Skip to contents

Non-Inferiority Test

Source: R/efficacy_responder.R
test_non_inferiority.Rd

Tests non-inferiority of treatment vs comparator.

Usage

test_non_inferiority(
  data,
  outcome_var,
  trt_var,
  ref_group,
  ni_margin,
  type = c("continuous", "binary"),
  higher_better = TRUE,
  conf_level = 0.975,
  method = c("wilson", "wald", "exact")
)

Arguments

data

Data frame

outcome_var

Character. Outcome variable name

trt_var

Character. Treatment variable name

ref_group

Character. Reference group name

ni_margin

Numeric. Non-inferiority margin (positive value)

type

Character. "continuous" or "binary"

higher_better

Logical. TRUE if higher values are better (default: TRUE)

conf_level

Numeric. One-sided confidence level (default: 0.975, equivalent to two-sided 95% CI)

method

Character. Method for binary endpoints only: "wald", "wilson", "exact" (default: "wilson"). Ignored for continuous endpoints.

Value

List with:

  • estimate: Point estimate of difference (trt - ref)

  • ci_lower: Lower bound of one-sided CI

  • ci_upper: Upper bound (may be Inf for one-sided)

  • ni_margin: The margin used

  • non_inferior: Logical. TRUE if non-inferiority is concluded. When higher_better=TRUE: ci_lower > -ni_margin. When higher_better=FALSE: ci_upper < ni_margin.

  • conclusion: Character summary

  • method: Method used

Details

For continuous endpoints: Tests if (mean_trt - mean_ref) + ni_margin > 0 For binary endpoints: Tests if (prop_trt - prop_ref) + ni_margin > 0

Non-inferiority is concluded if the lower bound of the one-sided CI for the treatment difference exceeds -ni_margin.

The direction of the test depends on higher_better:

  • If higher_better=TRUE (default): the lower CI bound must exceed -ni_margin

  • If higher_better=FALSE: the upper CI bound must be below ni_margin

For binary endpoints, the Wilson method uses the Newcombe-Wilson hybrid CI approach. The Wald method provides a simpler alternative. For regulatory submissions, verify the CI method aligns with agency preferences.

References

IQWiG Methods v8.0, Section 10.3.5, p. 217-218.

Examples

if (FALSE) { # \dontrun{
# Non-inferiority test for continuous endpoint
result <- test_non_inferiority(
  data = adeff,
  outcome_var = "CHG",
  trt_var = "TRT01P",
  ref_group = "Active Control",
  ni_margin = 0.5,
  type = "continuous"
)

# Non-inferiority test for binary endpoint
result <- test_non_inferiority(
  data = adrs,
  outcome_var = "AVALC",
  trt_var = "TRT01P",
  ref_group = "Active Control",
  ni_margin = 0.10,
  type = "binary"
)
} # }