RESUMO
<p><b>OBJECTIVE</b>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.</p><p><b>METHODS</b>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.</p><p><b>RESULTS</b>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 palliative surgery. 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.</p><p><b>CONCLUSION</b>The modified P-POSSUM and the modified Cr-POSSUM model provide an accurate prediction of inpatient mortality in Chinese colorectal cancer patients.</p>