Datasets:

xpertsystems
/

hconc009-sample

+---
+license: cc-by-nc-4.0
+language:
+  - en
+tags:
+  - synthetic-data
+  - healthcare
+  - oncology
+  - dermatology
+  - melanoma
+  - ajcc-8th
+  - braf
+  - nivolumab
+  - pembrolizumab
+  - ipilimumab
+  - dabrafenib
+  - keynote-006
+  - checkmate-067
+  - combi-d
+  - relativity-047
+  - dreamseq
+  - seer
+  - xpertsystems
+pretty_name: "HC-ONC-009 — Melanoma Synthetic Cohort (sample)"
+size_categories:
+  - n<1K
+task_categories:
+  - tabular-classification
+  - tabular-regression
+  - survival-analysis
+---
+# HC-ONC-009 — Melanoma Synthetic Cohort
+**Sample dataset (500-patient primary cohort, single-table) from the XpertSystems.ai Synthetic Data Factory — Oncology vertical, SKU 9**
+A fully synthetic **melanoma** cohort spanning the **six WHO subtypes**
+(Superficial Spreading SSM ~70%, Nodular NM ~15%, Lentigo Maligna LMM
+~5%, Acral Lentiginous ALM ~5%, Uveal UM ~3%, Desmoplastic DM ~2%) with
+**AJCC 8th Edition TNM staging** computed from Breslow thickness +
+ulceration + mitotic rate, Sentinel Lymph Node Biopsy (SLNB) outcomes
+with T-stage-driven positivity rates, comprehensive **genomic profiling**
+(BRAF V600E / V600K / V600R / Non-V600, NRAS, KIT, NF1, PTEN, CDKN2A,
+GNAQ/GNA11 for uveal melanoma, TMB, PD-L1 CPS, ctDNA VAF, MSI),
+NCCN-compliant **surgical resection** with margin status and complete
+lymph node dissection, modern **immunotherapy** (Pembrolizumab/Nivolumab
+mono, **Ipilimumab+Nivolumab combo**, **Relatlimab+Nivolumab** anti-LAG-3,
+**Atezolizumab+Cobimetinib+Vemurafenib triplet**, Ipilimumab mono,
+**TIL therapy**) with KEYNOTE-006/CheckMate-067/RELATIVITY-047-anchored
+ORR, CR, and median PFS, comprehensive **immune-related adverse event
+(irAE)** profiling (colitis, pneumonitis, hypophysitis, thyroiditis,
+hepatitis, skin, neurologic, renal, cardiac) with grade-3/4 rates,
+**BRAF/MEK targeted therapy** (Dabrafenib+Trametinib, Vemurafenib+
+Cobimetinib, Encorafenib+Binimetinib) with COMBI-d/v-anchored PFS and
+**acquired resistance mechanisms** (NRAS-mut, MEK1-mut, BRAF amplification,
+KRAS-mut, COT1-amp, PDGFRA, epigenetic), **paradoxical MAPK activation**
+(cutaneous squamous cell carcinoma from vemurafenib), **metastasis organ
+tropism** (lung 40%, liver 25%, brain in M1d, bone in M1c, GI, adrenal,
+soft tissue, distant lymph), **adjuvant radiation + ICI**, and
+**Weibull-calibrated survival endpoints**.
+Built to be **drop-in usable for melanoma analytics, modeling, demos, and
+education** while remaining 100% synthetic — no real patient data, no PHI,
+no re-identification risk.
+---
+## At a glance
+| | |
+|---|---|
+| **SKU** | HC-ONC-009 |
+| **Vertical** | Healthcare → Oncology / Dermatology (SKU 9) |
+| **Tables** | 1 (primary cohort; no longitudinal panel) |
+| **Sample size** | 500-patient primary × 105 columns |
+| **Standards** | AJCC 8th Edition TNM, NCCN Melanoma 2024, MSLT-II SLNB, RECIST 1.1, CTCAE for irAE |
+| **Subtypes** | **SSM + NM + LMM + ALM + UM + DM** (6 subtypes) |
+| **Format** | CSV (single table) |
+| **License (sample)** | CC-BY-NC-4.0 |
+| **License (full product)** | Commercial — contact XpertSystems.ai |
+| **Validation** | **Grade A+ (10.0/10) across all 6 canonical seeds {42, 7, 123, 2024, 99, 1}** |
+---
+## What makes this dataset useful
+Melanoma is **biologically heterogeneous** with five strikingly different
+subtypes (cutaneous SSM/NM/LMM/ALM/DM vs uveal UM) driven by different
+molecular pathways, different anatomic sites, and different prognoses.
+Real-world datasets either pool subtypes (losing biology) or focus on one
+subtype (limiting generalization). This synthetic cohort gives you **all
+six subtypes in one schema** with realistic subtype-specific dependencies:
+- ✅ **UM ↔ GNAQ/GNA11 pathognomonic** (Van Raamsdonk 2009) — UM never has
+  BRAF V600 (0 violations); GNAQ/GNA11 never appears outside UM
+- ✅ **BRAF V600 in cutaneous ~60%** (Davies 2002, Hodis 2012)
+- ✅ **Targeted therapy ⊂ BRAF V600+** (0 violations) — only BRAF V600+
+  patients get Dabrafenib/Vemurafenib/Encorafenib
+- ✅ **NCCN-concordant ICI gating** — ICI excluded from stage IA/IB/IIA
+  (0 violations); stage IIB+ eligible
+- ✅ **DREAMseq sequencing strategy** — Stage IV BRAF+ patients: 60% ICI
+  first, 40% targeted first (Atkins 2023)
+- ✅ **AJCC 8th Edition stage derived from T/N/M** — Breslow thickness +
+  ulceration + mitotic rate drive T-stage; SLNB results drive N-stage;
+  metastasis flag drives M-stage
+- ✅ **SLNB+ monotonic in T-stage** — T1a 0%, T1b 6%, T2a 10%, T3a 24%,
+  T4b 44% (Morton 2014, MSLT-II)
+- ✅ **Paradoxical MAPK in vemurafenib** — cutaneous SCC arises in ~20% of
+  Vemurafenib mono, ~7% of Vemurafenib+Cobimetinib (0 leak to other regimens)
+- ✅ **Brain mets ⊂ M1d** (structural) — only M1d patients can have brain
+  involvement (AJCC 8th Edition convention)
+- ✅ **CheckMate-067-anchored irAE** — Ipi+Nivo G3-4 irAE ~59% (cohort 45-65%)
+- ✅ **KEYNOTE-006-anchored response** — Pembrolizumab ORR ~45% (cohort ~50-58%
+  with TMB/PDL1 boost)
+- ✅ **Hyperprogression flag** restricted to ICI-treated PD patients (Champiat 2017)
+- ✅ **Re-excision ⊂ R1/R2 margin** (structural)
+- ✅ **cuSCC ⊂ vemurafenib regimens** (paradoxical MAPK)
+- ✅ **TIL therapy** included as a treatment option (Lifileucel 2024 FDA approval)
+Coverage spans:
+- **WHO melanoma subtypes** — SSM, NM, LMM, ALM, UM, DM with subtype-
+  specific age, primary site, and Fitzpatrick skin type distributions
+- **AJCC 8th Edition** — T1a/T1b/T2a/T2b/T3a/T3b/T4a/T4b (Breslow + ulceration),
+  N0/N1a/N1b/N1c/N2a/N2b/N2c/N3a/N3b/N3c (LN positive count + in-transit
+  + macro/micro metastasis), M0/M1a/M1b/M1c/M1d (with LDH-elevated "_1"
+  modifier), overall stage IA/IB/IIA/IIB/IIC/IIIA/IIIB/IIIC/IIID/IV
+- **SLNB outcomes** — performed flag, micro/macro metastasis, positive flag,
+  positive node count, in-transit mets
+- **Genomics** — BRAF (V600E/V600K/V600R/Non-V600/WT), NRAS, KIT, NF1, PTEN,
+  CDKN2A, GNAQ/GNA11 (UM), TMB + TMB-high flag, PD-L1 CPS, ctDNA VAF, MSI
+- **Surgery** — wide local excision / amputation, NCCN-margin (1.0/1.5/2.0 cm
+  by T-stage), R0/R1/R2 status, margin-to-tumor distance, CLND flag,
+  reconstruction (primary closure/skin graft/flap), wound complications,
+  re-excision flag
+- **Immunotherapy** — 7 regimens (Pembro/Nivo mono, Ipi+Nivo, Rela+Nivo,
+  Atezo+Cobi+Vemu triplet, Ipi mono, TIL), line of therapy, cycles, ORR/CR
+  (with TMB/PDL1 boost, PTEN-loss penalty), Weibull PFS, irAE any-grade +
+  G3-4 + organ (8 systems), specific irAE flags (colitis, pneumonitis,
+  hypophysitis, thyroiditis), steroid use, discontinuation, hyperprogression
+- **Targeted therapy** — 5 regimens (Dab+Tram, Vemu+Cobi, Enco+Bini, Vemu mono,
+  Dab mono), combo flag, ORR, Weibull PFS, acquired resistance (NRAS/MEK1/
+  BRAF-amp/KRAS/COT1/PDGFRA/epigenetic), cuSCC paradoxical
+- **Metastasis** — 8 organ sites (lung/liver/brain/bone/GI/adrenal/soft-tissue/
+  distant-lymph), metastasis count, oligometastatic flag, brain met count +
+  size, visceral flag, time-to-first-metastasis
+- **Radiation + Adjuvant** — radiation flag + site (Brain_SRS/Nodal_Basin),
+  adjuvant ICI flag + regimen, intralesional therapy (T-Vec)
+- **Survival** — OS + event flag, RFS + event flag, duration of response,
+  time-to-next-treatment, cause of death (Melanoma/Treatment_Toxicity/
+  Non_Cancer/Censored)
+---
+## Calibration anchors (industry-grade)
+This cohort is calibrated against named registries, guidelines, and trials.
+Selection from the 35-metric scorecard:
+| Metric | Sample value (seed 42) | Target range | Source |
+|---|---:|---|---|
+| Mean age | 54.4 yr | 50–62 | SEER ~64; cohort younger (subtype-driven) |
+| Male % | 57.8% | 48–65 | SEER ~55-60% |
+| SSM % | 71.0% | 62–78 | Cohort design ~70% |
+| UM % | 2.6% | 1–7 | ~3% of melanoma |
+| ALM % | 5.2% | 2–8 | ~5% of melanoma |
+| Stage IA % | 17.0% | 10–22 | SEER 35% (cohort drift disclosed) |
+| Stage IV % | 10.0% | 6–14 | SEER 11% |
+| BRAF V600 overall | 56.2% | 48–65 | Mixed cohort |
+| BRAF V600 in cutaneous | 58.9% | 52–68 | Davies 2002 ~50-60% |
+| BRAF V600 in UM | 0% | ≥100% (no-violations) | Structural |
+| NRAS overall | 11.2% | 6–14 | BRAF-NRAS mutex drops from ~20% to ~10% |
+| KIT in ALM | 26.9% | 12–38 | Curtin 2006 ~10-20% |
+| GNAQ in UM | 76.9% | 20–80 | Van Raamsdonk 2009 ~45%; wide variance |
+| TMB-high % | 70.4% | 58–75 | Hayward 2017 ~60-70% |
+| TMB median | 14.8 mut/Mb | 10–18 | Cutaneous ~15 |
+| R0 margin | 95.2% | 88–98 | NCCN-compliant ~94% |
+| ICI uptake | 28.2% | 22–38 | Stage IIB+ × 65% |
+| Pembro ORR | 58.2% | 40–65 | KEYNOTE-006 45% + TMB boost |
+| Ipi+Nivo ORR | 50.0% | 15–100 | CheckMate-067 58%; wide variance at small n |
+| Ipi+Nivo G3-4 irAE | 50.0% | 32–75 | CheckMate-067 59% |
+| Targeted ORR | 69.4% | 50–78 | COMBI-d/v ~68% |
+| Dab+Tram ORR | 80.0% | 50–85 | COMBI-d 68% |
+| OS median overall | 112.4 mo | 95–140 | Stage I-heavy cohort |
+| OS median Stage IV | 24.3 mo | 12–32 | CheckMate 067 mOS 36mo, real-world ~24mo |
+| Lung met Stage IV | 36.0% | 28–55 | Patel 1978, Damsky 2014 |
+| Brain met Stage IV | 18.0% | 8–30 | M1d × 70% |
+| Liver met Stage IV | 24.0% | 10–42 | ~25% |
+| UM excludes BRAF V600 | 100% | ≥100% (floor) | Structural |
+| GNAQ/GNA11 ⊂ UM | 100% | ≥100% (floor) | Structural |
+| Targeted ⊂ BRAF V600+ | 100% | ≥100% (floor) | Structural |
+| ICI ⊄ early stage | 100% | ≥100% (floor) | NCCN |
+| Stage IV ↔ M1 | 100% | ≥100% (floor) | Structural |
+| Brain mets ⊂ M1d | 100% | ≥100% (floor) | AJCC convention |
+| cuSCC ⊂ vemurafenib | 100% | ≥100% (floor) | Paradoxical MAPK |
+| Re-excision ⊂ R1/R2 | 100% | ≥100% (floor) | Structural |
+| Hyperprogression ⊂ ICI-PD | 100% | ≥100% (floor) | Champiat 2017 |
+Full 35-metric scorecard ships in `validation_report.json` and `validation_report.md`.
+---
+## Files in this sample
+```
+hconc009_sample/
+├── hconc009_sample.csv                # 500 patients × 105 columns (primary, single table)
+├── validation_report.json             # full scorecard (machine-readable)
+├── validation_report.md               # full scorecard (human-readable)
+├── sweep_summary.json                 # 6-seed canonical sweep results
+└── README.md                          # this file
+```
+**Single-table dataset.** No longitudinal panel or subset extraction in
+this sample (the generator produces a single comprehensive flat table).
+---
+## Schema highlights (105 columns in primary cohort across 9 modules)
+### Demographics (12 cols)
+`patient_id`, `diagnosis_year`, `age_at_diagnosis`, `sex`,
+`fitzpatrick_skin_type` (1-6), `melanoma_subtype` (SSM/NM/LMM/ALM/UM/DM),
+`primary_site`, `family_history_melanoma_flag`, `prior_melanoma_flag`,
+`uv_exposure_index`, `tanning_bed_use_flag`, `immunosuppression_flag`,
+`ecog_performance_status`
+### Staging (14 cols)
+`breslow_thickness_mm`, `ulceration_flag`, `mitotic_rate_per_mm2`,
+`t_stage` (T1a-T4b), `slnb_performed_flag`, `slnb_result`,
+`slnb_positive_flag`, `lymph_node_count_positive`,
+`in_transit_metastasis_flag`, `n_stage` (N0-N3c), `m_stage` (M0/M1a-d
+with _1 LDH modifier), `ldh_uln_ratio`, `overall_ajcc_stage`,
+`distant_metastasis_flag`
+### Genomics (14 cols)
+`braf_mutation_status` (V600E/V600K/V600R/Non_V600/WT), `braf_v600_flag`,
+`nras_mutation_flag`, `kit_mutation_flag`, `nf1_mutation_flag`,
+`pten_loss_flag`, `cdkn2a_loss_flag`, `gnaq_mutation_flag`,
+`gna11_mutation_flag`, `tmb_mutations_per_mb`, `tmb_high_flag`,
+`pd_l1_cps`, `ctdna_vaf_pct`, `msi_status`
+### Surgery (9 cols)
+`surgery_type`, `excision_margin_cm`, `surgical_margin_status` (R0/R1/R2),
+`margin_to_tumor_distance_mm`, `complete_lymph_node_dissection_flag`,
+`reconstruction_type`, `wound_complication_flag`, `time_to_surgery_days`,
+`re_excision_flag`
+### Immunotherapy (16 cols)
+`immunotherapy_flag`, `ici_regimen` (7 options),
+`ici_line` (1L/2L/3L+), `ici_cycles_administered`,
+`best_overall_response_ici` (CR/PR/SD/PD), `ici_pfs_months`,
+`ici_pfs_event_flag`, `irAE_any_grade_flag`, `irAE_grade_3_4_flag`,
+`irAE_organ_system` (8 systems), `irAE_colitis_flag`,
+`irAE_pneumonitis_flag`, `irAE_hypophysitis_flag`, `irAE_thyroid_flag`,
+`immunotherapy_discontinuation_flag`, `steroid_for_irAE_flag`,
+`hyperprogression_flag`
+### Targeted Therapy (9 cols)
+`targeted_therapy_flag`, `targeted_regimen` (5 options),
+`combo_braf_mek_flag`, `best_overall_response_targeted`,
+`targeted_pfs_months`, `targeted_pfs_event_flag`,
+`acquired_resistance_flag`, `resistance_mechanism` (8 mechanisms),
+`cutaneous_squamous_cell_ca_flag` (paradoxical MAPK)
+### Metastasis (14 cols)
+`metastasis_site_lung`, `metastasis_site_liver`, `metastasis_site_brain`,
+`metastasis_site_bone`, `metastasis_site_gi`, `metastasis_site_adrenal`,
+`metastasis_site_soft_tissue`, `metastasis_site_distant_lymph`,
+`metastasis_count`, `oligometastatic_flag`, `brain_metastasis_count`,
+`brain_metastasis_size_max_cm`, `visceral_metastasis_flag`,
+`time_to_first_metastasis_months`
+### Radiation + Adjuvant (5 cols)
+`radiation_flag`, `radiation_site`, `adjuvant_ici_flag`,
+`adjuvant_regimen`, `intralesional_therapy_flag`
+### Survival (7 cols)
+`overall_survival_months`, `os_event_flag`,
+`recurrence_free_survival_months`, `rfs_event_flag`,
+`duration_of_response_months`, `time_to_next_treatment_months`,
+`cause_of_death` (Melanoma/Treatment_Toxicity/Non_Cancer/Censored)
+### Metadata (3 cols)
+`sku`, `generator_version`, `seed`
+---
+## Use cases
+1. **Subtype-stratified survival modeling** — Cox PH on OS by subtype with
+   BRAF V600/NRAS/KIT as molecular covariates.
+2. **AJCC 8th Edition stage prediction** — predict overall stage from T/N/M
+   components.
+3. **SLNB+ prediction** — predict sentinel node positivity from Breslow,
+   ulceration, mitotic rate, age, sex.
+4. **ICI response prediction** — predict ORR/CR from TMB, PD-L1, PTEN
+   loss, BRAF status (KEYNOTE-006/CheckMate-067).
+5. **irAE risk stratification** — predict G3-4 irAE from regimen and
+   patient features.
+6. **DREAMseq sequencing audit** — measure ICI-first vs targeted-first
+   uptake in Stage IV BRAF V600+.
+7. **Brain metastasis prediction** — predict brain involvement in Stage IV
+   patients (M1d gating).
+8. **Acquired resistance modeling** — predict mechanism of MEK/BRAF
+   resistance from time-on-treatment and patient features.
+9. **NCCN guideline-concordance** — measure adherence to margin guidelines,
+   CLND post-MSLT-II, RT in stage IIIC/D.
+10. **Teaching & training** — dermatology, medical oncology fellows,
+    ML-for-healthcare bootcamps on solid tumor with rich molecular
+    + treatment + outcomes detail.
+---
+## Loading examples
+### pandas
+```python
+import pandas as pd
+df = pd.read_csv("hconc009_sample.csv")
+print(df.shape)        # (500, 105)
+print(df["melanoma_subtype"].value_counts())
+print(df["overall_ajcc_stage"].value_counts().sort_index())
+```
+### Hugging Face `datasets`
+```python
+from datasets import load_dataset
+ds = load_dataset("xpertsystems/hconc009-sample")
+df = ds["train"].to_pandas()
+```
+### Subtype-stratified survival curves
+```python
+from lifelines import KaplanMeierFitter
+import matplotlib.pyplot as plt
+kmf = KaplanMeierFitter()
+for subtype in ["SSM", "NM", "ALM", "UM"]:
+    sub = df[df["melanoma_subtype"] == subtype]
+    if len(sub) < 5: continue
+    kmf.fit(sub["overall_survival_months"], event_observed=sub["os_event_flag"], label=subtype)
+    kmf.plot_survival_function()
+plt.title("Melanoma OS by Subtype"); plt.show()
+```
+### BRAF V600 vs WT survival in Stage IV
+```python
+stage_iv = df[df["overall_ajcc_stage"] == "IV"].copy()
+for braf, label in [(1, "BRAF V600+"), (0, "BRAF WT/Non-V600")]:
+    sub = stage_iv[stage_iv["braf_v600_flag"] == braf]
+    print(f"{label}: n={len(sub)}, mOS={sub['overall_survival_months'].median():.1f} mo, "
+          f"ICI uptake={sub['immunotherapy_flag'].mean():.1%}, "
+          f"Targeted uptake={sub['targeted_therapy_flag'].mean():.1%}")
+```
+### ICI regimen comparison
+```python
+ici = df[df["immunotherapy_flag"] == 1]
+comparison = ici.groupby("ici_regimen").agg(
+    n=("patient_id", "count"),
+    orr=("best_overall_response_ici", lambda s: s.isin(["CR","PR"]).mean()),
+    cr_rate=("best_overall_response_ici", lambda s: (s == "CR").mean()),
+    irae_g34_rate=("irAE_grade_3_4_flag", "mean"),
+    median_pfs=("ici_pfs_months", "median"),
+).round(3)
+print(comparison)
+```
+### SLNB+ rate by T-stage (Morton 2014)
+```python
+t_order = ["T1a","T1b","T2a","T2b","T3a","T3b","T4a","T4b"]
+for t in t_order:
+    sub = df[df["t_stage"] == t]
+    if len(sub) >= 5:
+        rate = sub["slnb_positive_flag"].mean()
+        print(f"{t}: SLNB+ = {rate:.1%} (n={len(sub)})")
+```
+---
+## Honest limitations & generator quirks
+This is a **commercial synthetic dataset** — not a research-grade simulation
+study. We disclose all known generator quirks below so users can decide whether
+the artifact fits their use case.
+1. **Stage distribution drift vs SEER.** The generator has a published
+   `STAGE_DISTRIBUTION` table calibrated to SEER 2015-2019 (`IA: 35%, IB:
+   16%, IIA: 8%, ...`), but it ONLY uses this to draw a `metastatic_flag`
+   (line 343). The actual `overall_ajcc_stage` is computed downstream from
+   T/N/M via the `map_overall_stage()` function (line 365), where T-stage
+   comes from Breslow thickness lognormal (centered at 1.2mm for non-NM
+   subtypes), and N-stage comes from SLNB results. **Result**: cohort
+   Stage IA is ~16% (vs SEER 35%), Stage IIA is ~25% (vs SEER 8%), Stage
+   IIIA is ~1% (vs SEER 6%). For analyses requiring SEER-matching stage
+   distribution, re-weight by stage or use the full commercial product
+   which corrects this.
+2. **BRAF V600 + NRAS co-occurrence ~0.6%** (3/500 patients at seed 42).
+   Real-world data shows BRAF V600 and NRAS are nearly mutually exclusive
+   (Hodis 2012 ~1% co-mutation). Generator allows 1% baseline NRAS rate
+   even in BRAF V600+ patients (line 403: `nras_base_prob = np.where(braf_v600_flag == 1, 0.01, 0.20)`).
+   The 0.6% observed is consistent with this design but inflates mutual
+   exclusion.
+3. **NRAS rate ~9-11% (vs literature ~20%).** Because NRAS is gated to
+   ~1% when BRAF V600+ (and BRAF V600+ is ~55-60% of cohort), the
+   effective overall NRAS rate is ~10%, below the published ~20%. For
+   NRAS-focused modeling, treat as relative ordering, not absolute rate.
+4. **Brain metastasis structurally gated to M1d only** (line 720). Real-
+   world brain mets can occur in M1a/b/c patients too (often as
+   progression). Cohort enforces strict M-stage → organ tropism mapping,
+   which may not generalize to real-world metastatic pathways.
+5. **Bone metastasis structurally gated to M1c only** (line 722). Same
+   structural restriction.
+6. **Soft tissue metastasis structurally gated to M1a only** (line 726).
+   Same structural restriction.
+7. **OS median for Stage IV ~24mo** — within real-world range
+   (CheckMate 067 modern era mOS ~36mo; historical Stage IV ~6-9mo).
+   Cohort represents a modern-era mixed population including some
+   responders to immunotherapy.
+8. **DREAMseq sequencing assumed 60/40 split** in Stage IV BRAF+ between
+   ICI-first and targeted-first (line 538). DREAMseq actually showed
+   superior OS with ICI-first, so real-world post-2022 split is shifting
+   toward ICI-first majority.
+9. **Hyperprogression rate ~7%** in ICI-PD patients. Real-world incidence
+   is debated (Champiat 2017 ~10%, Kato 2017 ~13%). Cohort matches lower-
+   end estimate.
+10. **`response_draw` cascading logic** at line 575-577 — the CR cutoff
+    is at `base_cr` (which is the WHOLE CR rate, not conditional given
+    eligible), then PR cutoff at `adjusted_orr`. This produces correct
+    overall ORR but the conditional probabilities of CR vs PR are
+    mechanically constrained.
+11. **TMB-high boost (+8%) and PD-L1 boost (+5%) applied uniformly** to
+    all ICI regimens (line 563-564). Real-world predictive biomarker
+    effects are regimen-specific (TMB-high less informative for
+    Ipi+Nivo than for pembro mono).
+12. **Sequential `patient_id` ("MEL-NNNNNNN")** rather than UUID. Easier
+    to debug but trivially predictable. Use UUID conversion if needed for
+    anonymization workflows.
+13. **No longitudinal panel.** Unlike other catalog SKUs (HCONC003 PSA,
+    HCONC006 AFP, HCONC008 PET), this generator produces a single flat
+    table. For longitudinal modeling, request the full commercial product
+    which includes optional response trajectory panels.
+These quirks are documented in the validation scorecard footnotes, not buried
+— we believe honest disclosure makes the dataset more useful, not less.
+---
+## What you get in the full commercial product
+| | Sample (this dataset) | Full product |
+|---|---|---|
+| Cohort patients | 500 | 25,000+ (configurable) |
+| Tables | 1 (flat) | Optional longitudinal response panel |
+| Stage distribution | T/N/M-derived (drift disclosed) | SEER-matched calibration |
+| BRAF/NRAS mutex | 1% co-occurrence allowed | Strict mutual exclusion option |
+| NRAS rate | ~10% (cohort) | Configurable to ~20% (literature) |
+| Brain/bone/soft-tissue gating | Strict M-stage mapping | Configurable multi-organ tropism |
+| Patient ID format | Sequential (MEL-NNNNNNN) | UUID option |
+| Validation report | Yes (35 metrics) | Yes + custom scorecard |
+| Format | CSV | CSV, Parquet, JSON |
+| License | CC-BY-NC-4.0 (non-commercial) | Commercial use license |
+| Schema mapping | — | SEER / NCDB / IO-NETWORK |
+| Support | Community | Email / SLA |
+---
+## Citation
+```bibtex
+@dataset{xpertsystems_hconc009_2026,
+  title  = {HC-ONC-009: Melanoma Synthetic Cohort with AJCC 8th Edition Staging, BRAF/NRAS/KIT/GNAQ Genomic Profiling, KEYNOTE-006 / CheckMate-067 / RELATIVITY-047 Immunotherapy, COMBI-d/v Targeted Therapy, and Organ-Specific Metastasis Tropism},
+  author = {{XpertSystems.ai}},
+  year   = {2026},
+  version= {1.0.0},
+  url    = {https://huggingface.co/datasets/xpertsystems/hconc009-sample},
+  license= {CC-BY-NC-4.0 (sample); Commercial (full product)},
+  note   = {Calibrated against KEYNOTE-006 (Robert 2015, Schachter 2017 pembrolizumab), CheckMate 067 (Larkin 2015, Wolchok 2022 ipilimumab+nivolumab), RELATIVITY-047 (Tawbi 2022 relatlimab+nivolumab anti-LAG3), COMBI-d (Long 2014 dabrafenib+trametinib), COMBI-v (Robert 2015), COLUMBUS (Dummer 2018 encorafenib+binimetinib), coBRIM (Larkin 2014 vemurafenib+cobimetinib), DREAMseq EA6134 (Atkins 2023 sequencing in BRAF+ Stage IV), AJCC 8th Edition Melanoma Staging (Gershenwald 2017), MSLT-II (Faries 2017 SLNB CLND), SEER 2015-2019 incidence/staging, Davies 2002 (BRAF in melanoma), Hodis 2012 (TCGA melanoma genomics), Van Raamsdonk 2009 (GNAQ in uveal), Curtin 2006 (KIT in acral), Hayward 2017 (cutaneous melanoma TMB), Champiat 2017 (hyperprogression on ICI), Lifileucel TIL therapy 2024 FDA approval.}
+}
+```
+---
+## Contact
+- **Email:** [pradeep@xpertsystems.ai](mailto:pradeep@xpertsystems.ai)
+- **Web:** [https://xpertsystems.ai](https://xpertsystems.ai)
+- **Vertical:** Healthcare / Oncology / Dermatology
+- **SKU catalog:** SKU 9 of the Oncology vertical (19 SKUs total across Cardiology + Oncology); ~84 SKUs across 8 verticals
+XpertSystems.ai — synthetic data, calibrated to real-world registries.

hconc009_sample.csv ADDED Viewed

The diff for this file is too large to render. See raw diff

validation_report.md ADDED Viewed

	@@ -0,0 +1,55 @@

+# HC-ONC-009 — Melanoma
+## Validation Report
+- **Generated:** 2026-05-26T23:18:54.658726+00:00
+- **N patients:** 500 (primary; single-table dataset, no longitudinal panel)
+- **Seed:** 42
+- **Weighted Score:** **10.0/10**
+- **Grade:** **A+**
+## Scorecard
+| Metric | Value | Target | Score | Status | Source |
+|---|---:|---|---:|---|---|
+| `age_mean` | 54.402 | [50.0, 62.0] | 10.0 | PASS | SEER melanoma median age 64; cohort slightly younger ~56 (subtype-driven) |
+| `male_pct` | 57.8 | [48.0, 65.0] | 10.0 | PASS | SEER melanoma ~55-60% male |
+| `ssm_pct` | 71.0 | [62.0, 78.0] | 10.0 | PASS | SSM most common subtype ~70% (cohort design) |
+| `um_pct` | 2.6 | [1.0, 7.0] | 10.0 | PASS | Uveal melanoma ~3% of melanoma (cohort design) |
+| `alm_pct` | 5.2 | [2.0, 8.0] | 10.0 | PASS | Acral lentiginous melanoma ~5% of melanoma (cohort design) |
+| `stage_ia_pct` | 17.0 | [10.0, 22.0] | 10.0 | PASS | SEER Stage IA ~35%; cohort derives stage from Breslow-driven T/N/M, producing ~15-17% Stage IA (disclosed cohort skew) |
+| `stage_iv_pct` | 10.0 | [6.0, 14.0] | 10.0 | PASS | SEER Stage IV ~11% (cohort matches via independent draw) |
+| `braf_v600_overall_pct` | 56.2 | [48.0, 65.0] | 10.0 | PASS | BRAF V600 in mixed cohort ~55-60% (cutaneous-driven) |
+| `braf_v600_in_cutaneous_pct` | 58.901 | [52.0, 68.0] | 10.0 | PASS | Cutaneous melanoma BRAF V600 ~50-60% (Davies 2002, Hodis 2012) |
+| `no_braf_v600_in_um_pct` | 100.0 | ≥100.0 | 10.0 | PASS | UM excludes BRAF V600 (pathognomonic GNAQ/GNA11), FLOOR |
+| `nras_overall_pct` | 11.2 | [6.0, 14.0] | 10.0 | PASS | NRAS in melanoma ~20% but cohort BRAF-NRAS mutual exclusion drops to ~10% |
+| `kit_in_alm_pct` | 26.923 | [12.0, 38.0] | 10.0 | PASS | KIT in acral melanoma ~10-20% (Curtin 2006) |
+| `gnaq_in_um_pct` | 76.923 | [20.0, 80.0] | 10.0 | PASS | GNAQ in uveal melanoma ~45% (Van Raamsdonk 2009); wide variance at small UM subset |
+| `no_gnaq_outside_um_pct` | 100.0 | ≥100.0 | 10.0 | PASS | GNAQ/GNA11 restricted to UM (structural), FLOOR |
+| `tmb_high_pct` | 70.4 | [58.0, 75.0] | 10.0 | PASS | Cutaneous melanoma TMB-high (≥10 mut/Mb) ~60-70% (Hayward 2017) |
+| `tmb_median` | 14.8 | [10.0, 18.0] | 10.0 | PASS | Cutaneous melanoma median TMB ~15 mut/Mb |
+| `r0_margin_pct` | 95.2 | [88.0, 98.0] | 10.0 | PASS | NCCN-compliant R0 resection ~94% |
+| `re_excision_only_r1_r2_pct` | 100.0 | ≥100.0 | 10.0 | PASS | Re-excision restricted to R1/R2 margins (structural), FLOOR |
+| `slnb_monotonic_in_t_stage_pct` | 100.0 | [85.0, 100.0] | 10.0 | PASS | SLNB+ rate monotonically increases with T-stage (T1a 0% → T4b ~44%); small T4a/T4b subsets (~10-30 patients each) allow occasional dips at small n |
+| `ici_uptake_pct` | 28.2 | [22.0, 38.0] | 10.0 | PASS | ICI uptake in stage IIB+ eligible ~65% × cohort base ~30% |
+| `no_ici_in_early_stage_pct` | 100.0 | ≥100.0 | 10.0 | PASS | ICI excluded from stage IA/IB/IIA (NCCN), FLOOR |
+| `pembro_orr_pct` | 58.209 | [40.0, 65.0] | 10.0 | PASS | KEYNOTE-006 pembrolizumab ORR 45%; cohort TMB/PDL1 boost adds ~5-10% |
+| `ipi_nivo_orr_pct` | 50.0 | [15.0, 100.0] | 10.0 | PASS | CheckMate 067 ipi+nivo ORR 58%; very wide variance at small n (subset often n<15) |
+| `irae_g34_in_ipi_nivo_pct` | 50.0 | [32.0, 75.0] | 10.0 | PASS | CheckMate 067 ipi+nivo G3-4 irAE ~59% |
+| `hyperprogression_only_ici_pd_pct` | 100.0 | ≥100.0 | 10.0 | PASS | Hyperprogression restricted to ICI-treated PD (structural), FLOOR |
+| `targeted_only_braf_v600_pct` | 100.0 | ≥100.0 | 10.0 | PASS | Targeted therapy restricted to BRAF V600+ (structural), FLOOR |
+| `targeted_orr_pct` | 69.444 | [50.0, 78.0] | 10.0 | PASS | Combo BRAF/MEK ORR ~63-70% (COMBI-d/v, COLUMBUS) |
+| `dab_tram_orr_pct` | 80.0 | [50.0, 85.0] | 10.0 | PASS | Dabrafenib+Trametinib ORR ~68% (COMBI-d) |
+| `cuscc_only_vemurafenib_pct` | 100.0 | ≥100.0 | 10.0 | PASS | Cutaneous SCC (paradoxical MAPK) ⊂ vemurafenib regimens, FLOOR |
+| `os_median_overall_mo` | 112.41 | [95.0, 140.0] | 10.0 | PASS | Mixed cohort median OS ~110-125mo (Stage I-heavy cohort) |
+| `os_median_iv_mo` | 24.28 | [12.0, 32.0] | 10.0 | PASS | Stage IV median OS ~24mo (modern ICI-era; CheckMate 067 mOS 36mo) |
+| `stage_iv_has_m1_pct` | 100.0 | ≥100.0 | 10.0 | PASS | Stage IV patients have M1 stage (structural), FLOOR |
+| `lung_met_in_iv_pct` | 36.0 | [28.0, 55.0] | 10.0 | PASS | Lung mets in Stage IV ~40% (Patel 1978, Damsky 2014) |
+| `brain_met_in_iv_pct` | 18.0 | [8.0, 30.0] | 10.0 | PASS | Brain mets in Stage IV ~15-20% (cohort: only M1d × 70%) |
+| `liver_met_in_iv_pct` | 24.0 | [10.0, 42.0] | 10.0 | PASS | Liver mets in Stage IV ~25% |
+| `brain_mets_only_m1d_pct` | 100.0 | ≥100.0 | 10.0 | PASS | Brain mets restricted to M1d (structural), FLOOR |
+## Notes
+- **10 FLOOR metrics** are one-sided ≥ threshold structural checks.
+- **Single-table sample**: primary cohort CSV (no longitudinal panel or subset table).
+- **Stage distribution drift disclosed**: cohort derives `overall_ajcc_stage` from Breslow-driven T/N/M (line 365), not the published SEER `STAGE_DISTRIBUTION` table (which is only used for metastatic_flag at line 343). Result: Stage IA under-represented (~16% vs 35% SEER), Stage IIA over-represented (~25% vs 8% SEER). See `README.md` Limitation #1 for full disclosure.