About the Conclusion Generator

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 (confidence interval), and clinical relevance as specified by the user.

User instructions
  • Modes: Two interpretation modes are available:
    • Statistical mode: This mode provides an accurate statistical interpretation of results, with optional specification of superiority and noninferiority bounds.
    • Clinical mode: This mode evaluates the clinical relevance (as specified by the user) of the point estimate and confidence 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 input confidence interval is interpreted as a compatibility interval, indicating values highly compatible with the data rather than regions of confidence.
  • 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

The app was developed by Morten Schmidt, Mads Parner, and Erik Parner.

The minimum required inputs are the Measure of effect, Point estimate, Lower confidence limit, Upper confidence limit, Number of decimals, Direction of benefit, and Level of detail.

The Optional extension allows the user to specify a Superiority bound (minimal benefit) and non-inferiority bound (maximum acceptable risk). These bounds should be specified on the same scale as the point estimate.