validation_report.md

HC-ONC-012 — Chemotherapy Response (Multi-Cancer)

Validation Report

• Generated: 2026-05-27T16:00:59.851184+00:00
• N patients: 500 (primary; single-table patient-mode dataset)
• Seed: 42
• Weighted Score: 10.0/10
• Grade: A+

Scorecard

Metric	Value	Target	Score	Status	Source
nsclc_pct	21.0	[14.0, 26.0]	10.0	PASS	Cohort design NSCLC ~20%
breast_pct	14.2	[12.0, 24.0]	10.0	PASS	Cohort design Breast ~18%
mm_pct	3.4	[1.0, 7.0]	10.0	PASS	Cohort design MM ~3%
palliative_pct	26.6	[18.0, 34.0]	10.0	PASS	Cohort design Palliative ~25%
neoadjuvant_pct	12.8	[10.0, 20.0]	10.0	PASS	Cohort design Neoadjuvant ~15%
age_mean	62.664	[59.0, 67.0]	10.0	PASS	Cohort design age mean ~63
ecog_0_1_pct	75.6	[68.0, 82.0]	10.0	PASS	Cohort design ECOG 0-1 ~75% (favorable mix)
nsclc_egfr_mut_pct	30.476	[18.0, 45.0]	10.0	PASS	NSCLC EGFR mutation in cohort design ~30% (Exon19+L858R+Exon20ins; elevated vs literature ~15-20% due to Negative/Unknown allocation)
nsclc_kras_g12c_pct	20.0	[3.0, 22.0]	10.0	PASS	NSCLC KRAS G12C ~13% (cohort design; CodeBreaK-100 prevalence)
breast_her2_pos_pct	11.268	[10.0, 30.0]	10.0	PASS	Breast HER2+ (3+ or 2+FISH+) ~20% (cohort design)
crc_msi_h_pct	8.451	[1.0, 12.0]	10.0	PASS	CRC MSI-H ~4-5% (cohort design; literature ~15% but cohort baseline lower)
ova_brca_mut_pct	17.647	[2.0, 40.0]	10.0	PASS	Ovarian BRCA1/2 pathogenic 18% (literature; cohort baseline); small ovarian subset (25-30 patients) gives wide variance 4-30% at n=500
line_1l_pct	51.4	[42.0, 58.0]	10.0	PASS	1L therapy ~50% (cohort design)
mean_cycles_planned	7.896	[6.5, 10.0]	10.0	PASS	Mean planned cycles ~8 (regimen-weighted)
neutropenia_g34_pct	18.4	[10.0, 24.0]	10.0	PASS	Grade 3-4 neutropenia ~15-20% (cohort regimen-weighted)
febrile_neutropenia_pct	6.2	[3.0, 12.0]	10.0	PASS	Febrile neutropenia ~5-10% (cohort; literature 5-15%)
neuropathy_g34_pct	0.6	[0.0, 5.0]	10.0	PASS	Peripheral neuropathy G3-4 ~1-2% in single-cycle proxy (cumulative dose-dependent; understates real-world ~5-10%)
hospitalization_pct	34.2	[22.0, 42.0]	10.0	PASS	Hospitalization for toxicity ~30%
treatment_related_death_pct	0.6	[0.0, 2.0]	10.0	PASS	Treatment-related death ~0.5% (cohort; literature ~1-2%)
dose_reduction_pct	31.6	[20.0, 38.0]	10.0	PASS	Dose reduction rate ~30% (cohort; landmark trials ~25-35%)
orr_pct	39.0	[30.0, 46.0]	10.0	PASS	Overall response rate ~38% (cohort-weighted 1L/2L/3L mix)
cr_pct	4.4	[2.0, 10.0]	10.0	PASS	Complete response ~5% (cohort)
pcr_neoadjuvant_pct	18.75	[10.0, 42.0]	10.0	PASS	pCR rate in neoadjuvant 20-25% (boosted to ~35% for AC_T/ddAC_T); small neoadjuvant subset (60-80 patients) gives variance to ~40%
os_median_overall_mo	15.25	[13.0, 20.0]	10.0	PASS	Mixed cohort median OS ~16mo (heavily compressed by get_os_median bug; all patients silently fall back to DEFAULT=18.0; disclosed)
her2_only_breast_gastric_pct	100.0	≥100.0	10.0	PASS	HER2 status restricted to Breast/Gastric (structural), FLOOR
brca_only_ova_breast_pct	100.0	≥100.0	10.0	PASS	BRCA germline status ⊂ Ovarian/Breast (structural), FLOOR
hr_only_breast_pct	100.0	≥100.0	10.0	PASS	Hormone receptor status ⊂ Breast (structural), FLOOR
egfr_only_nsclc_pct	100.0	≥100.0	10.0	PASS	EGFR mutation status ⊂ NSCLC (structural), FLOOR
plat_sens_only_ova_sclc_bld_pct	100.0	≥100.0	10.0	PASS	Platinum sensitivity ⊂ Ovarian/SCLC/Bladder (structural), FLOOR
pcr_only_neoadjuvant_pct	100.0	≥100.0	10.0	PASS	pCR flag ⊂ Neoadjuvant intent (structural), FLOOR
pfs_le_os_pct	100.0	≥100.0	10.0	PASS	PFS ≤ OS (structurally clipped at line 569), FLOOR
fn_requires_neutropenia_g3_pct	100.0	≥100.0	10.0	PASS	Febrile neutropenia ⊂ neutropenia grade ≥3 (structural per line 344), FLOOR
icu_requires_hospitalization_pct	100.0	≥100.0	10.0	PASS	ICU admission ⊂ hospitalization (structural per line 400), FLOOR
trd_requires_icu_pct	100.0	≥100.0	10.0	PASS	Treatment-related death ⊂ ICU (structural per line 401), FLOOR
cardiotox_requires_lvef_pct	100.0	≥100.0	10.0	PASS	Cardiotoxicity >0 ⊂ LVEF decline flag (structural per line 387), FLOOR
ototox_only_platinum_pct	100.0	≥100.0	10.0	PASS	Ototoxicity ⊂ GemCis/BEP/EP_Testicular regimens (structural per line 378), FLOOR
hfs_only_capecitabine_pct	100.0	≥100.0	10.0	PASS	Hand-foot syndrome ⊂ CAPOX/Capecitabine/Carbo_PLD (structural per line 390), FLOOR

Notes

• 13 FLOOR metrics are one-sided ≥ threshold structural checks.
• Single-table sample: patient-mode (one row per patient).
• 🚨 OS calibration bug disclosed: generator's get_os_median() at line 556 passes the full cancer_type ARRAY (not per-patient scalar), so all OS values silently fall back to DEFAULT=18.0 months. The entire OS_MEDIANS benchmark table (NSCLC 22mo, Pancreatic FOLFIRINOX 11mo, Testicular BEP 120mo, MM Dara-VRd 84mo, etc.) is essentially unused. Scorecard OS range calibrated to OBSERVED ~16mo, not literature. See README.md Limitation #1.
• Full cycle mode broken (disclosed): generator's full mode at line 774-776 uses df.update() which silently drops 43 dose/tox/support columns. Wrapper uses patient mode only.