Exact Patient Omission Diagnostics for Declared Hypotheses
Source:R/leave-one-patient-out.R
LeaveOnePatientOut.RdComputes full tests once through ExactPatientTests, then evaluates explicitly
declared patient omissions through the same API. An incomplete full hypothesis
cannot become testable by omitting a missing or ineligible patient. No clinical
groups, independence, omission set or multiplicity policy is inferred.
Usage
LeaveOnePatientOut(
endpoints,
design,
hypotheses,
omissions,
missingness = "require_complete",
max_allocations = 1000000L
)Arguments
- endpoints, design, hypotheses
Inputs with the
ExactPatientTestsschemas. Selected columns must be dimensionless vectors. Extra columns are ignored.- omissions
Data frame with character
hypothesis_idandpatient_id. Each pair must be unique; the patient must belong to that hypothesis's contrast design. Zero rows are allowed with typed character columns.- missingness
Only
"require_complete"is supported.- max_allocations
Allocation ceiling passed unchanged to full and reduced exact tests. No random approximation or changed numerical rule is used.
Value
List with full_tests (the complete exact-test result) and omissions
(one row per supplied omission in its original order). Omission rows retain
hypothesis keys and full results, reduced expected/supplied/evaluable counts,
the omitted arm, an explicit empty_arm flag, reduced results and effect
direction diagnostics. reduced_status is UNAVAILABLE_FULL_INCOMPLETE,
UNAVAILABLE_EMPTY_ARM or EXACT, in that priority order. Reduced allocations,
grid step, effect and p value are NA when no reduced test is performed.
Supplied counts include present endpoint rows regardless of eligibility;
evaluable counts require a finite value and nonmissing TRUE eligibility.
Full-incomplete omissions do not execute reduced tests. Zero effects have
direction ZERO; direction preservation is NA when either effect is missing.
Arithmetic overflow in an effect change fails rather than returning Inf.
Examples
endpoints <- data.frame(endpoint_id = "e", patient_id = letters[1:5],
value = c(1, 2, 4, 6, 7), eligible = TRUE)
design <- data.frame(contrast_id = "c", patient_id = letters[1:5],
group = c("A", "A", "B", "B", "B"))
hypotheses <- data.frame(hypothesis_id = "h", endpoint_id = "e",
contrast_id = "c", family_id = "f")
omissions <- data.frame(hypothesis_id = "h", patient_id = letters[1:5])
out <- LeaveOnePatientOut(endpoints, design, hypotheses, omissions)
out$omissions[, c("omitted_patient", "mean_difference", "p_value")]
#> omitted_patient mean_difference p_value
#> 1 a -3.666667 0.2500000
#> 2 b -4.666667 0.2500000
#> 3 c -5.000000 0.3333333
#> 4 d -4.000000 0.3333333
#> 5 e -3.500000 0.3333333