Module 8Skeleton — synthesized cases

Pathology Normalization

Harmonize pathology interpretation across readers and sites. Three views: population-level consensus, tumor-board prep for high-risk cases, and per-patient deep history.

Cases under review

184

Across 6 contributing sites

Agreement κ

0.78

Cohen's kappa, 30-day rolling

Discordant pending

Awaiting consensus

Free-text reads normalized

412

Auto-mapped this quarter

Discordant reads — consensus queue

4 cases pending

Reader AReader BDeltaStatusConfidence

PATH-2186·2026-05-12·Head of pancreasneeds consensus

Reader ADr. K. Park

Pancreatic ductal adenocarcinoma, G2

Confidence 92%

Reader BDr. R. Adekoya

PanIN-3 with adenocarcinoma in situ

Confidence 79%

Why this matters: Invasion vs. in-situ — affects staging and treatment intent

PATH-2191·2026-05-13·Body of pancreasneeds consensus

Reader ADr. M. Liu

Adenocarcinoma, moderately differentiated, perineural invasion

Confidence 95%

Reader BDr. T. Velasquez

Adenocarcinoma, moderate-to-poorly differentiated, perineural and vascular invasion

Confidence 88%

Why this matters: Vascular invasion presence — affects risk stratification

PATH-2198·2026-05-15·Uncinate processin review

Reader ADr. K. Park

Intraductal papillary mucinous neoplasm, low grade

Confidence 81%

Reader BDr. J. Otieno

IPMN with focal high-grade dysplasia

Confidence 83%

Why this matters: Dysplasia grade — affects surgical decision

PATH-2203·2026-05-16·Liver segment IVescalated

Reader ADr. M. Liu

Metastatic pancreatic adenocarcinoma

Confidence 94%

Reader BDr. R. Adekoya

Cholangiocarcinoma (rule out)

Confidence 71%

Why this matters: Primary site of malignancy — pancreas vs biliary

Free-text pathology → normalized variables

Demo extraction

Sample synoptic on the left; normalized output on the right. Production version reads from LIS / EMR pathology streams and emits to Patient Review, Cohort Builder, Risk Engine, and Treatment Recommendation.

Surgical pathology synoptic · 2026-05-14 · synthesized

SPECIMEN: Pancreaticoduodenectomy specimen, head of pancreas.
MICROSCOPIC: Sections show invasive ductal adenocarcinoma, moderately differentiated (G2),
measuring 2.8 cm in greatest dimension. Tumor extends into peripancreatic adipose tissue.
Perineural invasion is present. Lymphovascular invasion not identified. Surgical margins:
pancreatic neck margin clear (R0), uncinate margin involved by tumor (R1).
Regional lymph nodes: 3 of 18 nodes positive.

Normalized output

DiagnosisPancreatic ductal adenocarcinoma
“…invasive ductal adenocarcinoma…”
97% confidence
GradeG2 (moderately differentiated)
“…moderately differentiated (G2)…”
95% confidence
Size (cm)2.8
“…2.8 cm in greatest dimension…”
99% confidence
Perineural invasionPresent
“…perineural invasion is present…”
97% confidence
Lymphovascular invasionAbsent
“…lymphovascular invasion not identified…”
93% confidence
Margin statusR1 (uncinate)
“…uncinate margin involved by tumor (R1)…”
96% confidence
Nodes positive / total3 / 18
“…3 of 18 nodes positive…”
98% confidence

Standardized outputs feed into

Patient Review

Integrated case synthesis

LLM Cohort Builder

Cohort-relevant variables

Risk Engine

Path features in scoring

Treatment Recommendation

Path-aware gating

Skeleton implementation: discordant cases above are synthesized to demonstrate the consensus workflow shape. Production version needs a Reader and AdjudicationEvent schema, real pathology stream ingest, and LLM-powered free-text mapping. Spec module 8 §"Reader disagreement detection and consensus review workflow".