Eligibility, Administrative data, and Critical AppraisalEligible Studies Identified During Systematic SearchFig. 1

PRISMA flow diagram documenting the selection process by which we arrived at the final sample of studies that met the eligibility criteria

We identified 526 studies during the initial database searches (Fig. 1) and screened 363 studies after duplicates were automatically removed by the Rayyan software. One hundred and forty-six studies were removed for not studying children, adolescents or youth living with HIV. An additional 61 studies were rejected because the cohort under study consisted primarily of adults—had a mean age > 26 years. Thirteen studies were excluded due to no cognitive assessment findings being presented. These were predominantly neuroimaging studies where the associations were with clinical variables. Similarly, we rejected 22 studies investigating only cognition without neuroimaging.

We excluded 28 studies that reported both neuroimaging and cognitive outcomes, but no associations between them, as well as 35 reviews and 6 studies in non-human subjects. Lastly, we excluded 19 reports with formats that did not meet the inclusion criteria: these included book chapters, conference proceedings, case reports, editorials, and protocols for prospective randomised trials.

At the end of the selection process, thirty-three (n = 33) studies were eligible for inclusion [15, 24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55].

Administrative Data, Sample Characteristics, and Study Design

We summarise the administrative data: short details of the studies, sample characteristics, and HIV transmission information of the infected children, adolescents, and youth in each study in Table 1. Ten studies (30%) were conducted in Africa (the majority in South Africa, n = 8, the remainder in Zambia), 8 (24%) in North America (USA), 7 (21%) in Asia, and 8 (24%) in Europe (Netherlands n = 6, England n = 1, Spain = 1). Seventeen studies (52%) described the socioeconomic status (SES) within which the cohorts lived; SES was usually assessed through annual household income and parental education levels. In these 17 studies, the cohorts were described as either residing in low-SES households or having SES lower than the general population of the country. Some of the studies included well-characterised cohorts, for example the neurological, cognitive, and visual performance in perinatally HIV-infected children cohort (NOVICE, n = 6), Cape Town adolescent antiretroviral cohort (CTAAC, n = 3), children on early antiretroviral therapy (CHER) trial (n = 3), adolescent master protocol (AMP) of the paediatric HIV/AIDS cohort study (PHACS, n = 3), and the paediatric randomized early versus deferred initiation in Cambodia and Thailand cohort (PREDICT, n = 1).

Considering the mean age of the participants, there were in total 20 (61%) studies in children (0–14 years), 10 (30%) in adolescents (15–18 years), and 3 (9%) in youth (19–26 years). Therefore, most of the eligible studies (91%) we found were conducted in children and adolescents. For the 3 studies in youth, one study included some participants aged between 26 and 30 years but had an overall mean age of 23 years [51]. The other 2 studies by Ashby et al. [30] and Martin Bejarano-Garcia et al. [50] had mean ages of 19.0 and 19.6 years, respectively, and included participants in their late adolescence. Ackermann, et al. [28] and Andronikou and colleagues [29] studied the youngest population, aged less than 4 years.

Many of the studies (91%) were of cross-sectional design, except for 3 (9%) that were longitudinal [42, 44, 52]. Most of the studies (82%) had relatively small sample sizes, i.e., less than 80 subjects overall; the lowest number of subjects was n = 8 [24] and the largest n = 374 [37]. However, since 76% of the studies included both children with HIV and controls or children living without HIV, large sample sizes were mostly due to a large sample of control children. The 8 studies (24%) that included no controls were Nozyce et al. [25], Gabis et al. [24], Ackermann, et al. [28], Andronikou et al. [29], Uban et al. [33], Herting et al. [31], Hoare et al. [32], and Hoare et al. [41]. The study with the largest number of children and adolescents living with HIV was by Nozyce et al. [25] (n = 274), followed by the study by Hoare and colleagues [41] (n = 125). In the other studies, the number of participants living with HIV was less than 65.

The HIV population in 97% of the studies (32 out of 33 articles) acquired HIV perinatally, i.e., via vertical transmission. Nagarajan et al. [27] had one out of 16 adolescents living with HIV horizontally infected, specified to be via blood transfusion at age 1. Therefore, the studied HIV populations generally lived with HIV from infancy.

Furthermore, with one exception, the sample populations were either all on cART or more than 71% of participants were on cART. One study specifically investigated outcomes of ART-naïve children [26]. The study samples had varying treatment regimens, the most common being a combination of thymidine, zidovudine or stavudine with cytidine, and efavirenz or nevirapine. In 67% (n = 22) of the studies, the children were on cART or ART, but the drugs/regimen were not specified by the researchers.

Table 1 Administrative data of the studies, showing the location where the research was conducted and where the study population lives, a short description of the articles, cohort name, study design, age range of the cohort, sample size, and the ART treatment regimen for the HIV sampleMethodological Quality of the Studies: Strengths, Limitations, and Risk of Bias

We present strengths and limitations in Table 2 as well as the assessment of the risk of bias in Fig. 2. The leading limitations that the authors specify are the relatively small sample sizes and the cross-sectional analyses of the data. All studies obtained ethical approval from their institutional regulatory boards. We obtained funding sources and potential conflicts of interest from all but 2 studies [24, 29]. Three co-authors for van den Hof et al. [52] declared potential conflicts, but these were not related to the work presented. We considered the abovementioned factors (limitations, strengths, reporting, sources of bias) along with other Quadas-2 checklist items during the assessment of quality of each study.

One study [40] did not sufficiently detail inclusion/exclusion criteria of the sample population and another one [49] did not describe subscales chosen/cognitive domains assessed. These missing details made their risks of bias, as assessed by Quadas-2 criteria (Fig. 2), unclear. One study [49] did not justify the exclusion of left-handed participants in their investigation.

Furthermore, we identified 3 studies [25, 28, 46] that used qualitative, rather than quantitative neuroimaging assessment. The scans in these studies were anonymised and reviewed by one or more neuroradiologists blinded to HIV and cognitive status. The neuroradiologist then identified lesions (‘lesion load’), calcifications, and anomalies in the images and provided a diagnosis of cortical atrophy, white matter signal abnormalities, or cerebrovascular disease, respectively. Diagnoses were then associated with cognitive outcomes. Ackermann and colleagues [28] and Nozyce et al. [25] used only one paediatric neuroradiologist blinded to the clinical findings at the time of referral, while Schneider et al. [46] used 2 neuroradiologists. Moreover, there was also one study [31] where the cognitive outcome (adaptive functioning) was based on a caregiver report rather than direct assessment.

Overall, there was a potential risk of bias in the studies that used qualitative rather than quantitative neuroimaging or cognitive assessment. We had applicability concerns in the domain of patient selection for the studies that included no controls as these studies did not avoid inappropriate exclusions. Furthermore, for 2 studies [40, 49] that used in children a cognitive assessment battery commonly applied in adult populations, there were applicability concerns in the index test domain. The remaining studies were judged as having low risks of bias.

Table 2 The strengths and limitations, ethical approval, description of inclusion/exclusion criteria, and conflicts of interests for each study Fig. 2

Methodological quality and risk of bias assessment of each study [15, 24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55] using the Quadas-2 checklist

Study Methodology: Neuroimaging Modalities and Cognitive Domains

Table 2 summarises the neuroimaging modalities and measures of interest, cognitive test batteries and cognitive domains of interest, as well as the statistical technique used to assess associations between imaging and cognitive outcomes.

Neuroimaging

The neuroimaging used in these studies was MRI, except for the oldest study Nozyce et al. [25] in which 91% of the children were scanned with computerized tomography (CT) and 9% with MRI. It is worth noting this was one of the three studies where neuroimaging was used qualitatively and that found no association between neuroimaging and behavioural problems. Seven of the studies (21%) exclusively used diffusion tensor imaging (DTI) to evaluate white matter microstructure; 8 (24%) only used