| Objective:To investigate the connection of the hematological indicators and the pathological progression of NMIEBC and establish a model of hematological indicators combining with tumor features to predict the pathological progression of NMIBC,and visualize the model and assess clinical utility.Methods:(1)Retrospective analyzing the data of patients who received the first TURBT in our hospital and the postoperative pathology confirmed as NMIBC and relapsed within 5 years after surgery,including the preoperative hematological data of TURBT and the pathological data at the time of fust and second.(2)According to the pathological results of the patient at the time of recurrence,the patients were divided into two groups,one group is that the pathological result is still NMIBC at the time of recurrence,and the other group is the pathological progression to MEBC.(3)Univariate analysis explores hematologic indicators and tumor features related to pathological progression.(4)Multivariate Logistic regression analysis found the independent risk factors predicting pathological progression,and constructing tumor models and combining models respectively.(5)Comparing the performance of the two models in training and test data,and visualizing the model with better prediction performance.(6)Try to build a predictive model based on machine learning methods,which is better than the traditional modeling method.Results:There were 263 patients in the training data,including 210 males and 53 females,with an average age of 59 years(23-87 years).The progression group consisted of 62 patients and the progression-free group consisted of 201 patients.En the testing data,there were 41 males and 7 females with an average age of 60.2 years(37-78 years).A total of 11 patients had a pathological progression to MEBC at the time of relapse.and 37 patients had no pathological progression.After t-test or chi-square test,there was no significant difference in all the observed indicators(P<0.05)between the training data and testing data,which was comparable.In the univariate analysis of hematological indicators of patients in the two groups of training data,we found that the patients with pathological progression were more leukocytes and neutrophils than the patients without progression(6.1=2.0×109/L vs 5.6= 1.5×109/L.P=0.36 and 4.0=1.8×109/L vs 3.5=1.3×109/L.P=0.45).Compared with patients without progression of pathology,plasma fibrinogen was higher and the difference was statistically significant(3.4= 1.0g/L vs 3.0= 0.8 g/L,P=0.01).However.ALB/FIB in patients with progression was lower than patients without progression(12.9=3.5 vs 14.9=4.8.P=0.001).and the difference was statistically significant.Analysis of pathological data showed that the patients of multifocality.tumor size>3cm,high grade tumor and T1 with pathological progiession was significantly higher than that of patients with no progression(38.8%vs 21.1%.P=0.01 and 38.9%vs 21.1%.P=0.02 and 34.1%vs 22.0%.P=0.04 and 35.1%vs.20.4%.P=0.02).The results of multivariate Logistics regression analysis showed that ALB/FIB.tumor size,tumor number,tumor grade,and T stage were five independent risk factors for predicting pathological progiession.of which the OR of ALB/FEB was 0.892(95%CI:0.815-0.977.P=0.014).Taking each increase of ALB/FEB by 6.6 as an interval,the patients were divided into 5 groups,and with the increase of ALB/FEB.the probability of pathological progression decreased gradually,and there was a significant statistical difference between groups(P<0.001).Tumor size OR was 2.403(95%CI:1.041-5.549.P=0.04).The number of tumors OR was 2.212(95%CI:1.075-4.551.P=0.03).The tumor OR was 2.131(95%CI:1.107-4.105.P=0.02).The T stage OR is 2.466(1.242-4.893.P=0.01).When the presence of one independent factor(either of them)increases to two(any combination of two),the paobability of pathological progression increasing,to three independent factors(any three combinations),and the probability will increase further,and when it is increased to four,its predicting value is the largest.Logistic regression analysis based on tumor characteristies based on the prediction model AUC of 0.67(95%CI:0.59-0.75.P<0.001),the χ2 in the Hosmer-Lemeshow test was 4.23(P=0.24>0.05).the difference was not statistically significant indicating that the model fit was good.The predictive model AUC was 0.73(95%CI:0.66-0.80.P<0.001).that was constructed by hematological indicators and tumor features.The Hosmer-Lemeshow test evaluates the fit of this predictive model,and itsχ2 is 6.405,P=0.60>0.05.the difference is not statistically significant indicating that the model fit is also good.ROC curves and DCA curves were applying to evaluate the model.Both the training data(AUC=0.73.95%CI:0.66-0.80,P<0.001)and the test data(AUC=0.75.95%CI:0.59-0.84,P<0.001)showed higher differentiation than tumor models.Both the tumor model and the combining model had a significant portion of the DCA curve located on the None and ALL lines,both of which can benefit the patient population when predicting the probability threshold of the characteristics(10%-79%and 15%-71%.respectively).The benefit threshold probability interval for a single-union model is wider and the benefit is greater.This suggests that combining model is more clinically useful.The AUC of optimal model established by Xgboost algorithm by hyperparameter optimization is 0.81(95%CI:0.694-0.870).the prediction accuracy is 73%(95%CI:0.690-0.844).the sensitivity is 0.76.and the specificity is 0.72.When applied to the testing data,the AUC was 0.778(95%CI:0.694-0.825).the prediction accuracy was 0.69(95%CI:0.690-0.844).the sensitivity was 0.64.and the specificity was 0.70.Comparing the Nomogram model,we could see that the model performance has been greatly improved.Conclusion:1.ALB/FIB is an independent predictor of postoperative pathological progression of NMIBC.2.The progression prediction model of hematological indicators combined with tumor feature constrution has better differentiation and clinical practicability than tumor model.3.Xgboost may be able to further improve the accuracy of the model. |
PDF Full Text Request