Synthetic biomarker study: results

Biomarker results for a made-up study of 200 participants, twelve biomarkers and five visits, with one row per participant, biomarker and visit. It is the one table every biomarker chart needs, and it exists so that a test can assert an answer that is known in advance.

Usage

Synthetic_Results

Format

A data frame with 11472 rows and 6 columns. The column names are the defaults safety.viz uses for results data.

USUBJID: character Participant identifier, BIO-001 to BIO-200.
VISIT: character Visit name: Baseline, Week 2, Week 4, Week 8 or Week 12.
VISITNUM: integer Visit order, the week number: 0, 2, 4, 8 or 12.
TEST: character Biomarker name, in ASCII: IL-6, CRP, TNF-alpha, IL-10, IFN-gamma, IL-2, IL-1beta, IL-8, VEGF, Ferritin, D-dimer or LDH.
STRESU: character Unit of the result.
STRESN: numeric Result, to three decimal places; missing where a sample was not analysable.

Source

Generated by data-raw/synthetic-study.R. The same table is in the package as system.file("extdata", "synthetic_results.csv", package = "gsm.bio").

Details

Everything in it is synthetic. The values are random draws from the model in the script data-raw/synthetic-study.R, run with the seed recorded in Synthetic_Truth. No real study data was used and nothing was copied from any real study; the biomarker names are generic analyte names, with levels chosen only to look plausible.

Planted effects

Two of the three planted effects are in this table. Their true sizes are parameters of the generating model, held in Synthetic_Truth.

A difference between two groups in change from baseline. In the Treatment arm, IL-6 is lower by 1.5 pg/mL at every visit after Baseline. So the difference between the arms in mean change from Baseline, Treatment minus Placebo, is -1.5 pg/mL at each of Week 2, Week 4, Week 8 and Week 12. The arms do not differ at Baseline.
A correlation between two biomarkers. TNF-alpha and IL-10 are drawn from a bivariate normal distribution with a Pearson correlation of 0.6 at each visit, on the scale stored here.

The third, a survival difference between high and low Baseline CRP, is in Synthetic_Outcomes. Every other biomarker is null by construction: unrelated to arm, to every other biomarker and to the outcome. Each planted biomarker carries its one effect only.

The true sizes describe the model before the stored values were rounded to three decimal places, which moves none of them by an amount a test can see.

Model

Each biomarker has a participant-level part, 60 percent of its variance, and a visit-level part. IL-6, TNF-alpha, IL-10, IL-2, IL-1beta, VEGF and LDH are normal on the stored scale; CRP, IFN-gamma, IL-8, Ferritin and D-dimer are log-normal, so right-skewed.

Imperfections

Baseline is complete: every participant has all twelve biomarkers. After Baseline, two things are missing completely at random, as they would be in a real study, so that a chart has dropped rows to count:

a participant misses a whole visit with probability 0.05, and that visit has no rows at all;
a result is not analysable with probability 0.02, and its row is present with a missing STRESN.