Andrew F. Auld, Andrew D. Kerkhoff, Yasmeen Hanifa, Robin Wood, Salome Charalambous, Yuliang Liu, Tefera Agizew, AnikieMathoma, Rosanna Boyd, Anand Date, Ray W. Shiraishi, George Bicego, Unami Mathebula-Modongo, Heather Alexander, Christopher Serumola, GoabaoneRankgoane-Pono, PontshoPono, Alyssa Finlay, James C. Shepherd, Tedd V. Ellerbrock, Alison D. Grant, Katherine Fielding.
Abstract
Among people living with HIV (PLHIV), more flexible and sensitive tuberculosis (TB) screening tools capable of detecting both symptomatic and subclinical active TB are needed to (1) reduce morbidity and mortality from undiagnosed TB; (2) facilitate scale-up of tuberculosis preventive therapy (TPT) while reducing inappropriate prescription of TPT to PLHIV with subclinical active TB; and (3) allow for differentiated HIV–TB care.
Introduction
Tuberculosis (TB) remains the most common cause of death among people living with HIV (PLHIV), with 251,000 HIV-associated TB deaths in 2018, over 95% of which occurred in low- and middle-income countries (LMICs) [1]. Among PLHIV who die from TB, TB often remains undiagnosed at the time of death [2,3]. The World Health Organization (WHO) recommends a 4-symptom TB screening rule (i.e., for cough, weight loss, night sweats, or fever) to determine which PLHIV need investigation for active TB and which are eligible for immediate tuberculosis preventive therapy (TPT) [4]. WHO 4-symptom TB screening rule is recommended for LMIC regardless of expected prevalence of active TB, setting (e.g., high or low TB incidence settings), or antiretroviral therapy (ART) status (ART-naive or ART-experienced) [4].
Method
We used data from the Xpert Package Rollout Evaluation using a stepped wedge design (XPRES) trial conducted in Botswana to derive the predictive TB clinical score [15,16]. We split XPRES cohort data geographically into 11 southern and 11 northern clinics to serve as an internal derivation and validation dataset, respectively. We used 2 different but complementary modeling approaches to generate a parsimonious TB clinical risk score comprised of variables easily available in a resource-constrained clinic setting: (1) logistic regression models; and (2) random forest machine learning models. Random forest machine learning models are particularly useful for identifying important nonlinear associations between predictors and outcomes because the modeling approach does not rely on assumptions of average linear or curvilinear associations [17]. Having derived the clinical score, we then used data from 3 other settings to externally validate the derived clinical score.
Discussion
To our knowledge, this study provides new information by deriving and externally validating an initial clinical score for active TB among both ART-naive and ART-experienced adult PLHIV that includes but does not rely solely on WHO TB symptom screening and allows flexibility in choosing the desired sensitivity, specificity, NPV, PPV, and NNS across a range of cutoffs, depending on the setting, use case scenario, and population served. In addition, following further validation and evaluation steps, the screening tool could potentially be used to reduce the likelihood of missing subclinical TB, which accounted for 6% to 27% of all TB cases across studies; this could potentially help reduce morbidity and mortality due to late or missed TB diagnosis and reduce TPT prescription to PLHIV needing a full TB treatment course. Similarly, following further validation efforts, the screening tool’s differentiation of 3 risk groups could be used to inform differentiated care in LMIC clinic settings, which could potentially improve efficiency and impact morbidity and mortality.
Citation: Auld AF, Kerkhoff AD, Hanifa Y, Wood R, Charalambous S, Liu Y, et al. (2021) Derivation and external validation of a risk score for predicting HIV-associated tuberculosis to support case finding and preventive therapy scale-up: A cohort study. PLoS Med 18(9): e1003739. https://doi.org/10.1371/journal.pmed.1003739.
Academic Editor: Sydney Rosen, Boston University School of Public Health, UNITED STATES
Received: November 24, 2020; Accepted: July 21, 2021; Published: September 7, 2021.
Copyright: This is an open access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication.
Data Availability: The authors confirm that, for IRB-approved reasons, some access restrictions apply to the data underlying the findings. Although the patient-level data do not include patient names, this IRB decision is in the interest of ensuring patient confidentiality. An individual may email the CDC Division of Global HIV & TB science office (gapmts@cdc.gov) to request the data.
Funding: This research has been supported by the President’s Emergency Plan for AIDS Relief (PEPFAR) through the US Centers for Disease Control and Prevention. The funder had no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: I have read the journal’s policy and the authors of this manuscript have the following competing interests: AG has received research grants (not related to this work) from the Medical Research Council, Economic and Social Research Council, Bill and Melinda Gates Foundation, National Institute of Allergy and Infectious Diseases, Wellcome Trust, Research England, and USAID.
Abbreviations: AIC, Akaike information criteria; ART, antiretroviral therapy; AUROC, area under the receiver operating characteristic; BIC, Bayesian information criteria; BMI, body mass index; CDC, Centers for Disease Control and Prevention; COVID-19, Coronavirus Disease 2019; CRP, C-reactive protein; CRT, cluster randomized trial; CXR, chest radiography; EMR, electronic medical record; HRDC, Health Research and Development Committee; IRB, Institutional Review Board; LMIC, low- and middle-income country; MFP, multivariable fractional polynomial; NNS, number needed to screen; NPV, negative predictive value; PEPFAR, US President’s Emergency Plan for AIDS Relief; PLHIV, people living with HIV; POC, point-of-care; PPV, positive predictive value; SA, South Africa; SOC, standard of care; SSA, sub-Saharan Africa; TB, tuberculosis; TBFT, TB Fast Track; TPT, tuberculosis preventive therapy; TRIPOD, Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis; WHO, World Health Organization.
