目的  通过与NRS 2002 和营养评定工具PG-SGA 的比较，探讨PNI 在胃肠肿瘤外科患者中的应用价值。方法  前瞻性应用营养筛评工具，非定点连续性方法进行入院营养风险筛查和评定及PNI 测定，以NRS 2002 筛查结果为诊断营 养风险的标准，绘制 PG-SGA 和 PNI 受试者工作特征曲线 ROC，分析各工具及诊断切点的灵敏度、特异度、阳性预测值、 阴性预测值、约登指数以及曲线下面积，了解PG-SGA 和 PNI 对营养风险预测评估的最佳截点；并比较区分组患者的营养 相关指标及围术期并发症发生率差异。Kappa 检验比较不同工具间诊断的一致性。结果  PG-SGA 的 ROC 曲线下面积 0.936 （95%CI=0.893~0.979），约登指数最大0.778，临界值为7.5，该临界值诊断营养不良的灵敏度84.6%，特异度93.2%，阳 性预测值为97.3%，阴性预测值为52.8%。与NRS 2002 的一致性检验P=0.000。PNI 最大值75.6，最小值29.35。PNI 的 ROC 曲线下面积为0.584（95%CI=0.480~0.688），约登指数最大为0.196，PNI 的临界值为53.8，以此切点诊断营养不良 的灵敏度为89.7%，特异度为29.9%，阳性预测值为63.2%，阴性预测值为13.3%。一致性检验P=0.015。以PNI=53.8 作 为临界值，将患者分为两组间的体质量、身高、ALb 及 TLC 差异具有统计学意义（P ＜ 0.05）。PG-SGA 临界值（取整数 ≥ 8）分组间患者的 BMI、TLC 具有统计学差异（P ＜ 0.05）。PG-SGA 与 PNI 两种判定方法的一致性 P=0.000。三种工具 区分两组间的总体并发症发生率未见统计学差异（P ＞ 0.05）。结论  择期手术胃肠肿瘤外科患者PNI=53.8 可作为营养不 良风险的诊断临界值。未见 NRS 2002、PG-SGA 和 PNI 三种工具区分的组间术后总体并发症发生率的明显差异。

 Objective  To explore the application value of prognostic nutrition index (PNI) in gastrointestinal cancer patients by comparing with nutritional screening tool NRS 2002 and nutrition assessment tool PG-SG. Methods  A prospective application of nutritional screening or assessment tools PNI, NRS 2002 and PG-SGA, into a group of unconsecutive cases of gastrointestinal cancer patients were carried. To draw the receiver operating characteristic (ROC) curve of PG-SGA and PNI according to NRS 2002 results, analyzing the difference between tools involving sensitivity and specificity, positive predictive value, negative predictive value, Yueden index and the area under the curve. To compare the nutrition related indicators between the two groups divided by the cutoff point. The Kappa test was used to compare the diagnostic consistency between the different diagnosing tools. Results  The area under the PG-SGA curve is 0.936 (95%CI=0.893~0.979), the Yueden index is 0.778, the cut-off value is 7.5, the sensitivity is 84.6%, the specificity is 93.2%, the positive predictive value is 97.3%, and the negative predictive value is 52,8%. Consistency was checked P=0.000. Maximum value of PNI was 75.6, and minimum value 29.35. ROC curve showed that area under the PNI curve was 0.584 (95%CI=0.480~0.688), the Yueden index is 0.196, the cut-off value of PNI is 53.8, the diagnostic sensitivity was 89.7% by this cutoff point , specificity was 29.9%, the positive predictive value was 63.2% and negative predictive value was 13.3%. The consistency test with NRS 2002 P=0.015. The patients were divided into two groups (by cut-off value PNI=53.8), there were statistically significant difference of weight, height, ALb and TLC between the two groups (P<0.05). While BMI and TLC had significant difference between the two groups devided by PG-SGA≥8 (P < 0.05). The consistency between the two methods of PG-SGA and PNI is P=0.000. There was no statistically significant difference about the total incidence of complications between the two groups divided by three evaluation tools (P>0.05). Conclusions  For patients with surgical gastrointestinal tumor, PNI=53.8 can serve as a cut-off diagnostic threshold for nutritional risk with more significant difference in related nutritional indexes. There was no significant difference in the total incidence of postoperative complications between the two groups districted by the three tools respectively.