Skip to contents

Compute and tabulate estimates for log-log regression

Source: R/plot_doseprop.R
df_doseprop.Rd

Computes per-metric log-log regression statistics and returns them in a cacheable, replottable container. The returned object is a pmx_stats() container with class c("doseprop_stats", "pmx_stats") and three slots — stats (per-metric regression body), obs (filtered observation rows used for the scatter overlay), and config (regression configuration: metric_name_var, metric_value_var, dose_var, ci, method) — so that plot_doseprop() / plot_build_doseprop() can render directly from this object without re-fitting any regressions.

Usage

df_doseprop(
  data,
  metrics,
  metric_name_var = "PPTESTCD",
  metric_value_var = "PPORRES",
  dose_var = "DOSE",
  method = "normal",
  ci = 0.9,
  sigdigits = 3
)

Arguments

data

Input dataset for log-log regression. Default expected format is output from PKNCA::pk.nca() (i.e., SDTM PP formatting).

metrics

character vector of exposure metrics in data to plot

metric_name_var

Column in data containing the metric names listed in metrics. Accepts bare names or strings. Default is PPTESTCD.

metric_value_var

Column in data containing the exposure metric values (dependent variable). Accepts bare names or strings. Default is PPORRES.

dose_var

Column in data containing the dose (independent variable). Accepts bare names or strings. Default is DOSE.

method

character string specifying the distribution to be used to derive the confidence interval. Options are "normal" (default) and "tdist".

ci

confidence interval to be calculated: a single numeric value in (0, 1), e.g. 0.90 (default) or 0.95.

sigdigits

number of significant digits for the reported Intercept, StandardError, Power, LCL, UCL, and PowerCI columns. PowerCI pads every value to this many digits so the facet labels line up. The Proportional / Interpretation verdict is taken from the unrounded confidence bounds, so it never moves with sigdigits.

Value

A doseprop_stats container (subclass of pmx_stats) with three slots:

stats

One row per metric and columns Intercept, StandardError, CI, Power, LCL, UCL, Proportional, PowerCI, Interpretation, plus the column named in metric_name_var. Interpretation is "Dose-proportional" when the ci interval for Power contains 1, and less or greater than dose-proportional when it lies wholly below or above 1. It is a screening label: a CI that contains 1 does not establish proportionality, and no predefined equivalence bounds are applied.

obs

The filtered observation rows used for the plot scatter overlay.

config

Named list with metric_name_var, metric_value_var, dose_var, ci, method.

Pass directly to plot_doseprop() or plot_build_doseprop() to replot without refitting.

See also

Other dose proportionality: is_doseprop_stats(), plot_build_doseprop(), plot_doseprop(), style_doseprop()

Examples

df_doseprop(dplyr::filter(data_sad_nca, PART == "Part 1-SAD"),
            metrics = c("aucinf.obs", "cmax"))
#> <doseprop_stats>
#>   stats: 2 rows x 10 columns
#>   obs:   60 rows
#>   config: metric_name_var = PPTESTCD, metric_value_var = PPORRES, dose_var = DOSE, ci = 0.9, method = normal
#> 
#>   stats body:
#>   Intercept StandardError  CI Power   LCL  UCL Proportional
#> 1      3.97        0.0437 90% 0.979 0.907 1.05         TRUE
#> 2      1.02        0.0597 90% 1.060 0.959 1.16         TRUE
#>                            PowerCI    Interpretation   PPTESTCD
#> 1 Power: 0.979 (90% CI 0.907-1.05) Dose-proportional aucinf.obs
#> 2  Power: 1.06 (90% CI 0.959-1.16) Dose-proportional       cmax
#> 
#>   Use `x$obs` for the observation overlay.