Clinical trials and biomedical studies produce results that cannot be compared directly because each endpoint reports evidence differently.
Evidence ratio provides a single, likelihood-based evidence scale that works across analyses without changing existing statistical methods. Results can then be compared, stored, and reused consistently across endpoints, studies, databases, and regulatory review.
This allows trial and observational results to be exchanged, stored, and reused using a single, consistent reporting structure across statistical analyses, clinical databases, regulatory submissions, and interoperability standards such as HL7 FHIR.
Evidence ratio adopts a shared evidence standard. Foundational tools must be unrestricted so results can move safely between laboratories, hospitals, and national programmes.
Read moreEvidence ratio does not change how analyses are run. It standardises how their evidence is reported.
Analyses continue to produce effect estimates and uncertainty intervals. Evidence ratio adds a shared evidence measure that travels cleanly between endpoints and institutions.
R package installation
Source: cran.r-project evidenceratio
Install with R or RStudio directly from CRAN. For more information see the R vignette and R reference manual pages.
install.packages("evidenceratio")
library(evidenceratio)
help(package = "evidenceratio")
Where evidence ratio fits
Clinical analyses already produce results using appropriate statistical models. Evidence ratio operates after analysis, translating results into a shared evidence scale.
Analysis → effect estimate and interval → Evidence ratio → reusable reporting.
Discovery and modelling remain unchanged. Reporting becomes consistent.
What evidence ratio does not do
Evidence ratio does not replace statistical models.
It does not define decision thresholds.
It does not assess clinical or practical importance.
It does not perform multiplicity correction.
It does not replace effect sizes or uncertainty intervals.
Evidence ratio operates only at the level of reporting.
Conceptual role
Each statistical result contains three distinct components.
- Magnitude is captured by the effect estimate.
- Precision is captured by the uncertainty interval.
- Evidence is captured by a likelihood ratio comparing an explicit effect model with a no effect model.
Evidence ratio formalises this evidence component and reports it on a common log10 scale. Existing analyses remain unchanged.
Why this matters in practice
Clinical trials routinely analyse heterogeneous endpoints, making results difficult to compare side by side.
Evidence ratio gives every endpoint the same evidence unit, allowing heterogeneous results to be aligned directly in trial tables, dashboards, and databases.
Regulatory review, platform trials, and meta-analysis can therefore compare evidential support without reinterpretation.
Results can then be stored and reused consistently across clinical databases, trial platforms, hospital systems, and national clinical data infrastructures using a single reporting structure.
This structure can be exchanged across environments such as REDCap, electronic health records, and interoperability standards including HL7 FHIR.
Example output
Each result is reported using the same three quantities.
| Effect estimate | Uncertainty interval (95 percent) | log10 evidence ratio |
|---|---|---|
| −1.18 | [−1.62, −0.74] | 12.87 |
The effect and interval remain on their native scale. The evidence ratio provides a directly comparable measure of evidence support.
Unified reporting across analyses
Clinical studies use many different statistical analyses depending on outcome type, including mean comparisons, regression models, contingency analyses, and survival models. These analyses remain unchanged, but their results can now be reported using a shared evidence scale.
| Analysis type | Effect estimate | Uncertainty interval (95 percent) | log10 E(x) |
|---|---|---|---|
| One sample mean test | 0.42 | [0.21, 0.63] | 2.91 |
| Two sample mean test | −1.18 | [−1.62, −0.74] | 12.87 |
| Binary outcome association | 0.36 | [−0.22, 0.94] | 0.44 |
| Linear regression | −0.51 | [−1.19, 0.17] | 0.24 |
| Regression coefficient | −5.41 | [−6.18, −4.63] | 16.95 |
| Time to event analysis | −0.22 | [−0.52, 0.08] | 0.58 |
| Survival analysis | 0.88 | [0.61, 1.15] | 6.82 |
Different outcome types retain their native effect measures while sharing a common evidence scale.
Method overview
The evidence ratio is defined as a likelihood ratio comparing an effect model with a no effect model. It is reported on the log10 scale. Under the null model, large evidence values are rare. This provides finite sample validity without introducing decision thresholds. Further details are provided in the accompanying manuscript.
Standards and implementation
Evidence ratio is implemented in the evidenceratio R package.
It conforms to the Evidence Ratio Reporting Standard (SGA-ERRS-1.0), published by the Swiss Genomics Association:
view the ERRS standard.
