Criteria for considering studies for this review
Types of interventions
Types of interventions to be excluded:
Interventions concerned with retention in care between HIV diagnosis and ART initiation.
Decentralisation of care and task-shifting interventions. These interventions are covered in existing Cochrane reviews (Kredo 2012a, Kredo 2012b).
Types of outcome measures
Retention in care after ART initiation where retention is defined by a patient who is still on HIV treatment (assessed at clinically appropriate intervals, e.g. 6, 12, 24, 36, 48, 60 months) and has not (1) died, (2) transferred out, (3) stopped treatment, or (4) been lost-to follow-up.
Loss to follow-up
Adherence to ART
Search methods for identification of studies
See search methods used in reviews by the Cochrane Collaborative Review Group on HIV Infection and AIDS.
We will formulate a comprehensive and exhaustive search strategy in an attempt to identify all relevant studies regardless of language or publication status (published, unpublished, in press and in progress). Full details of the Cochrane HIV/AIDS Review Group methods and the journals hand-searched are published in the section on Collaborative Review Groups in The Cochrane Library.
Journal and trial databases
We will search the following electronic databases, in the period from 1 January 1996 to the search date:
CENTRAL (Cochrane Central Register of Controlled Trials)
Web of Science / Web of Social Science
World Health Organization (WHO) Global Health Library, which includes references from AIM (AFRO), LILACS (AMRO/PAHO), IMEMR (EMRO), IMSEAR (SEARO), and WPRIM (WPRO).
Along with appropriate MeSH terms and relevant keywords, we will use the Cochrane Highly Sensitive Search Strategy for identifying reports of randomised controlled trials in MEDLINE (Higgins 2008), and the Cochrane HIV/AIDS Group's validated strategies for identifying references relevant to HIV infection and AIDS. The search strategy will be iterative, in that references of included studies will be searched for additional references. All languages will be included.
See Appendix 2 for our PubMed search strategy, which will be modified and adapted as needed for use in the other databases.
We will search conference abstract archives on the web sites of the Conference on Retroviruses and Opportunistic Infections (CROI), the International AIDS Conference (IAC), and the International AIDS Society Conference on HIV Pathogenesis, Treatment and Prevention (IAS), for abstracts presented at all conferences from 1996 through 2013.
Searching other resources
In addition to searching electronic databases, we will contact individual researchers, experts working in the field and authors of major trials to address whether any relevant manuscripts are in preparation or in press. The references of published articles found in the above databases will be searched for additional pertinent materials.
We will search WHO’s International Clinical Trials Registry Platform (ICTRP) to identify ongoing trials.
Data collection and analysis
The methodology for data collection and analysis will be based on the guidance of Cochrane Handbook of Systematic Reviews of Interventions (Higgins 2008). Two authors will independently examine abstracts of all studies identified by electronic or bibliographic scanning. Where necessary, we will obtain the full text to determine the eligibility of studies for inclusion.
Selection of studies
One author will perform a broad first cut of all downloaded material from the electronic searches to exclude citations that are plainly irrelevant. Two authors will read the titles, abstracts and descriptor terms of the remaining downloaded citations to identify potentially eligible reports. We will obtain full text articles for all citations identified as potentially eligible, and two authors will independently inspect these to establish the relevance of the article according to the pre-specified criteria. Where there is uncertainty as to the eligibility of the record, we will obtain and review the full article.
Two authors will independently apply the inclusion criteria, and any differences arising will be resolved by discussion with a neutral arbiter. We will review studies for relevance based on design, types of participants and outcome measures.
Data extraction and management
Two authors will independently extract data into a standardised, pre-piloted data extraction form. The following characteristics will be extracted from each included study:
Study details: Complete citation, start and end dates, location, study design characteristics and other relevant details.
Details of the study: Study design; location and time-frame in which it was conducted; type of facility; investigators; other publications associated with the study; funding sources; etc.
Details of participants: Age range, gender, clinical staging, CD4 count, other pertinent details
Outcome details: Numerators and denominators associated with each outcome; effect estimates provided in papers; definitions of outcomes provided in papers; details of how outcomes were assessed.
Methodologic details: Trial design, recruitment, method of randomisation, the numbers of participants entering the trial, trial inclusion and exclusion criteria, length of follow up, losses to follow up, withdrawals or drop-outs.
Bias assessment data: Other details necessary to perform a bias risk assessment using the Cochrane tool described below.
Assessment of risk of bias in included studies
Two review authors will independently assess risk of bias for each study using the bias assessment tool described in the Cochrane Handbook (Higgins 2008). We will resolve any disagreement by discussion or by involving a neutral third party to adjudicate.
The Cochrane approach assesses risk of bias in individual studies across six domains: sequence generation, allocation concealment, blinding, incomplete outcome data, selective outcome reporting and other potential biases.
Sequence generation (checking for selection bias)
Low risk: investigators described a random component in the sequence generation process, such as the use of random number table, coin tossing, card or envelope shuffling.
High risk: investigators described a non-random component in the sequence generation process, such as the use of odd or even date of birth, algorithm based on the day or date of birth, hospital or clinic record number.
Unclear risk: insufficient information to permit judgment about the sequence generation process.
Allocation concealment (checking for selection bias)
Low risk: participants and the investigators enrolling participants cannot foresee assignment (e.g., central allocation; or sequentially numbered, opaque, sealed envelopes).
High risk: participants and investigators enrolling participants can foresee upcoming assignment (e.g., an open random allocation schedule, a list of random numbers), or envelopes were unsealed, non-opaque or not sequentially numbered.
Unclear risk: insufficient information to permit judgment of the allocation concealment or the method not described.
Blinding (checking for performance bias and detection bias)
Low risk: blinding of the participants, key study personnel and outcome assessor and unlikely that the blinding could have been broken. Not blinding in the situation where non-blinding is unlikely to introduce bias.
High risk: no blinding or incomplete blinding when the outcome is likely to be influenced by lack of blinding.
Unclear risk: insufficient information to permit judgment of adequacy or otherwise of the blinding.
Incomplete outcome data (checking for possible attrition bias through withdrawals, dropouts, protocol deviations)
Low risk: no missing outcome data, reasons for missing outcome data unlikely to be related to true outcome or missing outcome data balanced in number across groups.
High risk: reason for missing outcome data likely to be related to true outcome, with either imbalance in number across groups or reasons for missing data.
Unclear risk: insufficient reporting of attrition or exclusions.
Low risk: if a protocol is available, primary outcomes in the final trial report correspond closely to those presented in the protocol
High risk: the primary outcomes differ between the protocol and final trial report.
Unclear risk: no trial protocol is available or there is insufficient reporting to determine if selective reporting is present.
Other forms of bias
Low risk: there is no evidence of bias from other sources.
High risk: there is potential bias present from other sources (e.g., early stopping of trial, fraudulent activity, extreme baseline imbalance or bias related to specific study design).
Unclear risk: insufficient information to permit judgment of adequacy or otherwise of other forms of bias.
For blinding and incomplete outcome data, multiple entries can be made if more than one outcome (or time points) is involved.
We will assess the quality of evidence across the body of evidence using the GRADE approach (Guyatt 2011), which defines the quality of evidence for each outcome as “the extent of our confidence that the estimates of effect are correct." The quality rating across studies has four levels: high, moderate, low or very low. Randomised trials are considered to be of high quality but can be downgraded for any of five reasons; similarly, observational studies are considered to be of low quality, but can be upgraded for any of three reasons. The five factors that decrease the quality of evidence are 1) risk of bias; 2) indirectness of evidence; 3) unexplained heterogeneity or inconsistency of results; 4) imprecision of results and 5) high probability of publication bias. The three factors that can increase the quality level of a body of evidence are 1) large magnitude of effect; 2) if all plausible confounding would reduce a demonstrated effect and 3) the presence of a dose-response gradient.
Measures of treatment effect
For randomised controlled trials, we will calculate and present summary statistics for the risk ratio (RR) for dichotomous outcomes and the weighted-mean difference for continuous outcomes, using the 95% confidence interval (CI).
We will use the Review Manager 5 software (RevMan 2013) provided by the Cochrane Collaboration for statistical analysis and GRADEpro software (GRADEpro 2008) to produce GRADE Summary of Findings tables and GRADE evidence profiles.
If possible, we will calculate summary statistics using meta-analytic methods. To summarise evidence quality, we will present findings in GRADE Summary of Findings tables and GRADE Evidence Profiles for all outcomes of interest.
Unit of analysis issues
The unit of analysis will be the individual study participant. However, if studies have been clustered randomised by health facilities, such as clinics, for example, we will examine those separately in the case that individual-level data are not available.
Dealing with missing data
We will contact study authors if it is necessary to obtain data missing from published reports.
Assessment of heterogeneity
We will use the I2 statistic to measure heterogeneity among the trials in each analysis. If we identify substantial heterogeneity (I2 greater than 40%), we will explore it by pre-specified subgroup analysis. If heterogeneity persists, we will perform sensitivity analyses, present results separately and report reasons for the observed heterogeneity.
Assessment of reporting biases
Where we suspect reporting bias we will attempt to contact study authors and ask them to provide missing outcome data. Where this is not possible, and the missing data are thought to introduce serious bias, we will explore the impact of including such studies in the overall assessment of results by a sensitivity analysis.
We will assess the potential for publication bias for the studies using funnel plots. We will attempt to minimise the potential for publication bias by our comprehensive search strategy that includes evaluating published and unpublished literature.
We will conduct meta-analysis, if appropriate, using Cochrane's Review Manager software (RevMan 2013) and present results using the Mantel-Haenszel risk ratio. We will use both fixed and random effects models and conduct sensitivity analysis to explore differences between the two models. If meta-analysis is not possible, a narrative synthesis of studies will be undertaken. Data will also be presented using the GRADEpro software (GRADEpro 2008). GRADE evidence profiles and summary of findings tables will be generated.
When interventions and study populations are sufficiently similar across the different studies, we will pool the data across studies and estimate summary effect sizes using both fixed- and random-effects models. We intend to compare the estimates from fixed- and random-effects models in an attempt to explore the influence of small-study effects on results of a meta-analysis with intra-study heterogeneity. Specifically, we will estimate the log (risk ratio) for each included study and use the inverse variance method to calculate study weights. The inverse variance method assumes that the variance for each study is inversely proportional to its importance, therefore more weight is given to studies with less variance than studies with greater variance. If the estimates between the two modeling approaches are similar, then we can assume effects from small-studies only slightly affect the intervention's summary estimate. If estimates from random-effects are qualitatively substantially more beneficial than fixed-effects estimates, we will investigate whether the interventions were more effective in smaller studies than in larger studies. If upon reviewing the methodologies of the included studies we conclude that the larger studies were more rigorous, we may consider presenting only results from larger studies in a meta-analysis. As such, we intend to explore potential methodologic reasons for those differences in fixed- or random-effects estimates.
We will summarise the quality of evidence for the studies separately for each outcome for which data are available in GRADE Summary of Findings tables and GRADE evidence profiles (Guyatt 2013a, Guyatt 2013b).
Subgroup analysis and investigation of heterogeneity
Heterogeneity will be explored by analyses of the:
Type of Intervention
Region (e.g. sub-Saharan Africa, Southeast Asia etc.)
If pooled results are heterogeneous for the selected studies, we will conduct sensitivity analyses to identify studies with outlying results for further examination.