BACKGROUND & AIMS: Malnutrition in cancer patients is a common but under-diagnosed condition that has negative effects on clinical outcomes. The development of an easy and reliable malnutrition assessment tool is thus critical for identification and nutritional support. We aimed to develop a phase angle (PA)-based prediction model of malnutrition and evaluate it in patient prognosis.

METHODS

A retrospective cohort of data consisting of demographic, clinical parameter and PA test from 702 adult hospitalized cancer patients between June 2020 to February 2021 was analysed. PAs for 6 body sites were measured by a body composition analyser. Patient-generated subjective global assessment (PG-SGA) scale was used as the diagnostic standard of nutritional status (PG-SGA ≥ 4 points defined as malnutrition). Decision tree, mean decrease accuracy of random forest, stepAIC strategy and test of generalized likelihood ratio were employed to select important variables and develop models for predicting PG-SGA binary classification (PG-SGA < 4 or ≥ 4 as a split). Survival curves were plotted by using the Kaplan-Meier method.

RESULTS

In all, 490 (69.8%) patients were malnourished according to their actual PG-SGA scores. Except for age, tumor type and body mass index (BMI), PA of the left arm was found to influence malnutrition classification and incorporated in the final predictive model. The model achieved good performance with an AUC of 0.813, 75.9% sensitivity and 73.3% specificity. The actual and predicted survival curves were almost overlapped.

CONCLUSION

This study provides a simple nutritional assessment tool which may be used to facilitate oncology physicians to identify cancer patients at nutritional risk and potentially implement nutritional support.

CLINICAL TRIAL NO

ChiCTR2100047858.