One row per column of the matrix synpmx_pca_summarize() decomposed: one
per baseline covariate, and one per endpoint per retained nominal time. This
is the grid the whole method sits on, so it is where to look first when a
generated dataset is missing a visit or a covariate.
Arguments
- x
A dataset from
synpmx_pca(), or its trial summary.
Value
A data frame with feature, kind, endpoint, time, covariate,
level, patients, center, scale and transform. Marked
"restricted_not_releasable": the counts and moments are read from real
data.
Details
center and scale are the column's mean and standard deviation on the
modelling scale, which is the log scale for a positive endpoint — see
transform in the same row. patients is how many subjects hold an
observation in that cell; cells held by fewer than min_column_patients were
dropped rather than modelled, so they do not appear here at all.
Examples
data <- pmx_simulated_fixture(60)
roles <- pmx_roles(
id = "ID", time = "TIME", nominal_time = "NTIME", dv = "DV", amt = "AMT",
evid = "EVID", cmt = "CMT", dvid = "DVID", mdv = "MDV"
)
head(pca_features(synpmx_pca_summarize(data, roles)))
#> feature kind endpoint time covariate level patients center
#> 1 dv_cp__1 endpoint_cell cp 0.25 <NA> <NA> 60 0.1798121
#> 2 dv_cp__2 endpoint_cell cp 1.00 <NA> <NA> 60 2.1006899
#> 3 dv_cp__3 endpoint_cell cp 2.00 <NA> <NA> 60 1.4318031
#> 4 dv_cp__4 endpoint_cell cp 6.00 <NA> <NA> 60 0.5278136
#> 5 dv_cp__5 endpoint_cell cp 12.25 <NA> <NA> 60 0.3338168
#> 6 dv_cp__6 endpoint_cell cp 13.00 <NA> <NA> 60 2.1598688
#> scale transform
#> 1 0.07150904 log_offset
#> 2 0.08380604 log_offset
#> 3 0.08179566 log_offset
#> 4 0.07574147 log_offset
#> 5 0.07356425 log_offset
#> 6 0.08392740 log_offset