The standard stress test — standard

A recommended, citable set of families with fixed settings, so that a statistical analysis plan can say "the standard stress test of ballast" and mean one thing. It contains, in order:

Usage

standard_plan(
  spec,
  site = NULL,
  strata = site,
  baseline = NULL,
  deltas = NULL,
  shift_visits = NULL,
  ice = NULL,
  m = 25,
  n_perm = 2000,
  max_k = 5
)

Arguments

spec: A stress_spec().
site: Optional site or centre column (deletions, random intercept, re-randomisation strata unless strata is given).
strata: Optional stratification column for the re-randomisation test.
baseline: Optional baseline column for cLDA.
deltas: Optional shifts for the tipping-point family.
shift_visits: Visits whose imputed values the tipping family shifts; NULL shifts the contrast visit only (the plan text states the visits; pass every visit level to shift all imputed values).
ice: Optional intercurrent-event table for stress_estimand().
m: Imputations for the multiple-imputation families.
n_perm: Permutations for the re-randomisation test.
max_k: Maximum removals in the deletion search.

Value

A stress_plan.

Details

covariance structure: us, ar1h, ar1, cs, csh, toep, toeph;
inference: Satterthwaite, Kenward-Roger, linear Kenward-Roger, residual, sandwich with between-within df, jackknife sandwich;
covariate set: each covariate dropped, all dropped;
LS-mean weighting: equal and observed margins;
leave-one-out: every subject, and every level of site if given;
re-randomisation test: 2,000 permutations, within strata if given;
reference-based multiple imputation: MAR, J2R, CR, CIR, 25 imputations;
delta-adjusted multiple imputation over deltas if given;
estimator class: GEE independence, ANCOVA, robust ANCOVA, plus a random intercept for site and cLDA for baseline if given; the pre-specified MMRM on complete cases is also fitted, drawn and tabulated, as a diagnostic that is not counted (it answers a different question from the filed analysis, so it does not enter the verdict);
greedy deletion search for the loss of significance, up to 5 subjects;
intercurrent-event strategies if ice is given (treatment policy, hypothetical, while on treatment).

Examples

spec <- stress_spec(mmrm::fev_data, "FEV1", "USUBJID", "AVISIT", "ARMCD", "PBO",
                    covariates = ~ RACE + SEX + FEV1_BL)
standard_plan(spec, site = "RACE", baseline = "FEV1_BL")
#> Standard stress test (ballast 0.7.4)
#> 
#> Pre-specified analysis: FEV1 ~ RACE + SEX + FEV1_BL + ARMCD + AVISIT + ARMCD:AVISIT + us(AVISIT | USUBJID); contrast TRT minus PBO at VIS4; Kenward-Roger df, Kenward-Roger-Linear variance; alpha 0.05; equal LS-mean weights.
#> 
#> Perturbation families, run after unblinding exactly as listed; every refit is judged against the pre-specified estimate,
#> its 95% confidence interval and alpha.
#> Classification: fragile if any counted refit (an alternative analysis or a deletion) lands outside the pre-specified 95% confidence interval; otherwise sensitive if any crosses alpha; otherwise robust.
#> Specifications reached by two families are counted once; tipping-point grids are searches and are not counted; deletions are
#> counted separately.
#> 
#> 1. Within-subject covariance structure: structures = us, ar1h, ar1, cs, csh, toep, toeph
#> 2. Degrees-of-freedom method and variance estimator: defaults
#> 3. Covariate set: defaults
#> 4. Least-squares-mean weighting (equal versus observed margins): defaults
#> 5. Leave-one-out deletions: defaults | by = RACE
#> 6. Re-randomisation test: n = 2000; strata = RACE
#> 7. Reference-based multiple imputation (MAR, J2R, CR, CIR): m = 25
#> 8. Estimator class: estimators = complete_cases, gee_independence, ancova, robust_ancova, random_intercept, clda; random = RACE; baseline = FEV1_BL; diagnostic = complete_cases. The complete cases refit is fitted, drawn and tabulated as a diagnostic and is not counted.
#> 9. Smallest set of subjects that changes the conclusion: target = significance; max_k = 5