Background
In recent years, despite the introduction of combination antiretroviral therapy (ART), a high but varying prevalence of cognitive impairment (CI) has been reported among HIV-positive individuals. Robertson et al. [
1] described a 26 % prevalence of CI among HIV-positive subjects who had received ART for at least 20 weeks, while Heaton et al. [
2] reported a prevalence of 36 % among ART-treated asymptomatic individuals. Among those with long-standing viral suppression, Winston et al. [
3] and Simioni et al. [
4] reported prevalences of 51 and 74 %, respectively.
The optimal screening tools to identify CI are unknown. The current European AIDS Clinical Society (EACS) guidelines recommend three questions as one form of assessment to guide the initial evaluation of HIV-positive individuals with suspected cognitive problems [
5]. These patient-reported measures may be a quick and practical method of screening for cognitive impairment in clinical practice, however, doubts about their utility remain [
6]. When cognitive function is assessed through objective neuropsychological tests, differences in the reported prevalence of CI may be due to the use of different diagnostic criteria. Three criteria in current use include the ‘Frascati’ criteria, proposed by Antinori et al. [
7] and also known as HIV-associated neurocognitive disorder (HAND) criteria, the global deficit score (GDS) [
8] and the multivariate normative comparison (MNC) [
9]. These criteria differ in the way they combine scores from a battery of neuropsychological tests to classify subjects as either cognitively impaired or normally functioning. Several studies have reported contrasting results when using different criteria of CI on the same set of patients [
10,
11]. However, they only assessed prevalence rates by criteria and did not specifically investigate whether they identified the same people as impaired/not impaired.
The associations between CI and patient-reported outcomes of cognitive function have not been fully investigated and remain unclear. Some studies have found a relationship between subjective cognitive complaints and actual impairment in neuropsychological tests [
12,
13], while other studies have not [
14,
15]. While these studies tended to include high rates of untreated or unsuppressed subjects, there is currently a lack of data relating cognitive function with self-reported cognitive complaints in populations of well-treated HIV-positive individuals.
The aims of this study are threefold. Firstly, to compare the prevalence of CI in HIV-positive individuals over 50 years of age and demographically matched HIV-negative controls according to the Frascati criteria, GDS and MNC and combinations of these definitions. Secondly, to assess the level of agreement between these three criteria when identifying HIV-positive people with CI. Finally, to investigate the association between different definitions of CI and their combination with patient-reported symptoms of cognitive dysfunction.
Discussion
HIV-positive individuals exhibit poorer cognitive function when compared to an appropriate HIV-negative control group. Although the difference in the overall cognitive score is statistically significant, this would not be considered clinically meaningful (for T-scores, a 5-point difference is usually considered relevant from a clinical point of view [
28]). The prevalence of CI in older HIV-positive individuals may vary from 35 to 22 % depending on the criteria used.
Commonly-used criteria of CI show fair agreement, especially Frascati and GDS. However, this agreement is mainly driven by the ability of criteria to exclude CI in the same set of subjects rather than their ability to identify CI. As expected, subjects meeting all the criteria have generally poorer cognitive function (with particularly poor performances in the psychomotor and visual attention domains) and are more likely to experience cognitive problems. Similar cognitive scores were also observed in subjects classified as impaired by Frascati and GDS (but not MNC).
Consistent with several published studies [
14,
15], but contrary to others [
12,
13], the associations with self-reported cognitive symptoms are generally poor for all the three definitions of CI considered. Whilst CI, defined with all three criteria, correlate with memory loss, this is not the case for either attention or reasoning problems. These results, based on a cohort of mainly treated and virally-suppressed subjects, shed further light on the association between patient-reported and objective cognitive impairment in the post-ART era. The lack of association found may reflect the pattern of cognitive changes we observed, namely, poorer verbal learning (which may relate to patient reported memory problems) but no significant differences in executive function (reasoning) or working memory (attention). These results suggest a potential lack of a clear relationship between subjective measures of cognitive function and more objective measures based on neuropsychological tests, particularly in those with mild impairment. Several reasons may account for this lack of association, such as the over-reporting of cognitive symptoms and the subjectivity of the EACS questions. Moreover, depressive disorders have been previously reported to affect both subjective and objective cognitive function [
29], and can therefore confound the association between the two. We did not co-vary depression in our analyses, but from preliminary analysis, depression did not seem to change the associations between definitions of CI and patient-reported symptoms (data not shown).
A missing answer to questions about cognitive complaints may, in theory, be informative as it may indicate poor understanding or attention. However, almost 95 % of data was complete so it’s unlikely that the exclusion of this small group of individuals from analyses has introduced substantial bias. Although ART prescribing was in line with national guidelines, 3.1 % of enrolled HIV-positive individuals were not receiving suppressive antiretroviral therapy at study entry. This is justified by the aim of the study of recruiting a ‘real-world’ sample of PLWH in UK and Ireland. Nonetheless those not receiving ART had a median CD4+ cell count of 664 cells/μL, which makes it unlikely that lack of suppression of HIV replication in a minority of the sample has led to substantial bias to our findings.
Given the lack of a gold standard in defining CI, it is difficult to ascertain the validity of different definitions. An optimal definition of CI would capture subjects with the lowest cognitive performance scores and the greatest number of cognitive symptom complaints. In our study we observed the lowest overall cognitive score in subjects meeting all three definitions (median global T-score of 36.9 for those meeting Frascati, MNC and GDS). Moreover, in this group of subject, the number reporting cognitive complaints was highest (41.5 %). Given these findings, we consider further work to assess longitudinal outcomes in HIV-positive individuals meeting this definition of CI within the POPPY study, compared to other definitions of cognitive impairment, is justified.
Other definitions of cognitive impairment in PLWH, such as the Frascati criteria, have been criticised for overcalling the number of PLWH with cognitive deficits [
11]. On the converse argument, the Frascati criteria has attempted to define the presence of cognitive deficits prior to the onset of clinical symptomatology and classifies large numbers of subjects with CI. The rationale behind such criteria being interventions at an early stage of disease, if effective, may prevent the onset of clinically apparent conditions. However, to date, longitudinal data both in the HIV-field and in other neurodegenerative diseases have not provided convincing evidence for the diagnosis of a pre-morbid cognitive state [
30,
31].
There are several other problems when utilising non-stringent definitions of CI. Firstly, as large numbers of patients will meet a diagnostic criteria, unnecessary anxiety for patients could be created. Secondly, non-stringent definitions will include subjects with cognitive impairment but will also include subjects who do not have cognitive impairment. Within interventional trials, this may lead to the null hypothesis being proven when in fact an intervention does work. By including patients without a disease state in an interventional study, the results of the study may suggest an effective interventions has no effect whereas, if the effective intervention was trialled in the diseased population, an effect may be observed.
Acknowledgements
POPPY Management Team: Marta Boffito, Paddy Mallon, Frank Post, Caroline Sabin, Memory Sachikonye, Alan Winston
Scientific Steering Committee: Jane Anderson, David Asboe, Marta Boffito, Lucy Garvey, Paddy Mallon, Frank Post, Anton Pozniak, Caroline Sabin, Memory Sachikonye, Jaime Vera, Ian Williams, Alan Winston
POPPY Sites and Trials Unit:
• Elton John Centre, Brighton and Sussex University Hospital (Martin Fisher, Amanda Clarke, Jaime Vera, Andrew Bexley, Celia Richardson)
• St Stephen’s Centre, Chelsea and Westminster Hospital (Marta Boffito, David Asboe, Anton Pozniak, Chris Higgs, Elisha Seah, Stephen Fletcher, Michelle Anthonipillai, Ashley Moyes, Katie Deats)
• Homerton Sexual Health Services, Homerton University Hospital (Jane Anderson, Sifiso Mguni, Rebecca Clark, Rhiannon Nevin-Dolan)
• Caldecot Centre, King’s College Hospital (Frank Post, Lucy Campbell, Selin Yurdakul, Sara Okumu, Louise Pollard)
• HIV Molecular Research Group, Mater Misericordiae University Hospital (Paddy Mallon, Alan Macken, Bijan Ghavani-Kia, Joanne Maher, Maria Byrne, Ailbhe Flaherty)
• Department of Infection and Population Health, University College London (Ian Williams, Damilola Otiko, Laura Phillips)
• St. Mary’s Hospital London, Imperial College Healthcare NHS Trust (Alan Winston, Lucy Garvey, Matthew Stott, Linda McDonald)
• Imperial Clinical Trials Unit, Imperial College London (Andrew Whitehouse, Laura Burgess, Daphne Babalis)