Defining cognitive impairment in people-living-with-HIV: the POPPY study

The Pharmacokinetic and Clinical Observations in People Over Fifty (POPPY) study is a prospective, multicentre, observational study that aims to examine the effects of ageing on the clinical outcomes of people living with HIV in UK and Ireland. To address its aims the study has established cohorts of HIV-positive people aged over 50, younger HIV-positive controls less than 50 years old and demographically matched HIV-negative controls aged over 50 years. For the present analysis only the two older cohorts were considered as the purpose of the younger HIV-positive cohort is to provide a younger control group which is not directly relevant to this analysis. HIV-positive participants were recruited at HIV outpatient clinics around UK and Ireland. Inclusion criteria were: documented presence of HIV infection, self-defined white or black-African ethnicity, likely route of HIV acquisition via sexual exposure (either by male to male exposure if white or by heterosexual exposure if white or black-African) and ability to comprehend the study patient information leaflet. HIV-negative controls were frequency matched to the HIV-positive group on gender, ethnicity, sexual orientation and location (in or out of London) and were recruited from sexual health clinics affiliated with the HIV clinics, as well as from community events, churches, adverts in targeted publications and community groups. Recruitment was from January 2013 to September 2014. The study was approved by the UK National Research Ethics Service (NRES; Fulham London, UK number 12/LO/1409). Written informed consent was obtained from all participants prior to undertaking any study specific procedures.

All enrolled participants completed questionnaires detailing physical and mental health status. In particular, participants answered the three questions on cognitive symptoms described by the EACS guidelines [5] regarding memory loss (do you experience frequent memory loss [e.g., do you forget the occurrence of special events even the more recent ones, appointments, etc.]?), reasoning (do you feel that you are slower when reasoning, planning activities, or solving problems?) and attention (do you have difficulties paying attention [e.g., to a conversation, book or movie]?). Individuals answering ‘Yes, definitely’, as opposed to ‘Never’ and ‘Hardly ever’ were classified as experiencing the related cognitive symptom. In addition a positive answer to at least two of the three questions was considered indicative of self-reported cognitive problems.

Assessment of cognitive function was performed using the CogState battery [16], a computerized battery of neuropsychological tests that has been used in different clinical settings [17‐20], including HIV-positive cohorts [21‐24]. The battery covered six cognitive domains commonly affected by HIV-associated CI, including visual learning, psychomotor function, visual attention, executive function, verbal learning and working memory (see Additional file 1: Table S1 for details of individual tests and how they map onto cognitive domains). Raw test scores were log-transformed or arcsine root–transformed where necessary (as recommended by the CogState guidelines for analysis) and converted into demographically-adjusted T scores (with a mean of 50 and a standard deviation of 10) using the scores of the HIV-negative group as normative scores. Briefly, a linear regression was fit for each test in order to estimate regression coefficients for age, gender, ethnicity and education using scores from the HIV-negative group. These regression coefficients were then used to determine the normative scores depending on subjects’ age, gender, ethnicity and education. The difference between the normative score and the actual score for each subject was then standardized into T-scores. A single T-score was calculated for each of the 6 cognitive domains by averaging individual test T-scores within each domain. A global T-score was also obtained by averaging T-scores across the six domains. For all T-scores a higher value indicates better cognitive function.

For each subject, the T-scores were then used to determine if the individuals met three definitions of CI, Frascati, GDS and MNC, using published methods. Frascati and GDS are the most extensively adopted definitions in previous studies of CI in HIV, while the MNC is a relatively newer approach showing promising results in reducing the false positive rate [10]. First, according to the Frascati criteria [7], CI was defined as at least two cognitive domain T-scores below 40 (i.e., one or more standard deviations below the average normative score). Second, a GDS [8] was computed for each subject by converting domain T-scores into deficit scores (0: T-score ≥ 40, 1: 34 < T-score < 40, 2: 29 < T-score ≤ 34, 3: 24 < T-score ≤ 29, 4: 19 < T-score ≤ 24, 5: T-score ≤ 19). An overall GDS was obtained by averaging domain deficit scores and CI was defined as a mean score equal or greater than 0.5. Finally, the MNC method [9] was applied. The MNC is a statistical method that simultaneously compares multiple cognitive scores of each study participant to the average scores of the same tests in the control group (in our case the HIV-negative group), taking into account the variances and covariance between all scores. For each participant, a continuous measure of the deviation of the participant’s cognitive profile from the average cognitive profile in the control group is then obtained. If this deviation (also called Hotelling’s T²) exceeds a critical value associated with a 5 % significance the individual is classified as cognitively impaired (so that the chance of erroneously concluding that an individual has CI while this is not the case, i.e., the false positive rate, is approximately 5 %). Definitions of CI included the three criteria listed above plus all combinations of patients meeting individual criteria and two or all three of these criteria.

Group comparisons of baseline characteristics were assessed using Chi-square, Wilcoxon rank-sum and t-tests (two-tailed) as appropriate. Comparisons of the prevalence of CI in HIV-positive and HIV-negative persons were performed using the Chi-square test with odds ratios used to provide a comparative estimate of the prevalence in the two groups. The agreement between criteria of CI was assessed using Cohen’s κ statistics [25] and interpreted following Landis and Koch [26] guidelines. The null hypothesis that Cohen’s κ equals zero (i.e., no agreement between criteria other than what would be expected by chance) was tested using the asymptotic test [27]. Internal consistency between the three patient-reported cognitive problems was assessed using Cronbach’s α. Associations between each definition of CI and self-reported cognitive problems were assessed using the Chi-square test. All analyses were performed using SAS v9.4 with p-values <0.05 considered as statistically significant.

A total of 290 HIV-positive and 97 HIV-negative participants were enrolled into the study between January 2013 and September 2014 and completed the CogState battery. Demographic, lifestyle and HIV-related characteristics are summarised in Table 1. Groups were highly comparable in terms of age (median age [IQR] was 57 [53, 62] and 58 [54, 62] years in HIV-positive and HIV-negative participants, respectively; p = 0.22), ethnicity (p = 0.47), country of birth (p = 0.92), educational attainment (p = 0.14), alcohol consumption (p = 0.13) and smoking status (p = 0.79). HIV-positive participants were more likely to be male (88.3 % vs. 66.0 %, p < 0.01), gay or homosexual (71.7 % vs. 41.2 %, p < 0.01) and to have reported recreational drug use in the 6 months preceding study entry (27.9 % vs. 12.4 %, p < 0.01) compared to HIV-negative controls. HIV-positives had been diagnosed with HIV for a median (IQR) of 16.8 (10.2, 22.9) years previously, and around 96.9 % were on ART with a median (IQR) CD4⁺ cell count of 610 (478, 780) cells/μL.

Overall performance of HIV-positive subjects was poorer than HIV-negative controls with a median (IQR) global T-score of 48.6 (43.5, 52.3) compared to 50.8 (46.0, 55.7) for the controls (p < 0.01). Significantly lower scores in the HIV-positive group, as compared to the HIV-negative group, were found for the cognitive domains: psychomotor function (median [IQR]: 48.4 [40.5, 54.4] vs 50.8 [45.4, 57.0], p < 0.01), visual attention (48.6 [39.7, 55.0] vs 50.9 [44.5, 57.0], p = 0.03) and verbal learning (47.0 [38.9, 53.6] vs 51.6 [45.3, 57.2], p < 0.01). In contrast, no significant group difference was found for the cognitive domains of visual learning (49.3 [42.7, 54.0] vs 50.8 [46.0, 55.7], p = 0.06), executive function (50.5 [45.9, 55.4] vs 50.4 [46.2, 54.8], p = 0.83) and working memory (50.1 [43.9, 54.4] vs 51.4 [46.0, 54.9], p = 0.15). The prevalence of CI in the HIV-positive group varied from 34.5 % according to GDS, 30.0 % according to Frascati and 22.1 % for the MNC. Similarly, the prevalence of CI varied in the HIV-negative group from 14.4 % (GDS) to 16.5 % (Frascati) and 7.2 % (MNC). According to all the three criteria, HIV-positive participants were significantly more likely of having CI than HIV-negative controls; OR (95 % CI) were 2.17 (1.20–3.92, p = 0.01) for Frascati, 3.12 (1.69–5.78, p < 0.01) for GDS and 3.64 (1.61–8.24, p < 0.01) for MNC.

Overlap in the classification of CI between the three criteria and Cohen’s κ statistics are reported in Fig. 1. Overall 169 (58.3 %) HIV-positive individuals did not meet any of the three definitions of CI while 41 (14.1 %) were classified as cognitively impaired by all the three definitions. Thirty-six (12.4 %) met Frascati and GDS only, 10 (3.4 %) met GDS and MNC only and 2 (0.7 %) met Frascati and MNC only. Frascati and GDS showed a substantial agreement (κ = 0.74, p < 0.01) while Frascati and MNC, and GDS and MNC showed moderate agreement (κ = 0.42 and κ = 0.48, respectively; p < 0.01 for each). Definitions showed better agreement when tested on the HIV-negative group: κ = 0.84 (p < 0.01) for Frascati and GDS, κ = 0.57 (p < 0.01) for Frascati and MNC and κ = 0.53 (p < 0.01) for GDS and MNC. Cognitive scores in individual domains of HIV-positive individuals meeting each of the three definition were similar (Fig. 2).

Overall, 14, 15 and 17 HIV-positive individuals did not answer or had missing information on memory loss, reasoning and attention problems, respectively. Among those with complete information, 79 (28.6 %) reported frequent memory loss, 105 (38.2 %) reported reasoning problems and 79 (28.9 %) attention problems; moreover 90 (32.5 %) reported at least two of the three problems. Internal consistency of the three patient-reported measures was excellent with a Cronbach’s α (95 % CI) of 0.9 (0.82, 0.98). A significant association was found between memory loss and all the three definitions of CI (Table 2): using the Frascati criteria, 38 % of subjects with CI reported frequent memory loss, while only 25 % did so among those without CI (p = 0.02), with GDS the proportions were 40 and 23 % (p < 0.01) for those with and without CI, respectively, and with MNC 41 and 25 % (p = 0.02). There was no strong evidence for associations between reasoning problems and self-reported attention problems with definitions of CI, regardless of the definition (Reasoning: p = 0.77, p = 0.24 and p = 0.42 for Frascati, GDS and MNC, respectively; Attention: p = 0.59, p = 0.07 and p = 0.28 for Frascati, GDS and MNC, respectively).

Those meeting all three definitions (n = 41, 14.1 %) had a low median global T-score (36.9, Fig. 3), indicating poorer cognitive function, and approximately 41.5 % reported at least two of memory loss, reasoning and attention problems. In particular, performances in the psychomotor (median score: 26) and visual attention (33.1) domains were particularly poor. The 36 (12.4 %) subjects with CI according to Frascati and GDS (but not MNC) demonstrated similar cognitive function: 13 of them (36.1 %) reported at least two cognitive problems and the median global T-score was 41.9. On the other hand, the 11 (3.8 %) subjects identified as cognitively impaired only by MNC performed generally better (median global T-score equals to 48.0) and a lower proportion (18.2 %) reported two or more cognitive problems.