Protocol development and registration

This review was designed in accordance with preferred methods of reviewing available Systematic Review and Meta-analysis (SRM) studies and the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [10, 11]. First, a similar review was checked on PROSPERO, and no similar studies were found. The protocol of this review was subsequently summarized and registered as CRD42024499263 in PROSPERO. PROSPERO registration -related information is available upon reasonable request from the primary author. This systematic review and meta-analysis focused on a systematic synthesis of existing studies on the virological outcomes of third-line ART in a global context.

Search strategy and information sources

A comprehensive literature search was conducted for studies reported virological outcomes of third-line ART in the Embase, Web of Science, PubMed, Scopus, International Scientific Indexing (ISI), and Google Scholar databases. using the PICO frameworks. Combinations, keywords and MeSH terms were used to retrieve the studies. In addition, the snowballing technique was used to retrieve additional studies from the citation lists of the articles found in the available databases. Gray literature and manual searches were also performed to find unindexed/not published/researched articles. The search strategies were drafted using concepts and key search terms. The first concept: virological outcome: “treatment outcome”, “treatment responses”, “effectiveness”, “virological suppression”, and “outcome”. The second concept included the following: “third-line regimen”, “third-line therapy”, “third-line antiretroviral therapy”, “third-line highly active antiretroviral therapy”, and “third-line HAART”. Literature searches were independently conducted by two authors (TAK and RNH). Any inconsistency was resolved by agreement. In the case of articles with incomplete information, the primary authors of the respective article were contacted. We used the search terms “OR” or “AND” independently and/or in combination. In addition, ‘related article’ and ‘citied by’ features of the PubMed database were used to find articles from the included studies.

Eligibility criteriaInclusion criteria

A study reporting the virological outcome of third line ART written in the English language was included. For respective study to be considered for this systematic review and meta-analysis, it should fulfill the following prioritized criteria. Condition: The outcome of interest should be measured as the virological outcome of third line ART. Context: Setting can be anywhere. Cross-sectional studies and cohort studies (prospective and retrospective) were eligible. Population: HIV-1 infected individuals who were receiving third line ART. All population group can be included without age restrictions.

Exclusion criteria and definitions

Articles were excluded for one of the following reasons: (1) did not measure the outcome of interest for this systematic review and meta-analysis, (2) were written in languages other than English; or (3) were narrative reviews, expert opinions, case reports, editorials, correspondences, abstracts, or methodological studies.

Data extraction and management

Two authors (BBA and RNH) conducted the data extraction independently using a standardized extraction form. The title and abstract were first screened and selected, after which the full texts were reviewed. In the case of disagreement, discussion with other reviewers was performed to determine the final selection of articles to include in this review. After the systematic search was complete, potentially eligible articles were imported to EndNote 21. Duplicated studies were removed if two or more articles had common characteristics. The structured data were extracted in the form of a Microsoft Excel spreadsheet. The extracted data included the following: (1) study identification (last name of the primary author and year of publication), (2) setting, (3) sample size, (4) study design, (5) setting, (6) age range of the participant, (7) virological suppression, (8) cutoff point for virological suppression (viral load (VL) copies/ml), and (9) time at which virological suppression was detected. In addition, the percentage of virological suppression from each study was computed using the number of participants declaring suppressed viral RNA/ml as the numerator and the total number of sample sizes as the denominator. The corresponding author was contacted when any difficulties were encountered during data extraction.

Risk of bias assessment

The Joanna Briggs Institute (JBI) critical appraisal tool provided a measure of methodological quality for this systematic review and meta-analysis. Two independent reviewers evaluated each study using a series of “Yes,” “No,” or “Unclear” questions. To ensure objectivity, any disagreements were resolved through consensus among the authors and an independent reviewer. A scoring system (1 for “Yes,” 0 for “No,” U for “Unclear”) was applied, with final scores converted into percentages for risk-of-bias ranking: ≤49% (high), 50–69% (moderate), and above 70% (low). Only studies scoring at least 50% (indicating moderate or low risk of bias) were included. For ongoing reviewer disputes, individual ratings were averaged. The quality of each primary study’s results was documented in a dedicated column within the data extraction form to facilitate further analysis.

Statistical analysis

Once the data extraction was completed in Microsoft Excel, the data were imported to STATA version 17 software for analysis. Qualitative and narrative methods were employed to summarize the estimates of the included studies. When two or more estimates on the same topic were found, the range of the estimate and/or pooled estimate was used. The standard error was computed by considering a binomial distribution formula. The overall virological outcome (suppression) was pooled using a random effects model [12]. In addition, the pooled estimates were presented by using a forest plot. Cochrane’s Q statistics (chi-square), inverse variance (I2) and p-values [13] were computed to show the level of heterogeneity between studies. Zero invers variance (I2) revealed true homogeneity, whereas 25%, 50% and 75% had low, moderate and high heterogeneity, respectively [14, 15]. Subgroup analysis was performed according to publication year, age of the participant, study quality (JBI quality score), country income level, duration of viral load suppression and number of viral RNA copies. Leave one out (sensitivity) meta-analysis was performed to determine the effect of a single study on the overall pooled estimation. A funnel plot was constructed, and Egger’s regression test was used to determine publication bias [16].

Prediction interval

A prediction interval was computed to estimate how much variation we can expect in the results of a new study if that study was randomly chosen from the same group of studies used in the current analysis. This interval helps us understand how much the combined result might vary depending on the specific new study included [17].

Assessment of the quality of evidence

We employed a powerful tool called the GRADE tool (Grading of Recommendations Assessment, Development and Evaluation) to check how confident the pooled estimate is. Evidence was assessed based on five main domains (risk of bias, consistency, directness, precision, and publication bias) [18]. (Table 1).

Table 1 GRADE Quality of Evidence Interpretations