Of the 38 million people worldwide living with Human Immunodeficiency Virus (HIV), it is estimated that 2.8 million are children (<19 years) . According to the standard of care for clinical and laboratory monitoring of pediatric HIV infection, markers such as plasma HIV RNA and CD4+ T-lymphocyte count should be assessed routinely . However, the majority of HIV-infected children live in sub-Saharan Africa , where such routine monitoring is often unavailable [4, 5]. Due to prohibitive equipment and reagent costs, as well as a lack of laboratory facilities and trained personnel, clinics in resource-limited settings are often ill-equipped to measure these markers as regularly as recommended . Furthermore, even in places where facilities are available, most patients cannot access them consistently . The absence of such longitudinal marker measurements makes it difficult to monitor HIV-infected children on highly active antiretroviral therapy (HAART) regimens, which typically last multiple years. Therefore, methods enabling the prediction of post-HAART outcomes from existing patient data are of vital necessity as they provide a cost-effective alternative to routine monitoring in resource-limited settings. Predictive models of HIV disease prognosis will be more clinically useful if they account for the effect opportunistic co-infections, such as tuberculosis (TB) .
HIV and TB co-infection, in particular, poses a significant global health challenge as one in three HIV-positive individuals is estimated also to be infected with TB . Sub-Saharan Africa bears the brunt of these dual epidemics, accounting for 79% of all TB co-infected patients worldwide [8, 9]. Furthermore, the two pathogens act synergistically to worsen patient outcomes: TB is the most common opportunistic disease and leading cause of death amongst HIV-infected individuals; correspondingly, areas with the highest prevalence of HIV infections have seen the greatest increase in the incidence of TB over the past 20 years [10, 11]. The nature of these diseases to potentiate one another changes the approaches one must take when attempting to treat both infections concurrently.
Previous medical research has highlighted many challenges in treating HIV in the presence of TB co-infection . For instance, drug-to-drug interactions between rifampin, one of the most commonly used TB antibiotics worldwide, and various HIV highly active antiretroviral therapies (HAARTs) can yield unintended therapeutic consequences . While rifampin targets TB by inhibiting bacterial RNA polymerase, the drug also induces cytochrome P450 (CYP), a hemoprotein critical for the metabolism of drugs and other foreign molecules [14, 15]. This can accelerate the degradation of HIV-targeting protease inhibitors, resulting in subtherapeutic concentrations of HAARTs in TB co-infected patients . Furthermore, the existence of multiple drug-resistant (MDR) TB strains , varied TB presentations (including miliary, exudative pleuritis, and tracheobronchial ), and differential clinical manifestations in pediatric patients  add additional challenges to this problem. Therefore, to account for the inherent complexity of TB co-infection, models of HIV disease prognosis must be designed for specifically defined patient populations.
The majority of findings regarding TB co-infection were derived from studies of HIV-positive adult cohorts; however, investigations on TB co-infection in children are still lacking . Therefore, the main objective of our study was to determine if the impact of pre-HAART CD4+ T-lymphocyte count and percentage on pediatric immune recovery varied based on TB co-infection status. Using data from a cohort of HIV-positive children from Accra, Ghana, we assessed the ability of CD4+ T-lymphocytes to act as biomarkers for immune recovery. Precisely, CD4+ T-lymphocyte counts and percentages measured before the initiation of treatment were used. Using the Receiver Operating Characteristic (ROC) curves, we measured the diagnostic performances of these markers following adjustments for TB co-infection status, the primary covariate of interest. Additional concomitant variables such as age were also accounted for separately.