E1.S4

DATASET DESCRIPTION

Disease Area

Thoracic Oncology — Non-Small Cell Lung Cancer (NSCLC)

Specific Research Focus

Early-stage lung cancer treatment response prediction and post-treatment recurrence forecasting

Dataset Theme / Edition

Biomarker-Guided Precision Oncology — Synthetic Clinical Dataset, Edition v1.0.0

Short Dataset Summary

A fully synthetic, clinically grounded dataset of 1,000 early-stage NSCLC patients with 42 cross-sectional features and 13,970 longitudinal clinical visit records. Built to encode real-world molecular marker relationships (EGFR, ALK, KRAS, PD-L1, STK11), treatment eligibility rules, and five disease trajectory archetypes across active treatment, surveillance, and long-term follow-up phases.

Intended Research Applications

Treatment response classification, recurrence risk regression, progression-free and overall survival modeling, acquired therapy resistance detection, biomarker-guided XAI/SHAP analysis, SBRT vs. surgery comparative effectiveness research, federated learning benchmarking, and TimeGAN-based synthetic data augmentation

Total Number of Patients

1,000 synthetic patients (cross-sectional) generating 13,970 longitudinal visit records

Number of Features / Parameters

42 cross-sectional features + 25 longitudinal temporal columns = 67 total variables across both dataset components

Data Modalities Included

Demographics and risk profile, clinical symptoms and functional status (ECOG), imaging and pathology findings (CT, tumor size, histology, staging), molecular biomarkers (EGFR/ALK/KRAS/ROS1/PD-L1/STK11/TMB), treatment assignments (surgery, chemotherapy, radiation, targeted therapy, immunotherapy), and longitudinal visit-level clinical time-series

Recommended AI Tasks

Multi-class classification, binary and continuous regression, time-series sequence modeling (LSTM, Transformer, TimeGAN), survival analysis (Cox PH, DeepHit), feature importance and SHAP explainability, federated learning, and multi-task learning

Dataset Size

Cross-sectional CSV: 277 KB — Longitudinal CSV: 2.0 MB — NumPy tensors: 1.7 MB (sequences) + 79 KB (attention mask) — Total package: ~6.3 MB across all output formats

Release Date

May 2026 

Why This Dataset Exists

Real oncology datasets are heavily restricted due to patient privacy regulations (HIPAA, GDPR), making it extremely difficult for researchers, startups, and academic teams to develop and benchmark clinical AI models without institutional data access agreements. This dataset provides an open, privacy-safe, ML-ready alternative that faithfully encodes the clinical complexity of NSCLC management — including biomarker-guided treatment logic, longitudinal disease progression, and realistic missing data — without exposing any real patient information.

Real-World Problem Addressed

In early-stage NSCLC, treatment selection is increasingly driven by molecular profiling (EGFR, ALK, KRAS mutations), yet predicting which patients will achieve complete response, develop acquired resistance, or relapse within two years remains a significant clinical challenge. This dataset enables researchers to build and validate AI models that could ultimately support oncologists in personalizing treatment decisions, anticipating resistance mechanisms, and improving long-term survival outcomes for the approximately 2.5 million people diagnosed with lung cancer globally each year.