Related Experiment Videos
Small proportions: what to report for confidence intervals?
Hilde Tobi1, Paul B van den Berg, Lolkje T W de Jong-van den Berg
1Department of Social Pharmacy, Pharmacoepidemiology and Pharmacotherapy, Groningen University Institute for Drug Exploration, Groningen, The Netherlands. h.tobi@rug.nl
Pharmacoepidemiology and Drug Safety
|February 19, 2005
Summary
For small proportions, the Exact method and Score method with continuity correction (CC) provide reliable confidence intervals (CIs), outperforming standard methods in pharmaco-epidemiology research.
Area of Science:
- Biostatistics
- Pharmacoepidemiology
- Statistical Methods
Background:
- Confidence intervals (CIs) are more informative than point estimates or p-values in pharmaco-epidemiology.
- Calculating CIs for small proportions is challenging and methods are inconsistently reported.
Purpose of the Study:
- To identify the most suitable method for calculating confidence intervals (CIs) for small proportions.
- To compare seven approximate methods against the Clopper-Pearson Exact method.
Main Methods:
- A simulation study evaluated 90%, 95%, and 99% CIs with sample size 1000 and proportions from 0.001 to 0.01.
- Key quality criteria included coverage probability and interval width.
- Methods were illustrated using real-world pharmaco-epidemiology data.
Main Results:
- Standard Wald methods demonstrated insufficient coverage probability.
- The Exact method and the Score method with continuity correction (CC) exhibited superior performance.
- Real-world examples confirmed variations in results across different methods.
Conclusions:
- The Exact method and the Score method with continuity correction (CC) are recommended for CIs with small proportions (pi ≤ 0.01).
- These methods offer improved accuracy and reliability in pharmaco-epidemiological analyses.