Criteria for considering studies for this review
Types of studies
Randomised controlled trials (RCTs)
Prospective cohort studies which followed children from enrolment to start of cARTand for at least a median of one year on cART
Non-comparative cohort studies in which children are all commenced on cART at enrolment regardless of clinical-immunological condition were excluded.
Types of participants
Children with confirmed HIV infection (HIV PCR-positive if diagnosed at age<18 months or HIV ELISA-positive if diagnosed after 18 months of age) aged between 24 and 59 months (two to five years in age) who are treatment-naïve (except for exposure to drugs to prevent mother-to-child transmission) at commencement of the study.
Studies which included the above age group within a larger age range were included. We contacted trial authors to obtain the specific age range 24 to 59 months.
Studies in children with tuberculosis and other opportunistic illnesses were included.
Studies focused only on infants (less than one year in age) were excluded. These are included in another Cochrane review (Penazzato 2012).
For RCTs, at study entry, children must not have fulfilled guideline criteria for starting antiretroviral treatment recommended at the time of the study.
Types of interventions
Studies which compare initiating triple cART at different thresholds.
In RCTs, the intervention group must have initiated triple cART irrespective of clinical stage or CD4 count (immediate initiation). The control group must have initiated cART using clinical and immunological criteria to determine the time for initiation (deferred initiation) as recommended at the time of the study.
In cohort studies, children starting triple cART at different CD4 percent or CD4 counts must have been compared. Analysis must have included adjustment for time-dependent confounding and lead-time bias specifically to evaluate the effect of timing of cART initiation on outcomes.
Types of outcome measures
Clinical occurrence of new HIV-related events (death or AIDS-defining illness)
Time to event of new HIV-related events (death or AIDS-defining illness)
Immunological response (change in mean or median CD4+ cell count (mean or median relative change (percent) or mean or median absolute change, compared with baseline, and standard deviation or range as appropriate)
Virologic response (proportion of patients achieving and maintaining an undetectable viral load, as defined by the investigators; change in HIV-RNA levels (mean relative change (percent) or mean absolute change, compared with baseline, and standard deviations)
HIV drug resistance
Severe adverse events were reported. If classified according to grade 1 to 4 of the Adverse Event Toxicity Scale, we reported grade 3 and 4 events. Using this scale, grade 1 and 2 denote mild to moderate symptoms, grade 3 denote serious symptoms and grade 4 denote life-threatening events requiring significant clinical intervention. Grade 5 denotes death (DAIDS 2009).
Search methods for identification of studies
See: HIV/AIDS Collaborative Review Group search strategy. The search aimed to be comprehensive and included published and unpublished studies and was not limited to any language.
We developed the search strategy with the assistance of the HIV/AIDS Review Group Trials Search Co-ordinator. We formulated a comprehensive and exhaustive search strategy in an attempt to identify all relevant randomised controlled trials and cohort 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 inThe Cochrane Library.
For the RCT search, we combined the RCT strategy developed by The Cochrane Collaboration and detailed in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2009) in combination with terms specific to initiation of antiretroviral therapy and children. The search was iterative and a number of trial searches were run first as there are no database-specific terms for 'initiation' of treatment and so we used many free text terms. This increased the yield and hence the search sensitivity but reduced the precision.
As there are no validated search strategies for cohort studies, our strategy was informed by the cohort strategy developed by BMJ Clinical Evidence (http://clinicalevidence.bmj.com/x/set/static/ebm/learn/665076.html).
We searched the following databases:
1. Journal databases
For RCT Identification:
Medline via Pubmed - see search strategy conducted on 27 June 2012 in Appendix 1
EmBase - see search strategy conducted on 27 June 2012 in Appendix 2
Cochrane Central Register of Controlled Trials (CENTRAL) - see search strategy conducted on 27 June 2012 in Appendix 3
For cohort identification:
2. Conference databases:
We searched the AEGIS database (www.aegis.org) on 24 July 2012. www.aegis.org contained abstracts from the following major related conferences: 1st-5th International AIDS Society (IAS) Conference on HIV Pathogenesis and Treatment and Prevention (2001-2009); 10th-17th International AIDS Conference (IAC) (1994-2008); 1st-16th Conference of Retrovirus and Opportunistic Infections (CROI) (1994-2009); US National HIV Prevention Conference; 7th-14th British HIV Association (2001-2008); and 8th-9th European AIDS Clinical Society Conference (2001, 2003).
AEGIS does not allow for multiple search strings and so, in order to search the database effectively, a separate search was done for relevant terms combined with the term [random*] to identify relevant trials:
ANTIRETROVIRAL* AND RANDOM* AND (child* OR infant* OR toddler* OR pediatric* OR paediatric*)
The more recent conferences (up to 2012) were covered by searching the conference web-sites of the International AIDS Society, the International AIDS Conference and the CROI. The abstracts of the 1st, 2nd, 3rd and 4th International Workshop on HIV Pediatrics were hand-searched in Reviews in Antiviral Therapy & Infectious Diseases where they are published following the conference.
3. Ongoing trials:
To identify ongoing RCTs we searched ClinicalTrials.gov (http://clinicaltrials.gov/) on 9 July 2012 and the World Health Organization International Clinical Trials Registry Platform (http://apps.who.int/trialsearch/) on 12 July 2012.
Searching other resources
We also checked the reference lists of all studies identified by the above methods and examined any systematic reviews, meta-analyses, or guidelines we identified during the search process for references.
We were in close contact with individual researchers working in the field, and policymakers based in inter-governmental organizations including the World Health Organization (WHO) and UNAIDS.
We did not conduct hand-searching of specific journals other than those searched by the Cochrane HIV/AIDS Review Group and already included in CENTRAL.
Data collection and analysis
Selection of studies
For the RCT search of journal databases, NS and ME read the titles, abstracts and descriptor terms of all downloaded material from the electronic searches to identify potentially eligible reports. Full text articles were be obtained for all citations identified as potentially eligible and NS and ME independently inspected these to establish the relevance of the article according to the pre-specified criteria. Where there was any uncertainty as to the eligibility of the record, we obtained the full article.
NS and ME independently applied the inclusion criteria, and any differences arising were resolved by discussions with the third reviewer, MD. Studies were reviewed for relevance based on study design, types of participants, exposures and outcome measures.
For the additional cohort search of journal databases, NS read the titles, abstracts and descriptor terms of all downloaded material from the electronic searches to identify potentially eligible reports. After identification of potentially eligible articles, ME checked these and any uncertainty was discussed and resolved by discussions with the third review author, MD.
For all conference abstract searching, NS read the titles, abstracts and descriptor terms of all downloaded material and manually hand-searched conference abstract books when it was not possible to search the web-site effectively. NS contacted authors of abstracts to confirm whether the study had been published and/or whether a final report was available. Abstracts which described studies already identified in the electronic searches were linked to these.
Data extraction and management
NS independently extracted data into a standardised data extraction form. ME checked the data extraction independently. The following characteristics were extracted from each included study.
Administrative details: Trial or study identification number; author(s); published or unpublished; year of publication; number of studies included in paper; year in which study was conducted; details of other relevant papers cited;
Details of the study: Study design; type, duration and completeness of follow-up; country and location of study (e.g. higher-income vs. lower-income country); informed consent and ethics approval;
Details of participants: Setting, numbers, relevant baseline characteristics including CD4 count and viral load;age range;
Details of intervention: CD4 count and age at which treatment was initiated; drug combinations; additional co-interventions; and
Details of outcomes: Mortality; HIV-related morbidity; HIV-RNA viral load measurements and proposed levels for suppression, as defined by the authors; clinical disease progression; CD4+ cell counts; adverse events and toxicity.
Details of the analysis: For RCTs, details of the type of analysis (intention-to-treat or per protocol); for cohort studies, details of the type of adjustment performed in the analysis.
Assessment of risk of bias in included studies
For both RCTs and cohort studies, NS and ME independently examined the components of each included study for risk of bias using a standard form. This includes information on the sequence generation, allocation concealment, blinding (participants, personnel and outcome assessor), incomplete outcome data, selective outcome reporting and other sources of bias. The methodological components of the trials were assessed and classified as at a high, low or unclear risk of bias as per the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2009). Where differences arose, these were resolved by discussions with the third reviewer, MD.
Adequate: investigators described a random component in the sequence generation process such as the use of random number table, coin tossing, cards or envelops shuffling etc
Inadequate: 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/date of birth, hospital or clinic record number
Unclear: insufficient information to permit judgment of the sequence generation process
Adequate: participants and the investigators enrolling participants cannot foresee assignment, e.g. central allocation; or sequentially numbered, opaque, sealed envelopes.
Inadequate: participants and investigators enrolling participants can foresee upcoming assignment, e.g. an open random allocation schedule (e.g. a list of random numbers); or envelopes were unsealed or nonopaque or not sequentially numbered
Unclear: insufficient information to permit judgment of the allocation concealment or the method not described
Adequate: blinding of the participants, key study personnel and outcome assessor, and unlikely that the blinding could have been broken. Or lack of blinding unlikely to introduce bias. No blinding in the situation where non-blinding is not likely to introduce bias.
Inadequate: no blinding, incomplete blinding and the outcome is likely to be influenced by lack of blinding
Unclear: insufficient information to permit judgment of adequacy or otherwise of the blinding
Incomplete outcome data
Adequate: 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
Inadequate: 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: insufficient reporting of attrition or exclusions
Adequate: a protocol is available which clearly states the primary outcome as the same as in the final trial report
Inadequate: the primary outcome differs between the protocol and final trial report
Unclear: no trial protocol is available or there is insufficient reporting to determine if selective reporting is present
Control of time-dependent confounding (cohort studies only)
The use of standard regression models for the analysis of cohort studies with time-updated measurements may result in biased estimates of treatment effects if time-dependent confounders are present (Robins 2000). In cohorts of HIV-infected patients CD4 cell percentage or CD4 cell counts are measured regularly to assess the patients' eligibility for cART and, once on cART, to monitor therapy. CD4 count is a time-dependent confounder because it predicts both future cART and outcome, and is influenced by past antiretroviral therapy. In other words, CD4 count is on the causal pathway between treatment and the outcome.
Adequate: appropriate methods were used to control for time-dependent confounding (e.g. marginal structural models)
Inadequate: there was no control for time-dependent confounding
Unclear: insufficient reporting to determine whether there was control for time-dependent confounding
Other forms of bias
Adequate: there is no evidence of bias from other sources
Inadequate: 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: insufficient information to permit judgment of adequacy or otherwise of other forms of bias
Measures of treatment effect
Data analysis was conducted using (RevMan) version 5.1.7.
For RCT data, outcome measures for dichotomous data (e.g. death, virological suppression) were calculated as relative risks with 95% confidence intervals. For continuous data (e.g. CD4+ cell counts, HIV-RNA viral loads) we calculated the mean difference and standard deviation where means were reported. Where medians and ranges were reported, we extracted the data directly from the report.
For cohort data, we reported on the adjusted analysis using the estimate of effect reported in the study. Where the adjusted estimate of effect was reported with 95% Confidence Intervals (CI), we calculated the Standard Error in order to enter the data into RevMan, using the following formula for ratio measures:
lower limit = ln(lower confidence limit given for HR)
upper limit = ln(upper confidence limit given for HR)
intervention effect estimate = lnHR
SE = (upper limit – lower limit) / 3.92
Unit of analysis issues
We did not anticipate cluster or cross-over trials, with the potential for unit of analysis issues, being conducted to investigate this question.
Dealing with missing data
Where data was missing or required clarification, we contacted study authors to request additional data. Should this not have been possible, we would have stated explicitly where calculations were based on assumptions regarding missing data.
In trials which included participants from within a larger age range, we contacted study authors to obtain the specific age range 24 to 59 months.
Assessment of heterogeneity
Where RCTs were found to be methodologically or clinically comparable, we pooled trial results in a meta-analysis. As we anticipated the presence of statistical heterogeneity we combined the data using the random effects model throughout.
Only one cohort study was included in the review, so it was not applicable to assess heterogeneity. In future updates of this review, where cohort studies are found to be methodologically or clinically comparable, we will pool the results in a meta-analysis using the generic inverse variance function in RevMan to allow adjusted data to be used in the analysis. As we anticipate heterogeneity due to the likelihood of different analytical techniques and different adjusted variables, we will combine studies using the random effects model.
For the RCT meta-analyses, we formally tested for statistical heterogeneity using the Chi-square test for statistical homogeneity with a 10% level of significance as the cut-off. The amount of any statistical heterogeneity was quantified using the I² statistic (Higgins 2002).
Where studies do not have combinable outcomes, we provide the data in a narrative form.
For meta-analysis of RCTs, we combined the results and calculated the overall relative risk and 95% confidence intervals using the Mantel-Haenszel random-effects model. For continuous data, we combined the mean differences to calculate a weighted mean difference and standard deviation, using the inverse-variance random-effects model. Peto fixed-effects models were used in sensitivity analyses. If time-to-event data is available in future updates, we will combine the hazard ratios reported in the RCTs using the inverse-variance random-effects model.
For cohort studies in future updates, we will combine the adjusted estimates of effects using the inverse-variance random-effects model.
Subgroup analysis and investigation of heterogeneity
We anticipated statistical heterogeneity due to differences between studies conducted in resource-constrained compared with resource-rich settings, and planned to present the results according to these sub-groups should this have been necessary.
We performed subgroup analyses restricted to children less than 5 years old who had a CD4 above 25% to best inform the current (2012) WHO guideline processes.
For RCTs, we planned to explore the effect of study quality on the results by excluding those studies where allocation concealment was unclear or inadequate from the meta-analysis and assessing the effect of this on the overall results. For cohort studies we planned to examine the effect of adjustments for confounding and in particular confounding by indication.
We also conducted meta-analyses using the Peto Odds Ratio in addition to the RR for those comparisons where the event rate was extremely low.
GradePro 2008 was used to create Summary of Findings and Evidence Profile tables. GradePro software was developed as part of a larger initiative led by the Grading of Recommendations Assessment, Development and Evaluation (GRADE) Working Group. GRADE offers a system for rating quality of evidence in systematic reviews and guidelines and grading strength of recommendations in guidelines (Guyatt 2011). Use of GradePro within a Cochrane systematic review facilitates the process of presenting and grading evidence transparently (http://ims.cochrane.org/revman/other-resources/gradepro/about-gradepro).
In determining the level of evidence for each outcome, both the efficacy results and the assessment of the risk of bias was integrated into a final assessment of the level of evidence and full details of the decision was provided in footnotes.