PET-CT-based host metabolic (PETMet) features are associated with pathologic response in gastroesophageal...

¹⁸F-FDG PET-CT-based host metabolic (PETMet) profiling of non-tumor tissue is a novel approach to incorporate the patient-specific response to cancer into clinical algorithms.

A prospectively maintained institutional database of gastroesophageal cancer patients was queried for pretreatment PET-CTs, demographics, and clinicopathologic variables. ¹⁸F-FDG PET avidity was measured in 9 non-tumor tissue types (liver, spleen, 4 muscles, 3 fat locations). Logistic and Cox regression were used to model pathologic response (PR) and overall survival (OS) respectively. Classification and regression tree (CART) and random forest modeling were employed to create decision trees and identify PETMet features associated with outcome.

Two-hundred and one patients with distal gastroesophageal (48 %) or gastric (52 %) adenocarcinoma were included. PET-CT-derived scores were independently associated with PR after adjusting for clinical variables. CART and Random Forest methods identified critical split points of non-tumor tissue ¹⁸F-FDG avidity that can classify patients and predict PR. PET-CT risk groups created from decision trees predicted PR significantly better than the clinical model (p < 0.001). Specifically, an elevated erector spinae-to-gluteal fat ¹⁸F-FDG avidity ratio (≥2.7) combined with low ¹⁸F-FDG avidity in the spleen (<2.9) and rectus femoris (<0.52) predict PR. No advantage of PET-CT risk groups was seen for predicting OS (p = 0.155).

Pretreatment host PETMet features may be useful for predicting PR after neoadjuvant therapy in gastroesophageal cancer. Unsupervised decision trees indicate that low ¹⁸F-FDG avidity in visceral fat, subcutaneous fat, and muscle result in the most favorable PR, suggesting that systemic hypermetabolism adversely impacts prognosis.