---
license: cc-by-nc-4.0
language:
- en
tags:
- synthetic
- healthcare
- hemophilia
- hematology
- rare-disease
- gene-therapy
- bleeding-disorders
- clinical-trial
pretty_name: "Hemophilia A/B: Bleeding, Factor Replacement, Joint Damage, Inhibitors & Gene Therapy (Sample)"
size_categories:
- n<1K
configs:
- config_name: patients
  default: true
  data_files: hc_gen_003_sample.csv
- config_name: bleeding_events
  data_files: hc_gen_003_bleeding_events.csv
- config_name: joint_longitudinal
  data_files: hc_gen_003_joint_longitudinal.csv
- config_name: factor_pk_visits
  data_files: hc_gen_003_factor_pk_visits.csv
---

# HC-GEN-003 — Hemophilia A/B: Bleeding Episodes, Factor Replacement, Joint Damage, Inhibitors & Gene Therapy (Sample)

Synthetic patient-level dataset for **hemophilia A and B**, spanning genotype/severity classification, prophylaxis and factor-replacement pharmacology, non-factor/advanced therapies (emicizumab, fitusiran, marstacimab, concizumab), inhibitor development and immune tolerance induction, bleeding-episode analytics, joint-damage progression, biomarkers, and clinical outcomes — with three companion longitudinal tables.

This is a **500-patient sample** of the full HC-GEN-003 product. It is **synthetic** — generated by a calibrated simulation engine. It contains **no real patient data**.

> **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, diagnosis, or treatment decisions.

## Calibration anchors

Sample-level observed values (seed 42, n=500). Some metrics are calibrated to the engine's **observed** deterministic output rather than the literature target the engine was aiming for; where they differ, the gap is disclosed under Limitations.

| Metric | Observed | Target | Anchor |
|---|---|---|---|
| Severe-hemophilia fraction | 0.456 | 0.40–0.50 | WFH Annual Global Survey |
| Hemophilia A fraction | 0.800 | 0.76–0.84 | WFH registry A:B ratio |
| Inhibitor prevalence | 0.152 | 0.10–0.18 | USHTCN / CDC surveillance |
| Female fraction | 0.034 | 0.00–0.06 | X-linked inheritance |
| Mean ABR | 9.98 | 7.5–11.5 (observed) | engine intent: PedNet/WFH 5–8 |
| Prophylaxis rate | 0.548 | 0.45–0.62 (observed) | engine intent: WFH 70–80% |
| Mean Pettersson score | 23.8 | 20–27 (observed) | engine intent: 10–15 |
| Emicizumab rate (Hem A) | 0.276 | 0.20–0.50 | HAVEN program uptake |
| Factor-activity gap (Mild − Severe) | 21.8 pp | ≥15 (floor) | WFH severity definition |
| Prophylaxis ABR reduction | 15.6 | ≥8 (floor) | Manco-Johnson NEJM 2007 |
| Inhibitor excess (severe Hem A) | 0.353 | ≥0.20 (floor) | RODIN / USHTCN |
| Gene-therapy rate | 0.000 | =0 (floor; disclosed bug) | dead eligibility gate |

Validation: **Grade A+ (10.00/10)** across all six canonical seeds (42, 7, 123, 2024, 99, 1), deterministic — including all three companion tables (byte-identical across runs).

## Tables

This SKU ships a relational set keyed on `patient_id`:

- **`hc_gen_003_sample.csv`** — main patient table, 500 rows × 102 columns.
- **`hc_gen_003_bleeding_events.csv`** — per-bleed event log (~5,000 rows): type, joint, severity, trigger, duration, treatment response, hospitalization, ISTH severity, across 20 follow-up years.
- **`hc_gen_003_joint_longitudinal.csv`** — annual joint-damage trajectory (10,000 rows = 500 × 20 yrs): Pettersson, HJHS, synovitis grade, ankle/knee/elbow ROM, new target joints.
- **`hc_gen_003_factor_pk_visits.csv`** — quarterly PK monitoring visits (20,000 rows = 500 × 40 quarters): trough, peak, half-life, aPTT, dose-adjustment, adherence.

## Schema highlights (main table) by module

Demographics & genotype (`hemophilia_type` A/B/C, `severity_class`, `factor_activity_pct`, `mutation_type`, `inhibitor_high_risk_mutation`); treatment & prophylaxis (product type, dose, frequency, trough, half-life, adherence, annual IU & cost, home infusion); non-factor/advanced therapies (emicizumab regimen & ABR reduction, fitusiran, marstacimab, concizumab); inhibitors (titer, class, onset, ITI protocol/success/duration, bypassing agent, thrombosis); bleeding (ABR, joint-bleed rate, target joints, CNS/GI/psoas, ISTH-BAT, HJHS, WFH scores); joint damage (synovitis, Pettersson, MRI score, cartilage/osteophyte/joint-space, ROM, replacement, gait); labs (aPTT, VWF, recovery, thrombin generation/ETP, D-dimer, CRP, IL-6, viral serologies); gene therapy (product, vector dose, multi-year expression, ABR reduction, NAb, response — see disclosed bug); outcomes (mortality, hospitalization, SF-36, PROMIS, Haem-A-QoL, PHQ-9, GAD-7, employment).

## Loading

```python
import pandas as pd
patients = pd.read_csv("hc_gen_003_sample.csv")
bleeds   = pd.read_csv("hc_gen_003_bleeding_events.csv")
joints   = pd.read_csv("hc_gen_003_joint_longitudinal.csv")
pk       = pd.read_csv("hc_gen_003_factor_pk_visits.csv")

# Join example: total bleeds per patient vs recorded ABR
bleed_counts = bleeds.groupby("patient_id").size().rename("logged_bleeds")
patients = patients.merge(bleed_counts, on="patient_id", how="left")
```

```python
from datasets import load_dataset
patients = load_dataset("xpertsystems/hc-gen-003-sample", "patients")
bleeds   = load_dataset("xpertsystems/hc-gen-003-sample", "bleeding_events")
```

## Use cases

- Bleed-rate and joint-arthropathy progression modeling (cross-sectional + longitudinal tables).
- Prophylaxis vs on-demand treatment-effect simulation; PK-guided dosing.
- Inhibitor-risk and immune-tolerance-induction outcome modeling.
- Relational/longitudinal ML pipeline prototyping across a multi-table star schema.
- Synthetic-data methodology benchmarking.

## Limitations (honestly disclosed)

These are real defects/behaviors of the underlying engine, surfaced rather than hidden. The scorecard is calibrated to the engine's **observed** deterministic output, and misses against the engine's own stated targets are documented here.

- **Gene-therapy module is non-functional.** The eligibility gate requires `age_at_dx >= 18`, but `age_at_dx` is drawn from `exponential(scale=2)` (neonatal-peaked), so the condition effectively never holds: `gene_therapy_flag == 0` for **every** patient across **every** seed. All `gt_*` columns are present but uniformly `N/A`. This is scored only as a known-state integrity floor (rate must be exactly 0), never as a populated GT-outcome metric. **Do not use this sample for gene-therapy modeling.**
- **Historical viral-contamination rates are miscalibrated.** HIV/HCV positivity keys off `age_at_dx < 35` as a pre-1985/1992 birth-cohort proxy, but `age_at_dx` is *age at diagnosis*, which is ~universally below 35 — so the elevated cohort rates (HIV ≈0.20, HCV ≈0.35) are applied to nearly all patients. `hiv_positive` and `hcv_positive` are therefore **excluded from validation scoring**; treat them as unreliable.
- **ABR, prophylaxis rate, and Pettersson score miss the engine's printed literature targets.** Observed mean ABR ≈9–10 (engine intent 5–8), prophylaxis ≈0.50–0.57 (intent 0.70–0.80), mean Pettersson ≈23 (intent 10–15; inflated by accumulating a 20-year annual delta onto the baseline). The scorecard ranges reflect the observed, deterministic values; the literature intent is shown above as "engine intent."
- **Synthetic, partly summarized longitudinal design.** The main table holds annualized summaries and baselines; true longitudinal structure lives in the three companion tables.
- **Marginal calibration, not full joint fidelity.** Univariate prevalences and the engineered structural separations (severity factor-activity gap, prophylaxis ABR reduction, inhibitor concentration in severe Hem A) are anchored; higher-order correlations beyond those engineered are not independently validated.
- **Small-sample variance.** At n=500 some rates carry sampling variance; scorecard ranges accommodate this without masking real misses, and structural floors are weighted to dominate.

## Commercial / full version

| | Sample (this) | Full (commercial) |
|---|---|---|
| Patients | 500 | 20,000+ (configurable) |
| Main columns | 102 | 102 |
| Companion tables | 3 (bleeds / joints / PK) | 3, full depth |
| Formats | CSV | CSV / JSON / Parquet |
| Engine fixes | as-is, fully disclosed | GT eligibility gate & viral-cohort logic patched on request; ABR/Pettersson recalibration available |
| Seeds / reproducibility | 6 canonical | Unlimited |
| License | CC-BY-NC-4.0 | Commercial |
| Support | — | SLA, custom calibration, vertical extensions |

Contact **pradeep@xpertsystems.ai** · https://xpertsystems.ai

## Citation

```bibtex
@dataset{xpertsystems_hcgen003_2026,
  title  = {HC-GEN-003: Synthetic Hemophilia A/B Dataset --- Bleeding Episodes,
            Factor Replacement, Joint Damage, Inhibitors & Gene Therapy (Sample)},
  author = {XpertSystems.ai},
  year   = {2026},
  publisher = {Hugging Face},
  note   = {Synthetic data. Not for clinical use. Calibration anchors:
            WFH Annual Global Survey; USHTCN / CDC inhibitor surveillance;
            Manco-Johnson et al. NEJM 2007 (prophylaxis vs on-demand);
            HAVEN program (emicizumab); RODIN study (inhibitor risk);
            HOPE-B (etranacogene); GENEr8-1 (valoctocogene).
            Known disclosed defects: gene-therapy eligibility gate inactive;
            HIV/HCV cohort proxy miscalibrated; ABR/Pettersson/prophylaxis
            calibrated to observed output, not the engine's literature targets.},
  url    = {https://huggingface.co/datasets/xpertsystems/hc-gen-003-sample}
}
```