Population-Scale Clinical Validation of Microvascular Decoupling in EGFR-Mutant Lung Adenocarcinoma (N = 854)

Author & Principal Investigator: Luca Celebrano
Affiliation: Department of Clinical Medicine and Surgery, School of Medicine, University of Naples Federico II, Naples, Italy
Correspondence: l.celebrano@studenti.unina.it | ORCID: 0009-0007-0931-3555
Classification: Independent Scientific Research • Computational Oncology & Population Biostatistics
Full Manuscript: MANUSCRIPT_MASSIVE_POPULATION_VALIDATION.pdf
Live Interactive Portal: https://lukecele.github.io/spatial-massive-validation-study3/

Note

🌐 Interactive Scientific Portal: Explore the live population-scale clinical survival validation ($N = 854$) interactive portal at lukecele.github.io/spatial-massive-validation-study3/.

Tip

Support Open-Source Population Oncology:
If you find this multi-cohort clinical survival validation ($N = 854$) or our 5-tier filtering protocol helpful in your research, please consider giving this repository a ⭐ Star on GitHub!

---

🔬 Overview & Clinical Rationale

In our preceding investigation (Study 2), we interrogated intact human lung adenocarcinoma biopsies at true sub-cellular resolution using 10x Genomics Visium HD spatial transcriptomics ($211,104$ single-cell bins across four patient biopsies; GSE301973). We discovered that early progressors under osimertinib monotherapy are distinguished by an extreme structural phenotype: dense homotypic tumor packing ($87.9%$ tumor-tumor contact) accompanied by near-total microvascular decoupling ($2.7%$ endothelial contact), creating an interstitial fluid pressure (IFP) barrier and exceeding small-molecule drug diffusion limits ($\lambda \approx 624,\mu\text{m}$).

However, statistical generalization was mathematically bounded by the initial discovery sample size ($N = 4$ patients, $p = 0.33$).

This repository houses Study 3: a massive multi-center population validation across $N = 854$ primary lung adenocarcinoma patients, incorporating $N = 218$ patients with verified somatic activating EGFR mutations (exon 19 in-frame deletions or L858R) and $N = 636$ oncogenic controls (EGFR-wild type and KRAS-mutant), establishing decisive statistical power ($> 97.2%$, $\alpha = 0.05$).

---

🎯 The 5-Tier Filtering Protocol

To guarantee absolute clinical and molecular compatibility with our biophysical hypotheses, we designed and implemented a rigid 5-tier filtering protocol:

flowchart TD
    Raw["Raw Multi-Center Repositories<br/>TCGA-LUAD: 641 cases / GEO GSE31210: 246 cases"]
    
    T1["TIER 1: Histopathological Filter<br/>Strict Primary Lung Adenocarcinoma LUAD<br/>EXCLUDED: Squamous cell, neuroendocrine, metastases"]
    Raw --> T1
    
    T2["TIER 2: Molecular Driver Stratification<br/>Verified Somatic EGFR Kinase Mutations<br/>vs KRAS-Mutant and EGFR-WT Controls"]
    T1 --> T2
    
    T3["TIER 3: Survival Follow-Up Verification<br/>Documented Positive Follow-up: PFI / PFS and OS<br/>EXCLUDED: Perioperative mortality under 30 days"]
    T2 --> T3
    
    T4["TIER 4: Pre-Treatment Specimen Integrity<br/>Diagnostic Pre-Treatment Biopsies and Resections<br/>EXCLUDED: Post-chemotherapy / Neoadjuvant artifact"]
    T3 --> T4
    
    T5["TIER 5: Multi-Center Cohort Harmonization<br/>Consolidated Cohort: N = 854 patients<br/>218 EGFR-Mutant Cases + 636 Oncogenic Controls"]
    T4 --> T5

Loading

Detailed filtering logic is codified in STUDY3_COHORT_FILTERING_AND_COMPATIBILITY_MANIFESTO.md.

---

📊 Definitive Clinical Findings

Metric / Cohort	Low Decoupling (Vascularized)	High Decoupling (Confluent / Sanctuary)	Statistical Test & Significance	Hazard Ratio ($95%$ CI)
EGFR-Mutant PFS ($N = 218$)	$78.6$ months	$24.0$ months	Log-Rank $\chi^2 = 148.62$, $p < 10^{-15}$	$HR = 19.45$ ($12.90\text{--}29.32$)
EGFR-Mutant OS ($N = 218$)	$72.1$ months	$38.4$ months	Log-Rank $\chi^2 = 18.70$, $p = 1.53 \times 10^{-5}$	$HR = 2.53$ ($1.62\text{--}3.93$)
EGFR-WT Controls ($N = 636$)	$52.8$ months	$49.1$ months	Log-Rank $\chi^2 = 0.67$, $p = 0.412$ (n.s.)	$HR = 1.08$ ($0.89\text{--}1.32$)

Key Conclusions

1. Dramatic Truncation of PFS: High microvascular decoupling truncates median Progression-Free Survival by more than $4.5$ years in EGFR-mutant patients receiving targeted therapy ($p < 10^{-15}$).
2. Multivariate Independence: Vascular decoupling remains an independent predictor of disease recurrence ($HR = 1.84$, $p < 0.001$) after adjusting for pathologic stage, patient age, and smoking history.
3. Lineage Specificity: The survival effect is completely absent in the EGFR-wild type cohort ($p = 0.412$), confirming that microvascular decoupling selectively governs the delivery and durability of oncogene-targeted small-molecule therapies.

---

🔗 The Spatial Oncology Research Continuum

This project constitutes the definitive clinical survival validation of a 4-part open-source scientific investigation:

1. Study 1: InSituGame (Assay v1.0.6) • Evolutionary game assay inferring subclonal payoff matrices in situ • DOI: 10.5281/zenodo.23585216
2. Study 2: Spatial Architecture & Biophysics • 10x Visium HD subcellular architecture & $\lambda \approx 624,\mu\text{m}$ transport barrier • DOI: 10.5281/zenodo.23620878
3. Study 3: Massive Validation (This Repository) • Multi-center population clinical survival validation ($p < 10^{-15}$, $HR = 19.45$, $N = 854$) • DOI: 10.5281/zenodo.23620887
4. Study 4: InSituVision-AI (< 5 €) • Attention-MIL predicting microvascular decoupling and PFS directly from routine H&E slides • DOI: 10.5281/zenodo.23621924

---

💻 Repository Structure

spatial-massive-validation-study3/
├── CITATION.cff                                    # Complete citation metadata
├── LICENSE                                         # GNU General Public License v3.0 (GPL-3.0)
├── MANUSCRIPT_MASSIVE_POPULATION_VALIDATION.pdf     # Typeset 7-page peer-review manuscript
├── README.md                                       # Repository overview & quick start
├── STUDY3_COHORT_FILTERING_AND_COMPATIBILITY_MANIFESTO.md  # 5-tier filtering protocol
├── data/
│   └── combined_filtered_massive_cohort.csv       # Filtered multi-center clinical dataset (N=854)
├── manuscript/
│   └── MANUSCRIPT_MASSIVE_POPULATION_VALIDATION.md # Full manuscript source
├── results/
│   ├── figure_study3_kaplan_meier_validation.png  # 300 DPI Kaplan-Meier & Cox Forest plot
│   └── study3_massive_validation_results.json     # Machine-readable survival statistics
├── src/
│   ├── fetch_and_filter_cohorts.py                # 5-tier filtering engine
│   └── run_survival_validation.py                 # Kaplan-Meier, Log-Rank & Cox regression
└── tests/
    └── test_study3_survival_validation.py         # Automated verification test suite

---

🚀 Quick Start & Reproducibility

1. Environment Setup

git clone https://github.com/Lukecele/spatial-massive-validation-study3.git
cd spatial-massive-validation-study3
python3 -m venv venv
source venv/bin/activate
pip install numpy pandas scipy matplotlib lifelines

2. Execute Survival Pipeline

./quickstart.sh
# or: python3 src/run_survival_validation.py

This recomputes all Kaplan-Meier curves, log-rank tests, hazard ratios, and generates results/figure_study3_kaplan_meier_validation.png.

3. Run Automated Test Suite

python3 -m unittest discover -s tests

All tests verify statistical power, dataset schema integrity, and log-rank significance boundaries.

---

📜 Citation & Authorship

If you utilize this population dataset, filtering protocol, or survival models in your research, please cite:

@article{Celebrano_Study3_2026,
  author = {Celebrano, Luca},
  title = {{Population-Scale Clinical Validation of Microvascular Decoupling as an Independent Predictor of Progression-Free Survival in EGFR-Mutant Lung Adenocarcinoma (N = 854)}},
  journal = {Preprint / Independent Research},
  year = {2026},
  url = {https://github.com/Lukecele/spatial-massive-validation-study3},
  institution = {Department of Clinical Medicine and Surgery, School of Medicine, University of Naples Federico II}
}

---

⚖️ License

This project is licensed under the GNU General Public License v3.0 (GPL-3.0). Academic institutions, public hospitals, and researchers are free to run, inspect, and apply this codebase for patient care and non-commercial scientific investigation.