| --- |
| license: cc-by-nc-4.0 |
| language: |
| - en |
| tags: |
| - synthetic-data |
| - healthcare |
| - oncology |
| - metastatic-cancer |
| - distant-metastasis |
| - organ-tropism |
| - advanced-disease |
| - palliative-care |
| - oligometastatic |
| - sbrt |
| - srs |
| - breast-cancer |
| - nsclc |
| - mcrpc |
| - mhspc |
| - mpdac |
| - mmrcc |
| - acquired-resistance |
| - xpertsystems |
| pretty_name: "HC-ONC-014 — Metastatic Cancer Synthetic Cohort (sample)" |
| size_categories: |
| - n<1K |
| task_categories: |
| - tabular-classification |
| - tabular-regression |
| - survival-analysis |
| --- |
| |
| # HC-ONC-014 — Metastatic Cancer Cohort |
|
|
| **Sample dataset (500-patient single-table cohort) from the XpertSystems.ai Synthetic Data Factory — Oncology vertical, SKU 14 (FINAL)** |
|
|
| 🎉 **This is the fourteenth and final SKU in the XpertSystems Oncology vertical**, closing out a 14-SKU catalog covering Breast, Lung, Prostate, Colorectal, Pancreatic, Liver/HCC, Leukemia, Lymphoma, Melanoma, Ovarian, Multi-Cancer Tumor Progression, Chemotherapy Response, Immunotherapy/CPI Response, and now Metastatic Cancer. |
|
|
| A fully synthetic **metastatic cancer** cohort spanning **13 metastatic |
| cancer types** — breast_met 20%, nsclc_met 18%, crc_met 14%, prostate_met |
| 12%, melanoma_met 8%, pancreatic_met 6%, renal_met 6%, gastric_met 5%, |
| ovarian_met 5%, bladder_met 2%, hcc_met 2%, sclc_met 1%, sarcoma_met 1% — |
| with **12-organ tropism matrices** (liver, lung, bone, brain, adrenal, |
| peritoneum, pleura, distant lymph nodes, skin, spinal, pericardium, bone |
| marrow) calibrated to **Disibio 2008 autopsy series** and cancer-specific |
| biology (Bubendorf 2000 mPC bone, Coleman 2001 mBC bone, Damsky 2014 |
| mMel CNS), **synchronous vs metachronous metastasis** (de novo Stage IV |
| vs late recurrence), comprehensive **biomarker panels by cancer type** |
| (EGFR/ALK/KRAS-G12C/PD-L1 in NSCLC; HER2/HR/BRCA/PD-L1-CPS in breast; |
| RAS/BRAF/MSI/HER2 in CRC; BRAF-V600E/NRAS in melanoma; AR splice variants/ |
| PSA in prostate; HRD/CA-125/BRCA in ovarian; IMDC risk/VHL in RCC), |
| **4-line treatment sequences** with biomarker-routed regimens |
| (Palbociclib/Ribociclib/Abemaciclib for HR+HER2- mBC, Trastuzumab- |
| Deruxtecan/Tucatinib for HER2+ mBC, Sacituzumab/Pembro+Chemo for TNBC, |
| Osimertinib/Alectinib for EGFR/ALK NSCLC, Sotorasib/Adagrasib for KRAS-G12C, |
| Enzalutamide/Abiraterone/Lu-177-PSMA for mCRPC, Dabrafenib+Trametinib/ |
| Nivo+Ipi for mMel, FOLFIRINOX/NALIRIFOX for mPDAC, Cabozantinib/ |
| Pembro+Axitinib for mRCC, EV+Pembro/Erdafitinib for mUC, Atezo+Bev for HCC, |
| Atezo-EP for SCLC), **acquired resistance mechanisms** (EGFR T790M, ESR1 |
| mutation, MET amplification, KRAS secondary, BRCA reversion, AR-V7 splice |
| variant, PD-L1 loss, PTEN loss), **oligometastatic-directed local therapy** |
| (SBRT 3/5/8/12 fractions, SRS for ≤3 CNS lesions, WBRT, surgical |
| metastasectomy, RFA, TACE-Y90 for HCC, intrathecal therapy for |
| leptomeningeal disease, palliative radiation), **bone-modifying agents** |
| (denosumab, zoledronic acid) for SREs (skeletal-related events: fracture, |
| spinal cord compression, hypercalcemia, bone pain), comprehensive |
| **supportive care** (palliative care consult, hospice referral, opioid |
| analgesia, pain VAS scoring, nutrition support, chronic corticosteroids |
| for CNS edema), and **Weibull-anchored survival endpoints** calibrated to |
| 13 cancer-type × 3 line-of-therapy benchmarks (CLEOPATRA HER2+ mBC, MONALEESA-3 |
| HR+ mBC, KEYNOTE-189 mNSCLC, OAK, FOLFOX/FOLFIRI mCRC, ENZAMET/LATITUDE |
| mHSPC, CheckMate-067 mMel, FOLFIRINOX mPDAC, KEYNOTE-426 mRCC, CASPIAN |
| mSCLC). |
| |
| Built to be **drop-in usable for metastatic cancer outcomes analytics, |
| treatment sequencing modeling, oligometastatic intervention research, and |
| palliative care quality benchmarking** while remaining 100% synthetic — |
| no real patient data, no PHI, no re-identification risk. |
| |
| --- |
| |
| ## At a glance |
| |
| | | | |
| |---|---| |
| | **SKU** | HC-ONC-014 | |
| | **Vertical** | Healthcare → Oncology / Metastatic Disease (SKU 14, FINAL) | |
| | **Tables** | 1 (primary cohort, single flat table with first-3-lines flattened) | |
| | **Sample size** | 500-patient primary × 109 columns | |
| | **Cancer types** | **13 metastatic types** | |
| | **Organ sites** | **12 tropism columns** (liver/lung/bone/brain/adrenal/peritoneum/pleura/lymph_distant/skin/spinal/pericardium/bone_marrow) | |
| | **Treatment lines** | **Up to 6 lines per patient** (first 3 captured in flat columns) | |
| | **Standards** | RECIST 1.1, AJCC 8th, NCCN Metastatic 2024, Disibio 2008 | |
| | **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 |
| |
| Metastatic disease accounts for ~90% of cancer mortality and drives the |
| largest share of oncology spending — yet it's the hardest stage to model |
| because treatment sequencing, oligometa interventions, organ-specific |
| tropism, and palliative care decisions all interact in complex ways. This |
| SKU gives you a **comprehensive metastatic dataset spanning 13 cancer |
| types with full treatment sequences, organ tropism, biomarkers, and |
| survival outcomes** in one schema with strong biology-preserving |
| constraints: |
| |
| - ✅ **14 zero-violation structural identities** across all 6 seeds |
| - ✅ **Prostate ↔ Male 100%** (sex coupling) |
| - ✅ **Ovarian ↔ Female 100%** (sex coupling, 98% design target) |
| - ✅ **De novo metastatic → Stage IV 100%** (clinical logic) |
| - ✅ **De novo metastatic → time_primary_to_met=0 100%** (synchronous definition) |
| - ✅ **CNS metastasis ↔ CNS lesion count consistency 100%** (clinical hierarchy) |
| - ✅ **Skeletal-related events ⊂ bone metastases 100%** (SRE definition) |
| - ✅ **Treated patients → L1 response not NA 100%** (data completeness) |
| - ✅ **Best Supportive Care → 0 treatment lines 100%** (intent → action) |
| - ✅ **Best Supportive Care → L1 regimen='NA' 100%** (consistency) |
| - ✅ **SRS eligible ⊂ ≤3 CNS lesions + no leptomeningeal disease 100%** (NCCN gating) |
| - ✅ **EGFR mutation status ⊂ NSCLC 100%** (test panel scope) |
| - ✅ **BRAF V600+ ⊂ melanoma/CRC 100%** (test panel scope) |
| - ✅ **HER2+ status ⊂ breast/gastric/CRC 100%** (test panel scope) |
| - ✅ **Cancer-specific organ tropism matches Disibio 2008**: |
| - Breast bone mets ~70% (Coleman 2001 target) |
| - Prostate bone mets ~88% (Bubendorf 2000 target ~85-90%) |
| - Melanoma CNS mets ~58% (advanced disease target) |
| - ✅ **CLEOPATRA-anchored mBC OS ~53-59mo** matches HER2+ era |
| - ✅ **mNSCLC OS ~23-32mo** matches KEYNOTE-189 era |
| - ✅ **mPDAC OS ~10-14mo** matches FOLFIRINOX (Conroy 2011, mOS 11.1mo) |
| - ✅ **L1 ORR ~51-56%** matches landmark benchmarks (breast 52%, melanoma 58%) |
| - ✅ **De novo metastatic ~25%** matches mixed cohort (PDAC 80%, breast 6%) |
| - ✅ **CNS metastasis ~20%** matches advanced cohort (Disibio 2008) |
| - ✅ **Acquired resistance mechanisms** modeled at progression (T790M, ESR1, MET-amp, AR-V7, etc.) |
| - ✅ **Oligometa SBRT/SRS local therapy** gated by burden + ECOG + lesion count |
| - ✅ **Bone-modifying agents** (denosumab vs zoledronic acid) routed to bone-met patients |
| |
| Coverage spans: |
| - **Demographics** — age, sex (cancer-coupled), ECOG, BMI, Charlson |
| Comorbidity Index (CCI), de novo vs metachronous, time from primary |
| to metastatic dx, synchronous/metachronous/late recurrence classification, |
| stage at initial dx, prior adjuvant/neoadjuvant therapy flags |
| - **Metastatic Burden** — 12 organ-site flags (liver/lung/bone/brain/ |
| adrenal/peritoneum/pleura/lymph_distant/skin/spinal/pericardium/ |
| bone_marrow), total n_metastatic_sites, burden class (Oligo ≤3 / Poly |
| 4-9 / Diffuse >9), CNS lesion count (Poisson 3.5, capped 1-15), |
| leptomeningeal disease flag, SRS eligibility flag, SRE category |
| (None/Fracture/Spinal cord compression/Hypercalcemia/Bone pain), PCI |
| score (peritoneal carcinomatosis index), hepatic tumor burden %, |
| visceral crisis flag, malignant effusion flag |
| - **Biomarkers (cancer-stratified)** — EGFR (5 variants), ALK, KRAS-G12C, |
| HER2 (4 grades), BRAF V600E, BRCA1/2, MSI/MMR, PD-L1 TPS/CPS, TMB, |
| ctDNA VAF baseline, liquid biopsy flag, repeat biopsy flag, tumor |
| heterogeneity score, AR splice variant (CRPC), PSA baseline, HRD |
| status (ovarian), CA-125 baseline, NRAS (melanoma), VHL/IMDC risk |
| (RCC), acquired resistance mechanism (populated at progression) |
| - **Treatment Lines (1-3 flattened, up to 6 total)** — regimen name, best |
| response per line, PFS per line (months), IO flag, targeted flag, |
| combination flag, time to next treatment, line modality (Chemo/IO/ |
| Targeted/Combo), dose reduction flag/%, treatment switch reason, |
| depth of response %, clinical trial enrollment, compassionate use, |
| treatment holiday, progression site (same/new/CNS/both), CNS |
| progression flag, pseudoprogression flag, hyperprogression flag |
| - **Local Therapy** — SBRT flag with dose fractions (3/5/8/12) + lesion |
| count, SRS flag (≤3 CNS lesions + no LMD), WBRT flag, surgical |
| metastasectomy flag, RFA flag (liver), TACE-Y90 flag (HCC liver), |
| intrathecal therapy flag (LMD), palliative radiation flag, |
| bone-modifying agent (denosumab/zoledronic_acid/none) |
| - **Supportive Care** — palliative care consult flag, hospice referral, |
| opioid analgesic flag, pain score VAS (0-10), nutrition support, |
| chronic corticosteroid flag (CNS edema), palliative surgery flag |
| - **Survival** — Overall Survival (Weibull-anchored to landmark trials, |
| burden + ECOG + oligometa-local-therapy modifiers), OS event flag, |
| cause of death (Disease progression/Toxicity/Intercurrent illness/ |
| Unknown/Censored), 1-year/2-year/5-year OS landmark flags, n_lines |
| received, treatment-free interval (IO responders) |
| |
| --- |
| |
| ## Calibration anchors (industry-grade) |
| |
| This cohort is calibrated against landmark metastatic trials and autopsy |
| series. Selection from the 46-metric scorecard: |
| |
| | Metric | Sample value (seed 42) | Target range | Source | |
| |---|---:|---|---| |
| | Breast met % | 20.6% | 14–26 | Cohort design 20% | |
| | NSCLC met % | 15.8% | 12–22 | Cohort design 18% | |
| | Prostate met % | 13.2% | 8–18 | Cohort design 12% | |
| | PDAC met % | 6.0% | 3–10 | Cohort design 6% | |
| | Age mean | 60.1 yr | 57–64 | Cohort design ~61 | |
| | ECOG 0-1 % | 65.8% | 58–75 | Cohort design ~65% | |
| | De novo % | 25.6% | 18–32 | Mixed cohort ~25% | |
| | Oligometa % | 73.0% | 60–85 | **Observed (disclosed; claim 25%)** | |
| | Diffuse % | 0.0% | 0–5 | **Observed (disclosed; should be ~20%)** | |
| | Mean n sites | 2.8 | 2.3–3.4 | Tropism-driven | |
| | CNS met % | 20.2% | 15–28 | Disibio 2008 | |
| | Bone met % | 43.4% | 34–52 | Mixed cohort | |
| | Breast bone % | 69.9% | 55–85 | Coleman 2001 ~70% | |
| | Prostate bone % | 87.9% | 78–98 | Bubendorf 2000 ~85-90% | |
| | Melanoma brain % | 57.4% | 45–75 | Advanced disease 50-60% | |
| | NSCLC EGFR+ % | 31.6% | 15–40 | Cohort design 29% | |
| | Breast HER2+ % | 17.5% | 12–28 | Cohort design 20% | |
| | Melanoma BRAF+ % | 38.3% | 25–65 | Davies 2002 ~45% | |
| | Mean n lines | 1.9 | 1.6–2.4 | Poisson 2.2 design | |
| | BSC % | 17.8% | 8–22 | Poor ECOG / visceral crisis | |
| | L1 IO % | 13.8% | 7–20 | Modern integration | |
| | L1 targeted % | 20.2% | 15–32 | Biomarker-routed | |
| | L1 ORR % | 50.6% | 42–62 | Landmark-weighted | |
| | SBRT % | 19.2% | 12–24 | Oligometa-driven | |
| | SRS % | 5.0% | 3–12 | ≤3 CNS lesions gating | |
| | Opioid analgesia % | 38.2% | 30–45 | Pain-driven | |
| | Palliative consult % | 19.2% | 10–24 | ECOG≥2 driven | |
| | **OS median overall** | **31.2 mo** | **28–38** | **Inflated by Oligometa miscalibration** | |
| | OS median breast | 57.3 mo | 42–65 | CLEOPATRA HER2+ ~57mo | |
| | OS median NSCLC | 23.1 mo | 20–36 | KEYNOTE-189 era ~28mo | |
| | OS median PDAC | 11.0 mo | 7–18 | FOLFIRINOX 11.1mo | |
| | 1yr OS % | 81.0% | 75–90 | High (oligo-driven) | |
| | 2yr OS % | 58.6% | 52–72 | High (oligo-driven) | |
| | Prostate ↔ Male | 100% | ≥100 (floor) | Structural | |
| | Ovarian ↔ Female | 100% | ≥96 (floor) | Structural | |
| | De novo → Stage IV | 100% | ≥100 (floor) | Structural | |
| | De novo → time=0 | 100% | ≥100 (floor) | Structural | |
| | CNS lesion consistency | 100% | ≥100 (floor) | Structural | |
| | SRE ⊂ bone | 100% | ≥100 (floor) | Structural | |
| | Treated → L1 not NA | 100% | ≥100 (floor) | Structural | |
| | BSC → 0 lines | 100% | ≥100 (floor) | Structural | |
| | BSC → L1 NA | 100% | ≥100 (floor) | Structural | |
| | SRS-elig consistent | 100% | ≥100 (floor) | NCCN gating | |
| | EGFR ⊂ NSCLC | 100% | ≥100 (floor) | Structural | |
| | BRAF+ ⊂ mel/CRC | 100% | ≥100 (floor) | Structural | |
| | HER2+ ⊂ breast/gastric/CRC | 100% | ≥100 (floor) | Structural | |
| |
| Full 46-metric scorecard ships in `validation_report.json` and `validation_report.md`. |
| |
| --- |
| |
| ## Files in this sample |
| |
| ``` |
| hconc014_sample/ |
| ├── hconc014_sample.csv # 500 patients × 109 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.** Up to 6 treatment lines per patient internally, |
| but only the **first 3 lines flattened to columns** (`line1_regimen`, |
| `line2_regimen`, `line3_regimen`, etc.). For full per-line trajectory, |
| use the commercial product's long-format treatment table. |
| |
| --- |
| |
| ## Schema highlights (109 columns across 9 modules) |
| |
| ### Module 1: Demographics (12 cols) |
| `patient_id`, `sku`, `cancer_type`, `primary_histology`, |
| `metastatic_treatment_intent`, `age_at_metastatic_dx`, `sex`, |
| `ecog_ps_at_met_dx`, `bmi_kg_m2`, `comorbidity_cci`, |
| `stage_at_initial_dx`, `de_novo_metastatic_flag`, |
| `time_primary_to_met_months`, `synchronous_vs_metachronous`, |
| `prior_adjuvant_therapy_flag`, `prior_neoadjuvant_therapy_flag` |
| |
| ### Module 2: Metastasis Pattern (18 cols) |
| `site_liver`, `site_lung`, `site_bone`, `site_brain`, `site_adrenal`, |
| `site_peritoneum`, `site_pleura`, `site_lymph_distant`, `site_skin`, |
| `site_spinal`, `site_pericardium`, `site_bone_marrow`, |
| `n_metastatic_sites`, `metastatic_burden_class`, `cns_metastasis_flag`, |
| `cns_lesion_count`, `leptomeningeal_disease_flag`, `srs_eligible_flag`, |
| `bone_metastasis_flag`, `skeletal_related_event`, `visceral_crisis_flag`, |
| `malignant_effusion_flag`, `peritoneal_carcinomatosis_flag`, |
| `peritoneal_ci_score`, `hepatic_tumor_burden_pct` |
| |
| ### Module 3: Biomarkers (15 cols, cancer-stratified) |
| `bio_egfr_status`, `bio_alk_status`, `bio_kras_status`, `bio_her2_status`, |
| `bio_braf_v600e_flag`, `bio_brca_status`, `bio_msi_mmr_status`, |
| `bio_pdl1_tps_pct`, `bio_pdl1_cps_score`, `bio_tmb_mut_per_mb`, |
| `bio_ar_splice_variant_flag`, `bio_ctdna_vaf_baseline`, |
| `bio_liquid_biopsy_flag`, `bio_repeat_biopsy_flag`, |
| `bio_tumor_heterogeneity_score`, `bio_hr_status`, `bio_nras_mutation_flag`, |
| `bio_psa_baseline_ng_ml`, `bio_vhl_mutation_flag`, `bio_imdc_risk`, |
| `bio_hrd_status`, `bio_ca125_baseline_iu_ml`, `bio_acquired_resistance_mechanism` |
| |
| ### Module 4: Local Therapy (12 cols) |
| `local_therapy_flag`, `sbrt_flag`, `sbrt_dose_fractions`, |
| `sbrt_treated_lesion_count`, `srs_flag`, `wbrt_flag`, |
| `surgical_met_resection_flag`, `rfa_flag`, `tace_y90_flag`, |
| `intrathecal_therapy_flag`, `palliative_radiation_flag`, |
| `bone_modifying_agent` |
| |
| ### Module 5: Supportive Care (7 cols) |
| `palliative_care_consult_flag`, `hospice_referral_flag`, |
| `opioid_analgesic_flag`, `pain_score_vas`, `nutrition_support_flag`, |
| `corticosteroid_chronic_flag`, `palliative_surgery_flag` |
| |
| ### Module 6: Treatment Lines (3 × 6 = 18 flat cols) |
| `line1_regimen`, `line1_best_response`, `line1_pfs_months`, |
| `line1_io_flag`, `line1_targeted_flag`, `line1_ttnt_months`, |
| (same for line2_*, line3_*) |
| |
| ### Module 7: Survival (7 cols) |
| `overall_survival_months`, `os_event_flag`, `cause_of_death`, |
| `landmark_1yr_os_flag`, `landmark_2yr_os_flag`, `landmark_5yr_os_flag`, |
| `n_lines_received`, `treatment_free_interval_months` |
| |
| --- |
| |
| ## Use cases |
| |
| 1. **Metastatic-stage survival modeling** — Cox PH on OS by cancer type, |
| biomarker, ECOG, burden class. |
| 2. **Treatment sequencing optimization** — compare L1 → L2 transitions |
| and their impact on OS. |
| 3. **Oligometastatic intervention analysis** — measure SBRT/SRS uptake and |
| survival benefit. |
| 4. **Organ-tropism prediction** — predict bone vs CNS vs liver mets from |
| cancer type + biomarkers. |
| 5. **CLEOPATRA replication** — HER2+ mBC OS by line of therapy. |
| 6. **mPDAC FOLFIRINOX benchmark** — Conroy 2011 mOS replication. |
| 7. **Acquired resistance modeling** — predict T790M / ESR1 / MET-amp from |
| prior treatment. |
| 8. **Bone-modifying agent uptake audit** — measure denosumab vs ZA |
| utilization in bone-met patients. |
| 9. **Palliative care quality benchmarking** — measure consultation rates |
| vs ECOG/intent guidelines. |
| 10. **NCCN guideline-concordance** — measure adherence to SRS-eligibility |
| criteria, opioid pain management. |
| 11. **Teaching & training** — medical oncology fellows, palliative care |
| fellows, ML-for-healthcare bootcamps on advanced-disease modeling. |
| |
| --- |
| |
| ## Loading examples |
| |
| ### pandas |
| ```python |
| import pandas as pd |
| df = pd.read_csv("hconc014_sample.csv") |
| print(df.shape) # (500, 109) |
| print(df["cancer_type"].value_counts()) |
| print(df["metastatic_burden_class"].value_counts()) |
| ``` |
| |
| ### Hugging Face `datasets` |
| ```python |
| from datasets import load_dataset |
| ds = load_dataset("xpertsystems/hconc014-sample") |
| df = ds["train"].to_pandas() |
| ``` |
| |
| ### Organ tropism heatmap (Disibio 2008 replication) |
| ```python |
| sites = ["site_liver","site_lung","site_bone","site_brain","site_adrenal", |
| "site_peritoneum","site_pleura","site_lymph_distant","site_skin"] |
| tropism = df.groupby("cancer_type")[sites].mean().round(2) |
| print(tropism) |
| ``` |
| |
| ### Cancer-specific survival |
| ```python |
| from lifelines import KaplanMeierFitter |
| import matplotlib.pyplot as plt |
| |
| kmf = KaplanMeierFitter() |
| for ct in ["breast_met","nsclc_met","prostate_met","pancreatic_met","melanoma_met"]: |
| sub = df[df["cancer_type"] == ct] |
| if len(sub) < 10: continue |
| kmf.fit(sub["overall_survival_months"], event_observed=sub["os_event_flag"], label=ct) |
| kmf.plot_survival_function() |
| plt.title("Metastatic Cancer OS by Type"); plt.show() |
| ``` |
| |
| ### Oligometastatic intervention impact |
| ```python |
| oligo = df[df["metastatic_burden_class"] == "Oligometastatic"] |
| local_therapy = oligo[oligo["sbrt_flag"] | oligo["surgical_met_resection_flag"]] |
| no_local = oligo[~(oligo["sbrt_flag"] | oligo["surgical_met_resection_flag"])] |
| print(f"Oligo + local: n={len(local_therapy)}, " |
| f"median OS = {local_therapy['overall_survival_months'].median():.1f} mo") |
| print(f"Oligo no local: n={len(no_local)}, " |
| f"median OS = {no_local['overall_survival_months'].median():.1f} mo") |
| ``` |
|
|
| ### Line-of-therapy response cascade |
| ```python |
| treated = df[df["n_lines_received"] >= 1] |
| l1_response = treated["line1_best_response"].value_counts(normalize=True).round(3) |
| l2_response = treated[treated["n_lines_received"] >= 2]["line2_best_response"].value_counts(normalize=True).round(3) |
| print("L1 response distribution:\n", l1_response) |
| print("L2 response distribution:\n", l2_response) |
| ``` |
|
|
| ### Bone-modifying agent audit |
| ```python |
| bone_pts = df[df["bone_metastasis_flag"] == 1] |
| bma = bone_pts["bone_modifying_agent"].value_counts(normalize=True).round(3) |
| print(f"BMA utilization in bone-met patients:\n{bma}") |
| # Expected: Denosumab ~50%, Zoledronic_acid ~38%, None ~12% |
| ``` |
|
|
| --- |
|
|
| ## 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. **🚨 Oligometastatic burden mis-classification.** Generator's burden |
| thresholds at line 320-325 are: |
| - `n_sites <= 3` → Oligometastatic |
| - `4-9` → Polymetastatic |
| - `>9` → Diffuse |
|
|
| But the 12 organ-tropism site probabilities (line 47-62) are each <0.90, |
| producing **mean n_sites ~2.8 per patient**. Result: ~73% of patients |
| land in Oligometastatic (vs cohort claim 25%), and ~0% in Diffuse (need |
| >9 sites). The full commercial product re-calibrates either the tropism |
| probabilities upward or the burden thresholds downward to match the |
| intended ~25/55/20 distribution. |
| |
| 2. **🚨 OS inflated by Oligometa miscalibration.** The survival formula |
| (line 666-683) applies a `1.18x` median OS multiplier to Oligometastatic |
| patients (intended to capture the better prognosis of low-burden |
| disease). Because ~73% of cohort is Oligo instead of intended 25%, the |
| 1.18x boost applies to far too many patients, inflating cohort median |
| OS to ~32mo (vs literature mixed-metastatic mOS ~18-24mo). Cancer- |
| specific OS values still match landmark benchmarks (breast ~57mo, |
| PDAC ~11mo, NSCLC ~23mo) because the BENCHMARK_OS table is properly |
| anchored. |
| |
| 3. **Patient IDs are 16-digit random integers** (line 703: `rng.integers(10**15, |
| 10**16-1)`). At n=500 collision probability is negligible (birthday |
| paradox: ~10^-11), but at n=25,000+ collisions become possible. Full |
| product offers UUID format. |
|
|
| 4. **No CTCAE toxicity grading.** Unlike HCONC012 (chemotherapy), this |
| SKU does not track per-line toxicity grades. Adverse events are |
| summarized only by dose reduction flag and treatment switch reason. |
|
|
| 5. **Treatment lines flattened to first 3.** The generator internally |
| creates up to 6 treatment lines (max_lines parameter), but only lines |
| 1-3 are surfaced in the output columns. For patients with 4+ lines, |
| the later lines are computed but not exposed. |
| |
| 6. **No imaging tumor sum tracking.** Unlike HCONC011 (multi-cancer |
| progression) and HCONC013 (CPI), this SKU does not include RECIST |
| target lesion measurements at imaging timepoints. Best response is |
| captured per line as CR/PR/SD/PD only. |
| |
| 7. **`acquired_resistance_mechanism` populated only at progression OR |
| line≥2** (line 562). De-novo MET-amp at L1 or other primary resistance |
| markers may not be surfaced. |
| |
| 8. **CCI distribution from a hardcoded `_cci_probs()` method** (line 302-305) |
| not validated against real metastatic cohort registries. |
| |
| 9. **De novo rates by cancer type** at line 270-273 are hardcoded |
| approximations. PDAC 80%, SCLC 70%, HCC 60% are reasonable; breast |
| 6%, prostate 5%, renal 8% match SEER-CONCORD estimates. |
| |
| 10. **Hospice referral gated only by ECOG ≥3** (line 658). Real-world |
| hospice patterns also reflect age, cancer trajectory, and patient |
| preference. |
| |
| 11. **No social determinants** — race/ethnicity, insurance status, |
| geography not captured. Full product offers SDOH module. |
| |
| 12. **Bone marrow site (`site_bone_marrow`)** modeled but no leukemia |
| distinction from solid-tumor marrow involvement. |
| |
| 13. **Pericardium site rare across all cancers** (probabilities 3-10%). |
| Pericardial effusion clinical workflow not captured. |
| |
| 14. **No germline testing capture beyond BRCA** — Lynch syndrome, |
| Li-Fraumeni, HBOC panel breadth not represented. |
| |
| 15. **No external validation** against real metastatic cancer registries |
| (CONCORD-3, SEER-Medicare, Flatiron Advanced) beyond cohort design |
| targets and landmark trial endpoints. |
| |
| 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) | |
| | Burden classification | Oligo ~73% (disclosed) | **FIXED** (Oligo 25%, Poly 55%, Diffuse 20%) | |
| | OS calibration | Inflated by miscalibration | **FIXED** (mixed mOS ~18-24mo literature-anchored) | |
| | Treatment lines exposed | First 3 flat | All 6 (wide or long format) | |
| | Patient ID format | 16-digit random | UUID option | |
| | CTCAE toxicity | Not captured | Per-line CTCAE v5.0 grades | |
| | Imaging RECIST sums | Not captured | Per-timepoint sums + waterfall | |
| | SDOH module | Not included | Race/ethnicity/insurance/geography | |
| | Germline panel | BRCA only | Full HBOC + Lynch + Li-Fraumeni | |
| | Validation report | Yes (46 metrics) | Yes + custom scorecard | |
| | Format | CSV | CSV, Parquet, JSON | |
| | License | CC-BY-NC-4.0 (non-commercial) | Commercial use license | |
| | Schema mapping | — | OMOP CDM / mCODE / Flatiron Advanced | |
| | Support | Community | Email / SLA | |
| |
| --- |
| |
| ## Citation |
| |
| ```bibtex |
| @dataset{xpertsystems_hconc014_2026, |
| title = {HC-ONC-014: Metastatic Cancer Synthetic Cohort spanning 13 Metastatic Cancer Types with 12-Organ Tropism, Biomarker-Routed Treatment Sequences with Acquired Resistance Mechanisms, Oligometastatic-Directed Local Therapy (SBRT/SRS/Surgical Resection/RFA/TACE), Bone-Modifying Agents, Comprehensive Supportive Care, and Weibull-Anchored Survival Endpoints calibrated to 13 Cancer × 3 Line-of-Therapy Landmark Trial Benchmarks}, |
| author = {{XpertSystems.ai}}, |
| year = {2026}, |
| version= {1.0.0}, |
| url = {https://huggingface.co/datasets/xpertsystems/hconc014-sample}, |
| license= {CC-BY-NC-4.0 (sample); Commercial (full product)}, |
| note = {Calibrated against CLEOPATRA (Swain 2020 pertuzumab+trastuzumab+docetaxel HER2+ mBC OS 57mo), MONALEESA-3 (Slamon 2020 ribociclib+fulvestrant HR+ mBC), KEYNOTE-189 (Gandhi 2018 pembro+chemo mNSCLC OS 22mo), OAK (Rittmeyer 2017 atezolizumab mNSCLC), CheckMate-067 (Wolchok 2017/2022 ipi+nivo mMel OS 72mo+), RELATIVITY-047 (Tawbi 2022 relatlimab+nivo mMel), COMBI-d (Long 2014 dabrafenib+trametinib BRAF+ mMel), ENZAMET (Davis 2019 enzalutamide mHSPC), LATITUDE (Fizazi 2017 abiraterone mHSPC), TITAN (Chi 2019 apalutamide mHSPC), ARASENS (Smith 2022 darolutamide+docetaxel mHSPC), FOLFIRINOX-PRODIGE (Conroy 2011 mPDAC OS 11.1mo), NAPOLI-3 (Wainberg 2023 NALIRIFOX mPDAC), KEYNOTE-426 (Rini 2019 pembro+axitinib mRCC), CABOSUN (Choueiri 2017 cabozantinib mRCC), CASPIAN (Paz-Ares 2019 durvalumab+EP mSCLC), Disibio 2008 (autopsy metastasis tropism), Bubendorf 2000 (prostate cancer bone mets ~85%), Coleman 2001 (breast cancer bone mets ~70%), Patel 1978 (melanoma metastasis sites), AJCC 8th Edition Staging, NCCN Metastatic Cancer Guidelines 2024.} |
| } |
| ``` |
| |
| --- |
| |
| ## 🎉 Oncology Vertical Complete |
| |
| This SKU **closes out the XpertSystems Oncology vertical** at **14 SKUs total**: |
| |
| 1. HC-ONC-001 — Breast Cancer |
| 2. HC-ONC-002 — Lung Cancer (NSCLC + SCLC) |
| 3. HC-ONC-003 — Prostate Cancer (with PSA longitudinal) |
| 4. HC-ONC-004 — Colorectal Cancer (with CEA longitudinal) |
| 5. HC-ONC-005 — Pancreatic Cancer (5 sub-tables) |
| 6. HC-ONC-006 — Liver Cancer / HCC (with AFP longitudinal) |
| 7. HC-ONC-007 — Leukemia (with MRD longitudinal) |
| 8. HC-ONC-008 — Lymphoma (with PET + CAR-T) |
| 9. HC-ONC-009 — Melanoma |
| 10. HC-ONC-010 — Ovarian Cancer (with CA-125 longitudinal) |
| 11. HC-ONC-011 — Multi-Cancer Tumor Progression (pan-cancer) |
| 12. HC-ONC-012 — Chemotherapy Response (18 cancer types × 60+ regimens) |
| 13. HC-ONC-013 — Immunotherapy Response (CPI + irAE) |
| 14. HC-ONC-014 — **Metastatic Cancer (this SKU)** |
| |
| Together these provide pan-cancer synthetic data infrastructure for outcomes research, clinical decision support, AI training, and education. All 14 SKUs validated at **Grade A+ across 6 canonical seeds**. |
| |
| --- |
| |
| ## Contact |
| |
| - **Email:** [pradeep@xpertsystems.ai](mailto:pradeep@xpertsystems.ai) |
| - **Web:** [https://xpertsystems.ai](https://xpertsystems.ai) |
| - **Vertical:** Healthcare / Oncology / Metastatic Disease |
| - **SKU catalog:** SKU 14 (FINAL) of the Oncology vertical (24 SKUs total across Cardiology + Oncology); ~89 SKUs across 8 verticals |
| |
| XpertSystems.ai — synthetic data, calibrated to real-world registries. |
| |
