interpret_noninferiority: Non-inferiority test interpretations of confidence intervals.
In confinterpret: Descriptive Interpretations of Confidence Intervals

Description Usage Arguments Details Value Examples

Conduct non-inferiority tests on confidence intervals using a standard set of interpretations. Takes a confidence interval around an effect size measure, for example from the results from a randomised controlled trial comparing the outcome for an intervention group to a control group.

1
2
3

interpret_noninferiority(ci, actual_null = 0, ni_margin = 0.1,
  groups = c("Control intervention", "Test intervention"),
  beneficial_outcome = TRUE)

`ci`	A single row from a matrix of the type returned by `confint()`, containing the confidence interval for the parameter estimate. The two columns provide the lower and upper confidence limits.
`actual_null`	The value that precisely zero difference would have in the parameter being examined. For an absolute measure this will typically be 0. For a relative measure it will typically be 1. This is the starting point that the `ni_margin` is applied to in order to establish the point for comparison.
`ni_margin`	Numerical value specifying the non-inferiority margin to be used. Provided as a positive number; the value of `beneficial_outcome` defines whether it is added to or subtracted from the `actual_null` value to position the boundary. See Details.
`groups`	A character vector of length 2 containing short descriptive names of the groups being compared, such as the names of the interventions being compared if the confidence interval is derived from an outcome effect size measure in a randomised controlled trial. Give the name of the intervention given to the comparison or control group first and the new or tested intervention second.
`beneficial_outcome`	Is the outcome to be treated as beneficial (i.e., a higher value of the outcome is superior)? For harmful outcomes (where lower numbers are better), set this to FALSE. If, for example, the outcome is measuring something like prevalence of patients recovering from a disease, that is likely to be beneficial; if it is measuring the prevalence of patients falling ill with a disease it is likely to be not beneficial.

Non-inferiority tests are typically specified in analysis plans where a new intervention is being compared to an existing one, especially if it has some benefit other than the effect being measured. For example, the new intervention might be cheaper than the old one, or have fewer side effects. In these circumstances, the new intervention may not need to prove itself more effective than the old one, but just to be not substantially worse - i.e., non-inferior.

When conducting non-inferiority tests, a non-inferiority margin is defined. This is effectively the leeway of small, practically insignificant differences by which the new intervention is allowed to under-perform the old one and still be considered non-inferior.

The non-inferiority margin is defined as being a small amount on the inferior side of an actual null result. If using beneficial_outcome = TRUE (the default), the non-inferiority margin will extend below actual_null; if beneficial_outcome = FALSE it extends above it.

You are able to supply descriptive names of the interventions being compared, and these will be inserted into the resultant interpretation. If the comparison / baseline intervention does not have a convenient name (such as "Placebo"), some of these might be suitable:

"Business as usual"
"Treatment as usual"
"No intervention"

(Whilst these may work well as short descriptions for outputting from this function, in your reporting you will still normally want to provide information about what exactly those in a comparison group got.)

This function is provided in the form of a convenience wrapper for confinterpret, using interpretations_noninferiority as its interpretation_set.

A list object of class interpretation_result with elements stating the interpretation in different formats, plus the parameters used to generate the interpretation.

# Establish a test confidence interval
ci_test <- matrix(c(-0.05, 0.05),
                  nrow = 1, dimnames = list("estimate",
                                            c("2.5 %", "97.5 %")))
interpret_noninferiority(ci_test, 0, 0.1, c("Treatment as usual",
                                            "New treatment"))