Confidence regions for multiple outcomes, effect modifiers, and other multiple comparisons

Area of Science:

• Epidemiology
• Biostatistics

Background:

• Confidence intervals are commonly used but may understate uncertainty when estimating multiple parameters simultaneously.
• This can lead to a false sense of precision, potentially causing irreproducible results or poor decision-making.

Purpose of the Study:

• To introduce and illustrate the application of confidence bands as an extension of confidence intervals for multiple parameters in epidemiology.
• To provide practical guidance on selecting appropriate confidence regions based on scientific interest.

Main Methods:

• The study demonstrates the use of sup-t confidence bands through three case studies involving publicly available data.
• Code examples are provided in SAS, R, and Python for practical implementation.

Main Results:

• Confidence bands effectively address the underestimation of uncertainty inherent in using multiple confidence intervals.
• The case studies cover estimation of multiple causal effects and effect measure modification by binary and continuous variables.

Conclusions:

• Sup-t confidence bands are preferred for situations involving multiple parameters, such as effect measure modification or dose-response analyses, due to their statistical properties and ease of use.
• The choice of confidence region should align with the scientific inquiry, prioritizing confidence bands for multi-parameter estimation.