Trajectory of early postoperative nutritional status and its predictors in patients with gastrointestinal cancer: a...

To explore the trajectory categories of early postoperative nutritional status in patients with gastrointestinal cancer and to analyze their predictive factors using both multiple logistic regression and decision tree models, providing a basis for implementing precise nutritional interventions.

A longitudinal study design was used, and 351 patients who underwent radical surgery for gastrointestinal cancer were conveniently selected as the study subjects. Nutritional status was assessed using the Patient-Generated Subjective Global Assessment (PG-SGA) at four time points: preoperatively (T1), 1 week postoperatively (T2), 1 month postoperatively (T3), and 3 months postoperatively (T4). Latent class growth modeling (LCGM) was employed to identify different nutritional status trajectory categories. Multiple logistic regression and CHAID decision tree models were used to analyze the predictive effects of demographic characteristics, disease characteristics, sense of coherence, and social support on trajectory categories.

Three distinct nutritional status trajectory categories were successfully identified: the "Moderate Malnutrition-Gradual Recovery" group (40.5%), the "Persistent Severe Malnutrition" group (11.7%), and the "Transient Moderate Malnutrition-Rapid Recovery" group (47.9%). Both multiple logistic regression and decision tree models confirmed that cancer type (esophageal cancer), low handgrip strength, and low level of sense of coherence were core predictive factors affecting trajectory categories. Additionally, logistic regression identified male and low level of social support as independent risk factors, while the decision tree model further revealed key pathways for identifying specific high-risk subgroups based on factors such as chemotherapy history, neoadjuvant therapy, age, and surgical approach.

Early postoperative nutritional status trajectories in patients with gastrointestinal cancer show heterogeneity. Integrating the two predictive models can create a more comprehensive risk assessment framework, allowing preliminary screening using factors identified by logistic regression and targeted preventative management for key high-risk groups based on pathways highlighted by the decision tree model.