AnodeAI/Synthetic-cancer-clinical-genomics

Synthetic Cancer Clinical Genomics Dataset

This repository contains multi-modal synthetic clinical-genomic data designed for oncology tracking, therapeutic progression analysis, and financial cost-modeling. The dataset is organized relationally under four core entities: Patients, Genomic Biomarkers, Clinical Encounters, and Financial Claims.

Dataset Structure

The dataset is provided as a unified JSON file containing four arrays of structured objects.

{
  "patients": [...],
  "genomic_biomarkers": [...],
  "clinical_encounters": [...],
  "financial_claims": [...]
}

---

Data Fields

1. Patients Table (patients)

Contains baseline demographics for each unique patient tracked across the clinical timeline.

Field Name	Type	Description	Example
patient_id	String	Unique identifier for the patient (Primary Key)	"PT-60900"
age_at_diagnosis	Integer	Patient's age at initial cancer diagnosis	83
biological_sex	String	Confirmed biological sex (Male, Female)	"Female"
race_ethnicity	String	Demographic group (White, Black, Asian, Hispanic, Other)	"Black"

2. Genomic Biomarkers Table (genomic_biomarkers)

Tracks sequencing pipeline metrics, targeted alterations, and immune checkpoint expression values.

Field Name	Type	Description	Example
genomic_id	String	Unique identifier for the molecular testing event (Primary Key)	"G-92403"
patient_id	String	Associated patient identifier (Foreign Key linking to patients)	"PT-60900"
sequencing_date	String	ISO 8601 date of the genomic profile extraction	"2025-08-23"
gene_mutated	String	Target mutated gene analyzed	"HER2"
variant_classification	String	Specific variant alteration category	"R273H"
tmb_score	Float	Tumor Mutational Burden score (mutations per megabase)	34.2
pd_l1_expression_pct	Integer	Tumor proportion score (TPS) percentage for PD-L1 expression	29

3. Clinical Encounters Table (clinical_encounters)

Captures serial physical assessments, clinical staging advancements, and real-world therapy switches.

Field Name	Type	Description	Example
encounter_id	String	Unique clinical interaction code (Primary Key)	"ENC-35247"
patient_id	String	Associated patient identifier (Foreign Key linking to patients)	"PT-49536"
date_of_encounter	String	ISO 8601 date of the specific medical appointment	"2024-11-14"
primary_icd10_code	String	ICD-10 code designating malignant neoplasm site location	"C34.90"
cancer_stage	String	Clinical disease stage evaluated during the encounter	"Stage IV"
treatment_line	Integer	Active line of systemic antineoplastic therapy	1
medication_rxnorm	String	RxNorm Concept Unique Identifier (CUI) for prescribed medication	"2145574"
progression_free_survival_status	Integer	Binary indicator for clinical progression event (1 = Progressed, 0 = Stable)	1

4. Financial Claims Table (financial_claims)

Captures cost profiles, contractually allowed medical insurance amounts, and liability balances.

Field Name	Type	Description	Example
claim_id	String	Unique invoice/billing system item (Primary Key)	"CLM-98596"
patient_id	String	Associated patient identifier (Foreign Key linking to patients)	"PT-49536"
encounter_id	String	Linked clinical encounter (Foreign Key linking to clinical_encounters)	"ENC-35247"
allowed_amount	Float	Maximum negotiated contract value allowed by the payor ($)	18622.89
out_of_pocket_cost	Float	Patient out-of-pocket financial liability ($)	677.35

---

Intended Use Cases

• Survival Analysis: Modeling Real-World Progression-Free Survival (rwPFS) trends utilizing biomarker cohorts (tmb_score, gene_mutated).
• Health Economics and Outcomes Research (HEOR): Aggregating longitudinal out-of-pocket balances and total allowed claims costs linked directly to specific therapy lines (treatment_line).
• Relational Feature Engineering: Evaluating categorical entity embeddings via multi-table merge architectures.

---

Data Loading Quickstart

Load the entire schema into separate tabular pandas DataFrames via Python:

import json
import pandas as pd

# Load relational file
with open("oncology_data_batch_4.json", "r") as file:
    data = json.load(file)

# Parse distinct arrays into relational DataFrames
df_patients = pd.DataFrame(data["patients"])
df_genomics = pd.DataFrame(data["genomic_biomarkers"])
df_encounters = pd.DataFrame(data["clinical_encounters"])
df_claims = pd.DataFrame(data["financial_claims"])

print(f"Loaded {len(df_patients)} patients and {len(df_genomics)} biomarker profiles.")