GASTROBRAIN trial is an international (2 countries) observational, multicenter (15 centers) retrospective cohort study designed to investigate local treatment strategies for brain metastases of gastric and esophageal cancer

Brain metastases (BM) arising from gastric cancer (GC) and esophageal cancer (EC) represent a rare but devastating complication of upper gastrointestinal malignancies. Historical estimates have placed the incidence of BM in this population at well below 1%, with some of the earliest large series reporting figures as low as 0.16% to 0.7%. This rarity, however, belies the clinical significance of the problem. Despite advances in systemic therapy, the prognosis for these patients has remained stubbornly poor, with median overall survival still measured in months.

The paucity of data directly resulting from this low incidence has created a critical knowledge gap. The current evidence base is fragmented, consisting predominantly of small, single-center case series. These studies are inherently limited by selection bias and lack the statistical power to draw definitive conclusions about the optimal sequencing and comparative effectiveness of various local treatment modalities.

This lack of high-level evidence has direct clinical consequences. Treatment decisions for patients with BM from GC and EC are frequently extrapolated from data on other solid tumors, such as lung or breast cancer, or are based on institutional preference and physician experience. This uncertainty underscores a profound unmet clinical need for large-scale, real-world data.

Investigators from a multi-institutional international consortium of cancer centers have therefore initiated a collaborative effort to create the largest and most comprehensive dataset of patients with gastric and esophageal cancer brain metastases described to date. Pooling clinical, pathological, molecular, and treatment-related data from patients diagnosed over the past quarter-century (2000-2025), this dataset will enable a robust analysis of modern local treatment patterns, long-term oncological outcomes, and prognostic factors - ultimately seeking to provide evidence-based insights to guide clinical decision-making for this rare and challenging patient population.

The primary objective of this study is to evaluate, in a large multi-institutional real-world cohort, the effectiveness of the following local treatment modalities for gastric and esophageal cancer brain metastases:

1. Neurosurgical resection (NRS):

   • En-bloc resection (EBR)
   • Total piecemeal resection (TPR)
   • Subtotal (SPR) or partial piecemeal resection (PPR)
   • Extent of resection unknown

2. Radiotherapy (RT):

   • Stereotactic radiosurgery (SRS)
   • Staged stereotactic radiosurgery (stSRS)
   • Stereotactic radiotherapy (hypofractionated) (SRT)
   • Whole-brain radiotherapy (WBRT)

3. Combined local treatment modalities:

   • NRS with postoperative WBRT
   • NRS with postoperative SRT/SRS
   • Preoperative SRS followed by NRS
   • Combination radiotherapy (SRS and WBRT, SRT and WBRT, SRS and SRT)

4. No local treatment

Data to be collected for each patient will include:

1. Demographics and Baseline Characteristics

   • Sex (male/female)
   • Age at primary cancer diagnosis (years, median with range)
   • Age at brain metastases (BM) diagnosis (years, median with range)

2. Primary Tumor Characteristics

   • TNM classification (8th edition: T, N, M categories with subcategories, including X when unknown)

   • Disease stage at initial diagnosis (I-II, III, IV, unknown)

   • History of primary tumor resection (yes/no/unknown)

   • Primary tumor location (Esophagus, Gastroesophageal Junction (GEJ), Stomach)

   • Histology (gastric adenocarcinoma, esophageal adenocarcinoma, esophageal squamous cell carcinoma, unknown)

   • Grade:

     • High-grade (poorly differentiated / G3)
     • Low-grade (well to moderately differentiated / G1-G2)
     • Unknown

   • Lauren classification (applicable to gastric adenocarcinoma): intestinal, diffuse, mixed, unclassified, unknown

3. Extracranial Disease Burden:

   • At initial cancer diagnosis:

     • Number of extracranial organ sites involved (1, 2, ≥ 3, unknown)
     • Location of extracranial metastases (liver, lung, peritoneum, bone, adrenal glands, soft tissues, distant lymph nodes, other; with specification of exclusive vs. combined involvement)

   • At the time of BM diagnosis:

     • Presence of extracranial metastases (yes/no/unknown)
     • Number of extracranial organ sites involved (1, 2, ≥ 3, unknown)
     • Location of extracranial metastases (liver, lung, peritoneum, bone, adrenal glands, soft tissues, distant lymph nodes, other; with specification of exclusive vs. combined involvement; including "brain-exclusive" disease)
     • Activity of extracranial disease (stable, progressing, brain-exclusive, unknown)

4. Intracranial Tumor Characteristics

   • Number of BM at diagnosis (median with range; categorized as solitary vs. ≥ 2)
   • Location relative to the tentorium (supratentorial, infratentorial, both)
   • For solitary BM: specific localization (frontal, temporal, parietal, occipital lobe; cerebellum; left/right hemisphere; unknown)
   • Cumulative intracranial tumor volume (CITV) (cm³, median with range; categorized into clinically relevant volume groups)
   • Largest intracranial tumor volume (LITV) (cm³, median with range; categorized into clinically relevant volume groups)
   • Maximum diameter of the largest intracranial tumor (cm, median with range; categorized into clinically relevant diameter groups)
   • Radiological features (present/absent/unknown): Perifocal edema, Mass effect, Dislocation syndrome, Ventricular compression (lateral, III, IV), Brainstem compression, Intratumoral hemorrhage, Intratumoral necrosis, Leptomeningeal disease (LMD), Occlusive hydrocephalus, Bone destruction, Cerebellar tonsil herniation

5. Clinical Presentation

   • Timing of BM diagnosis:

     - Synchronous (within 60 days of primary tumor diagnosis or as initial presentation)

     • Metachronous (> 60 days after primary tumor diagnosis)
     • Unknown

   • Neurological symptoms at BM diagnosis (symptomatic, asymptomatic, unknown)

   • Presence of neurological deficit at BM diagnosis (yes/no/unknown)

   • Neurological syndromes at BM diagnosis (global cerebral symptoms (headache, nausea, vomiting, dizziness), aphasic, dysarthric, pyramidal, sensory, extrapyramidal, cerebellar/vestibulo-ataxic, bulbar, pseudobulbar, paroxysmal, cognitive, brainstem, visual, meningeal, occlusive, behavioral, unknown)

   • Eastern Cooperative Oncology Group (ECOG) performance status at BM diagnosis (0-1, 2-3, unknown)

6. Molecular Profile (where available)

   • Primary tumor:

     - HER2 status (positive/negative/unknown)

     • CPS score (PD-L1 Combined Positive Score)
     • MSI status (Microsatellite Instability (MSI)/Microsatellite Stable (MSS)/unknown)
     • CLDN18.2 status (positive/negative/unknown)

   • Brain metastases:

     • HER2 status (positive/negative/unknown)
     • CPS score (PD-L1 Combined Positive Score)
     • MSI status (MSI-H/MSS/unknown)
     • CLDN18.2 status (positive/negative/unknown)

7. Treatment-Related Data

   • Systemic therapy prior to first local treatment (yes/no)

   • Lines of systemic therapy prior to BM diagnosis (number; categorized as 0, 1, 2, 3, 4, ≥ 5, unknown)

   • Systemic therapy after first local treatment (yes/no)

   • Lines of systemic therapy after first local treatment (number)

   • Number of local treatments per patient (median with range)

   • For radiotherapy:

     - Radiation therapy device (Leksell Gamma Knife, Accuray CyberKnife, Varian TrueBeam/Novalis Tx, non-stereotactic linac, other)

     - Biologically effective dose (BED) (Gy, median with range)

     • Equivalent dose (EQD2) (Gy, median with range)

8. Institutional and Period Data

   • Institution type (federal center, reference center, regional center)

   • Treatment period (by year of primary cancer diagnosis):

     • 2000-2010
     • 2011-2020
     • 2021-2025

Primary Endpoints:

1. Overall Survival (OS): Defined as the time from the date of brain metastasis (BM) diagnosis to the date of death from any cause or last follow-up (censored).

2. Time to Intracranial Progression (TTIP): Defined as the time from the date of initial gastric and esophageal cancer diagnosis to the date of first BM detection. Based on this interval, patients will be categorized into two groups:

   - Synchronous BM: BM diagnosed either prior to or within 2 months (≤ 60 days) of the primary tumor diagnosis.

   - Metachronous BM: BM diagnosed more than 2 months (> 60 days) after the primary tumor diagnosis.

3. Central Nervous System Progression-Free Survival (CNS-PFS): Defined as the time from the date of first local treatment for BM to the date of subsequent intracranial progression or last instrumental follow-up (censored). Subsequent intracranial progression includes:

   • Continued growth of the treated lesion (≤ 6 months thereof);
   • Local recurrence of the treated lesion (> 6 months thereof);
   • Development of new distant intracranial lesions.

Secondary Endpoints:

1. Overall Survival from Initial Diagnosis: Time from initial gastric and esophageal cancer diagnosis to death from any cause or last follow-up.
2. Cancer-Specific Survival (Competing Risks Analysis): Time from BM diagnosis to death from cancer progression (intracranial and/or extracranial), accounting for death from other causes as a competing event.
3. Cumulative Incidence of Death from Intracranial Progression: Death directly attributable to neurological complications resulting from progressive intracranial disease, analyzed using competing risks methodology with death from other causes as competing events.
4. Cumulative Incidence of Death from Extracranial Progression: Death attributable to systemic disease progression in the presence of controlled intracranial disease, analyzed using competing risks methodology.
5. Cumulative Incidence of Death from Other Causes: Death from causes unrelated to cancer progression, analyzed using competing risks methodology.
6. Cumulative Incidence of Repeat Local Interventions: The cumulative proportion of patients undergoing any additional local treatment (neurosurgery or radiotherapy) for intracranial disease progression following initial local therapy. This endpoint captures the total burden of repeat procedures required for recurrent or new intracranial lesions during follow-up. Death without a repeat intervention is treated as a competing event in the analysis.

Statistical Analysis. All statistical analyses will be performed using IBM SPSS Statistics (version 29.0) and STATA (version 17.0, StataCorp LLC). A two-sided p-value < 0.05 will be considered statistically significant.

• Descriptive Statistics. Categorical variables will be presented as absolute frequencies (n) and relative frequencies (%). Continuous variables will be assessed for normality. Normally distributed variables will be presented as mean with standard deviation (SD); non-normally distributed variables will be presented as median with interquartile range (IQR) or full range. The frequency of missing data will be reported for each variable. Where appropriate, multiple imputation using chained equations (MICE) will be considered to address missing data and minimize bias.
• Survival Analysis. Survival curves will be estimated using the Kaplan-Meier method. Median survival times with 95% confidence intervals (CI) will be reported. Comparison of survival curves between groups will be performed using the log-rank test and, where appropriate, the Breslow-Wilcoxon test.

For analyses of cause-specific death, competing risks methodology will be employed to estimate the cumulative incidence of death from intracranial progression, extracranial progression, and other causes, with death from competing causes treated as a competing event.

- Univariable and Multivariable Analysis. Univariable analysis will be performed to identify potential prognostic factors associated with survival outcomes. For categorical variables, the log-rank test will be used. For continuous variables, univariable Cox proportional hazards regression will be performed.

Variables with p < 0.10 on univariable analysis, as well as clinically relevant factors regardless of significance, will be entered into multivariable Cox proportional hazards regression models to identify independent prognostic factors. The proportional hazards assumption will be tested. Results will be presented as hazard ratios (HR) with 95% CI.

- Preplanned subgroup analyses will be performed based on key variables such as local treatment modality, molecular profile, primary tumor location, timing of brain metastases (synchronous vs. metachronous), and presence of extracranial disease.

Specific Analyses:

• Analysis of growth kinetics. In patients with available serial imaging, volumetric growth rate (VGR, cm³/day) and volume doubling time (VDT, days) will be calculated. These parameters will be correlated with clinical, molecular, and outcome variables.
• Radionecrosis analysis. In patients receiving radiotherapy, the cumulative incidence and time to radionecrosis (TTRN) will be assessed and correlated with treatment modality, dose, volume, and molecular profile.
• Temporal trends. Outcomes and treatment patterns will be compared across three time periods (2000-2010, 2011-2020, 2021-2025) to assess changes in practice and survival over time.
• Institutional analysis. Outcomes will be compared across institution types (federal, reference, regional centers) to explore potential disparities in care.
• GPA index validation. The prognostic accuracy of established GI-GPA (Gastrointestinal Graded Prognostic Assessment) index will be evaluated in this cohort and compared with any novel prognostic models developed from this data.

Handling of Missing Data. The proportion of missing data will be reported for all variables. Patterns of missingness will be explored. Where appropriate and assuming data are missing at random, multiple imputation using chained equations (MICE) will be performed to impute missing values for key variables in multivariable models.

The results of this large-scale, international study aim to provide high-level evidence to guide clinical decision-making and pave the way for personalized treatment approaches for this rare but challenging patient population.

Inclusion Criteria:

• Male and female patients aged 18 years or older;

• Histologically confirmed epithelial malignancy of the stomach (gastric adenocarcinoma), esophagus (esophageal adenocarcinoma), or gastroesophageal junction (GEJ adenocarcinoma). Patients with squamous cell carcinoma of the esophagus are also eligible;

• Radiologically ± histologically confirmed brain metastases, including:

  • Solid brain parenchymal metastases
  • Leptomeningeal disease (LMD)
  • Both solid metastases and LMD

Exclusion Criteria:

• Synchronous or metachronous multiple primary malignancies (MPM) involving sites other than the stomach, esophagus, or gastroesophageal junction;
• Primary tumor located outside the stomach, esophagus, or gastroesophageal junction;
• Histologically confirmed non-epithelial gastrointestinal malignancy (e.g., neuroendocrine tumors, sarcoma, gastrointestinal stromal tumor, lymphoma);
• Isolated spinal cord involvement without brain parenchymal metastases;
• Incomplete medical records precluding assessment of at least one primary endpoint (OS, TTIP, or CNS-PFS). Patients with available data for any primary endpoint are eligible, even if other clinical details are missing.
• Prior local treatment for brain metastases at an outside institution with no available records, precluding determination of the first local treatment modality

The study population consists of adult patients (≥ 18 years) with histologically confirmed adenocarcinoma of the stomach, adenocarcinoma or squamous cell carcinoma of the esophagus, or adenocarcinoma of the gastroesophageal junction, and radiologically ± histologically documented brain metastases. Patients referred after initial local treatment for brain metastases are included provided that complete documentation of the first treatment modality is available.

De-identified individual participant data collected for this study, including data dictionaries, as well as the study protocol and statistical analysis plan, will be available upon reasonable request from the corresponding author beginning 6 months after publication and ending 5 years following article publication. Data will be shared with researchers who provide a methodologically sound proposal for use approved by an independent review committee. Proposals should be directed to the corresponding author

Individual participant data and supporting information will become available 6 months after publication of the primary results and will remain available for 5 years following article publication

Access will be granted to qualified academic researchers who submit a methodologically sound research proposal for use approved by an independent review committee