Mikoshi Meta-Analysis

Meta-analysis, complete

A comprehensive Python toolkit for systematic reviews and meta-analysis. Fixed and random effects, Bayesian models, publication bias, diagnostics, forest plots, and automated reporting โ€” all in one package.

pip install Mikoshi Meta-Analysis

16
Modules
3,647
Lines of Code
99
Tests Passing
0
Dependencies*

*Beyond standard scientific Python (numpy, scipy, pandas, matplotlib, statsmodels)

Meta-analysis tools are fragmented

Researchers typically juggle multiple tools: R's metafor for random effects, separate scripts for bias detection, manual forest plot generation, and Word documents for reporting. There's no single Python package that does it all.

Mikoshi Meta-Analysis is that package. Everything from effect size estimation to PRISMA flow diagrams to Bayesian sensitivity analysis โ€” in one pip install.

Everything a systematic reviewer needs

๐Ÿ“ˆ

Fixed & Random Effects

DerSimonian-Laird, REML, ML, and fixed-effect models. Heterogeneity statistics (Iยฒ, Q, ฯ„ยฒ), prediction intervals, and confidence intervals.

๐Ÿ”ฌ

Meta-Regression

Weighted least squares and mixed-effects meta-regression. Hierarchical models with random slopes. Bubble plots for moderator analysis.

๐Ÿงช

Bayesian Models

Bayesian random-effects meta-analysis using analytical approximation (no PyMC dependency). Prior sensitivity analysis, Bayes factors, posterior forest plots.

๐Ÿ“‰

Publication Bias

Egger's test, trim-and-fill, PET-PEESE, selection models (Vevea & Hedges), p-curve analysis, z-curve analysis, and Rosenthal's fail-safe N.

๐Ÿ”

Diagnostics

Leave-one-out influence analysis, Baujat plots, GOSH plots, Cook's distance, DFBETAS, hat values, studentised residuals, and outlier detection.

๐Ÿ“Š

Plots

Forest plots, funnel plots (with contour), bubble plots, radial (Galbraith) plots, L'Abbรฉ plots, cumulative forest plots, network geometry plots. All publication-quality matplotlib.

๐Ÿ”

Cumulative Meta-Analysis

Add studies one at a time sorted by year, effect size, precision, or sample size. Visualise how the pooled estimate evolves as evidence accumulates.

๐Ÿ›ก๏ธ

Robust Variance

Cluster-robust variance estimation for dependent effects. Sandwich standard errors for meta-regression. Permutation-based p-values.

๐Ÿ“

Automated Reporting

Generate full HTML or LaTeX reports with embedded plots, summary tables, PRISMA flow diagrams, and APA-formatted citations. One function call.

โญ

GRADE Quality

Certainty of evidence assessment following GRADE methodology. Quality-stratified effects analysis and sensitivity analysis by risk of bias.

๐ŸŒ

Network Meta-Analysis

Basic network (multiple-treatments) meta-analysis with league tables and network geometry plots.

๐Ÿ–ฅ๏ธ

Interactive Dashboard

Built-in Plotly Dash dashboard. Upload data, run analyses, generate plots, and export reports โ€” all from a web browser. No code required.

Modular by design

Each module is independent. Use the full suite or import just what you need.

core
Fixed/random effects
68 lines
extensions
Meta-regression, hierarchical
171 lines
bayesian
Bayesian models & Bayes factors
265 lines
bias
Egger's test
40 lines
bias_extra
Trim-fill, PET-PEESE, p-curve
367 lines
plotting
Forest, funnel, bubble, radial
449 lines
diagnostics
LOO, Baujat, GOSH, influence
334 lines
cumulative
Cumulative meta-analysis
140 lines
prediction
Prediction intervals & power
143 lines
quality
GRADE certainty assessment
255 lines
reporting
HTML/LaTeX reports, PRISMA
294 lines
robust
Cluster-robust, sandwich, perm
206 lines
multivariate
Network meta-analysis
272 lines
cli
Command-line interface
257 lines
dashboard
Interactive Plotly Dash UI
253 lines
vendor/pymare
Vendored meta-regression core
BSD-3

Three ways to use Mikoshi Meta-Analysis

1. Python API

from Mikoshi Meta-Analysis import core, plotting, diagnostics, bias_extra

# Load your data
effects = [0.5, 0.8, 0.3, 1.1, 0.6]
variances = [0.04, 0.09, 0.02, 0.16, 0.05]
labels = ["Study A", "Study B", "Study C", "Study D", "Study E"]

# Random-effects meta-analysis
result = core.random_effects(effects, variances)
# โ†’ {'effect': 0.58, 'ci_lower': 0.31, 'ci_upper': 0.85, 'tau2': 0.04, 'I2': 62.3}

# Forest plot
plotting.forest_plot(effects, variances, labels, save="forest.png")

# Leave-one-out sensitivity
loo = diagnostics.leave_one_out(effects, variances, labels)

# Trim-and-fill
adjusted = bias_extra.trim_and_fill(effects, variances)

2. Command Line

# Run random-effects meta-analysis
pymetasuite run data.csv --effect effect_size --var variance

# Generate forest plot
pymetasuite forest data.csv --effect effect_size --var variance

# Funnel plot with trim-and-fill
pymetasuite funnel data.csv --effect effect_size --var variance

# Bayesian meta-analysis
pymetasuite bayesian data.csv --effect effect_size --var variance

# Launch interactive dashboard
pymetasuite dashboard --port 8050

3. Interactive Dashboard

Run pymetasuite dashboard and open your browser. Upload a CSV, select columns, choose analysis type, and generate publication-quality plots โ€” no code needed. Export results as HTML or PDF reports.

Install in 10 seconds

# From GitHub
pip install git+https://github.com/DarrenEdwards111/Mikoshi Meta-Analysis.git

# Or clone and install locally
git clone https://github.com/DarrenEdwards111/Mikoshi Meta-Analysis.git
cd Mikoshi Meta-Analysis
pip install .

Built by researchers, for researchers

๐ŸŽ“

Systematic Reviewers

Everything you need for a Cochrane-style review. PRISMA flow diagrams, GRADE certainty assessment, forest plots, and automated reporting in one toolkit.

๐Ÿ”ฌ

Clinical Researchers

Random-effects models, subgroup analysis, meta-regression, and publication bias detection. Supports both continuous and binary outcomes.

๐Ÿ“š

Graduate Students

Clear API, comprehensive documentation, and an interactive dashboard. Learn meta-analysis methodology by doing, not just reading about it.

Cite Mikoshi Meta-Analysis

Edwards, D. (2025). Mikoshi Meta-Analysis: A comprehensive Python meta-analysis toolkit
(Version 0.9.0) [Computer software].
GitHub. https://github.com/DarrenEdwards111/Mikoshi Meta-Analysis