Positive Prediction Model Of Resection Margin After Radical Prostatectomy Based On Bayesian Network

Background: Prostate cancer is the most common urological malignancy in elderly men and one of the top five causes of cancer death in the world.Radical prostatectomy is one of the main and effective methods for the treatment of prostate cancer,but there is still the possibility of positive resection margin,which affects the prognosis and treatment strategy.Therefore,if we can identify the high risk factors and establish a predictive model,it is the key to reducing the surgical margin positive rate to providing the most reasonable treatment plan.Previous studies focused on the relationship between tumor invasive factors such as PSA derivatives,Gleason score in puncture and marginal positivity.Few studies discussed the characteristics of MRI and the predictive value of the number of MRI positive pain versus marginal positivity,and most studies focused on the analysis of high risk factors,and less on building predictive models based on high risk factors.Based on the current situation,we attempt to explore the MRI imaging characteristics and the relationship between the number of puncture positive pin and margin positive.We invited a radiologist who had been reading MRI for more than 10 years to reread the MRI to find out the imaging characteristics that can describe the tumor load and evaluate the influence of MRI factors on margin positivity.At the same time,we use Bayesian network to establish the positive prediction model of tangent edge,and compare it with the nomogram model to evaluate the practical value of Bayesian prediction model.Objective: To analyze the independent risk factors of marginal positivity after radical prostatectomy,and to evaluate the predictive value of Bayesian network based predictive model.Methods: The clinical data of 238 patients who underwent laparoscopic radical prostatectomy(LRP)or robot-assisted radical laparoscopic prostatectomy(RALP)in Qingdao University Hospital from June 2018 to May 2021 were retrospectively analyzed.The general clinical data,PSA-derived index,puncture factors and MRI imaging characteristics of the two groups were included as predictive variables and were analyzed by single factor and multifactor analysis.Then the tree augmented naive(TAN)and naive Bayes model is established based on 15 predictive variables by using Bayesia Lab software,and the Bayesia Lab verification function is used to conduct a prior probability statistical analysis of predictive variables,and a posteriori analysis is made with the tangential margin positive as the target variable and the residual variable as the attribute variable.The importance of the polymorphic Birbnaum is analyzed and calculated according to the posteriori analysis results,and the importance order of the attribute variables is given.Finally,we draw the Bayesian model and the Nemesis model’s ROC and calculate the area under the ROC curve to evaluate the model’s advantages and disadvantages.Results: Among the 238 patients included in the study,103 had positive margins after surgery,with a positive rate of 43.3%.Among them,53 cases were positive for tip margin(51.5%),12 cases were positive for base margin(11.7%)and ≥ 2 cases were 38 cases were positive for base margin(36.9%).There were 35 cases of biochemical recurrence(35/103)with positive margin of resection in one year,and 8 cases(5.9%)with negative margin of resection in one year.Univariate analysis: TPSA(P = 0.08),PSAD(P = 0.02);Gleason score of puncture tissue(P = 0.002),puncture positive needle ratio(P < 0.001);preoperative staging(P < 0.001),abnormal signal location(P = 0.002),abnormal signal location(P = 0.009)in MRI were statistically significant positive to incision margin.Multivariate analysis: clinical stage(OR = 3.645,95% CI 11.401 ～ 9.482,p = 0.006),positive pin number ratio(OR = 4.970,95% CI 2.448 ～ 10.090,P < 0.001)were independent predictors of margin positivity.The AUC of the nomogram model based on independent risk factors was 73.80%,the AUC of the naive Bayesian model based on 15 variables was 82.71%,and the AUC of the TAN Bayesian model was 80.80%.Clinical staging,puncture positive pin ratio,PSAD,abnormal signal position,Gleason score,TPSA,abnormal signal position,PI-RADS score were in the first important interval,and PSAD was in the second important interval.Conclusion: 1.Clinical stage and number of positive needles were independent predictors of margin positivity.2.mp MRI features can predict the positive margins of incisions.It is helpful to evaluate the probability of positive margins by improving MRI and analyzing its features before operation.3.In the classification of Bayesian importance,clinical staging and positive pin ratio are in the first importance interval,which are the same as the results of independent predictors in multivariate analysis;PSAD is in the second importance interval,and abnormal signal position,Gleason score,TPSA,abnormal signal position and PI-RADS score are in the third importance interval.This was the same as a statistically significant factor in a single factor.The results show that the importance grading based on Bayesian network can evaluate whether the predictors have statistical significance.4.TAN Bayesian model showed that PSAD was correlated with F/TPSA,TPSA and positive pin ratio,positive pin ratio was correlated with clinical staging,maximum transverse diameter was correlated with abnormal signal location,abnormal signal location and PI-RADS score.5.The positive predictive model of incision margin based on Bayesian network is superior to the nomogram predictive model,and has high accuracy.It can be used as a method for predicting positive incision margin after radical prostatectomy.