Pharmacometric Events – Versions & Changelog

stubbi / Pharmacometric Events

Public

About dataset version

Pharmacometric Event Dataset

Subject-level event data for pharmacometric modeling, structured in a long format that mixes dosing and observation records distinguished by EVID. Multiple endpoints (e.g., plasma concentration vs. biomarker) are supported via DVID. The dataset is designed for:

Population PK and PK/PD model fitting (compartmental, mixed-effects)
Exposure–response analyses and simulations (e.g., VPCs)
Covariate exploration (e.g., body weight effects, concomitant meds)
Benchmarking classical vs. quantum/quantum-inspired algorithms for dose-finding and PK/PD tasks

Contents & schema

Shape: 2,820 rows × 11 columns (one row per subject–time–event).

Column	Type (suggested)	Units / Encoding	Description
`ID`	int	—	Subject identifier.
`BW`	float	kg	Body weight (baseline covariate).
`COMED`	int	0/1	Concomitant medication indicator (0 = No, 1 = Yes).
`DOSE`	float	mg	Nominal dose level associated with the record (arm/level). Prefer `AMT` for modeling.
`TIME`	float	hours	Time since the first drug administration.
`DV`	float	mg/L if `DVID=1`; ng/mL if `DVID=2`	Observed value of the dependent variable; may be missing if `MDV=1`.
`EVID`	int	0/1	NONMEM event type: `0` = observation, `1` = dosing event.
`MDV`	int	0/1	Missing-DV flag: `1` = DV missing/ignored, `0` = DV present.
`AMT`	float	mg (or μg per study design)	Actual dose amount for dosing events; `0` on observation rows.
`CMT`	int	model-dependent	Compartment index (e.g., 1 = central), to be mapped in your model.
`DVID`	int	1/2	Endpoint selector: `1` = concentration, `2` = biomarker.

Event semantics (important)

Dose rows: EVID=1, AMT>0, typically MDV=1; DV (if present) must be ignored.
Observation rows: EVID=0, AMT=0, typically MDV=0; DV carries the measurement.
Multi-endpoint: Split by DVID prior to modeling or evaluation.

Modeling tips & good practices

Time origin: TIME is since first administration. If your workflow needs per-period re-zeroing (e.g., multiple doses), realign within-subject accordingly.
Units: Keep unit discipline: DVID=1 → mg/L, DVID=2 → ng/mL. Avoid mixing during joins/aggregations.
Compartments: CMT indices are model-dependent; map them to your structural model (e.g., central/peripheral).
Missingness: Respect MDV=1—do not impute DV on dosing rows.
Covariates: Explore BW (e.g., allometric scaling on CL/V) and COMED as categorical covariates.
Diagnostics: For quick checks, stratify concentration–time plots by DOSE level and COMED, and inspect per-subject residuals.

Data quality & caveats

The table intentionally mixes dose and observation events—filter carefully for summaries and plotting.
DOSE reflects nominal dose level; prefer AMT for quantitative model inputs.
No BLQ/LLOQ flags are provided—apply your own censoring rules (e.g., M3 method) if required.
CMT semantics may vary by your structural model; document your mapping for reproducibility.

Provenance

Upstream repository (MIT License): Quantum-Innovation-Challenge /LSQI-Challenge-2025 on GitHub.
Challenge site: Quantum Innovation Challenge 2025 (Life Sciences & Quantum Innovation, LSQI).
This Aqora dataset mirrors the relevant challenge data and reshapes it for convenient PK/PD workflows while preserving the event semantics.

Contact

Questions or corrections? Please open a discussion on the dataset page.