A Multidisciplinary-Validated Radiographic Severity Score for Immune Stratification in HIV-Associated Tuberculosis: Development and Internal Validation
Abstract
Background: Human immunodeficiency virus (HIV)-associated tuberculosis (TB) remains the leading cause of death among people living with HIV in sub-Saharan Africa. Where CD4 testing is unavailable, a chest X-ray must substitute for flow cytometry. We developed and internally validated a Radiographic Severity Score (RSS) to identify HIV-positive patients with advanced immunosuppression using plain-film radiography.
Methods: Cross-sectional diagnostic accuracy study of 242 HIV-positive adults with newly diagnosed pulmonary TB at two sites in southern Nigeria (2017–2018). Two radiologists independently scored films using the RSS (12 zones, weighted for severity). Discriminative accuracy, calibration, and inter-reader reliability were assessed.
Results: Of 242 enrolled, 228 had complete data. Mean RSS was 2.2±1.5 (CD4 <100 cells/µL), 3.4±1.7 (CD4 100–199 cells/µL), 4.8±1.9 (CD4 200–349 cells/µL), and 6.1±1.8 (CD4≥350 cells/µL). The stepwise gradient was significant (Spearman rho =+0.72, p <0.001). At threshold RSS ≤4: AUC=0.84 (95% CI 0.76–0.92), sensitivity 68.3%, specificity 80.0%, PPV 79.5%, NPV 69.4%. Inter-reader agreement was almost perfect (Cohen kappa =0.81, ICC=0.92).
Conclusion: The RSS is a simple bedside tool for identifying HIV-positive patients with CD4 <200 cells/µL in settings without same-day CD4 testing. In this Nigerian cohort, the RSS demonstrated strong discriminative accuracy (AUC 0.84) and almost perfect inter-reader reliability (kappa 0.81). However, external validation in independent cohorts is required before the RSS can be recommended for widespread implementation in TB-HIV integrated programmes
Keywords
Download Options
Introduction
HIV-associated tuberculosis remains the leading cause of death among people living with HIV in sub-Saharan Africa [1]. Global epidemiologic trends indicate persistent heterogeneity in TB-HIV burden across high-income and intermediate-burden settings, underscoring the need for context-specific triage tools. Despite World Health Organization (WHO) treat-all recommendations and universal CD4 enumeration at antiretroviral therapy (ART) initiation, an estimated 40% of African clinics lack same-day access to flow cytometry [2]. In these settings, CD4 <200 cells/µL remains the threshold for intensive clinical monitoring and rapid ART escalation—decisions that cannot wait for laboratory turnaround. When the laboratory fails, clinicians must rely on clinical judgment and the chest radiograph alone. Yet the radiograph is often underused for immune stratification—read for disease burden rather than immunologic status.
Existing radiographic scoring systems, such as the Chest Radiograph Reading and Recording System (CRRS), quantify overall disease burden for epidemiological surveillance but do not target immune stratification [3,4]. A patient with advanced immunosuppression may have minimal radiographic findings yet require the most urgent clinical attention—a paradox that current tools fail to capture. The biological reality is that immunosuppressed patients do not have 'fewer' radiographic findings; they have different patterns reflecting failed granulomatous architecture [5]. In Nigerian cohorts, atypical radiographic patterns are significantly more common among HIV-infected patients with low CD4 counts, and chest radiograph findings correlate with immunologic status [6,7].
We developed the Radiographic Severity Score (RSS) in collaboration with radiologists and clinical colleagues—a structured plain-film grading system that uses the chest X-ray already obtained in routine care to identify patients with CD4 <200 cells/µL. To our knowledge, the RSS represents one of the first structured radiographic scoring systems designed specifically for immune stratification rather than disease burden quantification in HIV-associated TB. This distinction—leveraging the inverse relationship between radiographic extent and immune status for bedside triage—differs from existing tools that quantify overall disease severity. This study reports the development and internal validation of the RSS in a Nigerian cohort, with the specific objectives of: (i) determining the correlation between RSS and CD4 count; (ii) establishing the diagnostic accuracy of RSS ≤4 for identifying advanced immunosuppression; and (iii) assessing inter-reader reliability among radiologists with variable experience.
Conclusion
The RSS is a simple, readily available bedside tool that identifies HIV-positive patients with CD4 <200 cells/µL using routine chest radiography alone. In this Nigerian cohort, the RSS demonstrated strong discriminative accuracy (AUC 0.84) and almost perfect inter-reader reliability (kappa 0.81), making it suitable for settings where readers have variable experience.
In settings without same-day CD4 testing, the RSS offers a practical, no-cost triage alternative to guide prioritisation of patients for enhanced monitoring and rapid ART initiation. However, external validation in independent cohorts with diverse readers, healthcare settings, and disease prevalence is required before the RSS can be recommended for widespread implementation in TB-HIV integrated programmes.
References
[1] World Health Organization. (2024). Global tuberculosis report 2024. WHO.
[2] Hamada, Y., Lujan, J., Schenkel, K., Ford, N., & Getahun, H. (2018). Sensitivity and specificity of WHO's recommended four-symptom screening rule for tuberculosis in people living with HIV: A systematic review and meta-analysis. Lancet HIV, *5*(9), e515-e523. https://doi.org/10.1016/S2352-3018(18)30146-1
[3] Syarif, A. K., et al. (2018). Validation of Chest Radiograph Reading and Recording System (CRRS) in three sites in Indonesia. KnE Life Sciences, *3*(3), 3554. https://doi.org/10.18502/kls.v3i3.3554
[4] Kebede, W., Abebe, G., Gudina, E. K., Kedir, E., Tran, T. N., & Van Rie, A. (2021). The role of chest radiography in the diagnosis of bacteriologically confirmed pulmonary tuberculosis in hospitalised Xpert MTB/RIF-negative patients. ERJ Open Research, *7*(1), 00708-2020. https://doi.org/10.1183/23120541.00708-2020
[5] Chan, A. C. K., et al. (2024). Changes in the epidemiology and clinical manifestations of HIV-associated tuberculosis in Hong Kong, 2007-2020. Hong Kong Medical Journal, *30*(4), e516833. https://doi.org/10.12809/hkmj2310683
[6] Akhigbe, R. O., et al. (2022). Chest radiograph patterns and their correlation with CD4 count in adults with human immunodeficiency virus (HIV) in Lagos University Teaching Hospital, Nigeria. Delta University Journal of Physical and Natural Sciences, *8*(1a), 45-54. https://doi.org/10.4314/dujopas.v8i1a.5
[7] Agbor, E., et al. (2025). Molecular characterization of Mycobacterium tuberculosis in HIV/AIDS patients in the Centre Region of Cameroon. International Journal of Mycobacteriology, *14*(1), 23-31. https://doi.org/10.4103/ijmy.ijmy_101_24
[8] Bossuyt, P. M., Reitsma, J. B., Bruns, D. E., et al. (2015). STARD 2015: An updated list of essential items for reporting diagnostic accuracy studies. BMJ, *351*, h5527. https://doi.org/10.1136/bmj.h5527
[9] Collins, G. S., Reitsma, J. B., Altman, D. G., & Moons, K. G. (2015). Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): The TRIPOD statement. BMJ, *351*, g7594. https://doi.org/10.1136/bmj.g7594
[10] Riley, R. D., Ensor, J., Snell, K. I. E., et al. (2020). Calculating the sample size required for developing a clinical prediction model. BMJ, *368*, m441. https://doi.org/10.1136/bmj.m441
[11] Dhana, A., et al. (2021). Tuberculosis screening among ambulatory people living with HIV: A systematic review and individual participant data meta-analysis. Lancet Infectious Diseases, *21*(12), 1655-1667. https://doi.org/10.1016/S1473-3099(21)00387-X
[12] Broger, T., et al. (2019). Novel lipoarabinomannan point-of-care tuberculosis test for people with HIV: A diagnostic accuracy study. Lancet Infectious Diseases, *19*(8), 852-861. https://doi.org/10.1016/S1473-3099(19)30001-5
[13] Nel, M., Franckling-Smith, Z., Pillay, T., Andronikou, S., & Zar, H. J. (2022). Chest imaging for pulmonary TB – an update. Pathogens, *11*(2), 161. https://doi.org/10.3390/pathogens11020161
[14] Nguyen, D. T., Bang, N. D., Hung, N. Q., Beasley, R. P., Hwang, L. Y., & Graviss, E. A. (2016). Yield of chest radiograph in tuberculosis screening for HIV-infected persons at a district-level HIV clinic. International Journal of Tuberculosis and Lung Disease, *20*(2), 211-217. https://doi.org/10.5588/ijtld.15.0705
[15] Fehr, J., et al. (2021). Computer-aided interpretation of chest radiography reveals the spectrum of tuberculosis in rural South Africa. NPJ Digital Medicine, *4*, 106. https://doi.org/10.1038/s41746-021-00471-y
[16] Mwape, L., et al. (2021). Utility of chest radiography and clinical screening in identifying active tuberculosis among people living with HIV in Zambia. International Journal of Mycobacteriology, *10*(4), 398-404. https://doi.org/10.4103/ijmy.ijmy_112_21
[17] Ogbuabor, D. C., & Onwujekwe, O. E. (2025). Impact of the use of the ultra-portable digital x-ray with CAD4TB for active case finding for tuberculosis in Nigeria. Frontiers in Digital Health, *7*, 1559203. https://doi.org/10.3389/fdgth.2025.1559203
[18] Qin, Z. Z., Barrett, R., Ahmed, S., et al. (2022). Comparing different versions of computer-aided detection products when reading chest X-rays for tuberculosis. PLOS Digital Health, *1*(6), e0000067. https://doi.org/10.1371/journal.pdig.0000067
[19] Harris, M., Qi, A., Jeagal, L., et al. (2019). A systematic review of the diagnostic accuracy of artificial intelligence-based computer programs to analyze chest x-rays for pulmonary tuberculosis. PLoS ONE, *14*(9), e0221339. https://doi.org/10.1371/journal.pone.0221339
[20] Stevens, R. J., & Poppe, K. K. (2020). Validation of clinical prediction models: What, when, how. Archives of Public Health, *78*, 25. https://doi.org/10.1186/s13690-020-00412-6
[21] Moons, K. G. M., Wolff, R. F., Riley, R. D., et al. (2019). PROBAST: A tool to assess risk of bias and applicability of prediction model studies. Annals of Internal Medicine, *170*(1), 51-58. https://doi.org/10.7326/M18-1376
[22] Collins, G. S., Dhiman, P., Andaur Navarro, C. L., et al. (2024). Protocol for development of a reporting guideline (TRIPOD-AI) for multivariable prediction models that use artificial intelligence. BMJ. https://doi.org/10.1136/bmj-2023-076821
[23] Miyake, N., et al. (2020). IGRA positivity among foreign-born individuals in Japan: A retrospective cohort study. International Journal of Mycobacteriology, *9*(4), 361-367. https://doi.org/10.4103/ijmy.ijmy_115_20
[24] Prasetyo, R., et al. (2024). Implication of negative GeneXpert Mycobacterium tuberculosis/rifampicin for chest radiography scoring system in pulmonary tuberculosis diagnosis. International Journal of Mycobacteriology, *13*(2), 154-160. https://doi.org/10.4103/ijmy.ijmy_111_24
[25] Creswell, J., Qin, Z. Z., Gurung, R., Lamichhane, B., Yadav, D. K., Prasai, M. K., et al. (2018). The performance and yield of tuberculosis testing algorithms using microscopy, chest x-ray, and Xpert MTB/RIF. Journal of Clinical Tuberculosis and Other Mycobacterial Diseases, *14*, 1-6. https://doi.org/10.1016/j.jctube.2018.11.002.