This systematic review and meta-analysis adheres to the reporting guidelines of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) [25]. The protocol of the systematic review and meta-analysis was registered in the International Prospective Register of Systematic Reviews (PROSPERO) database with (https://www.crd.york.ac.uk/prospero/display_record.php? ID = CRD42024533269).

A comprehensive search was conducted in five selected databases, namely PubMed, EMBASE, Scopus, Web of Science, and Cochrane Central Library. In addition, a systematic search was conducted in Google, Google Scholar, and institutional repositories to access gray literature. A combination of keywords and controlled vocabularies such as Medical Subject Headings (MeSH) in PubMed and Emtree in Embase was used in the searching process. The terms (Retention, retain*, “Retention in Care,” Adolescent*, Young adult*, Youth*, “Living with HIV,” “HIV positive,” “HIV infected,” Transition, Transfer, Link, Movement, “Transition to Adult Care,” “Patient Transfer”) with Boolean operators (OR, AND, and NOT), field codes, wildcards, brackets, and quotation marks were used to search in each database (Supplementary Table S1).

This systematic review and meta-analysis included studies that met specific inclusion criteria. The study population consisted of adolescents and young adults aged 10–25 years living with HIV who had transitioned from pediatric to adult HIV care. Additionally, only studies with cross-sectional and cohort study designs were considered. Finally, only cohort studies that fulfilled the eligibility criteria were included. The review focused on studies conducted after January 2015 and required that all included studies be published in the English language.

Conversely, studies were excluded if they met any of the following conditions: first, adolescents and young adults aged 10–25 years living with HIV who had not transitioned from pediatric to adult HIV care were excluded, particularly those with missing information about their transition. Second, studies that utilized designs other than cross-sectional and cohort studies were not included. Third, all studies conducted before January 2015 were excluded. Fourth, studies published in languages other than English were not considered. Finally, studies that were not available in full-text or had insufficient data for analysis were also excluded. The reasons for excluding primary studies from the review and analysis are presented in Supplementary Table S2.

All articles retrieved from the databases and other sources were downloaded and exported into the Covidence software [26]. Covidence is well known for its user interface and functionalities that promote collaboration among reviewers. It facilitates the removal of records of aids, in screening titles abstracts, and full texts efficiently. Additionally, we used covidence for data extraction, quality of articles evaluation, and generation of PRISMA flow diagrams.

For this research project, two independent reviewers (MS and AG) screened titles, abstracts, and full texts. The screening process was streamlined using Covidence’s features such as blinded assessments and conflict resolution mechanisms. Any disagreements between the reviewers were resolved by a reviewer (TG) who made decisions to ensure consistency and accuracy, in the review process.

Data were extracted from articles that met the inclusion criteria. We used Covidence to extract data containing the following categories: Name of the Author, Year of publication, Study Setting, Study Design, Study Population, Total number of transitions, Total number of Retention, and Retention rate. The data were extracted by two reviewers together to facilitate data extraction and minimize errors. We used The Joanna Briggs Institute (JBI) critical appraisal tool to assess the quality of our papers [27, 28]. The tool has eleven items to measure the quality of cohort studies. An article with a total JBI quality score greater than 50% was considered good quality and was included in our data analysis (Supplementary Table S3).

The extracted data were exported to Stata version 18 for data synthesis and statistical analysis. The characteristics of the extracted data were presented in the table and a summary of the pooled estimate was presented graphically in a forest plot. The random effects model was used because of the high level of heterogeneity between studies. The presence of statistical heterogeneity among the included studies was determined using the Higgins I2 statistics and the Cochran-Q test, with I2 values of 25%, 50%, and 75% being low, medium, and high respectively. Additionally, subgroup analyses by geographic region, study design, publication, median age at the time of care transition, and population type were performed to examine the variation in estimates among the categories and sensitivity analyses were also conducted to check the effect of a single study on the overall pooled estimate. The Freeman-Tukey double arcsine transformation was used to generate the retention rate with a 95% confidence interval and to stabilize the variance of each study’s proportion [29]. Publication bias was assessed graphically using a funnel plot and statistically using Egger’s regression test. The results were presented using text, figures, and tables.

A total of 1964 studies (1955 studies from five databases and 9 studies from other sources) were identified. Of the total studies 1,036 studies were identified as duplicates (1,030 studies by COVIDENCE software (https://www.covidence.org/) and 6 studies manually) and 928 studies passed for title and abstract screening. Of the studies that passed the title and abstract screening 894 studies were excluded as irrelevant and 34 studies passed the full-text screening. Finally, out of the 34 studies, 15 studies were included in the systematic review and meta-analysis, with the remaining 19 studies excluded (eleven of them did not publish the results, five of them had the wrong population, one study did not have the full text, one had the wrong study design, and one was not in English) (Figure 1).

A total of fifteen studies that met the eligibility criteria were used for this systematic review and meta-analysis with a total of 13,764 study participants. Five studies were conducted in the United States [Griffith, 2019 #11; Hussen, 2017 #12; Hussen, 2023 #19; Nassau, 2022 #17; Tanner, 2018 #10], seven studies in Africa [Abebe, 2021 #8; Castelnuovo, 2018 #20; Davies, 2017 #21; Meloni, 2020 #9; Ouedraogo, 2024 #18; Tsondai, 2020 #16; Zanoni, 2021 #14], one study in Canada [Kakkar, 2016 #15], one study in Asia [Rungmaitree, 2022 #7], and one study in the United Kingdom [Foster, 2020 #13]. Of these four were prospective cohort studies and eleven were retrospective cohort studies. Nine studies were published in or after 2020 and the remaining six were published before 2020. The studies were conducted among adolescents and young adults aged 10–25 years using sample sizes ranging from 19 to 11,283 subjects. Twelve studies were conducted among adolescents and three of them among young adults. Thirteen studies reported the median age at transition of care, ranging from 12.9 years (11.4–15.3) to 24.4 years [24, 25]. The studies included a total of 13,756 individuals with 6,373 (46.3%) male subjects and 7,383 (53.7%) female subjects (Table 1).

Fourteen studies were used to estimate a pooled 1-year retention rate of adolescents and young adults after the transition to adult HIV care. The pooled retention rate was 81% (95% CI: 72%, 88%) with point retention rates ranging from 37% (95% CI: 29%, 46%) to 92% (95% CI: 86%, 97%). There was a statistically significant higher degree of heterogeneity among the studies (I2 = 96.73%, P < 0.001). Therefore we used a random effects model to adjust for the heterogeneity between studies (Figure 2).

Seven studies were used to estimate the pooled 2-year retention rate of adolescents and young adults after transition to adult HIV care. The pooled retention rate was 69% (95% CI: 53%, 83%) with point retention rates ranging from 45% (95% CI: 39%, 52%) to 93% (95% CI: 86%, 98%). There was a statistically significant higher degree of heterogeneity among the studies (I2 = 97.39%, P < 0.001). Therefore we used a random effects model to adjust for the heterogeneity between studies (Figure 3).

Sub-group analysis was performed by geographic regions (Africa, Asia, Europe, and North America), study design (Prospective or Retrospective), year of publication (before 2020 and after 2020), median age of transition (before 20 years or after 20 years) and population type (adolescents or young adults) (Supplementary Figures S1–S10). A statistically significant difference was observed only by geographical region for both the 1-year and 2-year retention rates (P < 0.08 and P < 0.01, respectively). The pooled 1-year retention rate was 85% (95% CI: 78%, 91%) in Africa, 87% (95% CI: 80%, 93%) in Asia, 88% (95% CI: 83%, 92%) in Europe, and 73% (95%CI: 55%, 88%) in North America while the pooled 2-year retention rate was 73% (95%CI: 44%, 94%) in Africa, 92% (95%CI: 86%, 97%) in Asia, and 57% (95%CI: 41%, 72%) in North America (Supplementary Figures S1, S2).

Furthermore, meta-regression was performed on both 1-year and 2-year retention rates to determine the effect of covariates on the pooled estimates and only study design was found to be a significant factor affecting the pooled estimates for 1-year retention rates. Retrospective studies had a positive relation with the pooled estimates (with coefficient = 0.429, and P-value = 0.035) (Supplementary Table S4).

Publication bias was assessed graphically using a funnel plot and statistically using Egger’s test for both 1-year and 2-year retention rates. Although a mild asymmetry was observed in the funnel plot (Figures 4, 5). Egger’s test showed the absence of publication bias with a p-value of (P = 0.6393 and P = 0.1878, in 1-year and 2-year retention rates respectively (Supplementary Tables S5–S6). Sensitivity analysis was performed to assess the effect of a single study on the pooled analysis. The results indicate that no single study significantly impacts the pooled estimates for both 1-year and 2-year retention rates, meaning that the overall pooled estimates remain stable across studies (Supplementary Figures S11, S12).

This systematic review and meta-analysis has several limitations. The original studies that were included had varying follow-up durations, leading to differences in outcomes related to healthcare retention. Additionally, differences in HIV care guidelines and recommendations between studies may have influenced treatment outcomes and retention rates. Furthermore, this review was potentially affected by bias due to the dominance of retrospective study designs, with limited representation from other designs, such as prospective cohort studies.

More than four out of five adolescents and young adults were retained in HIV care 1 year after transition to adult care, with retention dropping to two-thirds by the second year. This highlights the need for further efforts to achieve the ambitious 95-95–95 targets of the global ART program. Given changes in treatment eligibility criteria over time and varying definitions of healthcare retention, it is critical to standardize how retention is operationalized. Additionally, close and frequent follow-up, along with robust mechanisms to trace defaults, is strongly recommended to improve retention rates.