Study Registration and Strategy
In developing this protocol, the guidelines of The Preferred Reporting Items for Systematic reviews and Meta-analysis (PRISMA) was followed, as reported in the additional file 1. This protocol is registered in PROSPERO 2021 (CRD42021271358). The search terms will include HIV and neurocognitive disorder, as provided in detail in additional file 2. We will search Medline (OVID), Embase (OVID), PsycInfo (OVID), Web of Science, ProQuest, as well as OpenGrey and clinical trial databases to identify records relevant to our search terms, published from 1 January 2007 up to 31 May 2021, and select studies from. This primary search method will be complemented with hand-searching of the reference of each selected article will be assessed.
Eligibility Criteria
Studies reporting the proportion of NI using observational cross-sectional or cohort design will be included. Reviews, systematic reviews, meta-analyses, case reports, case series, and qualitative studies will be excluded. We will include studies examining adolescents (12 years or older) and adults (18 years or older). The exposure is HIV diagnosis or documented HIV status at study enrolment. The comparator is HIV negatives age 12 years or older, with contemporaneous study enrolment as the exposed group. The outcome of the studies is the proportion of NI by MoHE in the HIV seropositive group compared with that of the seronegative group. The NI should be assessed using validated neurocognitive tests and international standardised diagnostic criteria. Studies published from 2007, when the terminology of HAND was established, and in English will be included.
Study Selection
Titles and abstracts of unique citation records will be screened independently by two reviewers (AP, GG), supervised, and adjudicated by a third reviewer (AR) in the case of disagreement. Full-text articles will be obtained, and four reviewers will independently perform eligibility assessment, none of whom will be blinded to the bibliographic information of the study articles.
Extraction and Data Management
Four reviewers (APA, AP, AR, GG) will independently extract the following information from eligible studies onto a pre-piloted spreadsheet form:
-
Bibliographic information (title, year of publication, first author, country)
-
Participant eligibility, including the exclusion of neurological confounds (e.g., exclusion of participants due to a history of central nervous system disorders, major depression, or substance dependence)
-
Demographics (age, sex, years of education)
-
Comorbid conditions applicable to all participants
-
Type of controls, according to how the MoHE is defined in the studies, which can be:
-
'MoHE-naive controls': HIV-negative participants in undefined or unbalanced proportions of MoHE compared to their HIV-positive counterparts; or
-
'MoHE-adjusted controls': HIV-negative participants with a homogenous HIV risk aetiology in distribution with their HIV-positive counterparts, defined arbitrarily at ≥ 90% of participants from a single MoHE group in both serostatus groups.
-
MoHE where relevant (e.g., men who have sex with men, people who use drugs or alcohol, perinatally infected persons)
-
Number of outcomes by HIV status
-
Number of neurocognitive domains assessed
-
Diagnostic criteria for neurocognitive impairments
-
Smoking status
-
HIV characteristics, i.e., nadir CD4 count and proportions on antiretroviral treatment
-
Type of analysis adjustment (none, demographics, and/or comorbidities)
-
A measure of association in relative risks
To ensure consistency across reviewers, they should follow an exercise on using the spreadsheet form before extracting the data. Any disagreement will be resolved through discussion between reviewers involving an external arbitrator, who is a clinician expert in neurocognitive and HIV. If necessary, the article's corresponding author will be contacted for clarification.
If the study describes outcomes of the HIV-positive and HIV-negative groups in multiple strata, we will extract these by strata, thereby generating more than one observation from the study if each stratum has ≥ 50 participants in total across the serostatus groups. Should any stratum fail to achieve this minimum size, we will average the outcomes and cohort characteristics across the strata, weighting the statistics by stratum size or the inverse of variance for adjusted effect measures. For instance, a study reports the number of HIV-positive and HIV-negative participants, stratified by age group, as follows: 22 and 21 for those aged < 50 years (Stratum 1) and 25 and 26 for those aged ≥ 50 years (Stratum 2). The size of Stratum 1 (n = 43) is less than 50 so that the combined data to extract will be a total of 94 participants with 47 in each HIV serostatus group. If more than one subgroup is reported for an HIV serostatus, only the most similar subgroup conditions to the other HIV serostatus will be extracted. For example, if a study enrolled HIV-positive smokers but HIV-negative participants by smoking status, only data from the HIV-negative smokers will be extracted for analysis.
Sub-cohorts from a parent cohort reported in multiple studies will constitute independent observations representing valid MoHE groupings. In such cases, we will subtract the size of the larger cohort included in the analysis proportional to the size of the sub-cohorts and the associated outcome rates to avoid overlapped cohort membership and approximately correct the reported effect size.
Study authors will be contacted by email, in three attempts spanning over two weeks, for clarifications on study design, participant characteristics, or methods to compute outcomes to assess eligibility better or for more data that are needed for synthesis than what is reported in the text such as, for instance, unreported standard deviations. Should the contacted authors provide no or insufficient response, a consensus judgment will be exercised to determine the inclusion or exclusion of the study in question. The missing data will be treated as such. Disagreements among reviewers during data extraction will be resolved through consensus.
If studies report categorical levels of education, duration in years will be approximated by assigning a standard length of time for each education level, capped at university education if the highest level is 'university or above', and averaging it for participants in the HIV-positive and HIV-negative groups. Similarly, continuous age in years will be approximated from studies reporting in age groups by constructing the weighted mean, assuming a quarter of the age range up to the maximum age, or life expectancy if not reported, for the last age group.
Relative risks (RRs) and the corresponding standard errors will be computed from the standard formula using the frequency of cases and non-cases in each serostatus group or extracted from the reported proportions or regression outputs for adjusted effect estimates. We will provide indirect extraction of RRs for studies reporting performance test scores in > 1 neuropsychological domains by using the mean and the standard deviation of the normal distribution of test scores to estimate the frequency of cases in each serostatus group using the established or instrument-specific cut-off values for diagnosis of neurocognitive impairments. Scores from a neuropsychological test battery will be converted to an approximated normal distribution of z-scores and then to t-scores before averaging over the individual tests and by the serostatus group. In the second step, these global t-scores will be used to construct a normal distribution of participant values, which will be rescaled to deficit scores to apply the GDS diagnostic criteria of neurocognitive impairments at the recommended cut-off of ≥ 0.5. If raw scores are reported, we will use the HIV-negative group as the normative population to standardise test scores against studies enrolling matched cohorts or utilise regression adjustments on sex, age, and education to be consistent with the GDS approach.
We will place preference on the neuropsychological test battery over other instruments in validation studies and extract the diagnostic criterion with the strongest evidence of a statistical difference if more than one criterion is reported in the study. Reports of adjusted odds ratios (ORs) will be converted to RRs by dividing the OR by one minus the product of multiplication between an assumed outcome proportion in the control group and the difference in OR from unity.
Outcomes
The primary outcome will be the number of participants diagnosed with NI by the MoHE in seropositive and seronegative groups. The NI was assessed on more than two neurocognitive domains and in accordance with the following diagnostic criteria for HIV-associated neurocognitive disorders: a) Frascati (≥ 1 standard deviation [SD] below the mean of normative scores) (11); b) Gisslen (≥ 1.5 SD below the mean of normative scores) (4, 12); c) global deficit scores (3); d) clinical rating scale (3); e) multivariate normative comparisons (3); and f) screening analogues (e.g., CogState, Montreal Cognitive Assessment, International HIV Dementia Scale) with established or reported cut-off scores to delineate impairments (13–15). Studies assessing a single class of neurocognitive impairments (e.g., dementia only) will be excluded.
Risk of Bias
The Newcastle Ottawa Scale (NOS) (16) will be applied to assess each selected study's possible risk of bias. The items in the NOS for case-control and cohort studies will be selected according to the purpose of this systematic review and meta-analysis. Two reviewers (GG, GM) will independently assess the quality of included studies. Disagreements in the assessment process will be adjudicated by a third reviewer (AR). Quality scores will be used in meta-analysis to detect potential differences in effect sizes by study quality ranking.
Data Synthesis
We will first describe the included studies and summarize pertinent characteristic information. Next, we will pool the relative risks (RRs) and the standard errors using the inverse-variance random-effects (DerSimonian-Laird) method of meta-analysis, stratifying the studies by the type of control group as defined above. Hypothesis testing on the difference in RRs between the MoHE-naive and MoHE-adjusted strata will be performed. The I2 and Q statistics will also be computed to assess heterogeneity of the pooled RRs and their 95% confidence intervals for each stratum, with I2 > 50% signifying substantial heterogeneity. This main analysis will be presented in a forest plot. A leave-one-out analysis will be performed in pairs across the strata by removing studies individually in one stratum over a serial exclusion of the remaining studies in the other stratum until all possible pairs are exhausted.
The degree of symmetry in the distribution of log RRs will be visually inspected in a funnel plot and statistically tested using the Egger's test for each study stratum for assessment of small-study effects. The Duval and Tweedie trim-and-fill technique will be used to adjust the pooled RRs in the event of small-study effects.
In accordance with the Meta-analyses Of Observational Studies in Epidemiology (MOOSE) guideline, exploratory examinations of the effect of certain clinical subgroups and potential moderators will be performed and described in the following paragraphs.
Where studies are available in sufficient number in each stratum, we will perform subgroup analyses by age groups; selection of participants from a comorbid population; diagnostic criteria; geographical locations or country income groupings; outcome extraction methods (direct vs indirect); and adjustments to demographic and/or neurocognitive or other comorbid confounds. We will compare the pooled RRs in the two strata of HIV serostatus for all levels of subgroups and compute the p values for the difference, cautioning that these statistics may be inestimable or unreliably estimated in a number of subgroups due to study unavailability or few available studies in either stratum.
Finally, we will conduct random-effects meta-regression with the Knapp-Hartung modification to assess the difference in pooled RRs by type of control groups, adjusted for continuous or ordinal study-level covariates mean age; nadir CD4 count; proportions of HIV-positive participants on antiretroviral treatment; the number of neurocognitive domains; and the study quality ranking. Because we anticipate fewer eligible studies than would be minimally required for multivariate analysis, meta-regression adjustments for the type of control group will be conducted separately for each covariate. We will advance a second regression model in which MoHE-adjusted studies will be further distinguished into behavioural (e.g., men who have sex with men, people who use drugs or alcohol, high-risk heterosexuals) and non-behavioural HIV risk groupings (e.g., haemophiliac patients, perinatally infected persons) or a similar distinction to contrast the effect of different MoHEs on the excess risk of neurocognitive impairments. Any missing covariates will be imputed by the average value of the remaining studies in the same geographical location or country income grouping.