Meta-Analysis Q&A: Technical Answers on Heterogeneity, Cochrane Standards, Workflow and IQWiG Expectations

• I² gives the share of variability due to heterogeneity; τ² gives its absolute size, and a prediction interval shows where a new study’s effect may lie.
• MECIR now covers conduct only; Cochrane uses PRISMA 2020 for reporting.
• IQWiG’s model depends on the number of studies: fixed-effect for two, Bayesian random-effects for three or four, Knapp-Hartung with Paule-Mandel for five or more.

This page answers the questions researchers and HTA teams in Germany most often ask about meta-analysis. Answers are grouped into four topics and give the technical detail needed to act on each point, with the source for each claim.

1. What does the I² statistic measure?

The percentage of variability in effect estimates that is due to heterogeneity rather than chance, calculated as I² = max(0, (Q − df) / Q) × 100%, so negative values are set to zero3.

2. How is τ² different from I²?

τ² (tau-squared) is the absolute between-study variance on the effect scale; I² is a relative percentage that rises as study precision increases, even when τ² stays the same13.

3. What does a prediction interval show?

The range within which the true effect in a new, similar study is expected to lie, reflecting both τ² and uncertainty in the pooled estimate. It is typically reported when there are about five or more studies1.

4. Why is Cochran’s Q often tested at P < 0.10?

Because the test has low power when few studies are included, Cochrane suggests a P value of 0.10 rather than 0.051. IQWiG, by contrast, generally does not pool results if the heterogeneity test gives P < 0.054.

5. When is meta-regression appropriate?

When covariates are pre-specified and there are generally at least ten studies for each characteristic examined; the Handbook advises against it with fewer than ten studies1.

6. What does Chapter 10 cover?

Effect measures, fixed-effect and random-effects models, meta-analysis methods, heterogeneity, subgroups, missing data and sensitivity analysis1. Our Chapter 10 and MECIR guide maps each topic to its standard.

7. How are MECIR standards classified?

As mandatory or highly desirable conduct standards for Cochrane intervention reviews (C1–C75) and their updates. The reporting standards were retired in August 2023, and Cochrane now uses PRISMA 2020 for reporting2.

8. When should Mantel-Haenszel or Peto methods be used?

Mantel-Haenszel suits dichotomous data, especially with few events or small studies; Peto suits rare events with small effects and balanced groups1.

9. When is the standardised mean difference used?

When studies measure the same construct on different scales1.

10. Why is the DerSimonian-Laird method criticised?

Its τ² estimator can be biased and its confidence intervals too narrow, especially with few studies. Since 2024 the Handbook prefers estimators such as REML, now the RevMan default1.

11. What are the main stages of a meta-analysis?

PICO question, protocol and registration, searching, screening, data extraction, risk of bias, data synthesis, heterogeneity and sensitivity analyses, GRADE and PRISMA reporting. See the meta-analysis workflow flowchart for owners and durations.

12. Why register a protocol in PROSPERO?

Prospective registration reduces selective reporting and unplanned duplication of reviews.

13. Which risk of bias tools are used?

RoB 2 for randomised trials and ROBINS-I for non-randomised studies of interventions. Our risk of bias tool decision tree shows which to use.

14. What does GRADE assess?

Certainty of evidence for each outcome. Evidence is rated down for risk of bias, inconsistency, indirectness, imprecision and publication bias, and evidence from non-randomised studies may be rated up for a large effect, a dose-response gradient or plausible confounding that would reduce the effect.

15. How long does a systematic review with meta-analysis take?

Often more than a year. An analysis of reviews registered in PROSPERO found a mean of 67.3 weeks from registration to publication5.

16. What is IQWiG’s role?

It assesses the benefits and harms of interventions for the Federal Joint Committee (G-BA), including AMNOG early benefit assessments of new medicines4.

17. Which outcomes does IQWiG prioritise?

Patient-relevant outcomes: mortality, morbidity (symptoms and complications) and health-related quality of life, with harms such as adverse events assessed alongside4.

18. Which meta-analysis model does IQWiG usually apply?

It depends on the number of studies. For two studies, usually a fixed-effect model; for three or four, a Bayesian random-effects meta-analysis with an informative prior for heterogeneity; for five or more, a random-effects model with the Knapp-Hartung method and the Paule-Mandel heterogeneity estimator (General Methods 8.0, section 10.3.7)4. Read more in our IQWiG General Methods guide.

19. How does IQWiG express certainty?

As proof, indication or hint of a benefit or harm, from highest to lowest certainty4.

20. How does the EU HTA Regulation affect German assessments?

Joint clinical assessments have applied to oncology medicines and advanced therapy medicinal products since 12 January 2025, and Member States must give their reports due consideration in national assessments6.