xpertsystems/hc-gen-002-sample

HC-GEN-002 — Sickle Cell Disease: Vaso-Occlusive Crisis, Multi-Organ Damage, Transfusion & Novel/Gene Therapy (Sample)

Synthetic patient-level dataset for sickle cell disease (SCD), spanning genotype classification, vaso-occlusive crisis (VOC) and acute chest syndrome analytics, multi-organ damage (renal, cardiac, CNS, ophthalmic, hepatic, musculoskeletal, splenic), transfusion management and iron overload, and the full modern therapeutic landscape — hydroxyurea, L-glutamine, crizanlizumab, voxelotor, HSCT, and gene therapy.

This is a 500-patient sample of the full HC-GEN-002 product. It is synthetic — generated by a calibrated simulation engine anchored to published cohort and trial statistics. 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

The generator is calibrated so cohort-level statistics fall within ranges drawn from the literature. Sample-level observed values (seed 42, n=500):

Metric	Observed	Target	Anchor
HbSS mortality (follow-up)	0.083	0.05–0.30	CSSCD (Platt et al. NEJM 1994)
VOC episodes/yr (mean)	3.07	2.0–5.5	PUSH Cohort
HbF response on hydroxyurea (%)	16.5	12–22	BABY HUG / MSH
Hydroxyurea prevalence	0.65	0.55–0.75	ASH 2020 SCD Guidelines
Crizanlizumab VOC reduction	0.47	0.35–0.60	SUSTAIN (Ataga et al. NEJM 2017)
Gene-therapy HbAT/HbF rise (%)	41.6	25–52	Lovotibeglogene Phase III
HbSS overt-stroke prevalence	0.148	0.06–0.20	CSSCD / STOP
RBC alloimmunization	0.302	0.20–0.40	Chou et al. Blood 2013
Pulmonary hypertension (TRV)	0.210	0.12–0.28	Gladwin et al. NEJM 2004
Hgb gap (HbSC − HbSS)	2.32 g/dL	≥1.3 (floor)	Genotype-phenotype (Serjeant)
VOC gap (HbSS − HbSC)	2.28 /yr	≥1.0 (floor)	PUSH / CSSCD
Retinopathy excess (HbSC)	0.132	≥0.05 (floor)	HbSC proliferative-retinopathy predilection
HU HbF induction separation	16.5 pp	≥8 (floor)	MSH hydroxyurea induction
Functional-asplenia excess (HbSS)	0.469	≥0.25 (floor)	HbSS autosplenectomy
Female priapism violations	0	=0 (floor)	Sex-specific schema integrity

Validation: Grade A+ (10.00/10) across all six canonical seeds (42, 7, 123, 2024, 99, 1), deterministic on every scored metric.

Schema highlights by module

141 columns, single table. Module groups:

• Demographics & genotype — genotype (HbSS / HbSC / HbSβ⁰-thal / HbSβ⁺-thal / HbSD-Punjab / HbSE), age_at_cohort_entry_years, follow_up_years, alpha_thal_coinheritance, coinherited_g6pd_deficiency_flag.
• Hematology & biomarkers — baseline hemoglobin, HbF%, HbS%, reticulocytes, LDH, bilirubin, AST, creatinine, EPO, NT-proBNP, soluble VCAM-1, hs-CRP, and a 0–10 disease_severity_score composite.
• Pain crisis / VOC — voc_episodes_yr, hospitalization flag, location, trigger, severity (NRS), acute chest syndrome (ICU/ventilation), chronic pain, opioid type and MME/day, ED visits.
• Organ damage — sickle nephropathy stage / eGFR / UPCR / ESRD, pulmonary hypertension (TRV, BNP), LV dysfunction/LVH, TCD velocity, overt stroke / silent cerebral infarct / moyamoya / cognitive score, AVN (hip/shoulder, Ficat-Arlet stage), proliferative retinopathy, hepatopathy/cholelithiasis, priapism (males), spleen status / functional asplenia, leg ulcers, sepsis.
• Transfusion management — program type/indication, HbS target, lifetime PRBC units, alloimmunization & antibody count, DHTR, ferritin, liver iron concentration, myocardial iron, chelation agent/adherence/toxicity, exchange parameters.
• Novel therapies — hydroxyurea (dose, MTD, HbF response, WBC nadir, VOC reduction, adherence, toxicity), L-glutamine, crizanlizumab, voxelotor (Hgb rise, LDH normalization), HSCT (donor type, conditioning, engraftment, GVHD, TMA, OS estimate), gene therapy (product, CD34 dose, HbF/HbAT rise, VOC-free months, engraftment, hematologic response).
• Outcomes & QoL — mortality flag/cause, hospitalization, SF-36 PCS, PROMIS fatigue/pain, PHQ-9, GAD-7, school days missed, employment, vaccination/prophylaxis flags, palliative-care referral.

Files

• hc_gen_002_sample.csv — 500 synthetic patients, 141 columns.
• README.md — this file.

Loading

import pandas as pd
df = pd.read_csv("hc_gen_002_sample.csv")
print(df.shape)  # (500, 141)

# Genotype-stratified VOC burden
print(df.groupby("genotype")["voc_episodes_yr"].mean())

from datasets import load_dataset
ds = load_dataset("xpertsystems/hc-gen-002-sample")

Use cases

• SCD severity stratification and complication-risk modeling.
• Disease-modifying-therapy effect simulation (HU, crizanlizumab, voxelotor) and gene/cell-therapy trial design.
• Genotype-phenotype association studies on synthetic cohorts.
• Transfusion / iron-overload management and alloimmunization-risk modeling.
• Synthetic-data methodology benchmarking and schema/ETL prototyping (FHIR-export hooks present in the engine).

Limitations (honestly disclosed)

• Two engine fields are non-deterministic and are excluded from validation scoring. The engine derives voc_duration_days_median and time_to_mortality_months from NumPy's global RNG (np.random.weibull) rather than the seeded generator, so these two columns vary run-to-run even at a fixed --seed. They remain in the CSV for completeness/parity with the engine output, but should not be relied upon for reproducible analysis. Every scored metric is byte-stable across all six canonical seeds.
• Dead .isin code paths (harmless). Two genotype/transfusion-type filters reference a pandas-style .isin guarded by hasattr; because the underlying objects are NumPy arrays, the equivalent boolean-OR fallback always executes. The intended logic is preserved; no values are affected.
• Therapy fields are not NA-gated when a flag is 0. Categorical therapy descriptors (e.g. gene_therapy_product, hsct_donor_type, gt_hematologic_response) are populated for all patients regardless of the corresponding enrollment flag. Filter on the relevant *_flag before interpreting these fields.
• Synthetic, cross-sectional with summarized longitudinal rates. Each row is one patient; longitudinal disease course is encoded as annualized rates and baselines, not as visit-level records.
• Marginal calibration, not full joint fidelity. Univariate prevalences and the engineered structural separations (genotype Hgb/VOC/asplenia gaps, HU pharmacodynamic separation, retinopathy predilection) are anchored to literature; higher-order correlations beyond those engineered are not independently validated.
• Gene/cell-therapy parameters are forward-looking. SCD gene therapy is newly approved/investigational; pharmacodynamic values represent plausible trial-design ranges, not long-term real-world registry data.
• Small-sample variance. At n=500 some rates (HbSS mortality, PH, stroke) 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)
Columns	141	141
Formats	CSV	CSV / JSON / Parquet / FHIR R4 bundle
Determinism	Scored metrics fully deterministic	Global-RNG fields patched to seeded RNG on request
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

@dataset{xpertsystems_hcgen002_2026,
  title  = {HC-GEN-002: Synthetic Sickle Cell Disease Dataset --- Vaso-Occlusive
            Crisis, Multi-Organ Damage, Transfusion & Novel/Gene Therapy (Sample)},
  author = {XpertSystems.ai},
  year   = {2026},
  publisher = {Hugging Face},
  note   = {Synthetic data. Not for clinical use. Calibration anchors:
            CSSCD (Platt et al. NEJM 1994); PUSH Cohort; BABY HUG;
            MSH (Charache et al. 1995); SUSTAIN (Ataga et al. NEJM 2017);
            HOPE (Vichinsky et al. NEJM 2019, voxelotor);
            Lovotibeglogene autotemcel Phase III; Exagamglogene Phase I/II;
            STOP (Adams et al.); Gladwin et al. NEJM 2004 (pulmonary hypertension);
            Chou et al. Blood 2013 (alloimmunization);
            NHLBI SCD Guidelines 2014; ASH SCD Evidence-Based Guidelines 2020.},
  url    = {https://huggingface.co/datasets/xpertsystems/hc-gen-002-sample}
}