Methodology · effect measures
Choosing effect measures in Axelium
Effect measures determine how study results are represented and pooled. Axelium suggests defaults by endpoint type, and you can override them when your protocol requires a different choice.
Binary outcomes
For outcomes that represent occurrence vs. non-occurrence of an event:
- Risk Ratio (RR) — intuitive for relative risk comparisons; preferred for cohort and RCT designs where absolute risks are estimable.
- Odds Ratio (OR) — common in case-control and logistic regression models; natural output when absolute event rates are not directly available.
- Hazard Ratio (HR) — for time-to-event endpoints where censoring occurs; used in survival analysis contexts.
Modified Rankin Scale (mRS) as a binary outcome
Extraction accepts mRS in two structured shapes. When a paper reports the dichotomised split (mrs_pct_0_to_2 andmrs_pct_3_to_6), Axelium treats mRS 0–2 as a favourable-outcome event and pools it with RR or OR alongside other binary endpoints. When the full ordinal distribution is captured (ordinal_distribution.score_0 throughscore_6), the same 0–2 vs 3–6 cut is derived automatically; ordinal-aware pooling across all seven categories is planned but not yet available.
Continuous outcomes
For outcomes measured on a numeric scale:
- Mean Difference (MD) — used when all studies report the same units; preserves clinical interpretability.
- Standardized Mean Difference (SMD) — used when studies measure the same construct on different scales; expressed in standard deviation units.
Papers that report median with IQR or median with range can feed MD and SMD analyses directly — extraction stores the summary as given, and the math engine derives an approximate mean and SD using the Wan et al. and Hozo et al. estimators. No manual transcription to mean ± SD is required.
Single-group prevalence
For prevalence and incidence estimates without a comparator:
- Use pooled proportions with appropriate transformations (logit, Freeman-Tukey double arcsine) depending on whether extreme values are expected.
- Random-effects assumptions are usually appropriate when pooling prevalence across heterogeneous populations.
Practical selection checklist
- Match the measure to the endpoint definition and study design.
- Keep the measure consistent with your protocol registration.
- Document overrides and rationale in your audit trail.
- Consider clinical interpretability when choosing between RR and OR — when event rates differ substantially across populations, RRs are generally more informative.
How Axelium measures map to RevMan 5 (.rm5) export
When you download a review as a RevMan 5 file (Literature Search → Export RevMan), each extracted outcome is written into one of three RevMan outcome elements. The mapping is deterministic:
| Axelium measure | RevMan element | Data shape |
|---|---|---|
| RR, OR, RD | DICH_OUTCOME | events / total per arm (Mantel-Haenszel) |
| MD, SMD | CONT_OUTCOME | mean / SD / n per arm (inverse variance) |
| HR, IRR, Rate ratio | IV_OUTCOME | log(point estimate) + SE(log), derived from the 95% CI via SE = (ln(upper) − ln(lower)) / 3.92 |
| Proportion, Mean, Count, Themes | skipped | Single-arm / qualitative — no RevMan comparison equivalent. Listed under PUBLISHED_NOTES in the exported file. |
Risk-of-bias judgements (QUALITY_ITEMS) are emitted empty — Axelium does not yet capture RoB 2.0, so reviewers complete bias assessments inside RevMan after import. The exported file is valid for RevMan 5 Desktop and is also accepted by RevMan Web via its Import .rm5 workflow.