivity of 90% and up to 100% with the Xpert MTB/RIF Ultra assay (Xpert Ultra), a newer generation test with higher sensitivity.[ 19 – 21 ] This has led to reported reductions of TB testing costs of 57%-87% depending on the pool size.[ 22 ] A recent cost-effectiveness analysis reported a 34.9% decrease in costs when comparing individual to pooled Xpert testing.[ 23 ] The benefits of pooling are proportional to the prevalence of the disease in the target population. The lower the prevalence, the higher the theoretical savings. This raises the utility of pooling in high-throughput, low-yield approaches such as active case finding (ACF) campaigns where large numbers of individuals need to be screened and tested to detect a person with TB. ACF is an important component of all high burden countries’ national TB response and is critical to reach those people who are less likely or able to get care in public facilities. However, it is more expensive to conduct outreach, and ways to reduce costs are urgently needed to reach all persons with TB, and especially those with subclinical TB.[ 24 , 25 ] Here, we evaluated the impact of using AI outputs to inform different pooling strategies based on Xpert testing data collected during ACF campaigns in four high TB burden settings with the goal of modeling diagnostic savings and theoretical expansion of access to mWRDs. Methods Study design This was a retrospective analysis of ACF campaigns using AI probability scores to model the incremental reduction in Xpert cartridge consumption. Data sources Data were obtained from four implementers located in high TB burden countries of Bangladesh, Nigeria, Vietnam and Zambia. These data consisted of aggregate AI probability scores and Xpert test results from community- and facility-based case finding campaigns conducted between 2014-17 in Bangladesh and 2022 and 2023 all other countries. These campaigns targeted a heterogenous mix of vulnerable populations particular to each ACF setting and country. All countries used CXR with slightly different modalities to screen for TB. Screening methods have been described elsewhere but are summarized here.[ 26 – 29 ] In Bangladesh, standalone screening centers supported referrals of health-seeking individuals in outpatient care departments of public and private sectors to conduct CXR screening and Xpert testing in Dhaka. In Nigeria, mobile teams conducted active outreach events in remote rural areas using ultra-portable X-ray among people with limited access to healthcare services such as pastoralists and nomadic populations. Vietnam delivered community-based ACF campaigns in rural and urban areas focused on household contacts, older persons, urban poor, and people with a history of TB. Zambia used a portable X-ray system in health facilities to screen clinic attendees and household contacts.[ 30 ] Nigeria and Viet Nam used qXR v3, Zambia used CAD4TB7 for generation of the AI probability score. Bangladesh used qXR v3 and CAD4TB6 for the same dataset, but only qXR v3 results from the former were due to concord with the software used in Nigeria and Viet Nam. Xpert MTB/RIF (Bangladesh and Nigeria) or the newer Xpert MTB/RIF Ultra assay (Viet Nam and Zambia) was used for diagnostic testing of distinct individuals in all countries. Model structure Each participating country provided Xpert testing data aggregated by deciles ( D m , where m = 1 to 10) of AI probability scores in the range of 0-100 for CAD4TB, i.e., D 1 : AI score = 0–9, D 2 : 10–19 …, and, similarly, 0-0.99 for qXR. We calculated the positivity rate ( p m ) for each D m . We assigned testing thresholds for each country based on the decile ( D m ) where ~ 95% of cases would be detected ( C ) with Xpert testing starting at that decile to have only ~ 5% missed cases ( M ) as a result of not employing Xpert testing in the previous deciles. (Supplementary information 1–3) To model the theoretical savings, we calculated theoretical number of tests per person needed to achieve the same positivity through pooling for a two-step hierarchical testing strategy.[ 31 ] The difference between the total actual number of individual tests performed and the theoretical number of tests per person when employing the pooling strategy represented the number of tests saved. For the primary analysis, we assumed pool size of four based on prior studies and that both individual and pooled testing were 100% sensitive and specific.[ 32 , 33 ] We then modeled four ordinal screening and testing approaches with increasing complexity to estimate incremental savings compared to the previous approach (Fig. 1 ): Baseline approach: individual Xpert tests in all deciles as per the original datasets; CXR approach: individual Xpert tests in all deciles for which ΣM ~ 5%; Indiscriminate pooling approach: pooled Xpert tests in all deciles for which ΣM ~ 5%; AI-guided pooling approach: a combination of pooled and individual Xpert tests in all deciles for which ΣM ~ 5%, with individual Xpert testing in high deciles with a p m ≥20%. While we aimed to achieve a ΣC ~ 95% (≡ ΣM ~ 5%) for screening and testing approaches, the actual values were 95.7% (Bangladesh), 95.3% (Nigeria), 94.9% (Viet Nam), and 96.7% (Zambia). Data analysis We described number of tests performed, positive results, and positivity rates in total and for each AI-score decile. We further calculated the theoretical number and proportion of tests saved incrementally between each screening and testing approach and cumulatively over the baseline. To characterize the optimal approach in each country, we identified the deciles below which testing could be foregone and the deciles at which individual testing instead of pooled testing would save tests. The number of tests saved was multiplied by the current price of $ 7.97 per cartridge to calculate crude cost savings, and subsequently divided by the number of positive test results for a unit-cost estimate per persons diagnosed with TB.[ 9 ] To offer an alternative perspective to cost savings, we estimated the ratio of additional tests that could be performed with the savings over the theoretical number of tests needed for the current cohort as a measure of the extent to which access to mWRDs could have been expanded by employing the optimal screening and testing approach. For the sensitivity analyses, we modeled using pool size of three [ 19 ] and relaxed the 100% sensitivity and specificity assumption of the pooling method based on systematic review findings from pooling with Xpert Ultra as Xpert MTB/RIF will be discontinued in 2024 using the datasets from all countries.[ 20 ] We used binGroup2 package in R for the analysis.[ 34 ] Results Dataset characteristics In the two facility-based ACF settings, Bangladesh and Zambia performed 24,079 and 2,353 Xpert tests with a matching corresponding AI result, respectively. In the community-based ACF settings, Nigeria performed 1,021 and Viet Nam performed 5,074 tests. Positivity rates were 15.3% (3,679/24,079) in Bangladesh, 11.6% (273/2,353) in Zambia, 9.0% (455/5,074) in Viet Nam and 8.3% (85/1,021) in Nigeria (Figure 2). In terms of testing distribution by AI-score decile most countries exhibited similar U-shaped patterns except Zambia (Figure 3A). In Zambia, almost half (49.3%) of the tests were conducted in D 5 –D 6 . Meanwhile, 40.6% of Xpert tests in Bangladesh had an AI score in the lowest decile ( D 1 ), which was higher than Viet Nam (22.1%), Nigeria (17.4%) and Zambia (2.5%). Meanwhile, the testing rate in the AI-scores D 4 –D 5 , were higher in Nigeria ( D 4 :15.3% and D 5 :15.0%) compared to Bangladesh ( D 4 :4.1% and D 5 :3.9%) and Viet Nam ( D 4 :5.4% and D 5 :5.2%). In terms of positivity by AI-score decile (Figure 3B) the two facility-based ACF sites exhibited higher rates than the community-based counterparts. In Bangladesh, positivity was high in the high AI-score deciles of D 7 –D 10 than in the other countries with rates increasing from 12.3% to 70.0%. Zambia exhibited a similar pattern with a lower peak with rates ranging from 13.6% to 65.1% for D 7 and D 10 , respectively. In comparison, the respective positivity of D 7 and D 10 only rose from 4.8% to 28.7% in Nigeria and from 6.3% to 30.6% in Viet Nam. Tests cost savings in the model output All incremental screening and testing approaches resulted in additional savings except the indiscriminate pooling approach in Bangladesh and are presented for each country in Table 1. In Bangladesh, the CXR approach resulted in savings of 52.5% over baseline. Based on model parameters, these savings were realized at an AI threshold of 0.30-0.39, i.e., foregoing testing for D 1 –D 3 and individually testing everyone in higher deciles. Indiscriminately pooling all persons in D 4 –D 10 actually resulted in excess of 2.0% tests over the CXR approach. Using AI to indicate individual testing for high AI-score deciles D 9 –D 10 reversed this trend and increased cumulative savings to 61.5%. Table 1. Incremental and cumulative savings by country. Total tests Incremental savings Cumulative savings N N % N % Bangladesh Baseline approach 24,079 CXR approach 11,448 12,631 52.5% 12,631 52.5% Indiscriminate pooling approach 11,676 -228 -2.0% 12,403 51.5% AI-guided pooling approach 9,262 2,414 20.7% 14,817 61.5% Zambia Baseline approach 2,353 CXR approach 1,960 393 16.7% 393 16.7% Indiscriminate pooling approach 1,352 608 31.0% 1,001 42.5% AI-guided pooling approach 1,158 194 14.3% 1,195 50.8% Nigeria Baseline approach 1,021 CXR approach 738 283 27.7% 283 27.7% Indiscriminate pooling approach 459 279 37.8% 562 55.0% AI-guided pooling approach 434 25 5.4% 587 57.5% Viet Nam Baseline approach 5,074 CXR approach 2,917 2,157 42.5% 2,157 42.5% Indiscriminate pooling approach 2,110 807 27.7% 2,964 58.4% AI-guided pooling approach 1,955 155 7.3% 3,119 61.5% CXR=Chest X-ray; AI=Artificial Intelligence. Table 1 note The model assumes pool sizes of 4 with both pooling and individual testing sensitivity and specificity of 100% for a testing threshold resulting in missed cases of 4.4%, 4.7%, 5.1% for Bangladesh, Nigeria and Viet Nam using qXR v3, respectively, and 3.3% for Zambia using CAD4TB v7. Incremental savings indicate the difference from the prior case, whereas cumulative savings are calculated against the baseline approach. In Zambia, the CXR approach yielded savings of 16.7% over baseline at an AI threshold of 0.3. Indiscriminately pooling all persons in D 4 and higher produced incremental savings of 31.0% and cumulative savings over baseline of 42.5%. Per AI guidance, testing reverted to an individual basis in D 8 –D 10 to yield additional savings of 14.3% for cumulative savings of 50.8%. The CXR approach in Nigeria showed an initial savings of 27.7% in testing at an AI threshold of 0.3 above which indiscriminate pooling reduced an incremental 37.8% in testing for a cumulative savings of 55.0% over baseline. Similar to Viet Nam, there were diminishing returns from the AI-guided pooling approach as switching to individual testing in D 10 led to incremental and cumulative savings over baseline were only 5.4% and 57.5%, respectively. In Viet Nam, the CXR approach reduced testing by 42.5% at an AI threshold of 0.4. Meanwhile, indiscriminate pooling above the AI threshold saved an incremental 27.7% in testing over the CXR approach for a cumulative savings of 58.4% over baseline. The AI-guided pooling approach indicated individual testing in D 10 which added 7.3% in incremental test reductions for a cumulative savings of 61.5% (Figure 4). At the above-described pooling scenarios, in Bangladesh the number of tests needed declined from 24,079 to 9,262 for savings of 12,403-14,817 tests. This translated to $98,852-$118,091 in crude costs or $26.87-$32.10 per person with TB averted (Table 2). Alternatively, these savings could be redeployed for an expansion of mWRD access of 110%–160% with existing public health resources. In Zambia, the number of tests dropped from 2,353 to 1,158 for savings of 393-1,195 tests. This implies cost savings of $3,132-$9,524 or $11.47-$34.89 per person with TB for a theoretical expansion of mWRD access of 20%–103%. Based on test volume reductions from 1,021 to 434 in Nigeria, the total test and cost savings ranged from 283-587 and $2,256-$4,678 or $26.54-$55.04 per person with TB. This represented 38%–135% in potentially greater access to mWRD. Lastly, in Viet Nam the number of tests needed fell from 5,074 to 1,955 for 2,157-3,119 tests saved corresponding to crude savings or $17,191-$24,858 or $37.78-$54.63 per person with TB. This represented a potential mWRD expansion of 74%-160%. Table 2. Crude total and per-TB detection cost savings and mWRD access expansion by country. Crude cost savings ($) Per-TB cost savings ($) mWRD access expansion Incremental Cumulative Incremental Cumulative Incremental Cumulative Bangladesh Baseline approach CXR approach 100,669 100,669 27.36 27.36 110.3% 110.3% Indiscriminate pooling approach -1,817 98,852 -0.49 26.87 -2.0% 106.2% AI-guided pooling approach 19,240 118,091 5.23 32.10 26.1% 160.0% Zambia Baseline approach CXR approach 3,132 3,132 11.47 11.47 20.1% 20.1% Indiscriminate pooling approach 4,846 7,978 17.75 29.22 45.0% 74.0% AI-guided pooling approach 1,546 9,524 5.66 34.89 16.8% 103.2% Nigeria Baseline approach CXR approach 2,256 2,256 26.53 26.53 38.3% 38.3% Indiscriminate pooling approach 2,224 4,479 26.16 52.70 60.8% 122.4% AI-guided pooling approach 199 4,678 2.34 55.04 5.8% 135.3% Viet Nam Baseline approach CXR approach 17,191 17,191 37.78 37.78 73.9% 73.9% Indiscriminate pooling approach 6,432 23,623 14.14 51.92 38.2% 140.5% AI-guided pooling approach 1,235 24,858 2.72 54.63 7.9% 159.5% CXR=Chest X-ray; AI=Artificial Intelligence. Table 2 note Incremental savings indicate the difference from the prior case, whereas cumulative savings are calculated against the baseline approach. Crude cost savings are based on a cost of $7.97 per Xpert cartridge. Per-TB case cost savings are based on number of positive test results of 3,679 in Bangladesh, 273 in Zambia, 85 in Nigeria and 455 in Viet Nam. mWRD access expansion was calculated by dividing the number of cartridges saved by the number of cartridges used in each individual screening and testing approach. The sensitivity analyses did not show a substantial change of the results. Across different scenarios, the change in savings against the primary case (pool size of three with pooled sensitivity and specificity of 1) ranged from -2.6% to +3.8%. Testing in smaller pools increased number of tests. Reducing pooling sensitivity reduced test usage and so did increasing pooling specificity, which will result in an increase in missed cases and false positives, respectively, for the overall two-step hierarchical testing algorithm. (Supplementary table 4) In Bangladesh, we compared savings between qXRv3 and CAD4TB6 and found that differences were small (0.7%). CAD4TB scores resulted in 9,200 tests in the AI-guided pooling approach compared to 9,262 in qXRv3 against the base case of 24,079 test. (Supplementary table 5) Discussion Our modelling study results show that the combined use of computer-aided chest radiography and pooling may achieve compounding effects to achieve significant savings in diagnostic consumables that could be redeployed to increase the global coverage of mWRDs. We further found that the substantial heterogeneity in AI thresholds and impact of AI scores on subsequent pooling across the different settings will require differentiated deployment of these two innovations in order to optimize the potential gains. However, while countries and settings may differ, it appears that the various screening and testing approaches and particularly the AI-guided pooling approach may be able to achieve consistent results between the most advanced software platforms. Across the different countries, settings and approaches, the combination of using CXR with AI-scores to inform decisions on pooling sputum produced cumulative savings on testing of 50.8%-61.5% over baseline compared to universal testing of all individuals with signs of TB. In 2022, Bangladesh tested only 20% of people with TB with mWRDs as the first-line diagnostic test. While the gap was smaller in Nigeria, Zambia and Vietnam, three in ten persons with TB were not diagnosed using molecular diagnostics.[ 4 ] Our results suggest that more than twice as many people could be tested for the same diagnostic test costs using CXR and pooling as compared to testing all presumptive individuals. A major contributor to this potential capacity expansion was the use of AI and CXR. As the data of all countries originated from ACF campaigns rather than prevalence surveys, each dataset included an inherent level of preselection or pre-screening to raise the TB prevalence in the sample. This pre-filtering may have consisted of targeting of highly vulnerable populations such as contacts (Zambia), deploying in high-yield settings such as health facilities (Bangladesh) or having a previous binary read of the CXR by a potentially inexperienced human reader (Viet Nam).[ 35 ] Despite these methods of preselection, our study once again highlighted the well-documented effectiveness of CXR for screening and triaging for TB.[ 36 – 39 ] Beyond that, we observed incremental savings by leveraging the AI’s quantitative output to optimize CXR interpretation and forego testing in lower AI-score deciles. This approach was particularly useful in settings with a high testing proportion in the lowest decile such as Bangladesh and Viet Nam, the first screening and testing approach generated already generated cumulative savings of 52.5% and 42.5%, respectively. This optimization of testing volumes through computer-aided chest radiography was also concordant with available evidence.[ 27 , 40 ] Our model suggested that in each country pooling could be effective for generating additional savings in testing. However, our model also evinced differences across both pooling strategies based on the populations screened and the results of the CXR and AI. For instance, in the facility-based settings of Bangladesh and Zambia 65%-70% of individuals in the highest decile had bacteriologically confirmed TB compared to only 29%-31% in Nigeria and Viet Nam’s community-based ACF cohort. This dichotomy was reflected by the pooling approach, as the AI-guided pooling, i.e., reversion to individual testing in high deciles, continued to generated substantial savings in the facilities, while the model exhibited substantial diminishing returns in the low-yield community setting. Interestingly, while indiscriminate pooling saved tests in three countries, in Bangladesh it actually increased testing unless an AI-guided pooling approach was used. In Bangladesh, where reads for both qXR and CAD4TB were available, the performance on both platforms were highly concordant. However, each required adjustment of the testing threshold, highlighting the need for end-user optimization based on the local epidemiology and platforms used. The approaches may be further optimized based on where sample volumes allow more pools. In one such additional approach, the AI-guided cohort pooling case resulted in up to 2.5% additional savings over the AI-guided pooling case. (Supplementary information 5) Nevertheless, these finding inspire confidence in the high precision that characterizes many of today’s commercial AI solutions for TB.[ 41 ] This variability was encountered in different areas of our study and represented its key strength. For example, in contrast to the high savings achieved through the CXR approach in Bangladesh and Viet Nam, savings in Nigeria and Zambia were only 28% and 17%, respectively. Yet, both countries differed substantially in the additional gains from each incremental pooling approach. The testing distribution patterns also exhibited discordance, as all countries exhibited a U-shaped curve except Zambia, or in positivity by decile, whereby reversion to individual testing of the AI-guided pooling approach occurred in D 8 for the Zambian dataset, in D 9 for the data from Bangladesh and D 10 in Nigeria and Viet Nam. Particularly with respect to the optimal application of AI and its continuous outputs, there has been much discussion around using different AI threshold scores. However, the growing evidence base suggests variability in AI performance across demographic, clinical, or behavioral characteristics of the population, screening setting or even radiography equipment, which underscores the well-documented need for local calibration and threshold setting.[ 16 , 40 , 42 ] Similarly, while general pooling can save costs, optimizing a pooling strategy will depend on the use of local data, capacities and established practices. Our study has a limitation in that modelled results are fraught with assumptions about performance and each situation will differ. In the primary analysis, we assume perfect sensitivity and specificity for both pooling and individual testing, which is unlike real-world performance. While we attempt to mitigate this limitation in the sensitivity analysis by varying the pooling parameters, we do not model the overall performance of the diagnostic algorithm, e.g., the overall sensitivity and specificity of a combination of no testing, individual testing and pooled testing in the AI-guided pooling case. This will also impact the overall number of missed cases in the algorithm. We do not expect a substantial impact on the number of tests given the superior performance of Xpert Ultra in both individual and pooled testing documented previously, and encourage future prospective evaluations to factor in these considerations, in addition to other implementation challenges. Lastly, although the data are based on real-world ACF activities they did not contain the daily distribution of people with presumptive TB and their abnormality scores. Hence, for this analysis we assumed 100% completeness in pool sizes with four sputum specimen per pool. Pooling-decisions based on abnormality scores such as the AI-guided pooling approach would on occasion require pools of two or three specimen, which might impact overall savings. However, we conducted this sensitivity analysis, which also showed that the impact on overall savings was small. The real-world readiness for computer-aided chest radiography informed pooling in real-time in programmatic settings can be challenging. Despite the positive reception of pooling by laboratory technicians for its time-saving properties [ 19 ], pooled testing for TB is not feasible in all laboratories. For example, though specificity issues have not been widely reported, the manipulation of several samples bears the risk of contamination. CXR and AI are also not readily available in most high TB burden countries. Nevertheless, as pooling becomes established practice in more indications and access to AI improves, this type of pragmatic approach could mitigate commonly encountered cartridge shortages and provide more people with signs of TB access to molecular tests and should be evaluated prospectively to see how they work in real world settings. Conclusions To achieve End TB Strategy goals, it is necessary to optimize the use of available tools. Integrating computer-aided chest radiography and pooling into TB screening and testing algorithms has the potential to substantially reduce diagnostic testing, thus freeing up constrained financial and human public health resources to save costs and extend access to more people in need of high-quality, rapid molecular testing for TB. Abbreviations ACF Active Case Finding AI Artificial Intelligence CXR Chest X-ray mWRD molecular WHO-recommended Rapid Diagnostic NAAT Nucleic Acid Amplification Test TB Tuberculosis WHO World Health Organization Declarations Ethics approval and consent to participate Ethical approvals were obtained from respective review committees for the original data collection in the four participating countries. All data furnished for this study were aggregated and contained no personally identifiable information and thus this modelling study did not require additional ethical approval. Consent for publication Not applicable. Availability of data and materials All the data used for the analysis is available in the paper. The reproducible code for analysis is available in a public GitHub repository at https://github.com/peptonefizz/Xpert-pooling-analysis. Competing interests The authors declare no competing interests. Funding TW is supported by grants from: the Wellcome Trust, UK (209075/Z/17/Z); the Department of Health and Social Care (DHSC), the Foreign, Commonwealth & Development Office (FCDO), the Medical Research Council (MRC) and Wellcome, UK (Joint Global Health Trials, MR/V004832/1); the Medical Research Council (Public Health Intervention Development Award “PHIND”, APP2293); and the Medical Research Foundation (Dorothy Temple Cross International Collaboration Research Grant, MRF-131–0006-RG-KHOS-C0942). Authors' contributions JC, LNQV, AJC and TG conceptualized the manuscript. The methodology was developed by AJC and TG. Formal analyses and visualization were conducted by LNQV and TG. Resources were contributed by SB, AJC, LNQV, MM, SJ and SA. JC, LNQV, and TG wrote the original draft, while review and editing was done by all authors. Supervision was provided by JC. JC and LNQV were responsible for coordination with the co-authors and submission of the article. All authors have read and approved the final manuscript. Acknowledgements The authors would like to thank their affiliate institutions and NTP of the contributing countries. We also express our gratitude to public health staff and partners at primary and secondary care levels, and the communities and patients involved in the initial generation of the original data used for the secondary analysis of this study. Authors' information (optional) Not applicable. References Evans CA. GeneXpert—A Game-Changer for Tuberculosis Control? PLoS Med. 2011;8:e1001064. Nikam C, Jagannath M, Narayanan MM, Ramanabhiraman V, Kazi M, Shetty A, et al. Rapid Diagnosis of Mycobacterium tuberculosis with Truenat MTB: A Near-Care Approach. PLoS One. 2013;8:1–7. Branigan D. Tuberculosis Diagnostics Pipeline Report » 2023. 2023. World Health Organization. Global Tuberculosis Report 2023. Geneva, Switzerland; 2023. World Health Organization. WHO standard: universal access to rapid tuberculosis diagnostics. Geneva, Switzerland: World Health Organization; 2023. Daniel TM. The history of tuberculosis. Respir Med. 2006;100:1862–70. Albert H, Nathavitharana RR, Isaacs C, Pai M, Denkinger CM, Boehme CC. Development, roll-out and impact of Xpert MTB/RIF for tuberculosis: What lessons have we learnt and how can we do better? Eur Respir J. 2016;48:516–25. Engel N, Ochodo EA, Karanja PW, Schmidt B-M, Janssen R, Steingart KR, et al. Rapid molecular tests for tuberculosis and tuberculosis drug resistance: provider and recipient views. Cochrane Database Syst Rev. 2021;2021. Stop TB Partnership. November 2023 Diagnostics, Medical Devices & Other Health Products Catalog. 2023; November. Brümmer LE, Thompson RR, Malhotra A, Shrestha S, Kendall EA, Andrews JR, et al. Cost-effectiveness of Low-complexity Screening Tests in Community-based Case-finding for Tuberculosis. Clin Infect Dis. 2023;:1–10. Baik Y, Nakasolya O, Isooba D, Mukiibi J, Kitonsa PJ, Erisa KC, et al. Cost to perform door-to-door universal sputum screening for TB in a high-burden community. Int J Tuberc Lung Dis. 2023;27:195–201. World Health Organization. WHO consolidated guidelines on tuberculosis, Module 2: Screening, Systematic screening for tuberculosis disease. Geneva, Switzerland; 2021. Law I, Floyd K, Abukaraig EAB, Addo KK, Adetifa I, Alebachew Z, et al. National tuberculosis prevalence surveys in Africa, 2008–2016: an overview of results and lessons learned. Trop Med Int Heal. 2020;25:1308–27. Onozaki I, Law I, Sismanidis C, Zignol M, Glaziou P, Floyd K. National tuberculosis prevalence surveys in Asia, 1990-2012: An overview of results and lessons learned. Trop Med Int Heal. 2015;20:1128–45. Creswell J, Vo LNQ, Qin ZZ, Muyoyeta M, Tovar M, Wong EB, et al. Early user perspectives on using computer-aided detection software for interpreting chest X-ray images to enhance access and quality of care for persons with tuberculosis. BMC Glob Public Heal. 2023;1:30. Geric C, Qin ZZ, Denkinger CM, Kik S V., Marais B, Anjos A, et al. The rise of artificial intelligence reading of chest X-rays for enhanced TB diagnosis and elimination. Int J Tuberc Lung Dis. 2023;27:367–72. Daniel EA, Esakialraj L BH, S A, Muthuramalingam K, Karunaianantham R, Karunakaran LP, et al. Pooled Testing Strategies for SARS-CoV-2 diagnosis: A comprehensive review. Diagn Microbiol Infect Dis. 2021;101:115432. Ho J, Jelfs P, Nguyen PTB, Sintchenko V, Fox GJ, Marks GB. Pooling sputum samples to improve the feasibility of Xpert® MTB/RIF in systematic screening for tuberculosis. Int J Tuberc Lung Dis. 2017;21:503–8. Vuchas C, Teyim P, Dang BF, Neh A, Keugni L, Che M, et al. Implementation of large-scale pooled testing to increase rapid molecular diagnostic test coverage for tuberculosis: a retrospective evaluation. Sci Rep. 2023;13:1–10. Cuevas LE, Santos VS, Lima SVMA, Kontogianni K, Bimba JS, Iem V, et al. Systematic review of pooling sputum as an efficient method for xpert MTB/RIF tuberculosis testing during COVID-19 pandemic. Emerg Infect Dis. 2021;27:719–27. Chakravorty S, Simmons AM, Rowneki M, Parmar H, Cao Y, Ryan J, et al. The new Xpert MTB/RIF ultra: Improving detection of Mycobacterium tuberculosis and resistance to Rifampin in an assay suitable for point-of-care testing. MBio. 2017;8. Dos Santos PCP, Da Silva Santos A, De Oliveira RD, Da Silva BO, Soares TR, Martinez L, et al. Pooling Sputum Samples for Efficient Mass Tuberculosis Screening in Prisons. Clin Infect Dis. 2022;74:2115–21. Iem V, Bimba JS, Santos VS, Dominguez J, Creswell J, Somphavong S, et al. Pooling sputum testing to diagnose tuberculosis using xpert MTB / RIF and xpert ultra : a cost-effectiveness analysis. 2023;:1–11. Stop TB Partnership. The Global Plan to End TB 2023-2030. Geneva, Switzerland; 2022. Ismail N, Nathanson CM, Zignol M, Kasaeva T. Achieving universal access to rapid tuberculosis diagnostics. BMJ Glob Heal. 2023;8:8–10. Banu S, Haque F, Ahmed S, Sultana S, Rahman MM, Khatun R, et al. Social Enterprise Model (SEM) for private sector tuberculosis screening and care in Bangladesh. PLoS One. 2020;15:e0241437. John S, Abdulkarim S, Usman S, Rahman MT, Creswell J. Comparing tuberculosis symptom screening to chest X-ray with artificial intelligence in an active case finding campaign in Northeast Nigeria. BMC Glob Public Heal. 2023;1:17. Muyoyeta M, Kasese NC, Milimo D, Mushanga I, Ndhlovu M, Kapata N, et al. Digital CXR with computer aided diagnosis versus symptom screen to define presumptive tuberculosis among household contacts and impact on tuberculosis diagnosis. BMC Infect Dis. 2017;17:1–8. Nguyen LH, Codlin AJ, Vo LNQ, Dao T, Tran D, Forse RJ, et al. An Evaluation of Programmatic Community-Based Chest X-ray Screening for Tuberculosis in Ho Chi Minh City, Vietnam. Trop Med Infect Dis. 2020;5:185. Vo LNQ, Codlin A, Ngo TD, Dao TP, Dong TTT, Mo HTL, et al. Early Evaluation of an Ultra-Portable X-ray System for Tuberculosis Active Case Finding. Trop Med Infect Dis. 2021;6:163. Dorfman R. The Detection of Defective Members of Large Populations. Ann Math Stat. 1943;14:436–40. Abdurrahman ST, Mbanaso O, Lawson L, Oladimeji O, Blakiston M, Obasanya J, et al. Testing pooled sputum with Xpert MTB/RIF for diagnosis of pulmonary tuberculosis to increase affordability in low-income countries. J Clin Microbiol. 2015;53:2502–8. Iem V, Chittamany P, Suthepmany S, Siphanthong S, Siphanthong P, Somphavong S, et al. Pooled testing of sputum with Xpert MTB/RIF and Xpert Ultra during tuberculosis active case finding campaigns in Lao People’s Democratic Republic. BMJ Glob Heal. 2022;7:1–8. Bilder CR, Hitt BD, Biggerstaff BJ, Tebbs JM, Mcmahan CS. binGroup2: Statistical Tools for Infection Identification via Group Testing. 2018;XX:1–17. Qin ZZ, Sander MS, Rai B, Titahong CN, Sudrungrot S, Laah SN, et al. Using artificial intelligence to read chest radiographs for tuberculosis detection: A multi-site evaluation of the diagnostic accuracy of three deep learning systems. Sci Rep. 2019;9:15000. Shazzadur Rahman AAM, Langley I, Galliez R, Kritski A, Tomeny E, Squire SB. Modelling the impact of chest X-ray and alternative triage approaches prior to seeking a tuberculosis diagnosis. BMC Infect Dis. 2019;19:1–11. World Health Organization. Chest radiography in tuberculosis detection – summary of current WHO recommendations and guidance on programmatic approaches. Geneva, Switzerland; 2016. van Cleeff MRA, Kivihya-Ndugga LE, Meme H, Odhiambo JA, Klatser PR. The role and performance of chest X-ray for the diagnosis of tuberculosis: A cost-effective analysis in Nairobi, Kenya. BMC Infect Dis. 2005;5:1–9. Creswell J, Qin ZZ, Gurung R, Lamichhane B, Yadav DK, Prasai MK, et al. The performance and yield of tuberculosis testing algorithms using microscopy, chest x-ray, and Xpert MTB/RIF. J Clin Tuberc Other Mycobact Dis. 2019;14 November:1–6. Qin ZZ, Ahmed S, Sarker MS, Paul K, Adel ASS, Naheyan T, et al. Tuberculosis detection from chest x-rays for triaging in a high tuberculosis-burden setting: an evaluation of five artificial intelligence algorithms. Lancet Digit Heal. 2021;3:e543–54. Gelaw SM, Kik S V., Ruhwald M, Ongarello S, Egzertegegne TS, Gorbacheva O, et al. Diagnostic accuracy of three computer-aided detection systems for detecting pulmonary tuberculosis on chest radiography when used for screening: Analysis of an international, multicenter migrants screening study. PLOS Glob Public Heal. 2023;3:e0000402. Codlin AJ, Dao TP, Vo LNQ, Forse RJ, Van Truong V, Dang HM, et al. Independent evaluation of 12 artificial intelligence solutions for the detection of tuberculosis. Sci Rep. 2021;11:23895. Additional Declarations No competing interests reported. Supplementary Files SupplementaryTables.docx Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 20 Mar, 2024 Reviews received at journal 13 Mar, 2024 Reviewers agreed at journal 08 Mar, 2024 Reviews received at journal 06 Mar, 2024 Reviewers agreed at journal 21 Feb, 2024 Reviewers agreed at journal 15 Jan, 2024 Reviewers agreed at journal 12 Jan, 2024 Reviewers invited by journal 10 Jan, 2024 Editor assigned by journal 03 Jan, 2024 Submission checks completed at journal 03 Jan, 2024 First submitted to journal 27 Dec, 2023 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-3813705","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":265035914,"identity":"bee8867a-1ed6-4e0d-8996-7c7cf7aab726","order_by":0,"name":"Andrew James Codlin","email":"","orcid":"","institution":"Friends for International TB Relief","correspondingAuthor":false,"prefix":"","firstName":"Andrew","middleName":"James","lastName":"Codlin","suffix":""},{"id":265035915,"identity":"8ed1b318-2e03-4517-a2be-f47d6864a6d6","order_by":1,"name":"Luan Nguyen Quang Vo","email":"","orcid":"","institution":"Friends for International TB Relief","correspondingAuthor":false,"prefix":"","firstName":"Luan","middleName":"Nguyen Quang","lastName":"Vo","suffix":""},{"id":265035916,"identity":"360f3798-0db8-4878-8296-f31d1d247a6f","order_by":2,"name":"Tushar Garg","email":"","orcid":"","institution":"Stop TB Partnership","correspondingAuthor":false,"prefix":"","firstName":"Tushar","middleName":"","lastName":"Garg","suffix":""},{"id":265035917,"identity":"f92f874a-78bd-4116-8da9-0b05f625ceec","order_by":3,"name":"Sayera Banu","email":"","orcid":"","institution":"International Centre for Diarrhoeal Disease Research","correspondingAuthor":false,"prefix":"","firstName":"Sayera","middleName":"","lastName":"Banu","suffix":""},{"id":265035918,"identity":"85a8a6c6-ba78-47ef-862d-ddc9b36bab7c","order_by":4,"name":"Shahriar Ahmed","email":"","orcid":"","institution":"International Centre for Diarrhoeal Disease Research","correspondingAuthor":false,"prefix":"","firstName":"Shahriar","middleName":"","lastName":"Ahmed","suffix":""},{"id":265035919,"identity":"71fb1cc8-9525-462d-a413-8cf3b0051c19","order_by":5,"name":"Stephen John","email":"","orcid":"","institution":"Janna Health Foundation","correspondingAuthor":false,"prefix":"","firstName":"Stephen","middleName":"","lastName":"John","suffix":""},{"id":265035920,"identity":"7287aabb-e786-42bb-b60f-aedc661bf867","order_by":6,"name":"Suraj Abdulkarim","email":"","orcid":"","institution":"Janna Health Foundation","correspondingAuthor":false,"prefix":"","firstName":"Suraj","middleName":"","lastName":"Abdulkarim","suffix":""},{"id":265035921,"identity":"56fbfc43-e375-42d2-9187-790056315f84","order_by":7,"name":"Monde Muyoyeta","email":"","orcid":"","institution":"Centre for Infectious Disease Research in Zambia","correspondingAuthor":false,"prefix":"","firstName":"Monde","middleName":"","lastName":"Muyoyeta","suffix":""},{"id":265035922,"identity":"e831ddb2-c0a1-470d-9852-a74fc1fa0530","order_by":8,"name":"Nsala Sanjase","email":"","orcid":"","institution":"Centre for Infectious Disease Research in Zambia","correspondingAuthor":false,"prefix":"","firstName":"Nsala","middleName":"","lastName":"Sanjase","suffix":""},{"id":265035923,"identity":"1ea1d105-6a38-4e8e-82ac-cc5ddbdf048f","order_by":9,"name":"Tom Wingfield","email":"","orcid":"","institution":"Liverpool School of Tropical Medicine","correspondingAuthor":false,"prefix":"","firstName":"Tom","middleName":"","lastName":"Wingfield","suffix":""},{"id":265035924,"identity":"6ca5d356-aac7-4cca-b3b8-1905e68bcdcf","order_by":10,"name":"Vibol Iem","email":"","orcid":"","institution":"Liverpool School of Tropical Medicine","correspondingAuthor":false,"prefix":"","firstName":"Vibol","middleName":"","lastName":"Iem","suffix":""},{"id":265035925,"identity":"67129ec7-56c8-43db-802e-600dfac59edb","order_by":11,"name":"Bertie Squire","email":"","orcid":"","institution":"Liverpool School of Tropical Medicine","correspondingAuthor":false,"prefix":"","firstName":"Bertie","middleName":"","lastName":"Squire","suffix":""},{"id":265035926,"identity":"8a616e59-b624-4459-a040-797fcd7980d1","order_by":12,"name":"Jacob Creswell","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAxklEQVRIiWNgGAWjYBACxgYGBmYGBhswh5kULWkMPEB2M9E2AbUcJkELc/vhh48Las7b27P3Hn9cwGCTL+9AyGE9acbGM47dTuzhOZfYPIMhzXLjAUJaGnLYpHkbbifwSOQYNvMwHDYwbCCkpf8NSMs5exK0zADbcoCxB6ZFnoAOoJZnIL8kJ/acOZc4e4ZBmoEBIS2G/cmgELOzZ2/vPfC5oMLGQJ6Qw5BcDowZBqAVBgcIaEFyOQ9UhJAto2AUjIJRMOIAAHa8OzVZ6FeZAAAAAElFTkSuQmCC","orcid":"","institution":"Stop TB Partnership","correspondingAuthor":true,"prefix":"","firstName":"Jacob","middleName":"","lastName":"Creswell","suffix":""}],"badges":[],"createdAt":"2023-12-27 20:44:08","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-3813705/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3813705/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":49243460,"identity":"9cc74bb4-455e-4d0e-a239-664dcdc6ef43","added_by":"auto","created_at":"2024-01-05 18:39:50","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":157798,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eThe framework for analysis with the role of AI and its impact on Xpert testing\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-3813705/v1/ebe8ea9e404cb0f47236b981.png"},{"id":49242761,"identity":"c5dcb1df-8f45-40a4-9925-aba296c2693d","added_by":"auto","created_at":"2024-01-05 18:31:50","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":39536,"visible":true,"origin":"","legend":"\u003cp\u003eXpert testing data for the four countries\u003c/p\u003e\n\u003cp\u003eFigure 2 Note\u003c/p\u003e\n\u003cp\u003eThe percentages are reported with reference to number of tests performed. The Xpert positivity ranged from 8% in Viet Nam to 15% in Bangladesh.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-3813705/v1/e820f1f1da3c5f8a3345abe1.png"},{"id":49241066,"identity":"f532a13b-d8e0-4384-8cfe-88bde55c170f","added_by":"auto","created_at":"2024-01-05 18:23:50","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":54248,"visible":true,"origin":"","legend":"\u003cp\u003eDistribution of tests performed and test positivity across the AI-score deciles for the four countries\u003c/p\u003e\n\u003cp\u003eFigure 3 Note\u003c/p\u003e\n\u003cp\u003e“A. Testing distribution by AI-score decile” refers to the percentage of total tests performed within each AI-score decile. “B. Test positivity by AI-score decile” shows the proportion of positive results in each AI-score decile. The decile labels (D1 to D10) represent AI-score deciles; D1-D10 for 0-0.99 in 0.1 intervals for qXR score and D1-D10 for 0-99 in 10-point intervals for CAD4TB score.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-3813705/v1/9be34cc034375fd8ca0be4f4.png"},{"id":49241070,"identity":"091cc692-7f9c-4392-99a5-4ba9e4a67ce0","added_by":"auto","created_at":"2024-01-05 18:23:50","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":69525,"visible":true,"origin":"","legend":"\u003cp\u003eTests performed under different approaches in the four countries\u003c/p\u003e\n\u003cp\u003eFigure 4 Note\u003c/p\u003e\n\u003cp\u003eThe model assumes pool sizes of 4 with both pooling and individual testing sensitivity and specificity of 100% for a testing threshold resulting in missed cases of 4.4%, 4.7%, 5.1% for Bangladesh, Nigeria and Viet Nam using qXR v3, respectively, and 3.3% for Zambia using CAD4TB v7. The percentages are reported with reference to the base case for the respective countries.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-3813705/v1/1cd0ae79fe649ecfdc7716e1.png"},{"id":49243685,"identity":"367cddd9-98ad-4257-b9aa-8d6a5860f2d4","added_by":"auto","created_at":"2024-01-05 18:47:51","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":918505,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3813705/v1/5563a67b-e254-40c9-9721-267b3550b086.pdf"},{"id":49241068,"identity":"15fbcc8d-4eaf-494c-a021-20001d739da7","added_by":"auto","created_at":"2024-01-05 18:23:50","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":59565,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTables.docx","url":"https://assets-eu.researchsquare.com/files/rs-3813705/v1/45dce6f4c36b81fe69aaf8de.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Expanding molecular diagnostic coverage for tuberculosis by combining computer-aided chest radiography and sputum specimen pooling: a modeling study from four high burden countries","fulltext":[{"header":"Introduction","content":"\u003cp\u003eWhen the World Health Organization (WHO) recommended the Xpert MTB/RIF assay (Cepheid; Sunnyvale, CA, USA) for diagnosis of tuberculosis (TB) in 2010, it was heralded as a game-changer.[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e] This novel technology of cartridge-based nucleic acid amplification test (NAAT) ushered in a new era of progress in TB diagnostics and was followed shortly thereafter by the second molecular WHO-recommended rapid diagnostic test (mWRD), the Molbio Truenat.[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e] Today, the TB diagnostic pipeline is healthier than ever with at least 35 other NAATs for diagnosing TB in development.[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eDespite the bright future, only 47% of people newly diagnosed with TB received a mWRD as their initial test in 2022.[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e] Meanwhile, most persons with TB were still diagnosed by smear microscopy, the same method Robert Koch used to isolate \u003cem\u003eM. tuberculosis\u003c/em\u003e as the etiologic agent in 1882.[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e] Among the many causes for this diagnostic coverage gap, a major reason is cost.[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e] Despite recent price reductions of consumables and reagents[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e], ensuring universal mWRD coverage may cost over \u003cspan\u003e$\u003c/span\u003e1\u0026nbsp;billion per year.[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eTo mitigate high mWRD costs, a screening step can be used to rule out people with a low probability of TB disease.[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e] Among various options, chest X-ray (CXR) has become the screening tool of choice in many settings due to the ability to identify the large cohort of asymptomatic people with TB.[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e, \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e] More recently, computer-aided detection (CAD) and artificial intelligence (AI) platforms to support CXR reading such as qXR (Qure.ai; Mumbai, India), CAD4TB (Delft Imaging; Delft, the Netherlands), and other platforms, have gained popularity with value propositions ranging from capacity creation to overcome the lack of trained human readers to workload reduction through triaging of normal CXR images.[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e] For individuals 15 years and above, CAD/AI may be used in place of human reading for screening and triage, and working alongside of human readers to help automating and standardizing interpretation.[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e] CAD/AI offers a key advantage over human readers through its continuous score (0\u0026ndash;1 or 1\u0026ndash;100) which confers a probability of TB among people screened, and grants greater flexibility to tailor follow-on testing to limited public health budgets.[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eAnother recent process innovation to address high costs of laboratory tests in resource-constrained settings that was effectively employed during the COVID-19 pandemic is sample pooling. This method involves mixing specimen for a two-step hierarchical diagnostic algorithm that foregoes individual testing in the event of a negative pooled sample.[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e] While initial TB pooling studies had found that dilution of the bacterial load can lead to lower levels of detection and potentially missed diagnoses [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e], more recent evaluations have reported sensitivity of 90% and up to 100% with the Xpert MTB/RIF Ultra assay (Xpert Ultra), a newer generation test with higher sensitivity.[\u003cspan additionalcitationids=\"CR20\" citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e] This has led to reported reductions of TB testing costs of 57%-87% depending on the pool size.[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e] A recent cost-effectiveness analysis reported a 34.9% decrease in costs when comparing individual to pooled Xpert testing.[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eThe benefits of pooling are proportional to the prevalence of the disease in the target population. The lower the prevalence, the higher the theoretical savings. This raises the utility of pooling in high-throughput, low-yield approaches such as active case finding (ACF) campaigns where large numbers of individuals need to be screened and tested to detect a person with TB. ACF is an important component of all high burden countries\u0026rsquo; national TB response and is critical to reach those people who are less likely or able to get care in public facilities. However, it is more expensive to conduct outreach, and ways to reduce costs are urgently needed to reach all persons with TB, and especially those with subclinical TB.[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eHere, we evaluated the impact of using AI outputs to inform different pooling strategies based on Xpert testing data collected during ACF campaigns in four high TB burden settings with the goal of modeling diagnostic savings and theoretical expansion of access to mWRDs.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStudy design\u003c/h2\u003e \u003cp\u003eThis was a retrospective analysis of ACF campaigns using AI probability scores to model the incremental reduction in Xpert cartridge consumption.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003eData sources\u003c/h2\u003e \u003cp\u003eData were obtained from four implementers located in high TB burden countries of Bangladesh, Nigeria, Vietnam and Zambia. These data consisted of aggregate AI probability scores and Xpert test results from community- and facility-based case finding campaigns conducted between 2014-17 in Bangladesh and 2022 and 2023 all other countries. These campaigns targeted a heterogenous mix of vulnerable populations particular to each ACF setting and country.\u003c/p\u003e \u003cp\u003eAll countries used CXR with slightly different modalities to screen for TB. Screening methods have been described elsewhere but are summarized here.[\u003cspan additionalcitationids=\"CR27 CR28\" citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e] In Bangladesh, standalone screening centers supported referrals of health-seeking individuals in outpatient care departments of public and private sectors to conduct CXR screening and Xpert testing in Dhaka. In Nigeria, mobile teams conducted active outreach events in remote rural areas using ultra-portable X-ray among people with limited access to healthcare services such as pastoralists and nomadic populations. Vietnam delivered community-based ACF campaigns in rural and urban areas focused on household contacts, older persons, urban poor, and people with a history of TB. Zambia used a portable X-ray system in health facilities to screen clinic attendees and household contacts.[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e] Nigeria and Viet Nam used qXR v3, Zambia used CAD4TB7 for generation of the AI probability score. Bangladesh used qXR v3 and CAD4TB6 for the same dataset, but only qXR v3 results from the former were due to concord with the software used in Nigeria and Viet Nam. Xpert MTB/RIF (Bangladesh and Nigeria) or the newer Xpert MTB/RIF Ultra assay (Viet Nam and Zambia) was used for diagnostic testing of distinct individuals in all countries.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003eModel structure\u003c/h2\u003e \u003cp\u003eEach participating country provided Xpert testing data aggregated by deciles (\u003cem\u003eD\u003c/em\u003e\u003csub\u003e\u003cem\u003em\u003c/em\u003e\u003c/sub\u003e, where m\u0026thinsp;=\u0026thinsp;1 to 10) of AI probability scores in the range of 0-100 for CAD4TB, i.e., \u003cem\u003eD\u003c/em\u003e\u003csub\u003e\u003cem\u003e1\u003c/em\u003e\u003c/sub\u003e: AI score\u0026thinsp;=\u0026thinsp;0\u0026ndash;9, \u003cem\u003eD\u003c/em\u003e\u003csub\u003e\u003cem\u003e2\u003c/em\u003e\u003c/sub\u003e: 10\u0026ndash;19 \u0026hellip;, and, similarly, 0-0.99 for qXR. We calculated the positivity rate (\u003cem\u003ep\u003c/em\u003e\u003csub\u003e\u003cem\u003em\u003c/em\u003e\u003c/sub\u003e) for each \u003cem\u003eD\u003c/em\u003e\u003csub\u003e\u003cem\u003em\u003c/em\u003e\u003c/sub\u003e. We assigned testing thresholds for each country based on the decile (\u003cem\u003eD\u003c/em\u003e\u003csub\u003e\u003cem\u003em\u003c/em\u003e\u003c/sub\u003e) where ~\u0026thinsp;95% of cases would be detected (\u003cem\u003eC\u003c/em\u003e) with Xpert testing starting at that decile to have only\u0026thinsp;~\u0026thinsp;5% missed cases (\u003cem\u003eM\u003c/em\u003e) as a result of not employing Xpert testing in the previous deciles. (Supplementary information 1\u0026ndash;3)\u003c/p\u003e \u003cp\u003eTo model the theoretical savings, we calculated theoretical number of tests per person needed to achieve the same positivity through pooling for a two-step hierarchical testing strategy.[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e] The difference between the total actual number of individual tests performed and the theoretical number of tests per person when employing the pooling strategy represented the number of tests saved. For the primary analysis, we assumed pool size of four based on prior studies and that both individual and pooled testing were 100% sensitive and specific.[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e] We then modeled four ordinal screening and testing approaches with increasing complexity to estimate incremental savings compared to the previous approach (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e):\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eBaseline approach: individual Xpert tests in all deciles as per the original datasets;\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eCXR approach: individual Xpert tests in all deciles for which \u003cem\u003eΣM\u003c/em\u003e\u0026thinsp;~\u0026thinsp;5%;\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eIndiscriminate pooling approach: pooled Xpert tests in all deciles for which \u003cem\u003eΣM\u003c/em\u003e\u0026thinsp;~\u0026thinsp;5%;\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eAI-guided pooling approach: a combination of pooled and individual Xpert tests in all deciles for which \u003cem\u003eΣM\u003c/em\u003e\u0026thinsp;~\u0026thinsp;5%, with individual Xpert testing in high deciles with a \u003cem\u003ep\u003c/em\u003e\u003csub\u003e\u003cem\u003em\u003c/em\u003e\u003c/sub\u003e\u0026ge;20%.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003cp\u003eWhile we aimed to achieve a \u003cem\u003eΣC\u003c/em\u003e\u0026thinsp;~\u0026thinsp;95% (\u0026equiv; \u003cem\u003eΣM\u003c/em\u003e\u0026thinsp;~\u0026thinsp;5%) for screening and testing approaches, the actual values were 95.7% (Bangladesh), 95.3% (Nigeria), 94.9% (Viet Nam), and 96.7% (Zambia).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003eData analysis\u003c/h2\u003e \u003cp\u003eWe described number of tests performed, positive results, and positivity rates in total and for each AI-score decile. We further calculated the theoretical number and proportion of tests saved incrementally between each screening and testing approach and cumulatively over the baseline. To characterize the optimal approach in each country, we identified the deciles below which testing could be foregone and the deciles at which individual testing instead of pooled testing would save tests. The number of tests saved was multiplied by the current price of \u003cspan\u003e$\u003c/span\u003e7.97 per cartridge to calculate crude cost savings, and subsequently divided by the number of positive test results for a unit-cost estimate per persons diagnosed with TB.[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e] To offer an alternative perspective to cost savings, we estimated the ratio of additional tests that could be performed with the savings over the theoretical number of tests needed for the current cohort as a measure of the extent to which access to mWRDs could have been expanded by employing the optimal screening and testing approach.\u003c/p\u003e \u003cp\u003eFor the sensitivity analyses, we modeled using pool size of three [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e] and relaxed the 100% sensitivity and specificity assumption of the pooling method based on systematic review findings from pooling with Xpert Ultra as Xpert MTB/RIF will be discontinued in 2024 using the datasets from all countries.[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e] We used binGroup2 package in R for the analysis.[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003ch2\u003eDataset characteristics\u003c/h2\u003e\n\u003cp\u003eIn the two facility-based ACF settings, Bangladesh and Zambia performed 24,079 and 2,353 Xpert tests with a matching corresponding AI result, respectively. In the community-based ACF settings, Nigeria performed 1,021 and Viet Nam performed 5,074 tests. Positivity rates were 15.3% (3,679/24,079) in Bangladesh, 11.6% (273/2,353) in Zambia, 9.0% (455/5,074) in Viet Nam and 8.3% (85/1,021) in Nigeria (Figure 2).\u003c/p\u003e\n\u003cp\u003eIn terms of testing distribution by AI-score decile most countries exhibited similar U-shaped patterns except Zambia (Figure 3A). In Zambia, almost half (49.3%) of the tests were conducted in \u003cem\u003eD\u003csub\u003e5\u003c/sub\u003e\u0026ndash;D\u003csub\u003e6\u003c/sub\u003e\u003c/em\u003e. Meanwhile, 40.6% of Xpert tests in Bangladesh had an AI score in the lowest decile (\u003cem\u003eD\u003csub\u003e1\u003c/sub\u003e\u003c/em\u003e), which was higher than Viet Nam (22.1%), Nigeria (17.4%) and Zambia (2.5%). Meanwhile, the testing rate in the AI-scores \u003cem\u003eD\u003csub\u003e4\u003c/sub\u003e\u0026ndash;D\u003csub\u003e5\u003c/sub\u003e\u003c/em\u003e, were higher in Nigeria (\u003cem\u003eD\u003csub\u003e4\u003c/sub\u003e\u003c/em\u003e:15.3% and \u003cem\u003eD\u003csub\u003e5\u003c/sub\u003e\u003c/em\u003e:15.0%) compared to Bangladesh (\u003cem\u003eD\u003csub\u003e4\u003c/sub\u003e\u003c/em\u003e:4.1% and \u003cem\u003eD\u003csub\u003e5\u003c/sub\u003e\u003c/em\u003e:3.9%) and Viet Nam (\u003cem\u003eD\u003csub\u003e4\u003c/sub\u003e\u003c/em\u003e:5.4% and \u003cem\u003eD\u003csub\u003e5\u003c/sub\u003e\u003c/em\u003e:5.2%).\u003c/p\u003e\n\u003cp\u003eIn terms of positivity by AI-score decile (Figure 3B) the two facility-based ACF sites exhibited higher rates than the community-based counterparts. In Bangladesh, positivity was high in the high AI-score deciles of \u003cem\u003eD\u003csub\u003e7\u003c/sub\u003e\u0026ndash;D\u003csub\u003e10\u003c/sub\u003e\u003c/em\u003e than in the other countries with rates increasing from 12.3% to 70.0%. Zambia exhibited a similar pattern with a lower peak with rates ranging from 13.6% to 65.1% for \u003cem\u003eD\u003csub\u003e7\u003c/sub\u003e\u003c/em\u003e and \u003cem\u003eD\u003csub\u003e10\u003c/sub\u003e\u003c/em\u003e, respectively. In comparison, the respective positivity of \u003cem\u003eD\u003csub\u003e7\u003c/sub\u003e\u003c/em\u003e and \u003cem\u003eD\u003csub\u003e10\u003c/sub\u003e\u003c/em\u003e only rose from 4.8% to 28.7% in Nigeria and from 6.3% to 30.6% in Viet Nam.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003eTests cost savings in the model output\u003c/h2\u003e\n\u003cp\u003eAll incremental screening and testing approaches resulted in additional savings except the indiscriminate pooling approach in Bangladesh and are presented for each country in Table 1. In Bangladesh, the CXR approach resulted in savings of 52.5% over baseline. Based on model parameters, these savings were realized at an AI threshold of 0.30-0.39, i.e., foregoing testing for \u003cem\u003eD\u003csub\u003e1\u003c/sub\u003e\u0026ndash;D\u003csub\u003e3\u003c/sub\u003e\u003c/em\u003e and individually testing everyone in higher deciles. Indiscriminately pooling all persons in \u003cem\u003eD\u003csub\u003e4\u003c/sub\u003e\u0026ndash;D\u003csub\u003e10\u003c/sub\u003e\u003c/em\u003e actually resulted in excess of 2.0% tests over the CXR approach. Using AI to indicate individual testing for high AI-score deciles \u003cem\u003eD\u003csub\u003e9\u003c/sub\u003e\u0026ndash;D\u003csub\u003e10\u003c/sub\u003e\u003c/em\u003e reversed this trend and increased cumulative savings to 61.5%.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1. Incremental and cumulative savings by country.\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"595\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"39.59731543624161%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.093959731543624%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u003cstrong\u003eTotal tests\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"23.825503355704697%\" colspan=\"2\" valign=\"bottom\"\u003e\n \u003cp\u003e\u003cstrong\u003eIncremental savings\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.483221476510067%\" colspan=\"2\" valign=\"bottom\"\u003e\n \u003cp\u003e\u003cstrong\u003eCumulative savings\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"39.66386554621849%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.117647058823529%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u003cstrong\u003eN\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.453781512605042%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u003cstrong\u003eN\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.243697478991596%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u003cstrong\u003e%\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.941176470588236%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u003cstrong\u003eN\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.579831932773109%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u003cstrong\u003e%\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"39.66386554621849%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u003cstrong\u003eBangladesh\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.117647058823529%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.453781512605042%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.243697478991596%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.941176470588236%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.579831932773109%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"39.66386554621849%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;Baseline approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.117647058823529%\" valign=\"bottom\"\u003e\n \u003cp\u003e24,079\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.453781512605042%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.243697478991596%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.941176470588236%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.579831932773109%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"39.66386554621849%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;CXR approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.117647058823529%\" valign=\"bottom\"\u003e\n \u003cp\u003e11,448\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.453781512605042%\" valign=\"bottom\"\u003e\n \u003cp\u003e12,631\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.243697478991596%\" valign=\"bottom\"\u003e\n \u003cp\u003e52.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.941176470588236%\" valign=\"bottom\"\u003e\n \u003cp\u003e12,631\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.579831932773109%\" valign=\"bottom\"\u003e\n \u003cp\u003e52.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"39.66386554621849%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;Indiscriminate pooling approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.117647058823529%\" valign=\"bottom\"\u003e\n \u003cp\u003e11,676\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.453781512605042%\" valign=\"bottom\"\u003e\n \u003cp\u003e-228\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.243697478991596%\" valign=\"bottom\"\u003e\n \u003cp\u003e-2.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.941176470588236%\" valign=\"bottom\"\u003e\n \u003cp\u003e12,403\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.579831932773109%\" valign=\"bottom\"\u003e\n \u003cp\u003e51.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"39.66386554621849%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;AI-guided pooling approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.117647058823529%\" valign=\"bottom\"\u003e\n \u003cp\u003e9,262\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.453781512605042%\" valign=\"bottom\"\u003e\n \u003cp\u003e2,414\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.243697478991596%\" valign=\"bottom\"\u003e\n \u003cp\u003e20.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.941176470588236%\" valign=\"bottom\"\u003e\n \u003cp\u003e14,817\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.579831932773109%\" valign=\"bottom\"\u003e\n \u003cp\u003e61.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"39.66386554621849%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u003cstrong\u003eZambia\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.117647058823529%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.453781512605042%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.243697478991596%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.941176470588236%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.579831932773109%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"39.66386554621849%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;Baseline approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.117647058823529%\" valign=\"bottom\"\u003e\n \u003cp\u003e2,353\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.453781512605042%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.243697478991596%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.941176470588236%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.579831932773109%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"39.66386554621849%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;CXR approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.117647058823529%\" valign=\"bottom\"\u003e\n \u003cp\u003e1,960\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.453781512605042%\" valign=\"bottom\"\u003e\n \u003cp\u003e393\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.243697478991596%\" valign=\"bottom\"\u003e\n \u003cp\u003e16.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.941176470588236%\" valign=\"bottom\"\u003e\n \u003cp\u003e393\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.579831932773109%\" valign=\"bottom\"\u003e\n \u003cp\u003e16.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"39.66386554621849%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;Indiscriminate pooling approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.117647058823529%\" valign=\"bottom\"\u003e\n \u003cp\u003e1,352\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.453781512605042%\" valign=\"bottom\"\u003e\n \u003cp\u003e608\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.243697478991596%\" valign=\"bottom\"\u003e\n \u003cp\u003e31.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.941176470588236%\" valign=\"bottom\"\u003e\n \u003cp\u003e1,001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.579831932773109%\" valign=\"bottom\"\u003e\n \u003cp\u003e42.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"39.66386554621849%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;AI-guided pooling approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.117647058823529%\" valign=\"bottom\"\u003e\n \u003cp\u003e1,158\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.453781512605042%\" valign=\"bottom\"\u003e\n \u003cp\u003e194\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.243697478991596%\" valign=\"bottom\"\u003e\n \u003cp\u003e14.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.941176470588236%\" valign=\"bottom\"\u003e\n \u003cp\u003e1,195\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.579831932773109%\" valign=\"bottom\"\u003e\n \u003cp\u003e50.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"39.66386554621849%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u003cstrong\u003eNigeria\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.117647058823529%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.453781512605042%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.243697478991596%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.941176470588236%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.579831932773109%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"39.66386554621849%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;Baseline approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.117647058823529%\" valign=\"bottom\"\u003e\n \u003cp\u003e1,021\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.453781512605042%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.243697478991596%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.941176470588236%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.579831932773109%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"39.66386554621849%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;CXR approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.117647058823529%\" valign=\"bottom\"\u003e\n \u003cp\u003e738\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.453781512605042%\" valign=\"bottom\"\u003e\n \u003cp\u003e283\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.243697478991596%\" valign=\"bottom\"\u003e\n \u003cp\u003e27.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.941176470588236%\" valign=\"bottom\"\u003e\n \u003cp\u003e283\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.579831932773109%\" valign=\"bottom\"\u003e\n \u003cp\u003e27.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"39.66386554621849%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;Indiscriminate pooling approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.117647058823529%\" valign=\"bottom\"\u003e\n \u003cp\u003e459\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.453781512605042%\" valign=\"bottom\"\u003e\n \u003cp\u003e279\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.243697478991596%\" valign=\"bottom\"\u003e\n \u003cp\u003e37.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.941176470588236%\" valign=\"bottom\"\u003e\n \u003cp\u003e562\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.579831932773109%\" valign=\"bottom\"\u003e\n \u003cp\u003e55.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"39.66386554621849%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;AI-guided pooling approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.117647058823529%\" valign=\"bottom\"\u003e\n \u003cp\u003e434\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.453781512605042%\" valign=\"bottom\"\u003e\n \u003cp\u003e25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.243697478991596%\" valign=\"bottom\"\u003e\n \u003cp\u003e5.4%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.941176470588236%\" valign=\"bottom\"\u003e\n \u003cp\u003e587\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.579831932773109%\" valign=\"bottom\"\u003e\n \u003cp\u003e57.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"39.66386554621849%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u003cstrong\u003eViet Nam\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.117647058823529%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.453781512605042%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.243697478991596%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.941176470588236%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.579831932773109%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"39.66386554621849%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;Baseline approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.117647058823529%\" valign=\"bottom\"\u003e\n \u003cp\u003e5,074\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.453781512605042%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.243697478991596%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.941176470588236%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.579831932773109%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"39.66386554621849%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;CXR approach\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.117647058823529%\" valign=\"bottom\"\u003e\n \u003cp\u003e2,917\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.453781512605042%\" valign=\"bottom\"\u003e\n \u003cp\u003e2,157\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.243697478991596%\" valign=\"bottom\"\u003e\n \u003cp\u003e42.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.941176470588236%\" valign=\"bottom\"\u003e\n \u003cp\u003e2,157\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.579831932773109%\" valign=\"bottom\"\u003e\n \u003cp\u003e42.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"39.66386554621849%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;Indiscriminate pooling approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.117647058823529%\" valign=\"bottom\"\u003e\n \u003cp\u003e2,110\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.453781512605042%\" valign=\"bottom\"\u003e\n \u003cp\u003e807\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.243697478991596%\" valign=\"bottom\"\u003e\n \u003cp\u003e27.7%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.941176470588236%\" valign=\"bottom\"\u003e\n \u003cp\u003e2,964\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.579831932773109%\" valign=\"bottom\"\u003e\n \u003cp\u003e58.4%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"39.66386554621849%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;AI-guided pooling approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.117647058823529%\" valign=\"bottom\"\u003e\n \u003cp\u003e1,955\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.453781512605042%\" valign=\"bottom\"\u003e\n \u003cp\u003e155\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.243697478991596%\" valign=\"bottom\"\u003e\n \u003cp\u003e7.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.941176470588236%\" valign=\"bottom\"\u003e\n \u003cp\u003e3,119\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.579831932773109%\" valign=\"bottom\"\u003e\n \u003cp\u003e61.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cem\u003eCXR=Chest X-ray; AI=Artificial Intelligence.\u003c/em\u003e\u003c/p\u003e\n\u003ch3\u003eTable 1 note\u003c/h3\u003e\n\u003cp\u003eThe model assumes pool sizes of 4 with both pooling and individual testing sensitivity and specificity of 100% for a testing threshold resulting in missed cases of 4.4%, 4.7%, 5.1% for Bangladesh, Nigeria and Viet Nam using qXR v3, respectively, and 3.3% for Zambia using CAD4TB v7. Incremental savings indicate the difference from the prior case, whereas cumulative savings are calculated against the baseline approach.\u003c/p\u003e\n\u003cp\u003eIn Zambia, the CXR approach yielded savings of 16.7% over baseline at an AI threshold of 0.3. Indiscriminately pooling all persons in \u003cem\u003eD\u003csub\u003e4\u003c/sub\u003e\u003c/em\u003e and higher produced incremental savings of 31.0% and cumulative savings over baseline of 42.5%. Per AI guidance, testing reverted to an individual basis in \u003cem\u003eD\u003csub\u003e8\u003c/sub\u003e\u0026ndash;D\u003csub\u003e10\u003c/sub\u003e\u003c/em\u003e to yield additional savings of 14.3% for cumulative savings of 50.8%.\u003c/p\u003e\n\u003cp\u003eThe CXR approach in Nigeria showed an initial savings of 27.7% in testing at an AI threshold of 0.3 above which indiscriminate pooling reduced an incremental 37.8% in testing for a cumulative savings of 55.0% over baseline. Similar to Viet Nam, there were diminishing returns from the AI-guided pooling approach as switching to individual testing in \u003cem\u003eD\u003csub\u003e10\u003c/sub\u003e\u003c/em\u003e led to incremental and cumulative savings over baseline were only 5.4% and 57.5%, respectively.\u003c/p\u003e\n\u003cp\u003eIn Viet Nam, the CXR approach reduced testing by 42.5% at an AI threshold of 0.4. Meanwhile, indiscriminate pooling above the AI threshold saved an incremental 27.7% in testing over the CXR approach for a cumulative savings of 58.4% over baseline. The AI-guided pooling approach indicated individual testing in \u003cem\u003eD\u003csub\u003e10\u003c/sub\u003e\u003c/em\u003e which added 7.3% in incremental test reductions for a cumulative savings of 61.5% (Figure 4).\u003c/p\u003e\n\u003cp\u003eAt the above-described pooling scenarios, in Bangladesh the number of tests needed declined from 24,079 to 9,262 for savings of 12,403-14,817 tests. This translated to $98,852-$118,091 in crude costs or $26.87-$32.10 per person with TB averted (Table 2). Alternatively, these savings could be redeployed for an expansion of mWRD access of 110%\u0026ndash;160% with existing public health resources. In Zambia, the number of tests dropped from 2,353 to 1,158 for savings of 393-1,195 tests. This implies cost savings of $3,132-$9,524 or $11.47-$34.89 per person with TB for a theoretical expansion of mWRD access of 20%\u0026ndash;103%. Based on test volume reductions from 1,021 to 434 in Nigeria, the total test and cost savings ranged from 283-587 and $2,256-$4,678 or $26.54-$55.04 per person with TB. This represented 38%\u0026ndash;135% in potentially greater access to mWRD. Lastly, in Viet Nam the number of tests needed fell from 5,074 to 1,955 for 2,157-3,119 tests saved corresponding to crude savings or $17,191-$24,858 or $37.78-$54.63 per person with TB. This represented a potential mWRD expansion of 74%-160%.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2. Crude total and per-TB detection cost savings and mWRD access expansion by country.\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"747\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"27.540106951871657%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"26.470588235294116%\" colspan=\"2\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eCrude cost savings ($)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"21.925133689839573%\" colspan=\"2\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003ePer-TB cost savings ($)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"24.06417112299465%\" colspan=\"2\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003emWRD access expansion\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"27.613941018766756%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.075067024128685%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eIncremental\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.466487935656836%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eCumulative\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.589812332439678%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eIncremental\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eCumulative\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.734584450402144%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eIncremental\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eCumulative\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"27.613941018766756%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u003cstrong\u003eBangladesh\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.075067024128685%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.466487935656836%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.589812332439678%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.734584450402144%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"27.613941018766756%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;Baseline approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.075067024128685%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.466487935656836%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.589812332439678%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.734584450402144%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"27.613941018766756%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;CXR approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.075067024128685%\"\u003e\n \u003cp\u003e100,669\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.466487935656836%\"\u003e\n \u003cp\u003e100,669\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.589812332439678%\" valign=\"top\"\u003e\n \u003cp\u003e27.36\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e27.36\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.734584450402144%\"\u003e\n \u003cp\u003e110.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\"\u003e\n \u003cp\u003e110.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"27.613941018766756%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;Indiscriminate pooling approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.075067024128685%\"\u003e\n \u003cp\u003e-1,817\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.466487935656836%\"\u003e\n \u003cp\u003e98,852\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.589812332439678%\" valign=\"top\"\u003e\n \u003cp\u003e-0.49\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e26.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.734584450402144%\"\u003e\n \u003cp\u003e-2.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\"\u003e\n \u003cp\u003e106.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"27.613941018766756%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;AI-guided pooling approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.075067024128685%\"\u003e\n \u003cp\u003e19,240\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.466487935656836%\"\u003e\n \u003cp\u003e118,091\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.589812332439678%\" valign=\"top\"\u003e\n \u003cp\u003e5.23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e32.10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.734584450402144%\"\u003e\n \u003cp\u003e26.1%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\"\u003e\n \u003cp\u003e160.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"27.613941018766756%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u003cstrong\u003eZambia\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.075067024128685%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.466487935656836%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.589812332439678%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.734584450402144%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"27.613941018766756%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;Baseline approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.075067024128685%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.466487935656836%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.589812332439678%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.734584450402144%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"27.613941018766756%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;CXR approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.075067024128685%\"\u003e\n \u003cp\u003e3,132\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.466487935656836%\"\u003e\n \u003cp\u003e3,132\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.589812332439678%\" valign=\"top\"\u003e\n \u003cp\u003e11.47\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e11.47\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.734584450402144%\"\u003e\n \u003cp\u003e20.1%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\"\u003e\n \u003cp\u003e20.1%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"27.613941018766756%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;Indiscriminate pooling approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.075067024128685%\"\u003e\n \u003cp\u003e4,846\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.466487935656836%\"\u003e\n \u003cp\u003e7,978\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.589812332439678%\" valign=\"top\"\u003e\n \u003cp\u003e17.75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e29.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.734584450402144%\"\u003e\n \u003cp\u003e45.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\"\u003e\n \u003cp\u003e74.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"27.613941018766756%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;AI-guided pooling approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.075067024128685%\"\u003e\n \u003cp\u003e1,546\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.466487935656836%\"\u003e\n \u003cp\u003e9,524\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.589812332439678%\" valign=\"top\"\u003e\n \u003cp\u003e5.66\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e34.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.734584450402144%\"\u003e\n \u003cp\u003e16.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\"\u003e\n \u003cp\u003e103.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"27.613941018766756%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u003cstrong\u003eNigeria\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.075067024128685%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.466487935656836%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.589812332439678%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.734584450402144%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"27.613941018766756%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;Baseline approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.075067024128685%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.466487935656836%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.589812332439678%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.734584450402144%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"27.613941018766756%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;CXR approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.075067024128685%\"\u003e\n \u003cp\u003e2,256\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.466487935656836%\"\u003e\n \u003cp\u003e2,256\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.589812332439678%\" valign=\"top\"\u003e\n \u003cp\u003e26.53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e26.53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.734584450402144%\"\u003e\n \u003cp\u003e38.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\"\u003e\n \u003cp\u003e38.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"27.613941018766756%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;Indiscriminate pooling approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.075067024128685%\"\u003e\n \u003cp\u003e2,224\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.466487935656836%\"\u003e\n \u003cp\u003e4,479\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.589812332439678%\" valign=\"top\"\u003e\n \u003cp\u003e26.16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e52.70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.734584450402144%\"\u003e\n \u003cp\u003e60.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\"\u003e\n \u003cp\u003e122.4%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"27.613941018766756%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;AI-guided pooling approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.075067024128685%\"\u003e\n \u003cp\u003e199\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.466487935656836%\"\u003e\n \u003cp\u003e4,678\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.589812332439678%\" valign=\"top\"\u003e\n \u003cp\u003e2.34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e55.04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.734584450402144%\"\u003e\n \u003cp\u003e5.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\"\u003e\n \u003cp\u003e135.3%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"27.613941018766756%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u003cstrong\u003eViet Nam\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.075067024128685%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.466487935656836%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.589812332439678%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.734584450402144%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"27.613941018766756%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;Baseline approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.075067024128685%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.466487935656836%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.589812332439678%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.734584450402144%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"27.613941018766756%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;CXR approach\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.075067024128685%\"\u003e\n \u003cp\u003e17,191\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.466487935656836%\"\u003e\n \u003cp\u003e17,191\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.589812332439678%\" valign=\"top\"\u003e\n \u003cp\u003e37.78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e37.78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.734584450402144%\"\u003e\n \u003cp\u003e73.9%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\"\u003e\n \u003cp\u003e73.9%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"27.613941018766756%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;Indiscriminate pooling approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.075067024128685%\"\u003e\n \u003cp\u003e6,432\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.466487935656836%\"\u003e\n \u003cp\u003e23,623\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.589812332439678%\" valign=\"top\"\u003e\n \u003cp\u003e14.14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e51.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.734584450402144%\"\u003e\n \u003cp\u003e38.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\"\u003e\n \u003cp\u003e140.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"27.613941018766756%\" valign=\"bottom\"\u003e\n \u003cp\u003e\u0026nbsp; \u0026nbsp;AI-guided pooling approach\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"14.075067024128685%\"\u003e\n \u003cp\u003e1,235\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.466487935656836%\"\u003e\n \u003cp\u003e24,858\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"10.589812332439678%\" valign=\"top\"\u003e\n \u003cp\u003e2.72\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\" valign=\"top\"\u003e\n \u003cp\u003e54.63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.734584450402144%\"\u003e\n \u003cp\u003e7.9%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.26005361930295%\"\u003e\n \u003cp\u003e159.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cem\u003eCXR=Chest X-ray; AI=Artificial Intelligence.\u003c/em\u003e\u003c/p\u003e\n\u003ch3\u003eTable 2 note\u003c/h3\u003e\n\u003cp\u003eIncremental savings indicate the difference from the prior case, whereas cumulative savings are calculated against the baseline approach. Crude cost savings are based on a cost of $7.97 per Xpert cartridge. Per-TB case cost savings are based on number of positive test results of 3,679 in Bangladesh, 273 in Zambia, 85 in Nigeria and 455 in Viet Nam. mWRD access expansion was calculated by dividing the number of cartridges saved by the number of cartridges used in each individual screening and testing approach.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;The sensitivity analyses did not show a substantial change of the results. Across different scenarios, the change in savings against the primary case (pool size of three with pooled sensitivity and specificity of 1) ranged from -2.6% to +3.8%. Testing in smaller pools increased number of tests. Reducing pooling sensitivity reduced test usage and so did increasing pooling specificity, which will result in an increase in missed cases and false positives, respectively, for the overall two-step hierarchical testing algorithm. (Supplementary table 4) In Bangladesh, we compared savings between qXRv3 and CAD4TB6 and found that differences were small (0.7%). CAD4TB scores resulted in 9,200 tests in the AI-guided pooling approach compared to 9,262 in qXRv3 against the base case of 24,079 test. (Supplementary table 5)\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eOur modelling study results show that the combined use of computer-aided chest radiography and pooling may achieve compounding effects to achieve significant savings in diagnostic consumables that could be redeployed to increase the global coverage of mWRDs. We further found that the substantial heterogeneity in AI thresholds and impact of AI scores on subsequent pooling across the different settings will require differentiated deployment of these two innovations in order to optimize the potential gains. However, while countries and settings may differ, it appears that the various screening and testing approaches and particularly the AI-guided pooling approach may be able to achieve consistent results between the most advanced software platforms.\u003c/p\u003e \u003cp\u003eAcross the different countries, settings and approaches, the combination of using CXR with AI-scores to inform decisions on pooling sputum produced cumulative savings on testing of 50.8%-61.5% over baseline compared to universal testing of all individuals with signs of TB. In 2022, Bangladesh tested only 20% of people with TB with mWRDs as the first-line diagnostic test. While the gap was smaller in Nigeria, Zambia and Vietnam, three in ten persons with TB were not diagnosed using molecular diagnostics.[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e] Our results suggest that more than twice as many people could be tested for the same diagnostic test costs using CXR and pooling as compared to testing all presumptive individuals.\u003c/p\u003e \u003cp\u003eA major contributor to this potential capacity expansion was the use of AI and CXR. As the data of all countries originated from ACF campaigns rather than prevalence surveys, each dataset included an inherent level of preselection or pre-screening to raise the TB prevalence in the sample. This pre-filtering may have consisted of targeting of highly vulnerable populations such as contacts (Zambia), deploying in high-yield settings such as health facilities (Bangladesh) or having a previous binary read of the CXR by a potentially inexperienced human reader (Viet Nam).[\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e] Despite these methods of preselection, our study once again highlighted the well-documented effectiveness of CXR for screening and triaging for TB.[\u003cspan additionalcitationids=\"CR37 CR38\" citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e] Beyond that, we observed incremental savings by leveraging the AI\u0026rsquo;s quantitative output to optimize CXR interpretation and forego testing in lower AI-score deciles. This approach was particularly useful in settings with a high testing proportion in the lowest decile such as Bangladesh and Viet Nam, the first screening and testing approach generated already generated cumulative savings of 52.5% and 42.5%, respectively. This optimization of testing volumes through computer-aided chest radiography was also concordant with available evidence.[\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eOur model suggested that in each country pooling could be effective for generating additional savings in testing. However, our model also evinced differences across both pooling strategies based on the populations screened and the results of the CXR and AI. For instance, in the facility-based settings of Bangladesh and Zambia 65%-70% of individuals in the highest decile had bacteriologically confirmed TB compared to only 29%-31% in Nigeria and Viet Nam\u0026rsquo;s community-based ACF cohort. This dichotomy was reflected by the pooling approach, as the AI-guided pooling, i.e., reversion to individual testing in high deciles, continued to generated substantial savings in the facilities, while the model exhibited substantial diminishing returns in the low-yield community setting. Interestingly, while indiscriminate pooling saved tests in three countries, in Bangladesh it actually increased testing unless an AI-guided pooling approach was used. In Bangladesh, where reads for both qXR and CAD4TB were available, the performance on both platforms were highly concordant. However, each required adjustment of the testing threshold, highlighting the need for end-user optimization based on the local epidemiology and platforms used. The approaches may be further optimized based on where sample volumes allow more pools. In one such additional approach, the AI-guided cohort pooling case resulted in up to 2.5% additional savings over the AI-guided pooling case. (Supplementary information 5) Nevertheless, these finding inspire confidence in the high precision that characterizes many of today\u0026rsquo;s commercial AI solutions for TB.[\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]\u003c/p\u003e \u003cp\u003eThis variability was encountered in different areas of our study and represented its key strength. For example, in contrast to the high savings achieved through the CXR approach in Bangladesh and Viet Nam, savings in Nigeria and Zambia were only 28% and 17%, respectively. Yet, both countries differed substantially in the additional gains from each incremental pooling approach. The testing distribution patterns also exhibited discordance, as all countries exhibited a U-shaped curve except Zambia, or in positivity by decile, whereby reversion to individual testing of the AI-guided pooling approach occurred in \u003cem\u003eD\u003c/em\u003e\u003csub\u003e\u003cem\u003e8\u003c/em\u003e\u003c/sub\u003e for the Zambian dataset, in \u003cem\u003eD\u003c/em\u003e\u003csub\u003e\u003cem\u003e9\u003c/em\u003e\u003c/sub\u003e for the data from Bangladesh and \u003cem\u003eD\u003c/em\u003e\u003csub\u003e\u003cem\u003e10\u003c/em\u003e\u003c/sub\u003e in Nigeria and Viet Nam. Particularly with respect to the optimal application of AI and its continuous outputs, there has been much discussion around using different AI threshold scores. However, the growing evidence base suggests variability in AI performance across demographic, clinical, or behavioral characteristics of the population, screening setting or even radiography equipment, which underscores the well-documented need for local calibration and threshold setting.[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e, \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e] Similarly, while general pooling can save costs, optimizing a pooling strategy will depend on the use of local data, capacities and established practices.\u003c/p\u003e \u003cp\u003eOur study has a limitation in that modelled results are fraught with assumptions about performance and each situation will differ. In the primary analysis, we assume perfect sensitivity and specificity for both pooling and individual testing, which is unlike real-world performance. While we attempt to mitigate this limitation in the sensitivity analysis by varying the pooling parameters, we do not model the overall performance of the diagnostic algorithm, e.g., the overall sensitivity and specificity of a combination of no testing, individual testing and pooled testing in the AI-guided pooling case. This will also impact the overall number of missed cases in the algorithm. We do not expect a substantial impact on the number of tests given the superior performance of Xpert Ultra in both individual and pooled testing documented previously, and encourage future prospective evaluations to factor in these considerations, in addition to other implementation challenges. Lastly, although the data are based on real-world ACF activities they did not contain the daily distribution of people with presumptive TB and their abnormality scores. Hence, for this analysis we assumed 100% completeness in pool sizes with four sputum specimen per pool. Pooling-decisions based on abnormality scores such as the AI-guided pooling approach would on occasion require pools of two or three specimen, which might impact overall savings. However, we conducted this sensitivity analysis, which also showed that the impact on overall savings was small.\u003c/p\u003e \u003cp\u003eThe real-world readiness for computer-aided chest radiography informed pooling in real-time in programmatic settings can be challenging. Despite the positive reception of pooling by laboratory technicians for its time-saving properties [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e], pooled testing for TB is not feasible in all laboratories. For example, though specificity issues have not been widely reported, the manipulation of several samples bears the risk of contamination. CXR and AI are also not readily available in most high TB burden countries. Nevertheless, as pooling becomes established practice in more indications and access to AI improves, this type of pragmatic approach could mitigate commonly encountered cartridge shortages and provide more people with signs of TB access to molecular tests and should be evaluated prospectively to see how they work in real world settings.\u003c/p\u003e"},{"header":"Conclusions","content":"\u003cp\u003eTo achieve End TB Strategy goals, it is necessary to optimize the use of available tools. Integrating computer-aided chest radiography and pooling into TB screening and testing algorithms has the potential to substantially reduce diagnostic testing, thus freeing up constrained financial and human public health resources to save costs and extend access to more people in need of high-quality, rapid molecular testing for TB.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eACF Active Case Finding\u003c/p\u003e\n\u003cp\u003eAI Artificial Intelligence\u003c/p\u003e\n\u003cp\u003eCXR Chest X-ray\u003c/p\u003e\n\u003cp\u003emWRD molecular WHO-recommended Rapid Diagnostic\u003c/p\u003e\n\u003cp\u003eNAAT Nucleic Acid Amplification Test\u003c/p\u003e\n\u003cp\u003eTB Tuberculosis\u003c/p\u003e\n\u003cp\u003eWHO World Health Organization\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eEthics approval and consent to participate\u003c/h2\u003e\n\u003cp\u003eEthical approvals were obtained from respective review committees for the original data collection in the four participating countries. All data furnished for this study were aggregated and contained no personally identifiable information and thus this modelling study did not require additional ethical approval.\u003c/p\u003e\n\u003ch2\u003eConsent for publication\u003c/h2\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003ch2\u003eAvailability of data and materials\u003c/h2\u003e\n\u003cp\u003eAll the data used for the analysis is available in the paper. The reproducible code for analysis is available in a public GitHub repository at https://github.com/peptonefizz/Xpert-pooling-analysis.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003eCompeting interests\u003c/h2\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\n\u003ch2\u003eFunding\u003c/h2\u003e\n\u003cp\u003eTW is supported by grants from: the Wellcome Trust, UK (209075/Z/17/Z); the Department of Health and Social Care (DHSC), the Foreign, Commonwealth \u0026amp; Development Office (FCDO), the Medical Research Council (MRC) and Wellcome, UK (Joint Global Health Trials, MR/V004832/1); the Medical Research Council (Public Health Intervention Development Award \u0026ldquo;PHIND\u0026rdquo;, APP2293); and the Medical Research Foundation (Dorothy Temple Cross International Collaboration Research Grant, MRF-131\u0026ndash;0006-RG-KHOS-C0942).\u003c/p\u003e\n\u003ch2\u003eAuthors\u0026apos; contributions\u003c/h2\u003e\n\u003cp\u003eJC, LNQV, AJC and TG conceptualized the manuscript. The methodology was developed by AJC and TG. Formal analyses and visualization were conducted by LNQV and TG. Resources were contributed by SB, AJC, LNQV, MM, SJ and SA. JC, LNQV, and TG wrote the original draft, while review and editing was done by all authors. Supervision was provided by JC. JC and LNQV were responsible for coordination with the co-authors and submission of the article. All authors have read and approved the final manuscript.\u003c/p\u003e\n\u003ch2\u003eAcknowledgements\u003c/h2\u003e\n\u003cp\u003eThe authors would like to thank their affiliate institutions and NTP of the contributing countries. We also express our gratitude to public health staff and partners at primary and secondary care levels, and the communities and patients involved in the initial generation of the original data used for the secondary analysis of this study.\u003c/p\u003e\n\u003ch2\u003eAuthors\u0026apos; information (optional)\u003c/h2\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eEvans CA. GeneXpert\u0026mdash;A Game-Changer for Tuberculosis Control? PLoS Med. 2011;8:e1001064.\u003c/li\u003e\n\u003cli\u003eNikam C, Jagannath M, Narayanan MM, Ramanabhiraman V, Kazi M, Shetty A, et al. Rapid Diagnosis of Mycobacterium tuberculosis with Truenat MTB: A Near-Care Approach. PLoS One. 2013;8:1\u0026ndash;7.\u003c/li\u003e\n\u003cli\u003eBranigan D. Tuberculosis Diagnostics Pipeline Report \u0026raquo; 2023. 2023.\u003c/li\u003e\n\u003cli\u003eWorld Health Organization. Global Tuberculosis Report 2023. Geneva, Switzerland; 2023.\u003c/li\u003e\n\u003cli\u003eWorld Health Organization. WHO standard: universal access to rapid tuberculosis diagnostics. Geneva, Switzerland: World Health Organization; 2023.\u003c/li\u003e\n\u003cli\u003eDaniel TM. The history of tuberculosis. Respir Med. 2006;100:1862\u0026ndash;70.\u003c/li\u003e\n\u003cli\u003eAlbert H, Nathavitharana RR, Isaacs C, Pai M, Denkinger CM, Boehme CC. Development, roll-out and impact of Xpert MTB/RIF for tuberculosis: What lessons have we learnt and how can we do better? Eur Respir J. 2016;48:516\u0026ndash;25.\u003c/li\u003e\n\u003cli\u003eEngel N, Ochodo EA, Karanja PW, Schmidt B-M, Janssen R, Steingart KR, et al. Rapid molecular tests for tuberculosis and tuberculosis drug resistance: provider and recipient views. Cochrane Database Syst Rev. 2021;2021.\u003c/li\u003e\n\u003cli\u003eStop TB Partnership. November 2023 Diagnostics, Medical Devices \u0026amp; Other Health Products Catalog. 2023; November.\u003c/li\u003e\n\u003cli\u003eBr\u0026uuml;mmer LE, Thompson RR, Malhotra A, Shrestha S, Kendall EA, Andrews JR, et al. Cost-effectiveness of Low-complexity Screening Tests in Community-based Case-finding for Tuberculosis. Clin Infect Dis. 2023;:1\u0026ndash;10.\u003c/li\u003e\n\u003cli\u003eBaik Y, Nakasolya O, Isooba D, Mukiibi J, Kitonsa PJ, Erisa KC, et al. Cost to perform door-to-door universal sputum screening for TB in a high-burden community. Int J Tuberc Lung Dis. 2023;27:195\u0026ndash;201.\u003c/li\u003e\n\u003cli\u003eWorld Health Organization. WHO consolidated guidelines on tuberculosis, Module 2: Screening, Systematic screening for tuberculosis disease. Geneva, Switzerland; 2021.\u003c/li\u003e\n\u003cli\u003eLaw I, Floyd K, Abukaraig EAB, Addo KK, Adetifa I, Alebachew Z, et al. National tuberculosis prevalence surveys in Africa, 2008\u0026ndash;2016: an overview of results and lessons learned. Trop Med Int Heal. 2020;25:1308\u0026ndash;27.\u003c/li\u003e\n\u003cli\u003eOnozaki I, Law I, Sismanidis C, Zignol M, Glaziou P, Floyd K. National tuberculosis prevalence surveys in Asia, 1990-2012: An overview of results and lessons learned. Trop Med Int Heal. 2015;20:1128\u0026ndash;45.\u003c/li\u003e\n\u003cli\u003eCreswell J, Vo LNQ, Qin ZZ, Muyoyeta M, Tovar M, Wong EB, et al. Early user perspectives on using computer-aided detection software for interpreting chest X-ray images to enhance access and quality of care for persons with tuberculosis. BMC Glob Public Heal. 2023;1:30.\u003c/li\u003e\n\u003cli\u003eGeric C, Qin ZZ, Denkinger CM, Kik S V., Marais B, Anjos A, et al. The rise of artificial intelligence reading of chest X-rays for enhanced TB diagnosis and elimination. Int J Tuberc Lung Dis. 2023;27:367\u0026ndash;72.\u003c/li\u003e\n\u003cli\u003eDaniel EA, Esakialraj L BH, S A, Muthuramalingam K, Karunaianantham R, Karunakaran LP, et al. Pooled Testing Strategies for SARS-CoV-2 diagnosis: A comprehensive review. Diagn Microbiol Infect Dis. 2021;101:115432.\u003c/li\u003e\n\u003cli\u003eHo J, Jelfs P, Nguyen PTB, Sintchenko V, Fox GJ, Marks GB. Pooling sputum samples to improve the feasibility of Xpert\u0026reg; MTB/RIF in systematic screening for tuberculosis. Int J Tuberc Lung Dis. 2017;21:503\u0026ndash;8.\u003c/li\u003e\n\u003cli\u003eVuchas C, Teyim P, Dang BF, Neh A, Keugni L, Che M, et al. Implementation of large-scale pooled testing to increase rapid molecular diagnostic test coverage for tuberculosis: a retrospective evaluation. Sci Rep. 2023;13:1\u0026ndash;10.\u003c/li\u003e\n\u003cli\u003eCuevas LE, Santos VS, Lima SVMA, Kontogianni K, Bimba JS, Iem V, et al. Systematic review of pooling sputum as an efficient method for xpert MTB/RIF tuberculosis testing during COVID-19 pandemic. Emerg Infect Dis. 2021;27:719\u0026ndash;27.\u003c/li\u003e\n\u003cli\u003eChakravorty S, Simmons AM, Rowneki M, Parmar H, Cao Y, Ryan J, et al. The new Xpert MTB/RIF ultra: Improving detection of Mycobacterium tuberculosis and resistance to Rifampin in an assay suitable for point-of-care testing. MBio. 2017;8.\u003c/li\u003e\n\u003cli\u003eDos Santos PCP, Da Silva Santos A, De Oliveira RD, Da Silva BO, Soares TR, Martinez L, et al. Pooling Sputum Samples for Efficient Mass Tuberculosis Screening in Prisons. Clin Infect Dis. 2022;74:2115\u0026ndash;21.\u003c/li\u003e\n\u003cli\u003eIem V, Bimba JS, Santos VS, Dominguez J, Creswell J, Somphavong S, et al. Pooling sputum testing to diagnose tuberculosis using xpert MTB / RIF and xpert ultra : a cost-effectiveness analysis. 2023;:1\u0026ndash;11.\u003c/li\u003e\n\u003cli\u003eStop TB Partnership. The Global Plan to End TB 2023-2030. Geneva, Switzerland; 2022.\u003c/li\u003e\n\u003cli\u003eIsmail N, Nathanson CM, Zignol M, Kasaeva T. Achieving universal access to rapid tuberculosis diagnostics. BMJ Glob Heal. 2023;8:8\u0026ndash;10.\u003c/li\u003e\n\u003cli\u003eBanu S, Haque F, Ahmed S, Sultana S, Rahman MM, Khatun R, et al. Social Enterprise Model (SEM) for private sector tuberculosis screening and care in Bangladesh. PLoS One. 2020;15:e0241437.\u003c/li\u003e\n\u003cli\u003eJohn S, Abdulkarim S, Usman S, Rahman MT, Creswell J. Comparing tuberculosis symptom screening to chest X-ray with artificial intelligence in an active case finding campaign in Northeast Nigeria. BMC Glob Public Heal. 2023;1:17.\u003c/li\u003e\n\u003cli\u003eMuyoyeta M, Kasese NC, Milimo D, Mushanga I, Ndhlovu M, Kapata N, et al. Digital CXR with computer aided diagnosis versus symptom screen to define presumptive tuberculosis among household contacts and impact on tuberculosis diagnosis. BMC Infect Dis. 2017;17:1\u0026ndash;8.\u003c/li\u003e\n\u003cli\u003eNguyen LH, Codlin AJ, Vo LNQ, Dao T, Tran D, Forse RJ, et al. An Evaluation of Programmatic Community-Based Chest X-ray Screening for Tuberculosis in Ho Chi Minh City, Vietnam. Trop Med Infect Dis. 2020;5:185.\u003c/li\u003e\n\u003cli\u003eVo LNQ, Codlin A, Ngo TD, Dao TP, Dong TTT, Mo HTL, et al. Early Evaluation of an Ultra-Portable X-ray System for Tuberculosis Active Case Finding. Trop Med Infect Dis. 2021;6:163.\u003c/li\u003e\n\u003cli\u003eDorfman R. The Detection of Defective Members of Large Populations. Ann Math Stat. 1943;14:436\u0026ndash;40.\u003c/li\u003e\n\u003cli\u003eAbdurrahman ST, Mbanaso O, Lawson L, Oladimeji O, Blakiston M, Obasanya J, et al. Testing pooled sputum with Xpert MTB/RIF for diagnosis of pulmonary tuberculosis to increase affordability in low-income countries. J Clin Microbiol. 2015;53:2502\u0026ndash;8.\u003c/li\u003e\n\u003cli\u003eIem V, Chittamany P, Suthepmany S, Siphanthong S, Siphanthong P, Somphavong S, et al. Pooled testing of sputum with Xpert MTB/RIF and Xpert Ultra during tuberculosis active case finding campaigns in Lao People\u0026rsquo;s Democratic Republic. BMJ Glob Heal. 2022;7:1\u0026ndash;8.\u003c/li\u003e\n\u003cli\u003eBilder CR, Hitt BD, Biggerstaff BJ, Tebbs JM, Mcmahan CS. binGroup2: Statistical Tools for Infection Identification via Group Testing. 2018;XX:1\u0026ndash;17.\u003c/li\u003e\n\u003cli\u003eQin ZZ, Sander MS, Rai B, Titahong CN, Sudrungrot S, Laah SN, et al. Using artificial intelligence to read chest radiographs for tuberculosis detection: A multi-site evaluation of the diagnostic accuracy of three deep learning systems. Sci Rep. 2019;9:15000.\u003c/li\u003e\n\u003cli\u003eShazzadur Rahman AAM, Langley I, Galliez R, Kritski A, Tomeny E, Squire SB. Modelling the impact of chest X-ray and alternative triage approaches prior to seeking a tuberculosis diagnosis. BMC Infect Dis. 2019;19:1\u0026ndash;11.\u003c/li\u003e\n\u003cli\u003eWorld Health Organization. Chest radiography in tuberculosis detection \u0026ndash; summary of current WHO recommendations and guidance on programmatic approaches. Geneva, Switzerland; 2016.\u003c/li\u003e\n\u003cli\u003evan Cleeff MRA, Kivihya-Ndugga LE, Meme H, Odhiambo JA, Klatser PR. The role and performance of chest X-ray for the diagnosis of tuberculosis: A cost-effective analysis in Nairobi, Kenya. BMC Infect Dis. 2005;5:1\u0026ndash;9.\u003c/li\u003e\n\u003cli\u003eCreswell J, Qin ZZ, Gurung R, Lamichhane B, Yadav DK, Prasai MK, et al. The performance and yield of tuberculosis testing algorithms using microscopy, chest x-ray, and Xpert MTB/RIF. J Clin Tuberc Other Mycobact Dis. 2019;14 November:1\u0026ndash;6.\u003c/li\u003e\n\u003cli\u003eQin ZZ, Ahmed S, Sarker MS, Paul K, Adel ASS, Naheyan T, et al. Tuberculosis detection from chest x-rays for triaging in a high tuberculosis-burden setting: an evaluation of five artificial intelligence algorithms. Lancet Digit Heal. 2021;3:e543\u0026ndash;54.\u003c/li\u003e\n\u003cli\u003eGelaw SM, Kik S V., Ruhwald M, Ongarello S, Egzertegegne TS, Gorbacheva O, et al. Diagnostic accuracy of three computer-aided detection systems for detecting pulmonary tuberculosis on chest radiography when used for screening: Analysis of an international, multicenter migrants screening study. PLOS Glob Public Heal. 2023;3:e0000402.\u003c/li\u003e\n\u003cli\u003eCodlin AJ, Dao TP, Vo LNQ, Forse RJ, Van Truong V, Dang HM, et al. Independent evaluation of 12 artificial intelligence solutions for the detection of tuberculosis. Sci Rep. 2021;11:23895.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-global-and-public-health","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [BMC Global and Public Health](https://bmcglobalpublichealth.biomedcentral.com/)","snPcode":"44263","submissionUrl":"https://submission.springernature.com/new-submission/44263/3","title":"BMC Global and Public Health","twitterHandle":"@BMC_GPH","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"GeneXpert, Pooling, AI, CAD, X-ray, CXR, ACF, Active case finding, Tuberculosis diagnosis","lastPublishedDoi":"10.21203/rs.3.rs-3813705/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3813705/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eIn 2022, fewer than half of persons with tuberculosis (TB) have access to molecular diagnostic tests for TB due to their high costs. Studies have found that computer-aided detection using artificial intelligence (AI) for chest X-ray (CXR) and sputum specimen pooling can each reduce testing costs. We modeled the combination of both strategies to estimate potential savings in consumables that could be used to expand access to molecular diagnostics.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eWe obtained Xpert testing and positivity data segmented into deciles by AI probability scores for TB from community- and healthcare facility-based active case finding conducted in Bangladesh, Nigeria, Viet Nam and Zambia. AI scores in the model were based on CAD4TB version 7 (Zambia) and qXR (all other countries). We modeled four ordinal screening and testing approaches involving computer-aided CXR to indicate individual and pooled testing. Setting a false negative rate of 5%, for each approach we calculated additional and cumulative savings over the baseline of universal Xpert testing as well as the theoretical expansion in diagnostic coverage.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eIn each country, the optimal screening and testing approach was to use AI to rule out testing in deciles with low AI scores and guide pooled and individual testing in persons with moderate and high AI scores, respectively. This approach yielded cumulative savings in Xpert tests over baseline ranging from 50.8% in Zambia to 57.5% in Nigeria and 61.5% in Bangladesh and Viet Nam. Using these savings, diagnostic coverage theoretically could be expanded by 34\u0026ndash;160% across the different approaches and countries.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eUsing a combination of AI and CXR to inform different pooling strategies may optimize TB diagnostic test use, and could extend molecular tests to more people who need them. The optimal AI thresholds and pooled testing strategy varied across countries, which suggests that bespoke screening and testing approaches may be needed for differing populations and settings.\u003c/p\u003e","manuscriptTitle":"Expanding molecular diagnostic coverage for tuberculosis by combining computer-aided chest radiography and sputum specimen pooling: a modeling study from four high burden countries","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-01-05 18:23:45","doi":"10.21203/rs.3.rs-3813705/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2024-03-21T00:03:19+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-03-13T15:57:25+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"e02e5cae-eb6a-43a8-8fb2-54e4e5dc6ba7","date":"2024-03-08T13:13:32+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-03-06T20:32:13+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"94ffeb15-916b-4fb5-89cf-bf5809dce68e_SNPRID","date":"2024-02-22T00:18:59+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"20414413-a1fc-4083-8cc6-59bbc30005fc","date":"2024-01-15T13:28:16+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"004c0a4f-849e-49d4-b44f-65fb0fc5b8f0","date":"2024-01-12T05:22:03+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-01-10T06:10:50+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-01-03T11:56:11+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-01-03T11:40:45+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Global and Public Health","date":"2023-12-27T20:33:25+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-global-and-public-health","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"Learn more about [BMC Global and Public Health](https://bmcglobalpublichealth.biomedcentral.com/)","snPcode":"44263","submissionUrl":"https://submission.springernature.com/new-submission/44263/3","title":"BMC Global and Public Health","twitterHandle":"@BMC_GPH","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"d46d44aa-6d78-4a1a-a0ae-a84c581f0646","owner":[],"postedDate":"January 5th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2024-07-05T08:41:15+00:00","versionOfRecord":[],"versionCreatedAt":"2024-01-05 18:23:45","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-3813705","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3813705","identity":"rs-3813705","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}