PhD Biostatisticians and Cochrane-Standard Statistical Synthesis
Every synthesis we run follows the Cochrane Handbook for Systematic Reviews of Interventions, version 6.5 (Higgins JPT, Thomas J, et al., 2024) and is reported in line with the PRISMA 2020 Statement (Page et al., BMJ, 2021).
At Research Gold, our meta-analysis services cover every stage of quantitative synthesis across health sciences, nursing, psychology, education, social sciences, and beyond. We pool both randomized trials and observational studies, applying random-effects models, subgroup and meta-regression analyses, and ROBINS-I risk-of-bias assessment where the evidence base calls for it. We match you with a PhD biostatistician who has published in your field, drawn from a team of nine PhD methodologists spanning biostatistics and epidemiology. Our methodologists implement all analyses using R (metafor and meta packages) or Stata, and deliver fully reproducible code alongside publication-ready figures and tables. For projects that need dataset preparation and modeling before pooling, we also provide statistical analysis support for research teams.
Whether you need a standalone meta-analysis for an existing dataset or a combined systematic review and meta-analysis package, we deliver results that satisfy the most demanding peer reviewers and editorial boards. You can review our published meta-analyses, each with a verifiable digital object identifier, on our sample deliverables page.
What Is Included in Every Meta-Analysis
Effect Size Calculation and Standardization
We calculate or convert effect sizes from the data reported in primary studies. Depending on your outcome type, we compute:
- Odds ratios and risk ratios for dichotomous outcomes
- Mean differences and standardized mean differences (Cohen's d, Hedges' g) for continuous outcomes
- Hazard ratios for time-to-event data
- Correlation coefficients for association studies
When studies report insufficient data, we apply validated methods to estimate effect sizes from p-values, confidence intervals, t-statistics, or F-statistics. Explore how effect sizes are computed using our free advanced effect size calculator.
Forest Plots
Forest plots visualize the individual study effect sizes alongside the pooled estimate and its confidence interval. Every meta-analysis we deliver includes publication-quality forest plots for each outcome, formatted to the specifications of your target journal. Forest plots are delivered in high-resolution PNG and editable PDF format. You can preview the format using our free interactive forest plot generator.
Heterogeneity Assessment
Statistical heterogeneity measures the degree to which results vary across included studies beyond what would be expected by chance. We report:
- I-squared (I²): the percentage of variability attributable to between-study differences
- Tau-squared (τ²): the absolute between-study variance
- Cochran's Q statistic: the formal test for heterogeneity
- Prediction intervals: the range within which the true effect is expected to fall in a new study
The I-squared statistic follows the method of Higgins and colleagues for quantifying heterogeneity (Higgins et al., BMJ, 2003). When meaningful between-study variance is present, we pool effects with the DerSimonian and Laird random-effects model (DerSimonian and Laird, 1986) rather than a fixed-effect model, selecting the model to match your data rather than by default. When heterogeneity is substantial (I² > 50%), we investigate potential sources through subgroup analyses and meta-regression, depending on the number of studies and heterogeneity. Explore heterogeneity metrics using our free sample size estimation tool.
Publication Bias Detection
Funnel plots detect asymmetry that may indicate publication bias, small-study effects, or selective reporting. Where the number of studies is sufficient, typically ten or more, we test for small-study effects with the Egger test (Egger et al., BMJ, 1997). We apply:
- Egger's regression test for funnel plot asymmetry
- Begg's rank correlation test
- Trim-and-fill method to estimate the number of missing studies and adjust the pooled estimate
- Doi plot and LFK index as sensitivity measures
Visualize publication bias testing using our free our funnel plot generator.
Subgroup and Sensitivity Analyses
We conduct planned subgroup analyses to explore whether the treatment effect varies across predefined study-level characteristics (such as study design, population age, intervention dose, or follow-up duration). Sensitivity analyses include:
- Leave-one-out analysis: removing each study in turn to test the robustness of the pooled estimate
- Influence diagnostics: identifying outlier or overly influential studies
- Restricted analyses: excluding high-risk-of-bias studies or studies with imputed data
Meta-Regression
When heterogeneity sources need formal statistical investigation, we run meta-regression models using study-level covariates as moderators. This identifies whether variables such as sample size, publication year, intervention intensity, or baseline risk explain the observed variability in effect sizes.
GRADE Certainty of Evidence
For each critical outcome, we produce a GRADE assessment rating the certainty of evidence as high, moderate, low, or very low, following the GRADE approach (Guyatt et al., BMJ, 2008). We generate formatted summary of findings tables following the Cochrane Handbook for Systematic Reviews of Interventions, version 6.5 (Higgins JPT, Thomas J, et al., 2024). Explore the GRADE framework using our free our grade evidence certainty tool.
Reproducible Code
Every analysis is accompanied by fully annotated, reproducible R or Stata code. This allows you and your reviewers to verify every step, from data import to final forest plot. We use the metafor and meta packages in R, and the metan suite in Stata.
