A computer-aided detection (CAD)-guided chest x-ray screening strategy that referred screen-positive participants for confirmatory Xpert MTB/RIF Ultra testing detected more tuberculosis cases at a lower cost than a strategy that added point-of-care C-reactive protein (CRP) testing in a pragmatic community trial of more than 20,000 adults in Lesotho and South Africa. The CAD-plus-CRP strategy did not meet the prespecified criterion for noninferiority and cost 37% more per case detected, according to a study published in The Lancet Infectious Diseases.
The prospective TB TRIAGE+ trial used a paired screen-positive design, allowing investigators to compare the two population-based screening algorithms within the same participants and thereby increase statistical efficiency. Investigators enrolled household members aged 18 years or older from Sept. 26, 2022, through Sept. 4, 2024, in three neighborhoods in the uMgungundlovu District of South Africa and 182 villages in the Butha-Buthe District of Lesotho.
After exclusions for technical problems with an x-ray device or image acquisition and undisclosed tuberculosis treatment, the full analysis set included 20,023 participants: 6,986 in South Africa and 13,037 in Lesotho. Median age was 42 years, 62% were female, 23% had HIV, 8% reported a history of tuberculosis and 8% reported at least one of the four main tuberculosis symptoms.
Two screening strategies compared
All participants were intended to receive a digital chest x-ray analyzed with CAD4TBv7 computer-aided detection (CAD) software. In the CAD-only strategy, participants with a CAD score above a predefined threshold underwent confirmatory Xpert MTB/RIF Ultra testing.
The second strategy combined CAD with point-of-care C-reactive protein (CRP) testing. Participants with CAD scores in an intermediate prespecified range received a CRP test, followed by Xpert MTB/RIF Ultra testing when the CRP concentration exceeded 5 mg/L. Participants with CAD scores above a higher threshold proceeded directly to confirmatory testing. Positive Xpert MTB/RIF Ultra results were considered confirmatory for the primary outcome; trace results were excluded from the primary tuberculosis case definition.
The coprimary objectives were to determine whether the CAD-plus-CRP strategy was noninferior to CAD alone for tuberculosis case detection and whether it was more cost-effective. Because noninferiority was not demonstrated, the prespecified analysis called for a comparative cost analysis rather than the planned cost-effectiveness analysis.
CAD alone detected more cases
The primary outcome set included 73 participants with tuberculosis identified by at least one strategy who had complete data for both. The CAD-only algorithm identified tuberculosis in 69 of 73 participants, or 94.5%, compared with 60 of 73, or 82.2%, using CAD plus CRP. The difference for CAD plus CRP versus CAD alone was –12.3% (95% CI, –23.0 to –1.6). Thirteen participants with tuberculosis were identified exclusively by CAD alone, compared with four by CAD plus CRP.
The combined strategy did not meet the study’s prespecified 10% noninferiority criterion because the lower bound of the 95% CI extended below the –10% margin. In the discussion, the investigators reported that CAD alone identified 15% more participants with active tuberculosis under trial conditions and 20% more when participants with data from only one approach were included.
Preplanned sensitivity analyses produced consistent findings when investigators incorporated Xpert MTB/RIF Ultra trace results into different tuberculosis case definitions, used alternative CRP thresholds of 7 mg/L and 10 mg/L and addressed participants affected by recalibration of the CAD thresholds.
There was no noticeable effect modification by country, HIV status, age, or tuberculosis symptoms. Investigators reported some indication that CAD plus CRP might perform better than CAD alone among participants with a history of tuberculosis, but the number of observations in that subgroup was small, and the authors cautioned against overinterpreting the finding.
Adding CRP increased costs
The programmatic cost analysis included capital and recurrent screening costs but excluded research-related costs and costs associated with HIV or chronic disease care.
The estimated average cost per tuberculosis case detected was $5,454 with CAD alone and $7,486 with CAD plus CRP, making the combined strategy 37% more expensive. Total screening costs were $392,695 and $449,175, respectively, while costs per participant screened were $20 and $22. Costs were reported in 2023 US dollars.
Nearly half with confirmed tuberculosis had no main symptoms
Through day 56, a total of 100 of 20,023 participants had microbiologically confirmed tuberculosis, including cases identified through enrollment and follow-up testing. Of these participants, 46% had none of the four main tuberculosis symptoms. Among 78 participants identified as having tuberculosis at enrollment through the trial strategies or additional testing, 70, or 90%, were receiving tuberculosis treatment by day 56.
Trace Xpert MTB/RIF Ultra results also contributed to subsequent case identification. Of 35 participants with trace results, 21, or 60%, subsequently had a positive Xpert or Xpert MTB/RIF Ultra test at a health facility, and all 21 were receiving treatment at day 56.
Study limitations
Limitations included site-specific readjustment of CAD thresholds because x-ray equipment affected score distributions; missing outcomes for a few participants after the screening algorithms were reconstructed; and inconsistent referral documentation, which probably led to an overestimate of participants with presumptive tuberculosis. Cost estimates reflected trial rather than routine implementation conditions, and the economic component was limited to a within-trial analysis rather than a broader model-based evaluation. In addition, downstream health care costs were included only for participants with confirmed tuberculosis.
The authors concluded that CAD alone “has the potential to be an effective and economically viable tuberculosis screening strategy in areas with high disease burden.”
Tom Decroo, PhD, received funding from Global Health EDCTP3 for the TASP and SAHRI projects; Klaus Reither, PhD, received funding from the Swiss National Science Foundation for a BRIDGE project; and Niklaus Daniel Labhardt, MD, received a travel grant for conference attendance from Gilead Sciences. All other authors declared no competing interests. The study was funded by the European and Developing Countries Clinical Trials Partnership; the authors reported that the funder had no role in study design, data collection, data analysis, data interpretation or writing of the report.