Characteristics of included studies

The systematic search identified 8590 records, with 14 RCTs (2005–2024) [24,25,26,27,28,29,30,31,32,33,34,35,36,37] meeting inclusion criteria (PRISMA Flowchart, Fig. 1), collectively enrolling 2822 participants (Intervention: 1203 vs. Controls: 1619). Geographically, 12 trials originated from the United States, predominantly led by the University of Texas Medical Branch (UTMB)/Shriners Hospitals consortium (8 trials, including pivotal works by Herndon et al. 2005–2016) [30], with single contributions from Egypt (Ali et al. 2022, evaluating nandrolone decanoate) [35] and Indonesia (Gusti et al. 2022, testing oxandrolone + enhanced nutrition) [36]. Methodologically, 10 were single-center RCTs, while 2 represented multicenter designs: Wolf et al. [25] (2006; 14 U.S. burn centers, n = 81 adults) and Pham et al. [27] (2008; 4 centers, n = 113 elderly patients). Two were post-hoc analyses of prospective trials (Hundeshagen et al. 2024, re-analyzing Porro et al. 2012 data) [29, 37]. Pediatric populations dominated (9 trials, mean age 9 ± 5 years, TBSA 40–61%), 4 trials focused on adults (median age 54 years, TBSA 20–60%). Interventions primarily used oxandrolone monotherapy (9 trials: 0.1 mg/kg bid in pediatrics, 10 mg bid in adults), while 5 evaluated combinations: oxandrolone + propranolol (4 trials, e.g., Jeschke et al. 2007) [26], oxandrolone + propranolol + exercise (Gusti et al. 2022) [36], or nandrolone decanoate (50 mg/week IM; Ali et al. 2022) [35]. Controls received placebo (10 trials), standard care (3 trials), or propranolol alone (1 trial; Herndon et al. 2016) [30]. Treatment duration spanned acute phases (14 days–3 months; Wolf et al. 2006) [25] to long-term rehabilitation (6–24 months; Reeves et al. 2016). All trials reported ≥ 1 predefined outcome: Length of stay (LOS) (10 trials, e.g., Wolf et al. 2006) [25], LOS/TBSA (6 trials, e.g., Przkora et al. 2005), lean body mass (4 trials, measured via DEXA in recovery phase ≥ 14 days post-burn; Tuvdendorj et al., 2011), mortality (7 trials), surgical procedures (5 trials, including Wolf et al. reporting mean 2.2 vs. 4.0 operations), treatment-related side effects (9 trials, e.g., Przkora et al. (2005) [24] documenting perineal edema in 3 females; Wolf 2006 reporting ALT > 100 U/L in 19.1%), and infections (3 trials using ABA criteria; Jeschke et al. 2007 [26]). Funding sources were non-commercial (NIH/Shriners: 11 trials; government grants: 3), with no industry sponsorship.

Fig. 1

PRISMA Flow Diagram of Study Selection Process. This flow diagram illustrates the process of study selection for the meta-analysis, following the PRISMA guidelines. It shows the number of studies identified, screened, eligible, and included in the final analysis, along with the reasons for exclusion at each stage. The diagram provides a transparent overview of the study selection process and the number of studies remaining at each step

Risk of bias and quality assessment

Methodological rigor across all 14 included studies was comprehensively appraised per outcome using the Cochrane RoB 2.0 tool. The assessment revealed an overall robustness in the synthesized evidence, albeit with specific domain limitations detailed below and summarized in Fig. 2 and Table 2:

Fig. 2

Risk of bias charts. A Risk of bias in the included studies. B Risk of bias summary for all included studies

Table 2 Overall risk of bias

Randomization Process: Low risk was identified in 10 trials, characterized by computer-generated sequences with robust allocation concealment (e.g., Wolf et al. 2006: pharmacy-controlled randomization; Reeves et al. 2016: central web-based system). Some concerns arose in 4 trials: Gusti et al. (2022) utilized block randomization without allocator blinding; Ali et al. (2022) employed "sealed envelopes" without adequate documentation;

Deviations from Intended Interventions: Low risk predominated in 12 trials, primarily due to effective double-blinding with matched placebos (e.g., Przkora et al. 2005: identical tablets; Porro et al. 2012: blinded dispensers). Some concerns emerged in 2 trials: Hundeshagen et al. (2024; post-hoc analysis) due to incomplete adherence logs in its parent trial, and notably Pham et al. (2008) where an 18% crossover rate from control to intervention group occurred, unaddressed by intention-to-treat (ITT) analysis.

Missing Outcome Data: Low risk was observed in 13 trials, defined by attrition rates ≤ 10% with balanced causes across groups (e.g., Reeves et al. 2016: 6% loss in intervention vs. 8% in controls). A single trial, Tuvdendorj et al. (2011), exhibited high risk due to substantial (23%) and imbalanced attrition (3/13 controls vs. 1/13 oxandrolone group) without sensitivity analysis to assess its impact.

Outcome Measurement: Low risk was assessed for 12 trials utilizing objective, standardized metrics (e.g., DEXA for body composition, laboratory records for biomarkers, operative logs for procedures). Some concerns were noted in 2 unblinded trials: Ali et al. (2022) for subjective assessment of wound healing by unblinded assessors, and Gusti et al. (2022) for infection diagnosis lacking standardization against American Burn Association (ABA) criteria.

Selection of Reported Results: Low risk was attributed to 11 trials with prospective registration and protocol adherence (e.g., NCT00675714 for Reeves et al. 2016; NCT00003057 for Wolf et al. 2006). Some concerns were identified in 3 trials: Jeschke et al. (2007) omitted pre-specified lean mass outcomes from their unpublished protocol; Pham et al. (2008) failed to report a promised age-stratified analysis; Murphy et al. (2004) post-hoc changed the primary endpoint from muscle strength to lean body mass (LBM).

Overall Risk of Bias: Aggregating domain judgments (see Table 2 for study-level details), 5 trials (35.7%; Reeves et al. 2016, Wolf et al. 2006, Porro et al. 2012, Herndon et al. 2013, Hart et al. 2001) demonstrated low risk across all domains. Eight trials (57.1%; Przkora et al. 2005, Jeschke et al. 2007, Murphy et al. 2004, Hundeshagen et al. 2024, Pham et al. 2008, Ali et al. 2022, Gusti et al. 2022) were judged to have some concerns, often isolated to specific domains like outcome measurement (Ali/Gusti) or reporting transparency (Jeschke/Pham). Only one trial (7.1%; Tuvdendorj et al. 2011) exhibited high overall risk, primarily driven by problematic missing outcome data.

Critical Observations & Patterns: (1) Attrition bias was isolated to Tuvdendorj et al. (2011); (2) Performance bias was prominent in trials lacking assessor blinding (Ali/Gusti et al. 2022); (3) Reporting bias issues were more frequent in non-industry funded trials (Jeschke et al. 2007; Pham et al. 2008); (4) Trials led by the UTMB group (Wolf et al. 2006, Reeves et al. 2016, Herndon et al. 2013) consistently demonstrated exemplary methodological rigor.

This detailed, study-specific assessment, quantifying risks (e.g., crossover rates, attrition percentages) and highlighting methodological nuances (e.g., blinding methods, registration status), strengthens confidence in the validity of the primary findings while providing essential context for interpreting potential limitations within the evidence base.

Synthesis of resultsBody composition and muscular outcomes

Lean Body Mass (LBM) and Muscle Strength:

Oxandrolone consistently improved LBM across pediatric and adult populations. In children with severe burns (≥ 40% TBSA), 12-month treatment increased LBM by 25% vs. placebo (p < 0.05, Przkora et al. 2005), with synergistic effects when combined with propranolol (+ 12.3% vs. oxandrolone alone, Porro et al. 2012). Muscle strength gains peaked at 12 months (182 Nm/kg vs. 140 Nm/kg in controls; p < 0.05), but effects diminished post-discontinuation. Adults receiving acute-phase oxandrolone (10 mg bid) required fewer surgical procedures (2.2 ± 0.3 vs. 4.0 ± 0.6; p = 0.015), indicating accelerated wound healing and functional recovery (Wolf et al. 2006).

Protein Metabolism:

Stable isotope studies revealed oxandrolone’s mechanism: It enhanced net muscle protein deposition by suppressing breakdown during amino acid infusion (change: − 4% vs. + 36% in controls; p < 0.05), not by elevating synthesis (Tuvdendorj et al. 2011). Whole-body protein breakdown remained unaffected, highlighting tissue-specific anabolic effects.

Bone metabolism and growth

Bone Mineral Density (BMD):

Long-term oxandrolone (24 months) yielded superior skeletal outcomes vs. shorter regimens. In pediatric patients, lumbar spine BMD z-scores significantly increased (+ 0.055 vs. − 0.73 in controls; p = 0.0009), reducing osteoporosis risk (0% vs. 12% with z < − 2.0; p = 0.039) at 5 years post-burn (Reeves et al. 2016). This effect was amplified in children aged 7–18 years during growth spurts (p < 0.05). BMC gains persisted post-treatment, with 24-month therapy outperforming 12-month regimens by 38% (p < 0.05).

Linear Growth:

Height velocity increased by 15–20% during the first 2 post-burn years (p < 0.05), with growth retardation (> 2 SD below mean) resolving in 100% of treated children by 24 months vs. 24% in controls (Reeves et al. 2016). Bone age advanced concordantly with chronological age, confirming no growth plate acceleration.

Clinical and metabolic outcomes

Hospitalization and Recovery:

Acute-phase oxandrolone in adults (initiated 5 days post-burn) reduced length of stay by 28% (31.6 ± 3.1 vs. 43.3 ± 3.1 days; p = 0.042) and stay/TBSA by 30% (0.87 ± 0.05 vs. 1.24 ± 0.15 days/%; p = 0.032) (Wolf et al. 2006). Pediatric studies showed no significant effects on resting energy expenditure or scar scores.

Metabolic Hormones:

Serum IGF-1 and thyroid markers (T3 uptake, FTI) transiently increased during treatment (p < 0.05), but growth hormone, cortisol, and insulin remained unchanged (Przkora et al. 2005). Hepatic acute-phase proteins (prealbumin, haptoglobin) rose briefly without long-term hepatic size changes (Reeves et al. 2016).

Safety and adverse events

Hepatotoxicity: Transaminase elevations were age-dependent. Adults had higher ALT > 100 U/L incidence (19% vs. 5% in placebo; p = 0.002), requiring monitoring (Wolf et al. 2006). Pediatric cohorts showed no significant differences in ALT/AST, though alkaline phosphatase increased in both groups (Tuvdendorj et al. 2011).

Other Adverse Effects: (1) Pediatric: Perineal edema in 3 females (resolved post-discontinuation; Przkora et al. 2005); (2) Adult: Non-significant increases in skin complications (cellulitis, rash) and renal failure (Wolf et al. 2006). (3) No virilization, growth plate closure, or mortality differences in any study.

Combination therapies

Oxandrolone + propranolol outperformed monotherapy for LBM gains (+ 12.3% vs. + 8.1%; p < 0.05) in children (Porro et al. 2012). Nandrolone decanoate showed comparable efficacy to oxandrolone in adults but with higher hepatotoxicity risk (28% vs. 15% ALT elevation; Ali et al. 2022).

Meta-analysis findingsLean body mass in recovery phase

Three studies reported data on lean body mass in the recovery phase for burn patients treated with oxandrolone versus controls. Meta-analysis showed a non-significant SMD of 0.14 (95% CI − 2.02 to 2.29; p > 0.05, I2 = 93.8%) (Fig. 3A). Evaluation of publication bias using funnel plots indicated asymmetry (Fig. 3B), and Egger’s regression test confirmed significant bias (p < 0.001) (Fig. 3C). However, trim-and-fill adjustment did not alter the results materially (Fig. 3D), and leave-one-out sensitivity analysis indicated stable results (Fig. 3E).

Fig. 3

Meta-analysis of lean body mass in recovery phase for burn patients treated with oxandrolone versus controls. A Forest plot of SMD (95% CI) using the DerSimonian-Laird method (Overall SMD = 0.14; 95% CI − 2.02 to 2.29; p = 0.000, I2 = 93.8%). B Funnel plot showing asymmetry. C Egger’s regression test (p = 0.000). D Trim-and-fill funnel plot. E Leave-one-out sensitivity analysis results

We performed meta-regression analysis on the 3 included studies to evaluate the effect of study year on effect sizes. The results showed that when omitting different studies, the pooled effect sizes and their 95% CIs all included 0. The overall pooled effect size (0.136, 95% CI − 2.018 to 2.289) was statistically non-significant. In the meta-regression analysis, the REML estimate of between-study heterogeneity variance (tau2) was 6.743, and the I2 value was 96.82%, indicating substantial heterogeneity. The Adj R-squared was − 100.02%, implying study year did not explain between-study heterogeneity. Specifically, the regression coefficient for year was 0.0640 (SE = 0.337, P = 0.880), confirming no significant association.

In the meta-regression analysis, the REML estimate of between-study heterogeneity variance (tau2) was 6.743, and the I-squared_res value was as high as 96.82%, showing significant between-study heterogeneity. However, the Adj R-squared was − 100.02%, implying that study year couldn’t explain the between-study heterogeneity. Specifically, the regression coefficient of year was 0.0640 (standard error: 0.337), with a P-value of 0.880 and a 95% CI of − 4.214 to 4.342, indicating no significant linear relationship between year and effect size. Also, the constant term’s coefficient was − 128.851, with a P-value of 0.881 and a 95% CI of − 8757.856 to 8500.153, which was statistically non-significant. In summary, meta-regression analysis showed that study year wasn’t an effective covariate for explaining the heterogeneity among the included studies, and no significant overall effect was found in the included studies.

Side effects (mild liver dysfunction or local tissue edema)

Four studies reported data on side effects in the recovery phase for burn patients receiving oxandrolone versus controls. Meta-analysis showed a non-significantly increased RR in the treatment group (RR = 1.82; 95% CI 0.52–6.42; p = 0.34, I2 = 63.6%) (Fig. 4A). Publication bias assessment via funnel plots suggested asymmetry (Fig. 4B), and Egger’s regression test confirmed significant bias (p = 0.027) (Fig. 4C). However, trim-and-fill adjustment did not alter the results materially (Fig. 4D), and leave-one-out sensitivity analysis indicated stable results (Fig. 4E).

Fig. 4

Meta-analysis of lean body mass in recovery phase for burn patients treated with oxandrolone versus controls. A Forest plot of RR (95% CI) using the DerSimonian-Laird method (Overall RR = 1.82; 95% CI 0.52–6.42; p = 0.34, I2 = 63.6%). B Funnel plot showing asymmetry. C Egger’s regression test (p = 0.027). D Trim-and-fill funnel plot. E Leave-one-out sensitivity analysis results

We performed a meta-analysis using the Stata meta commands with a random-effects model. When each study was omitted one by one, the results were as follows: removing study 1 led to an estimated risk ratio of 1.470 (95% CI 0.393–5.502); without study 2, the RR was 4.030 (95% CI 0.478–33.962); excluding study 3 gave an RR of 4.194 (95% CI 0.568–30.993); leaving out study 4 resulted in an RR of 1.034 (95% CI 0.368–2.903); removing study 5 gave an RR of 1.822 (95% CI 0.517–6.424); same result was observed when study 6,7 or 9 was omitted; and excluding study 8 resulted in an RR of 1.373 (95% CI 0.382–4.935). The overall pooled RR was 1.822 (95% CI 0.517–6.424), which isn’t statistically significant.

Then, we ran a meta-regression with study year as a covariate. The analysis involved 5 observations. The REML estimate of between-study variance (tau2) was 0.57, and the I-squared_res was 48.33%, indicating a moderate level of heterogeneity. The Adj R-squared was 70.73%, showing that study year explains a large proportion of the between-study variance. However, the year coefficient was 0.180 (standard error 0.114), with a P-value of 0.213 and a 95% CI of − 0.183 to 0.543, which isn’t statistically significant. Similarly, the constant term had a coefficient of − 361.121 (standard error 229.362), a P-value of 0.213, and a 95% CI of − 1091.052 to 368.810, also not statistically significant. The meta-regression suggests that study year might not be a significant predictor of effect size in this set of studies.

Infection

Two studies provided data on infection rates. The pooled RR was 0.83 (95% CI 0.67–1.02; p = 0.639, I2 = 0.0%) (Fig. 5A), indicating no significant difference. Funnel plots appeared symmetrical (Fig. 5B), and Egger’s test found no significant bias (p > 0.05) (Fig. 5C). Trim-and-fill analysis and sensitivity analysis also supported result stability (Fig. 5D, E).

Fig. 5

Meta-analysis of side effects (mild liver dysfunction or local tissue edema) for burn patients treated with oxandrolone versus controls. A Forest plot of RR (95% CI) (Overall RR = 0.83; 95% CI 0.67–1.02; p = 0.639, I2 = 0.0%). B Symmetrical funnel plot. C Egger’s regression test (p > 0.05). D Trim-and-fill funnel plot. E Leave-one-out sensitivity analysis results

We started with a meta-analysis utilizing the Stata meta commands under a random-effects framework. Upon omitting each of the three studies individually, the resulting risk ratios (RREstimates) lay between 0.786 and 0.838, with corresponding 95% confidence intervals (CIs) that were relatively narrow, indicating more precision. The overall pooled risk ratio stood at 0.826 (95% CI 0.668–1.021), showing no statistically significant effect. Subsequently, we executed a meta-regression with study year as the predictor. The process hit a minor snag with a warning about approximate numerical derivatives and an encounter with a flat or discontinuous region, likely due to the very small dataset (three studies). The REML estimation yielded a zero between-study variance (tau2 = 0), and the I-squared_res metric was 0.00%, pointing to an absence of heterogeneity. The Adj R-squared was also 0%, indicating that study year doesn’t explain the variance between studies. The year variable had a coefficient of − 0.046 (standard error 0.049), which wasn’t significant (P = 0.522, 95% CI − 0.665 to 0.574). Similarly, the constant term wasn’t significant either (coefficient 91.145, P = 0.522). In essence, the meta-regression analysis demonstrates that the study year is not a significant predictor of effect size in the included studies. The lack of heterogeneity suggests that factors other than study year might be influencing the effect sizes, or that the effect is consistent across studies.

Mortality

Six studies reported data on mortality in burn patients treated with oxandrolone versus controls. Meta-analysis showed a non-significantly increased risk ratio in the treatment group (RR = 1.04; 95% CI 0.47–2.32; p = 0.913, I2 = 66.5%) (Fig. 6A). Publication bias assessment via funnel plots suggested asymmetry (Fig. 6B), and Egger’s regression test confirmed significant bias (p = 0.013) (Fig. 6C). However, trim-and-fill adjustment did not alter the results materially (Fig. 6D), and leave-one-out sensitivity analysis indicated stable results (Fig. 6E).

Fig. 6

Meta-analysis of infection rates for burn patients treated with oxandrolone versus controls. A Forest plot of RR (95% CI) using the DerSimonian-Laird method (Overall RR = 1.04; 95% CI 0.47–2.32; p = 0.913, I2 = 66.5%). B Funnel plot showing asymmetry. C Egger’s regression test (p = 0.013). D Trim-and-fill funnel plot. E Leave-one-out sensitivity analysis results

Our analysis hinged on the Stata meta commands, leveraging a random-effects model to accommodate study heterogeneity. The risk ratio estimates after sequential study omission ranged from 0.826 (omitting study 1) to 1.447 (omitting study 2), with corresponding 95% confidence intervals indicating variability in precision. The overall pooled RR of 1.043 (95% CI 0.468–2.323) precludes claims of a significant overall effect.

The meta-regression, with study year as the explanatory variable, estimated a between-study variance (tau2) of 0.7739 and presented an I2 statistic of 70.60%, evidencing substantial residual heterogeneity. However, the adjusted R2 of − 32.90% negates the explanatory power of study year regarding between-study differences. The year coefficient of 0.0289 (SE = 0.114) showed no significant association with effect size (P = 0.813, 95% CI − 0.288 to 0.346). Likewise, the constant term lacked significance (P = 0.813). Therefore, study year is an insignificant predictor of effect size in our meta-regression model, suggesting other unaccounted factors may underlie the observed heterogeneity.

Operation times

Four studies provided data on operation times in burn patients treated with oxandrolone versus controls. The pooled SMD was − 1.25 (95% CI − 2.45 to 0.04; p = 0.04, I2 = 97.2%) (Fig. 7A), indicating a significant reduction in surgical procedures in the treatment group. Funnel plots appeared symmetrical (Fig. 7B), and Egger’s test found no significant bias (p > 0.05) (Fig. 7C). Trim-and-fill analysis and sensitivity analysis also supported result stability (Fig. 7D, E).

Fig. 7

Meta-analysis of mortality for burn patients treated with oxandrolone versus controls. A Forest plot of SMD (95% CI) (Overall SMD = − 1.25; 95% CI − 2.45 to − 0.04; p = 0.04, I2 = 97.2%). B Symmetrical funnel plot. C Egger’s regression test (p > 0.05). D Trim-and-fill funnel plot. E Leave-one-out sensitivity analysis results

We performed meta-regression analysis on the 4 included studies with study year as a covariate. Leave-one-out sensitivity analysis showed consistent results (e.g., omitting study 2: SMD = − 1.78, 95% CI − 3.36 to − 0.19). The meta-regression revealed no significant association between study year and effect size (coefficient = 0.174, SE = 0.270, P = 0.586), with substantial residual heterogeneity (I2 = 98.11%).

LOS/TBSA(d/TBSA)

Five studies reported data on length of stay adjusted for total burn surface area in burn patients treated with oxandrolone versus controls. The pooled SMD was -1.07 (95% CI − 2.43 to 0.29; p = 0.007, I2 = 98.1%) (Fig. 8A), indicating a significant reduction in LOS/TBSA in the treatment group. Funnel plots suggested asymmetry (Fig. 8B), and Egger’s regression test confirmed significant bias (p = 0.000) (Fig. 8C). However, trim-and-fill adjustment did not alter the results materially (Fig. 8D), and leave-one-out sensitivity analysis indicated stable results (Fig. 8E).

Fig. 8

Meta-analysis of LOS/TBSA for burn patients treated with oxandrolone versus controls. A Forest plot of SMD (95% CI) using the DerSimonian-Laird method (Overall SMD = − 1.07; 95% CI − 2.43 to 0.29; p = 0.007, I2 = 98.1%). B Funnel plot showing asymmetry. C Egger’s regression test (p = 0.000). D Trim-and-fill funnel plot. E Leave-one-out sensitivity analysis results

The latest meta-regression analysis investigated the relationship between the effect sizes of five studies and the study year. The leave-one-out analysis revealed the following: excluding study 1 resulted in a pooled effect size of − 0.204 (95% CI − 0.496, 0.088); omitting study 2 gave − 1.369 (95% CI − 3.134, 0.396); leaving out study 3 yielded − 1.090 (95% CI − 2.651, 0.472); omitting study 4 led to − 1.310 (95% CI − 3.207, 0.587); and excluding study 5 resulted in − 1.261 (95% CI − 3.054, 0.532). The overall pooled effect size was − 1.072 (95% CI − 2.430, 0.285), which is not statistically significant.

The meta-regression results showed a REML estimate of between-study variance (tau2) of 2.682 and an I-squared_res of 98.27%, reflecting significant heterogeneity among the studies. The Adj R-squared was 4.03%, indicating that study year explains only a small portion of the heterogeneity. The coefficient for year was 0.251 (standard error 0.232), with a P-value of 0.358 and a 95% CI of − 0.486 to 0.989, showing no significant association with effect size. Similarly, the constant term had a coefficient of − 506.308 (standard error 465.914), a P-value of 0.357, and a 95% CI of − 1989.056 to 976.439, which was also not statistically significant. The meta-regression analysis found that study year is not a major factor in explaining the heterogeneity among the included studies.

LOS

Eight studies provided data on length of stay in burn patients treated with oxandrolone versus controls. The pooled SMD was 0.48 (95% CI − 0.59 to 1.54; p = 0.38, I2 = 99.1%) (Fig. 9A), indicating no significant difference. Funnel plots appeared symmetrical (Fig. 9B), and Egger’s test found no significant bias (p > 0.05) (Fig. 9C). Trim-and-fill analysis and sensitivity analysis also supported result stability (Fig. 9D, E).

Fig. 9

Meta-analysis of LOS for burn patients treated with oxandrolone versus controls. A Forest plot of SMD (95% CI) (Overall SMD = 0.48; 95% CI − 0.59 to 1.54; p = 0.38, I2 = 99.1%). B Symmetrical funnel plot. C Egger’s regression test (p > 0.05). D Trim-and-fill funnel plot. E Leave-one-out sensitivity analysis results

The most recent meta-regression analysis explored the connection between the effect sizes from eight studies and the year of the study. When each study was omitted one at a time, the results were as follows: removing study 1 led to a pooled effect size of 0.931 (95% CI − 0.109 to 1.970); without study 2, it was 0.765 (95% CI − 0.253 to 1.782); excluding study 3 gave 0.503 (95% CI − 0.678 to 1.684); leaving out study 4 resulted in 0.391 (95% CI − 0.832 to 1.614); removing study 5 gave 0.210 (95% CI − 0.955 to 1.375); without study 6, it was 0.545 (95% CI − 0.646 to 1.737); excluding study 7 resulted in 0.325 (95% CI − 0.970 to 1.619); and omitting study 8 gave 0.152 (95% CI − 0.860 to 1.164). The overall pooled effect size was 0.478 (95% CI − 0.588 to 1.544), which does not show statistical significance.

The meta-regression analysis indicated a between-study variance (tau2) of 1.393 via REML estimation and an I2 value of 98.21%, highlighting considerable heterogeneity across studies. The Adj R-squared value of 60.19% suggests study year may partially explain heterogeneity. The year coefficient was 0.209 (SE = 0.063, P = 0.016), indicating a temporal association with effect size, though substantial residual heterogeneity persists (I2 = 98.21%). In conclusion, the meta-regression analysis indicates that the study year is a significant factor in explaining the heterogeneity among the studies included.