# Inconclusive results and evidence against an effect

Source: https://onco.cc/terms/inconclusive-and-negative-evidence/  
OnCo record `inconclusive-and-negative-evidence` (Term). Data CC BY-NC 4.0, attribute "Data from OnCo (onco.cc)"; commercial use needs a licence.

## TL;DR

A study may fail to find an effect because the effect is small or because the study is not informative enough. Its estimate and uncertainty help distinguish those possibilities.

## Summary

A primary methodological analysis of replication projects, including cancer biology, showed that nonsignificant original and replication results can both be inconclusive. Failure to cross a significance threshold does not establish equivalence. Evidence against a meaningful effect needs a question, margin or model suited to that claim and enough information to distinguish relevant possibilities. A precise negative result and an untested idea should remain different evidence states.

## Fields

- Kind: Term
- Last checked: 2026-11-10
- Also known as: absence of evidence versus evidence of absence; interpretation of null results

## Notes

- “We did not find it” needs a description of where and how we looked. A literature search can establish that no relevant publication was found within its queries, synonyms, databases, access and cutoff date; it cannot substitute for a biological experiment. A negative assay means no signal met the detection rule in the tested preparation. Its interpretation depends on the material, recovery, units, controls, sensitivity, threshold, time and compartment sampled. A false negative occurs when the condition is present but the test says otherwise. A study missing a significance threshold asks a third question: whether the data met that analysis's decision rule. Its population, comparator, endpoint, effect scale, uncertainty, follow-up and missing-data rules determine what remains open.
- Read the effect estimate and confidence interval together. Too few events can leave a study with low power and an interval compatible with meaningful benefit and harm. Precision concerns sampling uncertainty under the analysis; it does not remove bias. Evidence excluding a large effect need not exclude a smaller one, and an unresolved effect is neither evidence of benefit nor proof of exactly zero effect. Equivalence testing asks whether an effect lies within justified lower and upper bounds, preferably specified before seeing results, using an appropriate analysis. Non-inferiority instead asks whether a treatment is worse than its comparator by more than a prespecified margin. Neither question is settled by failing to reject a zero difference, and meaningful margins depend on the outcome and setting.
- If no exact-variant drug-response study is found in the documented search, the proposed drug match remains unsupported. That absence establishes neither resistance nor benefit. A normal functional result also covers only the measured function under its assay conditions. A protein-function test may miss an effect on RNA processing. Disease mechanism, assay validity and the particular function tested must remain attached to the conclusion.
- In a fictional randomised study, the new treatment minus control difference in one-year event risk is -2 percentage points, with a 95% confidence interval from -12 to +8 points. The interval includes both appreciable benefit and harm; it does not support a claim of identical outcomes. It also crosses illustrative equivalence bounds of -5 and +5 points. Those bounds are invented for this example, rather than a universal definition of an unimportant effect. Even a narrower interval would need a sound design, justified prespecified bounds and the appropriate analysis. These numbers are not a patient estimate.

## Sources

- Null-result replication and evidence of absence (2024): https://doi.org/10.7554/eLife.92311
- Statistical tests, P values, confidence intervals, and power: a guide to misinterpretations.: https://pmc.ncbi.nlm.nih.gov/articles/PMC4877414/
- Equivalence Tests: A Practical Primer for t Tests, Correlations, and Meta-Analyses.: https://pmc.ncbi.nlm.nih.gov/articles/PMC5502906/
- Recommendations for application of the functional evidence PS3/BS3 criterion using the ACMG/AMP sequence variant interpretation framework.: https://pmc.ncbi.nlm.nih.gov/articles/PMC6938631/

## Connected records

- terms: [Evidence from measurement to patient benefit](https://onco.cc/terms/measurement-to-benefit-evidence/), [Non-inferiority margin and equivalence trials](https://onco.cc/terms/non-inferiority-margin/), [Reproducibility and negative results](https://onco.cc/terms/reproducibility/), [Statistical significance (P values, alpha, multiplicity)](https://onco.cc/terms/statistical-significance/)

---
JSON: https://onco.cc/api/v1/entities/inconclusive-and-negative-evidence.json