2025/08/07 by Lara D Veeken, Katherine C. Horton, Rein M G J Houben · 1 voice
Computer Science · Medicine · #Machine Learning in Healthcare #Mycobacterium research and diagnosis #Tuberculosis Research and Epidemiology
paper · doi:10.1093/infdis/jiaf419
openalex publication_date 2025/08/07 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/30
We thank McCreesh and colleagues for their interest in our review and contribution showing a linear trend in specificity across prevalence in prevalence surveys and community-wide screening studies. This supports the improved specificity found in community settings compared to clinics and confirms that the so-called “spectrum-bias” identified for other diseases [1] also applies to tuberculosis. One challenge in examining trends across prevalence comes from the various options of numerators and denominators in prevalence surveys. To our understanding, McCreesh et al defined prevalence as bacteriologically confirmed tuberculosis in the general population and in individuals with at least 1 available sputum specimen. We provide an alternative approach focusing on culture-positive prevalence in screen-positive individuals (with symptoms and/or chest radiographic abnormalities) to standardize metrics across all surveys and align prevalence estimates with the population in which specificity is assessed, as previously shown in the Supplementary Methods and Supplementary table 6 [2]. In line with results from McCreesh et al, this approach shows the strong negative association between GeneXpert specificity and tuberculosis prevalence. Our figure highlights decreasing specificity as culture-positive tuberculosis prevalence increases from unenriched populations in community-wide screening studies (prevalence 324 to 394 per 100 000) to enriched screened-positive individuals in prevalence surveys (prevalence 748 to 4481 per 100 000).