Objective: To explore the value of phase angle (PA) in constructing a predictive model of nutrition evaluation for tumor patients. Methods: A retrospective analysis was performed on 1 129 patients with malignant tumors hospitalized in the Cancer Center of Changzhi People's Hospital from June 2020 to February 2021. PA values of six parts of the body were measured by the body composition analyzer, including: left arm (LA), right arm (RA), left leg (LL), right leg (RL), the trunk (TR), and the whole body (WB). Patients' body mass index (BMI) was calculated and patient-generated subjective global assessment (PG-SGA) was assessed. The differences of PA values of six parts were compared and their correlations with BMI and PG-SGA in combination with age, gender and tumor disease types were analyzed, binary classification regression on BMI and PG-SGA was performed, and the functions of the best prediction model was fitted. Decision tree, random forest, Akaike information criterion in a Stepwise Algorithm (stepAIC) and generalized likelihood ratio test were used to select appropriate variables, and the logit logistic regression model was used to fit the data. Results: Comparing the PA values of six parts in pairs, it was found that the PA values of LA and RA, LL and RL, and TR and WB were linearly correlated and the coefficient was close to 1 (P<0.001). Binary classification regression was performed for BMI and PG-SGA, respectively. In order to make the data have clinical significance, 18.5 kg/m(2) was used as the classification point for BMI, 4 and 9 were used as the classification points for PG-SGA score, and the models of A, B and C were obtained. Suitable variables including PA-LA, PA-TR and tumor disease types were used as variables to fit BMI classification; BMI, PA-LA and age were used as variables to fit the PG-SGA model with 9 as the classification point. PA-LA, PA-TR, BMI, age and tumor disease types were used as variables to fit the PG-SGA model with 4 as the classification point. In this study, the predicted values of models A, B and C obtained by R-studio were imported into SPSS 26.0 software, and the cut-off values of classification were obtained by the receiver operating characteristic (ROC) curve. The ROC analytic results showed that the best cut-off values of Model A, B and C were 0.155, 0.793 and 0.295. Model A recommended when the probability is >0.155, a patient's nutritiond tatus should be classified as BMI < 18.5 kg/m(2) group. Model B recommended that PG-SGA<9 group be classified as the probability is >0.793. Model C recommended that PG-SGA < 4 group should be classified when probability is >0.295. Conclusions: The PG-SGA classification prediction model is simple to operate, and the nutritional status of patients can be roughly divided into three groups: normal or suspected malnutrition group (PG-SGA<4), moderate malnutrition group (4≤PG-SGA<9), and severe malnutrition group (PG-SGA≥9). This model can more efficiently predict the nutritional status of cancer patients, greatly simplify the nutritional assessment process, and better guide the standardized treatment of clinical malnutrition.
目的： 探讨相位角(PA)在构建肿瘤患者营养评估预测模型中的价值。 方法： 回顾性分析2020年6月至2021年2月于长治市人民医院住院的1 129例恶性肿瘤患者的临床病理资料。采用人体成分分析仪测量患者身体6个部位的PA值，6个部位分别为左上肢(LA)、右上肢(RA)、左下肢(LL)、右下肢(RL)、躯干(TR)、全身(WB)。计算患者的体质指数(BMI)，行患者主观整体评估(PG-SGA)。比较身体6个部位PA值的差异，并结合年龄、性别以及肿瘤类型分析PA与BMI和PG-SGA的相关性。分别对BMI及PG-SGA进行二分类，依次使用决策树、随机森林、stepAkaike信息准则以及广义似然比检验选择合适变量，并用logit逻辑回归模型对数据进行拟合，采用受试者工作特征(ROC)曲线以及模型预测准确率判断logit逻辑回归模型的效能。 结果： LA和RA、LL和RL、TR和WB的PA值成线性相关且系数约为1(P<0.001)。BMI以18.5 kg/m(2)为分界点，PG-SGA评分以4和9分为分界点，得到模型A、B、C。使用PA-LA、PA-TR及肿瘤疾病类型为变量拟合BMI分类模型(模型A)，使用BMI、PA-LA以及年龄为变量拟合PG-SGA以9分为分界点的模型(模型B)，使用PA-LA、PA-TR、BMI、年龄与肿瘤疾病类型为变量拟合PG-SGA以4分为分界点的模型(模型C)。ROC曲线显示，模型A、B、C的最佳临界值分别为0.155、0.793和0.295。模型A推荐当概率>0.155时归为BMI<18.5 kg/m(2)组，模型B推荐当概率>0.793时归为PG-SGA<9分组，模型C推荐当概率>0.295时归为PG-SGA<4分组。 结论： PG-SGA分组预测模型操作简单，可以将患者的营养状况大致划分为3个区间，分别为正常或可疑营养不良组(PG-SGA<4分)、中度营养不良组(4分≤PG-SGA<9分)和重度营养不良组(PG-SGA≥9分)，可以更加高效地预测肿瘤患者的营养状况，简化营养评估流程，更好地指导临床营养规范化治疗。.
Keywords: BMI; Malignant tumor; PG-SGA; Phase angle; Tumor nutrition evaluation.