Demographic and risk group heterogeneity across the UNAIDS 90-90-90 targets: a systematic review and meta-analysis protocol

We will include all observational cohort studies, case-control studies, cross-sectional studies, HIV surveillance reporting, and randomized control trials that report on one or more elements of the 90-90-90 HIV care cascade. We will include studies that report aggregate estimates or stratum-specific estimates. We will exclude studies that provide modeled estimates of testing, treatment, or virological suppression, systematic reviews and meta-analyses, and data collected exclusively prior to January 2014.

We will include any study of adult populations, defined as PLHIV aged 15 years or older, located in sub-Saharan Africa. We will include strata defined in Table 1.

The primary outcome is the prevalence of unsuppressed viral load by demographic and risk group, and risk factors associated with being unsuppressed. The labels and definitions of strata used in included studies will be recorded to document the diversity and similarity of strata categorization (e.g., mobile versus migratory populations). The secondary outcomes of interest are defined as the proportion, relative risk, or odds ratio of the following three groups: PLHIV who know their HIV status, PLHIV who know their status and are receiving ART, and PLHIV on ART who are virologically suppressed.

We will conduct four comprehensive literature searches, one for each of the 90-90-90 targets and a fourth search focusing on the entire HIV care cascade (Fig. 1). We will perform searches in PubMed and Embase and restrict the search to articles published in English between January 2014 and March 2018. We will limit our search to articles published after 2014 because the UNAIDS 90-90-90 targets were announced mid-2014 and the World Health Organization (WHO) universal test and treat guidelines were released in 2015 [11].

In addition to searching the published peer-review literature, we will conduct a comprehensive search of websites and databases for publicly available grey literature sources, such as WHO and UNAIDS reports, Ministries of Health websites and national surveillance reports, Médecins Sans Frontières (MSF) reports, the Integrated Bio-Behavioral Surveys (IBBS), U.S. President’s emergency plan for AIDS Relief (PEPFAR) country operational plans, the Analyzing Longitudinal Population-based HIV/AIDS data on Africa (ALPHA) Network, the National Technical Reports Library (NTRL), the Population-based HIV Impact Assessments (PHIA), AIDS Indicator Surveys (AIS), Demographic and Health Surveys (DHS), and other similar national HIV surveys (e.g., the South African National HIV Prevalence, Incidence and Behavior Survey) [12‐22]. We will also search HIV conference abstracts including the International AIDS Society (IAS), the Conference on Retroviruses and Opportunistic Infections (CROI), and the International Conference on AIDS and STIs in Africa (ICASA) [23‐25]. For eligible grey literature sources, we will review listed citations for additional data sources.

A detailed search strategy is presented in Additional file 2. In brief, MeSH and keywords will be included for HIV, such as “Human Immunodeficiency Virus” and “AIDS”, along with terms for each portion of the 90-90-90 cascade, and key terms for “Sub-Saharan Africa” and specific country names using Boolean “AND” and “OR” operators. For the first 90, we will include terms such as “testing”, “diagnosis”, and “serostatus”. The search for the second 90 will include terms such as “antiretroviral” and “treatment”, and the search for the third 90 will include terms such as “viral suppression” and “viral load”. Lastly, for our fourth search, we will include terms to capture the whole cascade, including “90-90-90”, “HIV care cascade”, and “universal test and treat”. Prior to publishing results, we will update our search to include any additional eligible papers that are published after March 2018.

Each of the separate four searches will be conducted and merged into a reference manager. We will record the number of duplicate records, which will then be removed prior to the selection process. We will use Microsoft Excel and Covidence, a systematic review management software, to document the outcome of the title screening, abstract screening, full text review, and data abstraction processes [26].

We will first screen records for inclusion based on title only, duplicated independently by two reviewers. In cases of disagreement, the record in question proceeds to the abstract review stage. Abstract review will also include duplicate reviews. Full texts of the records will then be obtained and reviewed for inclusion and will be conducted independently and in duplicate. Discrepancies from the abstract and full text review will be resolved through consensus or in discussion with a third independent reviewer. Records will be excluded from consideration at title, abstract, and full text review stages if they satisfy any of the following exclusion criteria: study only among HIV-negative persons, study on persons outside of sub-Saharan Africa setting, study on children aged under 15, study of an inappropriate design, study data collected exclusively prior to January 2014, or study does not report on at least one of the 90-90-90 outcomes (Table 2).

We will develop a data extraction form and pilot test this form on ten randomly selected publications which have been selected for data abstraction. This form will guide the collection of strata-specific estimates of inclusion across the three 90-90-90 targets, as well as strata definitions. Data extraction will be done independently by two reviewers, with discrepancies resolved by consensus or in consultation with a third reviewer.

The data items for extraction are informed by the study aims. We will include the following information from all studies:

We will summarize information on virological suppression, and strata associated with being virologically suppressed. Strata reported, definitions used to define strata and levels of the HIV care cascade, and study characteristics including geography and study design will be summarized in the narrative. Study characteristics such as year and study period will be reported separately for each included study. Our qualitative assessment of the comparability of these definitions will inform our quantitative analysis.

We will summarize information on strata associated with knowledge of HIV serostatus, enrollment on ART, and virological suppression. This will include synthesizing odds ratios, relative risks, and proportions of persons virologically suppressed. We will hand calculate measures of association, odds ratio and risk ratio point estimates without confidence intervals, for studies which do not report or calculate measures of association but provide sufficient data for their calculation. Where appropriate and feasible, we will use random effects meta-regression analyses to estimate the distribution of demographic and risk features among groups of combinable strata who are not virologically suppressed. We hypothesize that national-level epidemic characteristics such as HIV prevalence and progress towards the 90-90-90 goals will be critical contextual features to consider in quantitative meta-analyses.

In the case of missing or incomplete data, we will contact corresponding authors to request relevant information or for additional clarification. If the corresponding authors fail to respond, then additional authors will be contacted. A description of the missing data for each included study will be provided, along with the possible implications of missing data. We will assess potential publication bias through the use of funnel plots and a subjective assessment of asymmetry [18].

There will be three primary reviewers of included studies. At least two reviewers will assess the risk of selection bias, reporting bias, and attrition bias for each study included for data abstraction by using an adapted Cochrane Risk of Bias Tool including items for non-randomized studies [27]. Using this tool, reviewers will grade studies as having low, moderate, serious, or critical risk of bias. Conflicting ratings will be resolved through consultation of a third reviewer. We intend to assess the heterogeneity of study results by calculation of the I² statistic [28]. Studies with critical risk of bias will not be considered for quantitative synthesis.