E-value
The E-value is a measure that helps researchers understand how strong an unmeasured factor would have to be to explain away an observed treatment-effect relationship in observational studies. A high E-value means the observed effect is more robust to potential hidden biases like unmeasured confounding.
At a glance
Use when
Conducting causal inference in observational studies where unmeasured confounding is a concern; reporting results intended to support causal claims; performing sensitivity analysis for confounding bias
Avoid when
In randomized controlled trials with proper allocation concealment and low risk of bias; when the primary concern is measurement error or selection bias rather than unmeasured confounding; when effect estimates are on non-risk-ratio scales without appropriate transformation
Inputs
Observed risk ratio (or odds ratio approximated as risk ratio), confidence interval limits, direction of effect
Outputs
E-value for the point estimate and for the confidence interval limit closest to the null
How it works
The E-value quantifies the minimum strength of association, on the risk ratio scale, that an unmeasured confounder must have with both the exposure and the outcome to fully explain away an observed association, conditional on measured covariates. It is computed as the risk ratio that would nullify the observed effect estimate or the limit of its confidence interval closest to the null. It provides a formal sensitivity analysis for unmeasured confounding in causal inference from non-randomized studies.
- HTA domains
- Clinical Effectiveness
- Categories
- Heterogeneity
- Assumptions
- The unmeasured confounder affects both treatment assignment and outcome, operates independently of measured covariates, and acts multiplicatively on the risk ratio scale; no interaction between confounder and treatment on the outcome in the risk ratio metric
- Strengths
- Provides an intuitive, quantitative measure of robustness to unmeasured confounding,Easy to compute and interpret using only the reported effect estimate and confidence interval,Encourages transparency and critical appraisal of observational findings,Applicable across diverse fields using observational data for causal inference
- Limitations
- Assumes a single, binary unmeasured confounder with constant effect sizes,Relies on the assumption of no effect modification by the confounder,Does not account for complex confounding structures or multiple confounders,Interpretation may be less accurate when odds ratios are used as proxies for risk ratios in common outcomes
- Also known as
- E-value for unmeasured confounding, E-value sensitivity analysis
Questions this answers
- › How strong would an unmeasured confounder need to be to explain away the observed treatment-outcome association?
- › Is the observed association robust to potential unmeasured confounding?
- › What is the minimum bias needed to reduce the observed effect to null?
- › How should confidence intervals be interpreted in terms of sensitivity to confounding?
- › Can causal conclusions be justified despite the absence of randomization?
- › What level of confounding resistance does the estimate exhibit?
References & sources
Similar by meaning
Beta record. Generated from the primary source via AI extraction and independent audit, pending final human review.