Construction And Clinical Application Of Colorectal Cancer Auxiliary Diagnosis Prediction Model

Objectives: Based on the laboratory indicators commonly used in clinical practice,the risk factors of colorectal cancer are analyzed and screened,the CRC-assisted diagnosis prediction model is constructed,the nomogram is drawn and preliminary verified,and the high-risk population is effectively identified to guide clinical work.Methods: To retrospectively analyze and analyze the clinical data of 573 patients with colorectal disease admitted to the General Surgery Department of Shaanxi Provincial People’s Hospital from July 2015 to July 2018,mainly including basic clinical information,laboratory indicators and histopathological diagnosis results.573 patients were used as the training set,and were divided into 317 cases in the CRC group and 256 cases in the non-CRC group according to the results of colorectoscopy and histopathological diagnosis.Through single factor analysis and multi-factor logistic regression analysis,the possible risk factors were screened out,a CRC-assisted diagnosis prediction model was constructed,and a nomogram was drawn.The predictive power of the CRC-assisted diagnosis prediction model is evaluated by the model differentiation degree and model calibration degree.The model is internally verified by the non-parametric Bootstrap method.The clinical data of 255 patients with colorectal disease who were treated in the General Surgery Department of Shaanxi Provincial People’s Hospital from August 2018 to December 2019 were collected.The discriminating ability of the model.Results:1.A total of 828 patients were included in this study after inclusion criteria and exclusion criteria.The results of multivariate logistic regression analysis suggested that age,family history of colorectal cancer,living environment,NLR,PLR,FIB,CEA,and CA19-9 were predictive of CRC.Independent risk factors.Based on these risk factors,aCRC-assisted diagnosis prediction model is constructed,and further visualization is achieved by drawing a nomogram.2.The ROC curve analysis indicates that the best diagnostic threshold for NLR is2.40,sensitivity is 79.5%,specificity is 72.7%,and the area under the curve is 0.804;the best diagnostic threshold for PLR is 121.10,sensitivity is 72.2%,and specificity is60.3 %,The area under the curve is 0.707.3.The analysis of this study suggests that the area under the ROC curve of the CRC-assisted diagnosis prediction model is 0.908,and the calibration curve is close to the ideal curve.The non-parametric Bootstrap method is used to internally verify the model,and the calculated C statistic is 0.9026.In external verification,the area under the ROC curve of the verification set is 0.9149,which is basically consistent with the training set,indicating that the prediction model is well distinguished.The calibration curve is close to the ideal curve,indicating that the prediction model is well calibrated.Conclusion:1.In this study,the results of multivariate logistic regression analysis showed that:age,family history of colorectal cancer,living environment,NLR,PLR,FIB,CEA,CA19-9 were independent risk factors for predicting CRC;ROC curve analysis showed:NLR,PLR High value in the diagnosis of CRC.2.A CRC-assisted diagnosis prediction model based on clinical commonly used laboratory indicators was constructed,and a nomogram was drawn to further achieve visualization.3.The CRC-assisted diagnostic prediction model is verified and evaluated as having good predictive value,and it is worth further expanding the sample size for research.