--- license: cc-by-nc-4.0 language: - en tags: - synthetic - healthcare - clinical-trials - orphan-drug - rare-disease - regulatory - pharmacovigilance - health-economics pretty_name: "Orphan Drug Trial Dataset: Registry, Patients, Outcomes, Adverse Events, Regulatory & Health Economics (Sample)" size_categories: - 10K **Not for clinical use.** This dataset is for ML development, benchmarking, schema prototyping, and education only. It must not be used to inform real patient care, trial design decisions, or regulatory submissions. ## Unit of observation Unlike the patient-level HC-GEN SKUs, **the unit here is the trial**. The 50 trials in this sample expand into thousands of derived patients and tens of thousands of longitudinal outcome records across six relational tables keyed on `trial_id` (and `patient_id`). ## Calibration anchors Sample-level observed values (seed 42, 50 trials): | Metric | Observed | Target | Anchor | |---|---|---|---| | Phase III fraction | 0.340 | 0.18–0.40 | orphan trial phase mix | | Orphan designation rate | 1.000 | 0.98–1.00 | all trials carry designation | | Fast-track rate | 0.360 | 0.16–0.50 | FDA expedited-program rates (engine p=0.35) | | Approval rate | 0.320 | 0.18–0.45 | orphan-drug regulatory yield | | Mean actual enrollment | 98.0 | 60–130 | small-N pivotal rare-disease trials | | Serious-AE fraction | 0.433 | 0.38–0.48 | CTCAE grade≥3 / serious-flag | | Genetic characterization rate | ~0.65 | 0.58–0.72 | genetically defined enrollment | | Median ICER per QALY | $337,072 | $220k–420k | orphan-therapy cost-effectiveness | | Active−control responder separation | 0.26 | ≥0.15 (floor) | efficacy signal | | X-linked male skew | 0.45 | ≥0.35 (floor) | Hemophilia/Duchenne | | Pediatric age separation | 30 yr | ≥20 (floor) | Duchenne/Neuroblastoma/Dravet | | High-grade AE → drug action | 0.58 | ≥0.35 (floor) | grade≥4 withdrawal/hospitalization | | Patient→trial referential violations | 0 | =0 (floor) | integrity | | Trial count | 50 | =50 (floor) | portfolio-size contract | Validation: **Grade A+ (10.00/10)** across all six canonical seeds (42, 7, 123, 2024, 99, 1), deterministic. ## Tables Six relational CSVs keyed on `trial_id` / `patient_id`: - **`hc_gen_006_trial_registry.csv`** — 50 trials × 22 cols: phase, design, MoA, disease (Orpha/ICD-10/prevalence), jurisdiction, sponsor, designations (fast-track/breakthrough/accelerated), enrollment, sites, primary endpoint. - **`hc_gen_006_patients.csv`** — ~4,900 patients × 14 cols: demographics, genetic mutation, severity, treatment arm, dose, consent, country, site. - **`hc_gen_006_treatment_outcomes.csv`** — ~55,000 longitudinal visit records × 13 cols: primary endpoint trajectory, responder flag, two biomarkers, PK concentration (one-compartment model), QoL, current severity, compliance, status. - **`hc_gen_006_adverse_events.csv`** — ~4,000 AEs × 12 cols: MedDRA term/SOC, CTCAE grade, seriousness, onset/duration, causality, outcome, action taken. - **`hc_gen_006_regulatory_pathway.csv`** — ~97 submissions × 12 cols: jurisdiction, submission type, review pathway/status, PDUFA date, advisory-committee vote, post-market requirements, exclusivity, pediatric study plan. - **`hc_gen_006_health_economics.csv`** — 50 econ records × 9 cols: annual cost, QALY gain, ICER, eligible population, reimbursement status, payer, outcomes-based contract. ## Loading ```python import pandas as pd trials = pd.read_csv("hc_gen_006_trial_registry.csv") patients = pd.read_csv("hc_gen_006_patients.csv") outcomes = pd.read_csv("hc_gen_006_treatment_outcomes.csv") # Active vs control responder rate pt = patients[["patient_id","treatment_arm"]] oc = outcomes.merge(pt, on="patient_id") oc["active"] = oc.treatment_arm.isin(["Active_Drug","Active_Drug_High_Dose","Combination"]) print(oc.groupby("active")["responder_flag"].mean()) ``` ```python from datasets import load_dataset trials = load_dataset("xpertsystems/hc-gen-006-sample", "trial_registry") ``` ## Use cases - Orphan-drug trial-design and enrollment-forecasting models. - Longitudinal endpoint / biomarker / PK trajectory modeling and responder classification. - Pharmacovigilance and adverse-event signal detection (MedDRA SOC, CTCAE grade). - Regulatory-pathway and approval-likelihood modeling across FDA/EMA/PMDA. - Health-economic (ICER/QALY) and market-access analysis. - Relational/longitudinal ML pipeline prototyping over a six-table star schema. ## Limitations (honestly disclosed) - **Trial-level synthetic data; fictional drugs.** All trials, patients, products (`XS-*`), and outcomes are simulated. Disease portfolio weights and trial mix are illustrative, not market-representative. - **Legacy global-RNG engine (determinism handled by the wrapper).** The engine uses NumPy's legacy global `np.random` API seeded once. The sample wrapper drives reproducibility by re-seeding `np.random.seed(seed)` before each run, verified to produce byte-identical tables across repeated same-seed runs. Recommended full-product fix: migrate to `np.random.default_rng`. - **Control-arm responder flag is near-zero by construction.** The engine only sets the per-visit `responder_flag` for active-arm responders; control arms therefore show ~0 flagged responders. This produces a clean active-vs-control efficacy separation but means `responder_flag` should not be read as an unbiased control-arm response estimate. - **High serious-AE fraction.** The CTCAE grade distribution places ~35% of AEs at grade≥3, so the serious-AE fraction (~0.43) runs higher than some real rare-disease trials; it reflects the engine's safety model, not a specific program. - **Marginal calibration, not full joint fidelity.** Univariate rates and the engineered structural separations (active/control efficacy, X-linked sex skew, pediatric age, AE-grade action gradient) are anchored; higher-order correlations beyond those engineered are not independently validated. - **No engine defects identified.** All sampling is reproducible under fixed seeding; the simulation runs cleanly and writes six internally-consistent tables with full referential integrity. The only design note is the legacy-RNG approach above. - **Small-portfolio variance.** At 50 trials some trial-level rates (fast-track, phase mix, approval) carry binomial variance; scorecard ranges accommodate this without masking real misses, and structural floors are weighted to dominate. ## Commercial / full version | | Sample (this) | Full (commercial) | |---|---|---| | Trials | 50 | 150 (configurable) | | Derived patients | ~4,900 | ~15,000 | | Tables | 6 (full schema) | 6 (full schema) | | Formats | CSV | CSV / Parquet | | Determinism | seed-controlled via wrapper | migrated to `default_rng` on request | | Seeds / reproducibility | 6 canonical | Unlimited | | License | CC-BY-NC-4.0 | Commercial | | Support | — | SLA, custom portfolios, vertical extensions | Contact **pradeep@xpertsystems.ai** · https://xpertsystems.ai ## Citation ```bibtex @dataset{xpertsystems_hcgen006_2026, title = {HC-GEN-006: Synthetic Orphan Drug Trial Dataset --- Registry, Patients, Longitudinal Outcomes, Adverse Events, Regulatory Pathway & Health Economics (Sample)}, author = {XpertSystems.ai}, year = {2026}, publisher = {Hugging Face}, note = {Synthetic data. Not for clinical use. Fictional drugs. Calibration anchors: FDA/EMA/PMDA orphan-drug regulatory frameworks; CTCAE v5 adverse-event grading; MedDRA system-organ-class taxonomy; Orphanet/ICD-10 disease coding; orphan-therapy ICER/QALY literature.}, url = {https://huggingface.co/datasets/xpertsystems/hc-gen-006-sample} } ```