A simulated oncology study with crossover, dose reduction and compressed doses

Two hundred patients shaped like the phase 3 RECORD-1 trial of everolimus in metastatic renal cell carcinoma, following the tumour-growth model of Stein et al. (2012). Patients are randomised to everolimus 10 mg daily or to placebo; most placebo patients cross over to everolimus when their disease progresses, and about a quarter of everolimus patients reduce to 5 mg, some after a short interruption.

Usage

onc_sim

Format

A data frame with 3379 rows and 15 columns:

ID

Patient identifier, 1 to 200.

TIME

Study day.

NTIME

Nominal (protocol) day; scans are planned, so it equals TIME.

DV

Tumour size in cm or trough concentration in ng/mL, by NAME.

AMT

Daily dose in mg: 10, 5, or 0 while off treatment or on placebo.

EVID

1 on a dose record, 0 on an observation.

CMT

1 dosing, 2 concentration, 3 tumour size.

ADDL

Additional doses implied by this record.

II

Interdose interval in days, 1 on every dose record.

NAME

"SLD" or "Everolimus trough"; NA on dose records.

CENS

1 where a concentration is at the quantification limit.

ARM

Randomised arm, "Everolimus 10 mg" or "Placebo".

CROSSOVER

TRUE where a placebo patient crossed over to everolimus.

BSLD

Baseline tumour size in cm, the covariate the model scales on.

AGE, SEX

Baseline age in years and sex.

Source

Simulated by scripts/build-example-data.R, following Stein A, Wang W, Carter AA, Chiparus O, Hollaender N, Kim H, Motzer RJ, Sarr C. Dynamic tumor modeling of the dose-response relationship for everolimus in metastatic renal cell carcinoma using data from the phase 3 RECORD-1 trial. BMC Cancer 2012;12:311. doi:10.1186/1471-2407-12-311 . Not real patient data.

Details

It carries three shapes the public-data survey has no other example of. The sum of longest tumour diameters is a slow endpoint anchored on a per-patient baseline and measured over a year, where every other endpoint in the survey is a concentration or a fast pharmacodynamic signal. The dose changes within a patient for reasons that patient's own data explains. And the daily regimen is written with ADDL and II rather than one row per dose — 3379 rows instead of the 71096 the expanded form would need — which is the compressed dose encoding the survey states outright that it has no example of.

The tumour model is the paper's model 2 with its published parameters: $$dy/dt = r_i - E_{dose,i}\,y_i$$ where \(E_{dose}\) is \(E_{10}\) at 10 mg, \(E_5\) at 5 mg and zero off treatment, and each patient's growth rate and drug effect scale with baseline tumour size. Simulated one-year change from baseline is +139.8% on placebo, +21.6% at 5 mg and −13.8% at 10 mg, against the paper's reported +142.1%, +22.4% and −15.7%.

Pharmacokinetics do not drive the tumour, exactly as in the paper: the effect is indexed by the dose in force. Trough concentrations ride alongside as a second endpoint, simulated from a one-compartment oral model at everolimus's published disposition, so a pharmacokinetic fit of them means something.

See also

mixroute_sim