KEY POINTS
- This retrospective single-center French study included 222 patients with CNS metastases from gastrointestinal malignancies diagnosed between January 2000 and December 2022. Primary cancers included 156 colorectal tumors, 48 esophageal tumors, 16 gastric tumors, and two small-bowel tumors. Four prognostic tools were evaluated: recursive partitioning analysis (RPA), the general graded prognostic assessment (GPA), gastrointestinal-specific GPA (GI-GPA), and the Pietrantonio nomogram.
- Brain or leptomeningeal involvement generally occurred in patients with advanced systemic disease. CNS metastases were symptomatic in 95%, multiple brain metastases were present in approximately 50%, and 88.8% had extracranial metastatic disease. Median interval from primary cancer diagnosis to CNS involvement was 36.7 months for colorectal cancer versus 12.6 months for upper gastrointestinal tumors. Approximately 74% received local treatment, with whole-brain RT used in a substantial proportion of irradiated patients.
- Median overall survival following CNS metastasis diagnosis was only 3.7 months. Survival was 3.9 months for colorectal cancer versus 2.9 months for upper gastrointestinal primaries, without a statistically significant difference (p=0.9). Performance status was strongly associated with survival: median OS ranged from 7.75 months for ECOG 0 to 0.50 months for ECOG 4 (p<0.001). Better performance status remained independently prognostic in both colorectal and upper gastrointestinal subgroups.
- The four prognostic tools frequently differed from observed survival by more than three months. Such discrepancies occurred in 54.8% of patients with GI-GPA, 52.0% with the Pietrantonio nomogram, 42.7% with general GPA, and 31.5% with RPA. The median survival estimates assigned by these tools were 9.0, 6.0, 6.5, and 4.2 months, respectively, compared with the observed 3.7 months. The nonspecific RPA showed the lowest discrepancy frequency under the study's chosen metric.
- Additional prognostic factors varied by primary cancer site. In colorectal cancer, BRAF mutation was associated with worse OS in multivariable analysis (p=0.001), although molecular data were incomplete and many patients were treated before modern targeted therapy. In upper gastrointestinal tumors, lymph node metastases were associated with worse OS (p=0.01), while CNS metastasis diagnosis after 2018 was associated with longer survival (p=0.04). These subgroup findings are exploratory because of small numbers and treatment-era differences.
- The results do not validate RPA as a sufficiently accurate individual survival prediction tool. Even RPA differed from actual survival by more than three months in nearly one-third of patients. The study included two decades of systemic therapy evolution, a small upper gastrointestinal subgroup, incomplete biomarker information, and heterogeneous surgery and radiotherapy approaches. In addition, comparing predicted group-median survival with individual observed survival is not equivalent to a full calibration and discrimination analysis for contemporary prognostic modeling.
CLINICAL TAKEAWAY
Survival after CNS metastases from gastrointestinal malignancies was generally short, and established prognostic scores often produced clinically meaningful discrepancies from observed outcomes. Performance status remained the most consistent prognostic factor, while the historical RPA score performed better than newer gastrointestinal-specific tools under this study's discrepancy measure. Decisions regarding stereotactic treatment, surgery, or supportive care should not rely on prognostic scores alone, particularly without contemporary external validation.