Validate spec variables against skeleton data
Source:R/tteplan_validate_spec.R
tteplan_validate_spec.RdChecks that all implementation$variable references in the spec actually
exist as columns in the skeleton data.table. For categorical confounders,
also checks that the declared categories match the data. Collects all
issues before reporting.
Arguments
- spec
Parsed study specification from
tteplan_read_spec().- skeleton
A data.table skeleton (person-week panel) to validate against.
- skeleton_batch
Batch number of
skeleton. The prevalent-user warning reports it, so a reader knows which batch the check measured.
Details
It also checks the observation column of every enrollment that names one:
the column MUST exist in the skeleton, and it MUST be logical.
tteplan_read_spec() cannot run that check, because it reads no data. An
enrollment that declares the row_presence sentinel names no column, so
there is nothing to check.
The prevalent-user check
An enrollment gets its incident-user design from a washout rule, of type
no_prior_value or only_prior_value. The check warns when no washout
covers the enrollment's intervention level. All four rule blocks are
searched: inclusion_criteria$criteria, exclusion_criteria, and the
enrollment's additional_inclusion and additional_exclusion.
Coverage is measured on the weekly rows of skeleton. A prevalent week is a
week at the intervention level that follows an earlier week of the same
person at that level. A washout covers the enrollment when it makes every
prevalent week ineligible. The check evaluates the eligibility expression
the compiler builds, so it measures the column the skeleton will hold. A
first initiation stays eligible and is not an uncovered week. The enrollment
passes when at least one washout covers it.
The washout column and the treatment column often differ. A specification can wash out a whole drug class, then compare one sub-type of that class against no treatment. A test on the two column names cannot answer that, and it cannot see a washout that names the right column at the wrong level.
Set options(swereg.warn_prevalent_user = FALSE) to silence the warning.
Use it for a discontinuation or switching study, which enrols prevalent
users by design.