Cost-Effectiveness of Community Tuberculosis Screening in South Africa

Abstract

Rationale:

The World Health Organization recommends community-based tuberculosis active case finding using digital chest radiography with computer-aided detection (dCXR/CAD) and/or molecular diagnostics, but clinical and economic outcomes are unclear.

Objective:

To evaluate the cost-effectiveness of community-based tuberculosis screening strategies in South Africa.

Methods:

Using a microsimulation model, we evaluated three symptom-agnostic screening strategies among adult people without HIV (PWoH) and people with HIV (PWH): (1) No Screening; (2) sputum Xpert Ultra (Xpert); and (3) dCXR/CAD followed by confirmatory sputum Xpert (dCXR+Xpert). Base case tuberculosis prevalence was 0.64%-1.23%. Sensitivity/specificity/cost for dCXR/CAD were 77-90%/65-73%/$3.55; for Xpert Ultra, they were 69-91%/98-99%/$15.24. Model outcomes included life-years, costs, and incremental cost-effectiveness ratios (ICERs) (<$3,000/year-of-life saved [YLS] considered cost-effective). We conducted sensitivity analysis around key parameters, including test sensitivity, specificity, and cost.

Measurements and Main Results:

In the base case, Xpert identifies the most individuals with tuberculosis but produces the most false-positives and highest costs. Compared to Xpert, dCXR+Xpert identifies ~13% fewer individuals with tuberculosis while decreasing screening costs by ~45%. Given base case performance characteristics, at the lifetime horizon, dCXR+Xpert is cost-effective (ICER $610/YLS) while Xpert is not (ICER $3,460/YLS). dCXR+Xpert remains cost-effective relative to No Screening unless tuberculosis prevalence (PWoH/PWH) is ≤0.15%/0.45%, dCXR/CAD sensitivity is (PWoH/PWH) ≤20%/10%, dCXR/CAD cost is ≥$34.00, Xpert Ultra cost is ≥$135.00, or linkage to tuberculosis care is ≤15%.

Conclusion:

Digital chest radiography with computer-aided detection followed by confirmatory sputum Xpert Ultra would likely be a cost-effective strategy for tuberculosis screening in South Africa.

INTRODUCTION

Tuberculosis (TB) is the leading cause of death in South Africa, where TB prevalence in 2018 was 852 per 100,000 people.^1,2 High HIV prevalence in South Africa complicates TB control, as people with HIV (PWH) are more susceptible to TB disease and death, and several TB diagnostic tools perform worse in PWH than in people without HIV (PWoH).^3-7

Because many people with TB are asymptomatic or minimally symptomatic, TB can be challenging to detect.⁸ A South Africa national survey reported that over half of people with microbiologically-confirmed TB were asymptomatic, and in nationally representative surveys across high-incidence countries, over 80% of individuals with TB lacked persistent cough.^9,10 Although TB diagnostics have improved, limitations remain. Xpert Ultra, a molecular assay, is a recommended first-line test but is associated with logistical burdens, delay to result delivery, and cost, especially in rural and under-served areas.¹¹ The World Health Organization (WHO) suggests community-wide systematic screening using a sensitive tool such as chest radiography (CXR) followed by a more specific, confirmatory test (e.g., Xpert Ultra) – such a strategy might decrease the number of Xpert Ultra tests needed to be performed for screening and therefore decrease costs. Computer-aided detection (CAD) is an alternative to human interpretation of digital CXR (dCXR), helping alleviate personnel and financial burdens.^12,13

However, the cost-effectiveness of these tools for TB screening is unclear. Our objective was to project the clinical and economic outcomes and the cost-effectiveness of community-based TB screening using Xpert Ultra or a combination of dCXR/CAD and confirmatory Xpert Ultra in South Africa. Some results were previously reported in an abstract.¹⁴

METHODS

Analytic overview

We used the Cost-Effectiveness of Preventing AIDS Complications (CEPAC)-International microsimulation to project clinical and economic outcomes of community-based TB screening strategies for adults in South Africa.^15-18 We calculated screening strategy results and treatment referrals outside the CEPAC model (Supplement S1.1, Figures S1-S3) and then simulated TB treatment outcomes and other clinical and economic outcomes in CEPAC (Supplement S1.2). The simulated population comprised adults not presenting for clinical evaluation of symptoms suggestive of TB. We separately evaluated PWoH and PWH, conducting simulations of one million people in each group. Model results were combined using the ratio of adults in South Africa without and with HIV.¹⁹

Screening strategies

We modeled three TB screening strategies in our base case analysis: (1) No Screening, where there is no background screening and only passive case finding, reflecting common practice in South Africa; (2) sputum Xpert Ultra (Xpert), where people with a positive result are referred to TB treatment; (3) a combined strategy (dCXR+Xpert), where people first undergo dCXR/CAD, those with a positive result above a triaging threshold then undergo a confirmatory sputum Xpert Ultra test, and a positive Xpert Ultra result prompts treatment referral (Supplement S1.1, Figures S1-S3).

Outcomes of interest

We examined several clinical and economic outcomes at five years and over a lifetime. These included TB screening accuracy (true-positives, false-positives, true-negatives, false-negatives), TB treatment initiations from screening, TB deaths, all-cause deaths, screening costs, and costs of TB and HIV care from the healthcare sector perspective. We evaluated cost-effectiveness through the incremental cost-effectiveness ratio (ICER), the difference in costs divided by the difference in life-years between one strategy and the next least costly strategy.

Model overview

CEPAC is a validated Monte Carlo microsimulation model that includes detailed structure and parameters around TB and HIV natural history, diagnosis, treatment, and associated costs (Supplement S1.2-1.4).^16,18 Simulated individuals transition monthly between states of TB infection and disease and face mortality risks from TB, non-TB causes, and, for PWH, other opportunistic diseases (Figure S4).^15-17,20,21

Input parameters

Cohort characteristics

We applied cohort characteristics based on the Vukuzazi study in KwaZulu-Natal, South Africa (Table 1).²² Among 17,118 participants, 6,796 (34.2%) were PWH. The mean (standard deviation) age was 40.3 (13.1) years. Screening by symptoms, dCXR, and Xpert in Vukuzazi found a pulmonary TB prevalence of 0.64% among PWoH and 1.23% among PWH (99% of those enrolled in the study underwent TB testing) (Table 1, Supplement S1.3).^9,22,23 We weighted and combined model results for PWoH and PWH based on an estimated adult HIV prevalence of 17.1% in South Africa overall.¹⁹

Table 1. Model input parameters.

Parameter	Base case	Range	References
Cohort characteristics
Age, mean [SD], years	40.3 [13.1]		*
Males/Females, %
Among PWoH	32.1/67.9		22, *
Among PWH	23.1/76.9		22, *
HIV prevalence among simulated adult cohort, %	17.1		19
CD4 count among all PWH, mean [SD]	715 [339]		*
TB prevalence
TB prevalence among PWoH, %	0.64	0.51-0.76	9
TB prevalence among PWH, %	1.23	0.91-1.54	9,23
CD4 ≤ 200 cells/μL	4.14	3.05-5.23	9,23
CD4 200-350 cells/μL	1.83	1.40-2.31	9,23
CD4 351-500 cells/μL	1.34	0.99-1.70	9,23
CD4 > 500 cells/μL	0.95	0.70-1.20	9,23
Prevalence of MDR-TB among people with TB, %	3		64
TB natural history
Monthly TB infection incidence, %
Ages 18 to 25 years	0.55		26
Ages 26 to 45 years	0.79		26
Ages 46+ years	0.68		26
Monthly probability of progression to TB disease, %
PWoH	0.05		27
PWH (CD4-dependent)	0.05-0.69		27
Monthly probability of symptom development among people with TB disease, %
PWoH	8.2	6.8-9.9	28
PWH	22.7	15.1-33.0	28
Monthly probability of presenting for testing among symptomatic people with TB disease, %	32.1	0.25x-2x	65
Performance characteristics of screening tests
Proportion of people able to provide sputum, %	90	50-100	24
Screening test sensitivity/specificity, %
dCXR/CAD in PWoH	90/73	Sensitivity 80-96 Specificity 60-82	5
dCXR/CAD in PWH	77/65	Sensitivity 60-90 Specificity 55-77	4,5
Xpert Ultra in PWoH	91/99	Sensitivity 57-95 Specificity 91-99	6
Xpert Ultra in PWH	69/98	Sensitivity 57-95 Specificity 91-99	7
TB treatment characteristics
Linkage to TB treatment after TB diagnosis, %	75	55-85	25,Assumption
Monthly disengagement from care among people on TB treatment, %	3.6	0.5x-2x	29
Probability of treatment success (cure) among people who complete TB treatment regimen, %
ds-TB	95		66,67
MDR-TB	95		66,68
Mortality
Monthly probability of death from untreated TB among people with TB disease, %
Among people with prevalent TB disease at time of screening			54,55,69
PWoH	1.55	0.78 – 3.10
PWH	2.86	1.43 – 5.72
Among people with incident TB disease after initial screening			54,55,69
PWoH	1.80	0.90 – 3.60
PWH	3.32	1.66 – 6.64
Monthly probability of death from causes other than TB or HIV, % †	0.003-0.9		Derived from 70,71
Costs (2021 USD)
TB screening
dCXR/CAD	3.55	1.00-50.00	36,72
Xpert Ultra	15.24	10.00-75.00	36,72
TB treatment
ds-TB, per month	7.40		18,37
MDR-TB, per month	44.46		73
ART
First-line ART, per month	4.50		38
Second-line ART, per month	23.30		38

Abbreviations: SD, standard deviation; PWoH, people without HIV; PWH, people with HIV; ART, antiretroviral therapy; TB, tuberculosis; MDR-TB, multidrug-resistant tuberculosis; dCXR/CAD, digital chest radiography with computer-aided detection; ds-TB, drug-susceptible tuberculosis; USD, US dollars.

^*Personal communication, Dr. Mark Siedner.

^†These probabilities are age- and sex-dependent.

TB screening strategy performance

In the base case, sensitivity and specificity of dCXR/CAD are 90% and 73% among PWoH based on CAD4TB version 7 with threshold score 0.50.⁵ The corresponding values for PWH are 77% and 65%, derived from the reported relative performance of CAD4TB version 6 in PWH versus PWoH and applied to version 7.^4,5 Xpert Ultra sensitivity and specificity are 91% and 99% among PWoH, and 69% and 98% among PWH.^6,7 In the base case, 90% of people provide a sputum specimen for testing.²⁴ Among those with a positive screening result, 75% link to TB care and initiate treatment.²⁵

TB natural history and treatment

In all strategies, TB can be diagnosed later via passive case finding if simulated individuals develop symptoms. The monthly, age-stratified TB infection incidence is 0.55-0.79%, and the monthly probability of progression of TB infection to TB disease is 0.05-0.12%, dependent on HIV status and CD4 count.^26,27 After screening, the monthly probability of symptom development among people with TB disease who do not initiate treatment is 8.2% for PWoH and 22.7% for PWH.²⁸ Among symptomatic individuals, each month 32.1% present for testing, which is conducted using Xpert Ultra regardless of the initial screening strategy.⁷ People who initiate TB treatment, either from screening or through passive case finding, can disengage from treatment (3.6% probability per month) (Supplement S1.3).²⁹ People on TB treatment can experience toxicity (Table S1).^30,31 People with incident TB disease who do not receive treatment or receive inadequate treatment (e.g., treatment for drug-susceptible TB even though they have multi-drug resistant TB [MDR-TB]) have a monthly probability of death from TB: 1.80% for PWoH and 3.32% for PWH (Supplement S1.4).³² This probability is lower, 1.55% for PWoH and 2.86% for PWH, among individuals with prevalent TB disease at the time of screening, reflecting that these individuals are less symptomatic and likely have less severe disease. We validated mortality projections from these values and stratifications against published estimates in South Africa (Supplement S1.4).^2,33-35 Additional details are in Supplement S1.3.

HIV treatment

At model start, 82.3% of PWH are on antiretroviral therapy (ART) (Table 1). Mean (standard deviation) CD4 cell count is 715/μL (339/μL). They have a monthly probability of disengagement from HIV care which leads to non-adherence to ART. People not on ART, or on ineffective ART, face a probability of virologic non-suppression, a decrease in CD4 cell count, and an increase in opportunistic disease risks (Supplement S1.4, Table S1).

Costs

In the base case, we applied the following costs (all in 2021 US dollars [USD]): dCXR/CAD $3.55, Xpert Ultra $15.24, ds-TB treatment $7.40/month, ART $4.50/month (Table 1).^18,36-38 Cost estimates for dCXR/CAD and Xpert Ultra were based on a bottom-up ingredient approach where each individual test cost was determined by summing component costs, including labor, equipment, and consumables.³⁶ Scale-up costs of screening or treatment were not directly evaluated, but we performed sensitivity analysis around per-person costs.

Cost-effectiveness and budget impact analysis

We determined the ICER for each screening strategy using cost per year-of-life saved (YLS) at lifetime horizon. We considered an ICER below $3,000/YLS—approximating an opportunity cost-based threshold in South Africa—to be cost-effective.³⁹ We also conducted a five-year budget impact analysis where we considered the total TB and HIV care costs of performing each screening strategy among one million people. The calculation of ICERs used life-years and costs discounted 3% annually, whereas the budget impact analysis used undiscounted costs.

Sensitivity and scenario analysis

We conducted one-way sensitivity analyses comparing dCXR+Xpert to No Screening to evaluate the robustness of our results across variations in key parameters. The parameters varied included TB prevalence (0.51%-5.23%), sensitivity/specificity of dCXR/CAD (80%/60% to 96%/82% in PWoH; 60%/55% to 90%/77% in PWH), cost of dCXR/CAD per person ($1.00-$50.00), sensitivity/specificity of Xpert Ultra (57%/95% to 91%/99%), cost of Xpert Ultra per person ($10.00-$75.00), sputum provision (50-100%), linkage to TB treatment after screening (55%-85%), disengagement from TB treatment after screening (1.8%-7.2%/month), death due to untreated TB (1.66%-6.64%/month [0.5x-2x base case]), symptom development among people with TB disease (6.8%-33.0%/month), and presentation for testing among symptomatic people (8.0%-64.2%/month).^{4-7,9,23,28,36} The ranges of values were informed by published confidence intervals when available (Table 1). Given the possibility of higher Xpert Ultra specificity in community screening compared to clinic-based testing, we also assessed a specificity value of 99.4% during screening.⁴⁰ Where there were no changes in cost-effectiveness interpretations across the one-way sensitivity analysis ranges, we identified threshold values (outside of the ranges) where interpretations would change. We also conducted a sensitivity analysis evaluating the impact of incorporating non-HIV healthcare costs in our model analysis (Supplement S1.6).

Additionally, to account for existing parameter uncertainty, especially as new imaging and CAD technologies emerge, and simultaneous variations in sensitivity, specificity, and cost of dCXR/CAD, we conducted a multiway sensitivity analysis.^4,12,13 We sought to identify thresholds for dCXR/CAD performance (abnormality thresholds) and cost at which dCXR+Xpert would be cost-effective compared to No Screening. We varied the sensitivity (80-96% in PWoH, 60-90% in PWH) and specificity (60%-82% in PWoH, 55%-77% in PWH) of the dCXR/CAD component of dCXR+Xpert alongside different possible costs for dCXR/CAD ($3.55 [base case], $20.00, $30.00, and $40.00).

We also performed four alternate scenario analyses. One scenario includes empiric treatment in the dCXR+Xpert strategy. Specifically, 5% of those with positive dCXR/CAD and negative Xpert Ultra would be referred for empiric TB treatment, and 10% of those with positive dCXR and no Xpert Ultra result (due to sputum non-provision) would be referred for empiric TB treatment. The second scenario accounts for Xpert Ultra “trace call” results to reflect the proportion of Xpert Ultra “positive” cases that may be due to a “trace call” on testing; in this scenario, 10% of positive Xpert Ultra results are redefined as negative. In the third scenario, these 10% undergo repeat Xpert Ultra testing. The fourth scenario includes a theoretical dCXR/CAD-only strategy (dCXR), wherein individuals with a positive dCXR/CAD result are referred for empiric TB treatment without confirmatory testing. While it might not be appropriate to implement such a strategy, we include it here, as a counterfactual comparator, to help understand what an efficiency-driven analysis might recommend.

RESULTS

Base case

Screening results and costs

Model results are presented for one million simulated individuals (82.9% PWoH, 17.1% PWH, Table 2). Xpert would correctly identify 5,613 individuals with TB (true-positives) while dCXR+Xpert would correctly identify 4,882 (13% fewer than Xpert). However, Xpert would yield 11,936 false-positive results, whereas dCXR+Xpert would produce 3,466 false-positives. The ratio of false-positives to true-positives for Xpert and dCXR+Xpert would be approximately 2:1 and 0.7:1. dCXR+Xpert would correctly rule-out TB disease in more individuals than would Xpert (true-negatives, 989,161 versus 980,691) (Figure S5). The number of individuals initiating TB treatment (the sum of true-positive and false-positive individuals linking to care) would be 13,162 with Xpert and 6,261 with dCXR+Xpert. Costs of screening would be $13.7 million for Xpert and $7.5 million for dCXR+Xpert (45% lower than Xpert).

Table 2. Model-generated clinical and economic outcomes of tuberculosis screening strategies for one million simulated people in South Africa, over five years.

Outcomes	No Screening	Xpert	dCXR+Xpert
Screening results
True-positive	-	5,613	4,882
False-positive	-	11,936	3,466
True-negative	-	980,691	989,161
False-negative	-	1,760	2,491
False-positive: true-positive ratio, without empiric treatment	-	2:1	0.7:1
Treatment
Number of people starting TB treatment due to screening*	0	13,162	6,261
Mortality
TB deaths	4,123	3,330	3,437
Deaths from toxicity of TB treatment	9	16	13
All-cause deaths	51,099	50,356	50,452
Costs, 2021 USD
Cost of TB screening	-	$13.7 million	$7.5 million
Other TB- and HIV-related healthcare costs	$170 million	$171 million	$170 million
Total costs	$170 million	$185 million	$178 million

Abbreviations: TP, true-positive; FP, false-positive; TN, true-negative; FN, false-negative; TB, tuberculosis; USD, US dollars.

^*75% of people with a positive screening result will successfully link to TB care and start treatment.

Five-year clinical outcomes

Xpert would produce the fewest TB deaths (3,330) after five years, followed by dCXR+Xpert (3,437) and No Screening (4,123) (Table 2). Deaths due to toxicity from TB treatment would also be highest with Xpert (16). All-cause deaths would be lowest in Xpert (50,356), followed by dCXR+Xpert (50,452) and No Screening (51,099). Results stratified between PWoH and PWH are in Tables S2-3.

Five-year cost outcomes

When examining total TB and HIV care costs over five years (i.e., passive case finding for TB, ART, hospitalizations), No Screening, Xpert, and dCXR+Xpert would produce costs of $170 million, $185 million, and $178 million (Table 2). Results stratified between PWoH and PWH are in Tables S2-3.

Cost-effectiveness analysis

At the lifetime horizon, dCXR+Xpert would be cost-effective relative to No Screening (ICER $610/YLS, Table 3). Xpert would not be cost-effective relative to dCXR+Xpert, with its ICER of $3,460/YLS falling above the cost-effectiveness threshold of $3,000/YLS. Results stratified between PWoH and PWH are in Table S4. dCXR+Xpert would be cost-effective in both groups, while Xpert would be cost-effective only in PWH.

Table 3. Clinical and cost outcomes and cost-effectiveness of community-based tuberculosis screening strategies in South Africa.

Strategy*	Life-months per person, discounted† (undiscounted)	Costs per person, USD, discounted† (undiscounted)	ICER ($/YLS)‡
No Screening	233.11 (402.94)	552 (878)	-
dCXR+Xpert	233.26 (403.20)	560 (886)	610
Xpert	233.28 (403.25)	566 (893)	3,460

Abbreviations: USD, US dollars (2021); ICER, incremental cost-effectiveness ratio; YLS, year-of-life saved.

^*Order of strategies is based on ascending costs.

^†Outcomes discounted 3% per year.

^‡ICER is calculated based on precise (unrounded) values of per person average discounted life-years and discounted costs, but for visualization, rounded per person life-years and costs are shown here. ICERs are calculated compared to the next least expensive strategy.

Budget impact analysis

For a simulated population of one million, after 5 years most costs would be attributable to HIV management costs (including ART and treatment of opportunistic diseases), which would account for $178 million in each simulated strategy (Figure S6). Initial TB screening costs would be $13.7 million for Xpert and $7.5 million for dCXR+Xpert. TB management costs after screening (including TB diagnostics, TB treatment, and management of treatment toxicities) would account for approximately $3 million across all three strategies.

Sensitivity and scenario analysis

One-way sensitivity analysis

We determined the ICER of dCXR+Xpert relative to No Screening for each parameter change (Table 4). Relative to No Screening, dCXR+Xpert remains cost-effective across all parameter ranges tested in one-way sensitivity analyses, except for the upper value of the range for CXR/CAD cost. Increasing Xpert Ultra specificity during community screening to 99.4% has little impact on model projections of life expectancy, costs, and cost-effectiveness. In the sensitivity analysis that incorporates non-HIV healthcare costs, dCXR+Xpert remains cost-effective relative to No Screening (ICER $1,480/YLS) (Supplement S2.1).

Table 4. One-way sensitivity analysis around key model parameters: incremental cost-effectiveness ratio of dCXR+Xpert relative to No Screening, over a lifetime horizon.

Parameter*	Lower bound of parameter range					Upper bound of parameter range
	No Screening		dCXR+Xpert		ICER†	No Screening		dCXR+Xpert		ICER ($/YLS)†
	Life- months per person	Costs per person, USD	Life- months per person	Costs per person , USD		Life- months per person	Costs per person, USD	Life- months per person	Costs per person , USD
dCXR with CAD sensitivity/specificity (0.90/0.77: 0.80/0.60 - 0.96/0.90)‡	233.104	552	233.238	562	840	233.104	552	233.273	559	460
Xpert Ultra sensitivity/specificity (0.91,0.69: 0.57/0.91 - 0.95/0.99)‖	232.851	552	232.930	563	1,630	233.152	552	233.264	560	840
TB prevalence (0.0064: 0.005 - 0.008)¶	233.204	552	233.322	560	780	233.015	552	233.197	560	510
Proportion able to provide sputum (0.90: 0.50 - 1.0)	232.997	552	233.087	558	790	233.123	552	233.289	560	590
Probability of linkage to TB treatment (0.75: 0.55 - 0.85)	233.108	552	233.218	560	830	233.108	552	233.280	560	540
Probability of disengagement from TB treatment (0.036: 0.018 - 0.072)	233.138	552	233.294	560	590	233.048	552	233.189	560	650
Probability of symptom development (0.082: 0.068 - 0.099)+	232.954	552	233.124	560	550	233.239	552	233.376	560	670
Probability of presenting for TB testing among people with symptoms (0.25x - 2x of base case value)	232.276	551	233.522	558	390	233.312	553	233.440	560	720
Mortality from untreated TB (0.5x - 2x of base case value)	233.242	554	233.382	562	650	232.925	550	233.092	558	570
dCXR/CAD cost (3.55: 1.00 -50.00)	233.104	552	233.260	557	400	233.104	552	233.260	606	4,190
Xpert Ultra cost (15.24: 10.00 -75.00)	233.108	552	232.260	558	510	233.108	554	233.260	577	1,820

Abbreviations: USD, US dollars (2021); ICER, incremental cost-effectiveness ratio; YLS, year-of-life saved.; TB, tuberculosis; dCXR: digital chest radiography; CAD: computer-aided detection.

^*Parameter varied in sensitivity analysis (base case value for PWoH: lower bound of the parameter range used in sensitivity analysis for PWoH - upper bound of the parameter range used in sensitivity analysis for PWoH). Parameters ranges for PWH are in footnotes below and follow the same formatting.

^†ICER is calculated based on precise (unrounded) value of per person average discounted life-years and discounted costs, but for visualization purposes, per person life-years with three decimal places are shown here.

^‡PWH parameter range: 0.73/0.65: 0.60/0.55 - 0.82/0.77

^‖PWH parameter range: 0.68/0.98: 0.57/0.91 - 0.95/0.99

^¶PWH parameter range: 0.010-0.041: 0.007-0.031 - 0.012-0.052 (all values are stratified by CD4 count)

⁺PWH parameter range: 0.227 (0.151 - 0.33)

Threshold analysis

dCXR+Xpert is no longer cost-effective relative to No Screening if: TB prevalence (PWoH/PWH) is ≤0.15%/0.45%; dCXR/CAD sensitivity (PWoH/PWH) is ≤20%/10%; dCXR/CAD cost is ≥$34.00; Xpert Ultra sensitivity (PWoH/PWH) is ≤8%/11%; Xpert Ultra cost is ≥$135.00; linkage to TB care is ≤15%; or disengagement from TB care is ≥63%/month. Holding other parameters at base case values, Xpert would become cost-effective relative to dCXR+Xpert when dCXR/CAD cost is ≥$4.44.

Multiway sensitivity analysis

Across a range of values for dCXR/CAD performance characteristics and costs, dCXR+Xpert is generally cost-effective relative to No Screening. At higher dCXR/CAD cost (≥$30.00) and lower dCXR/CAD sensitivity (≤80% for PWoH and ≤60% for PWH), dCXR+Xpert is no longer cost-effective relative to No Screening. When dCXR/CAD cost is ≥$40.00, dCXR+Xpert is not cost-effective relative to No Screening at any of the sensitivity/specificity combinations evaluated. Figure 1 depicts results across combinations of dCXR/CAD sensitivity, specificity, and cost.

Figure 1. Multiway sensitivity analysis varying sensitivity, specificity, and cost of digital chest radiography with computer-aided detection: incremental cost-effectiveness ratio (ICER) of dCXR+Xpert relative to No Screening.

These are results from a multiway sensitivity analysis of the cost-effectiveness of the dCXR+Xpert screening strategy versus No Screening . Each row of heat maps represents the different cost scenarios for dCXR/CAD ($1.00, $3.55 [base case], $5.00, $10.00, $20.00, $30.00, and $40.00). The columns within each map represent different dCXR/CAD sensitivity levels, and the rows show different dCXR/CAD specificity levels. The sensitivity values for PWoH range from 80% to 96% while for PWH they range from 60% to 90%. The specificity values for PWoH range from 60% to 82% while for PWH they range from 55% to 77%. Red cells represent scenarios where dCXR+Xpert is not cost-effective, meaning that the ICER calculated exceeds a cost-effectiveness threshold of $3,000/YLS. Orange cells represent scenarios where dCXR+Xpert is cost-effective, meaning the ICER falls below a cost-effectiveness threshold of $3,000/YLS. Abbreviations: PWoH, people without HIV; PWH, people with HIV; ICER, incremental cost-effectiveness ratio; YLS, year-of-life-saved.

Scenario analysis

In the scenario with empiric treatment, dCXR+Xpert is cost-effective relative to No Screening (ICER $660/YLS). In the dCXR+Xpert strategy for this scenario, 21,677 individuals would initiate TB treatment after screening, 15,416 more than in the base case (Supplement S2.2).

In the scenario with Xpert Ultra “trace calls”, dCXR+Xpert is cost-effective relative to No Screening (ICER $690/YLS). In this scenario, fewer people in the dCXR+Xpert strategy initiate TB treatment than in the base case (5,635 compared to 6,261 individuals), and the ratio of false-positives to true-positives remains similar (0.7:1) (Supplement S2.3). Results for the scenario with repeat Xpert Ultra testing of “trace calls” and the scenario that includes the dCXR strategy are in Supplement S2.4-S2.5 and Table S5.

DISCUSSION

In our model analysis of community-based active screening strategies for TB among adults in South Africa, we found that, on a per-person basis, screening with dCXR/CAD followed by confirmatory sputum Xpert Ultra for those with positive dCXR/CAD (dCXR+Xpert) would be cost-effective, yielding fewer deaths and more life-years compared with no screening while maintaining an ICER below an opportunity cost-based threshold. dCXR+Xpert remains generally cost-effective in sensitivity analysis around dCXR/CAD performance characteristics and costs and other key parameters; it would no longer be cost-effective if dCXR/CAD sensitivity (PWoH/PWH) is ≤20%/10% or Xpert Ultra sensitivity (PWoH/PWH) is ≤8%/11%, or if dCXR/CAD costs ≥$34.00 or Xpert Ultra costs ≥$135.00. Screening all with Xpert Ultra (Xpert) would produce even fewer deaths and more life-years but at substantially higher costs than dCXR+Xpert, and it would not be cost-effective in our base case analysis.

These findings align with existing literature demonstrating the utility of community-based TB screening in high-burden settings.^16,41-44 Previous analyses suggest that using easier-to-deploy, sensitive screening tools, such as dCXR with CAD, followed by a more specific confirmatory test, such as Xpert Ultra, might facilitate diagnosis and treatment of more of the 3 million people with undetected TB worldwide in a cost-effective manner.⁴⁵ Our analysis expands on this premise by comparing a more specific, laboratory-based screening strategy (Xpert) and a combined strategy (dCXR+Xpert) against the current common practice of no active case finding. Although dCXR+Xpert identifies 13% fewer individuals with TB compared to Xpert, it decreases screening costs by 45%, largely by decreasing the number of Xpert Ultra tests performed. New technologies, such as molecular testing of tongue swabs, are on the horizon and might be relatively low-cost.⁴⁶ Once data are available about their performance in community screening, cost-effectiveness analysis can be performed.

TB prevalence varies substantially across regions in South Africa.⁴⁷ Considering local resource constraints, focusing community-based TB screening efforts on specific geographic areas may provide a feasible alternative to large-scale, difficult-to-implement screening initiatives. Given our base case performance characteristics of dCXR/CAD (sensitivity [PWoH/PWH] 90%/77% and specificity [PWoH/PWH] 73%/65%), and an estimated dCXR/CAD cost of $3.55 per person, a combined screening strategy using dCXR/CAD followed by confirmatory Xpert Ultra could be cost-effective even if TB prevalence is as low as 0.15%/0.45% (PWoH/PWH). Of note, this threshold prevalence is greatly exceeded among incarcerated populations in South Africa, in whom TB prevalence is substantially higher than in the general community.^48,49 Furthermore, appropriately linking those identified with TB disease to appropriate treatment and retaining them in treatment is critical and may be more challenging for people with asymptomatic or minimally symptomatic TB disease compared to those with symptomatic TB disease who present for testing.

Although there may remain uncertainty in performance characteristics or cost of dCXR/CAD in community settings in South Africa, our multiway sensitivity analyses demonstrate that if dCXR/CAD sensitivity (PWoH/PWH) is >80%/60%, specificity (PWoH/PWH) is >75%/70%, and cost ≤$30.00, then dCXR+Xpert would remain cost-effective. Additionally, ongoing advances in imaging and CAD could make dCXR+Xpert even more appealing for implementation in South Africa.^12,13 When more performance data of dCXR/CAD and Xpert Ultra at different risk thresholds become available, cost-effectiveness analyses of risk-based screening strategies can be performed.

We included incident TB disease in our model analysis, but we did not directly account for the potential of widespread screening to decrease TB incidence. There is likely a risk of transmission from people with asymptomatic TB, but the magnitude of that risk is unclear.^50,51 As such, we did not model the impact of earlier diagnosis and treatment on the onward effect of TB transmissions. Cost-effectiveness of one-time screening would likely improve if screening reduces transmissions and incident TB in the community, as our analysis does not capture these benefits.^52,53 Accordingly, our analysis might substantially underestimate the public health benefits of screening, and these model results may be interpreted as conservative estimates of the impact of more widespread screening.

This analysis has limitations. As with all model-based analyses, there is uncertainty in our model input parameters and assumptions. For example, the risk of mortality from untreated TB (i.e., the consequences of a false-negative TB screen in a person who is asymptomatic or minimally symptomatic) is challenging to define; previous modeling studies have attempted to estimate this based on historical data.^54-56 However, our model projections of TB deaths align well with South Africa estimates from the WHO and the Thembisa project.^2,35 Given data limitations, we did not differentiate between the prevalence of TB in people who are able to produce sputum for testing versus those who do not, although our base case analysis reflects sputum non-production by only 10% of individuals; given this small proportion amongst a screening population, there is unlikely to be meaningful impact on cost-effectiveness interpretations. We did not stratify the performance characteristics of the various screening modalities by state of HIV suppression, given data limitations for dCXR/CAD and conflicting data about CD4-stratified performance of Xpert.^57-59 Modeled cohort characteristics were primarily derived from the Vukuzazi cohort in KwaZulu-Natal and therefore may not be as broadly applicable to the general South African population. We did not account for the potential impact of earlier diagnosis and treatment on reducing post-TB lung disease given uncertainties. Prior work found that most of the disability-adjusted life-years lost from TB are due to mortality.^60-63 This was not a full economic analysis, and we did not include all scale-up costs of dCXR/CAD or Xpert Ultra in these screening programs, but our sensitivity analysis around the costs of dCXR/CAD and Xpert Ultra indirectly accounts for alternative scale-up costs on a per-person basis. The base case cost estimates used for dCXR/CAD and Xpert Ultra are from the same source and include labor, equipment, and consumables.³⁶ It would take both financial buy-in and community engagement efforts to implement widespread screening.

In summary, our model-based analysis demonstrates how an active screening strategy involving dCXR/CAD followed by confirmatory Xpert Ultra could increase the number of people referred for TB treatment, reduce TB deaths, and be cost-effective in South Africa, where there is high TB/HIV burden. This combined strategy is especially appealing in regions with testing and treatment resource constraints. Technologic advances in dCXR/CAD could render screening even more clinically effective and efficient, but appropriate considerations of scale-up and community engagement will continue to be important to achieve the full public health benefits of screening.