Development and internal/external validation of a prediction model for weight loss following gastric cancer surgery: A multicenter retrospective study

Abstract

Purpose: To develop an individualized model for predicting the extent of unintentional weight loss following gastrectomy in patients with gastric cancer based on related risk factors and to externally validate this model using multicenter clinical data in Korea.

Materials and Methods: Among gastric cancer patients who underwent curative gastrectomy at 14 different gastric cancer centers, clinical data from patients with more than one weight measurement during the three-year follow-up period were retrospectively collected. Risk factors associated with weight loss in gastric cancer patients after gastrectomy were analyzed, and a predictive model was developed. Internal and external validation were performed.

Results: The data from 2,649 patients were divided into a derivation set (n=1,420 from Seoul National University Hospital) and validation set (n=1,229 from 13 different gastric cancers). Postoperative duration (six vs. 12, 24, or 36 months), sex (female vs. male), age, preoperative body mass index, type of surgery (pylorus-preserving vs. total, distal or proximal gastrectomy), and cancer stage (I vs. II or III) were included in the final prediction model. The model showed approximately 20% accuracy in predicting weight loss at each period: R2 at six, 12, 24 and 36 months after gastrectomy in internal validation=0.20, 0.21, 0.17, and 0.18, respectively, and in external validation=0.20, 0.22, 0.18, and 0.18, respectively. Calibration slopes of internal and external validation were 0.95 and 1.0, respectively.

Conclusion: Although predictive accuracy of the model did not reach an acceptable level, repeated external validation measurements showed high reliability. The model may serve as a basic reference in clinical practice.