结直肠癌患者住院期间病死率新评分系统的建立和评价

Analysis of in-hospital morbidity and mortality of colorectal cancer and accuracy of POSSUM models for mortality risk

目的建立预测术后病死率的改良P-POSSUM和改良Cr-POSSUM评分系统并与POSSUM比较，评价其对结直肠癌患者住院期间病死率的预测能力。方法调查北京大学第三医院1992-2005年间结直肠癌经手术切除的903例患者资料，按70：30把本组病例分成建立模型样本和预测模型样本，用Logistic回归分析建立改良P-POSSUM和改良Cr-POSSUM，用ROC曲线分析判断改良P-POSSUM和改良Cr-POSSUM评分的判别能力，用Hosmer-Lemeshow检验判断评分的拟合优度，用不同危险因素群的O：E值判断评分的预测能力。结果本组患者住院期间的病死率为1．0％（9／903）。POSSUM、P-POSSUM和Cr-POSSUM评分预测的病死率明显高于实际病死率，O：E值分别为0．18、0．35和0．20。改良P-POSSUM除在急诊手术和姑息手术中判别能力较差外，在其他手术中都具有较好的判别能力，在所有手术中预测的死亡率与实际死亡率接近（O：E值为0．91）；改良Cr-POSSUM除在姑息手术中有很好的判别能力外，在评价模型样本和急诊手术预测的死亡率高于实际死亡率，但仍在实际死亡率95％的可信区间内（0：E值为0．78）。两者的预测能力都好于POSSUM。结论POSSUM、P-POSSUM和Cr-POSSUM在中国结直肠癌手术中预测的病死率高于实际病死率。改良P-POSSUM和改良Cr-POSSUM可较准确地预测中国结直肠癌患者手术住院期间病死率。

Objective To develop the modified P-POSSUM equation and the modified Cr- POSSUM equation and compare their performances with POSSUM in forecasting in-hospital morbidity and mortality of colorectal cancer. Methods Data of 903 patients undergone operation of colon and rectal cancers from 1992 to 2005 in our department were enrolled in this study. ROC curve was applied to judge the differentiation ability of each score. Model goodness-or-fit was tested by the Hosmer-Lemeshow statistic and subgroup analysis was performed by the ratio of observed to expected deaths （O：E ratio）. A 70：30 percent split-sample validation technique was adopted for model development and testing. Stepwise logistic regression was used to develop the modified P-POSSUM and the modified Cr-POSSUM. Their performance in validating sample, colonic cancer sample, rectal cancer sample, elective surgery sample, emergency surgery sample, curative surgery sample and palliative surgery sample was tested by ROC curve, Hosmer-Lemeshow statistic and O：E ratio. Results The modified P-POSSUM showed good discrimination in all samples except the emergency surgery and palliative surgery. The predicted mortality of modified P-POSSUM was very close to the observed mortality. However, the modified Cr-POSSUM showed good discrimination in all samples except the palliativesurgery. The predicted mortality was higher than the observed mortality, but still within the 95% confidence interval （CI） of the observed mortality. Both the modified models offered better accuracy than the P-POSSUM. Conclusion The modified P-POSSUM and the modified Cr-POSSUM model provide an accurate prediction of inpatient mortality in Chinese colorectal cancer patients.