The Conclusion Generator synthesizes a conclusion of study results as detailed in Schmidt M, Parner E. The Conclusion Generator. Ann Epidemiol. 2024. The auto-generated conclusions are based on the effect size (point estimate), random error (interval estimate: confidence interval or Bayesian credible interval), and clinical relevance as specified by the user.

• Modes: Two interpretation modes are available:
  • Statistical mode: This mode provides an accurate statistical interpretation of results, with optional specification of superiority and noninferiority margins.
  • Clinical mode: This mode evaluates the clinical relevance (as specified by the user) of the point estimate and interval limits.
• Validity assumption: Both modes assume no uncontrolled systematic errors (biases). If the input data are biased, the generated conclusion will also be biased.
• Measure of effect: Both absolute and relative measures of association are supported. Absolute measures include risk difference, while relative measures encompass risk ratio, odds ratio, hazard ratio, and incidence rate ratio.
• Uncertainty metrics: The framework does not require a particular interval type or level. Users should enter the interval estimate appropriate to their prespecified analytical framework, such as a 90%, 95%, or 99% confidence interval or a Bayesian credible interval. In frequentist analyses, confidence intervals are interpreted as compatibility intervals: values within the interval are more compatible with the data than values outside it, although values outside the interval are not necessarily incompatible. Compatibility is greatest near the point estimate and decreases towards and beyond the interval limits. Bayesian credible intervals can be classified in the same way according to their position relative to the prespecified decision margins but require a Bayesian interpretation. The generated classification is conditional on the type, level, and underlying assumptions of the interval entered. These should therefore be reported alongside the generated conclusion.
• Number of decimals: The number of decimals reported should be clinically relevant and statistically reasonable. Typically, no more than two decimals are recommended, and for absolute effect estimates, two significant digits are often sufficient.
• Direction of benefit: Specify the direction of a beneficial effect for accurate wording. For risks, a relative risk <1 indicates a beneficial effect, and >1 a harmful effect. Here, the direction of benefit will be “Reduced effect.”
• Hypothesis testing: Terminology implying dichotomization of results, such as statistical significance or nonsignificance, is avoided in line with recommendations from the American Statistical Association (2016). The type I error level is not included in the output but may be added manually.
• Level of detail: The Statistical mode allows the user to select the degree of detail in the output, appropriate to the context, e.g., concise for abstracts and elaborated version for full-text descriptions.
• Example: Each mode provides a pre-filled example of use.
• User costs: None