| Background and objectives:Colorectal cancer(CRC)is a malignant tumor with very high morbidity and mortality.With the insidious onset,it typically would have progressed into an advanced stage when clinical symptoms develop,which makes its early screening crucial.At present,the screening methods such as colonoscopy,fecal immunochemical test,and fecal DNA testing are widely recommended,yet each of them has its respective limitations in clinical application.With the development of precision medicine,clinical prediction models have been playing an increasingly important role in disease prediction and prognosis,etc.Studies in and out of China have suggested that prediction models based on clinical laboratory parameters(tumor markers,intestinal flora,genetic testing,etc.)are generally effective and stable,yet these parameters are relatively costly and not included in the routine laboratory tests for communities with subclinical symptoms and people receiving physical examination.To achieve effective early screening while reducing the cost to ease the economic burden on people being tested,our research group had previously established a prediction model based on the correlation between routine blood tests and CRC,which had demonstrated good sensitivity and accuracy.For further exploration,this study intends to construct a prediction model based on the previous research by further combining the potentially important parameters in biochemical examinations for CRC prediction;and verify the prediction effect of the model,in an effort to provide clinicians with a new approach to an early screening of CRC.Methods:This study was designed as a case-control study,The patients who were primarily diagnosed as CRC by colonoscopy and histopathology in The First Affiliated Hospital of Dalian Medical University from January 1,2019 to December 31,2021 were included in patient group.Persons for whom CRC was excluded by colonoscopy and histopathology in Physical Examination Center were included in control group.Routine blood test and biochemical parameters(WBC,Hb,FPG,ALT,AST,ALB,GGT,DBIL,Scr and SUA)were compared and analyzed between the two groups.All samples were randomly divided into training set and validation set according to 7:3.Multivariate analysis was performed by binary Logistic regression in training set.Main relevant factors of CRC were screened and the prediction models were constructed according to the test items and in combination with all parameters.Nomograms were created using the"rms"and"foreign"packages of R 4.2.3 software to visualize the model,and Bootstrap resampling was used to validate the calibration of the model.The predictive effects of each model were compared and verified in the validation set.The Receiver Operating Characteristic(ROC)curve and calibration curves were drawn.The area under ROC curve(AUC)were plotted to evaluate these parameters,and the sensitivity,specificity of each model were calculated.Results:(1)A total of 867 subjects range from 50 to 75 years were included in this study,with a mean age of 60.26±6.43 years.The mean age of patient group(62.01±6.62years)and control group(58.72±5.85 years)was statistically different(t=-7.710,p<0.001).Gender distribution was not significantly different(χ~2=1.640,p=0.200).(2)Comparison of routine blood test and biochemical parameters between CRC group and control group:Hb,ALT,AST,ALB,DBIL,Scr and SUA in CRC group were lower than those in control group,while WBC,FPG and GGT in CRC group were higher than those in control group,all of the differences were significant,p<0.05,except for FPG(p=0.500),GGT(p=0.749)and Scr(p=0.542).(3)For sample data in training set,multivariate analysis of gender and age with routine blood test(WBC and Hb),glucose(FPG),liver function(ALT,AST,ALB,GGT and DBIL),renal function(Scr and SUA)and all 10 routine blood test and biochemical parameters accordingly,and the binary Logistic regression models were also constructed.Finally,the models that routine blood test model,liver function model,renal function model and the total model were constructed successfully,and Nomograms were plotted for each model for visualization.The total model had the best predictive effect(AUC 0.932,95%CI:0.911-0.952),followed by liver function model(AUC 0.914,95%CI:0.891-0.937),routine blood test model(AUC 0.858,95%CI:0.828-0.889)and renal function model(AUC 0.739,95%CI:0.700-0.778).The predictive power of each model was higher than that of each single parameters.ALB had the best predictive power(AUC 0.908,95%CI:0.884-0.932),followed by Hb(AUC 0.766,95%CI:0.728-0.803).The calibration curves of each model showed that the predictive values were generally consistent with the actual values,indicating high accuracy.(4)All models were validated in validation set.The results were compared with that of training set.The sensitivity and specificity of the total model,liver function model,routine blood test model in the training set and validation set were relatively high,and the total model was the best.The renal function model showed low sensitivity in the validation set.Conclusion:The routine blood test and biochemical parameters are important in clinical prediction of colorectal cancer.The prediction models of colorectal cancer are successfully constructed,including the total model,liver function model,routine blood test model and renal function model.Except for the low sensitivity of the renal function model in the validation set,all of the total model,liver function model and routine blood test model show good discrimination and calibration,and have been validated with good prediction.The predictive power of the models constructed by combining routine blood test and biochemical parameters were higher than that of each single indicator,and the total model had the best predictive power.Among the single parameters,ALB and Hb also show good discrimination and calibration,especially ALB shows a high predictive power similar to the total model,which is of great significance in the clinical prediction of CRC. |
PDF Full Text Request