Applied Mathematical Models For Predicting The Dosage Of Heparin In Continuous Renal Replacement Therapy

Objective: Continuous renal replacement therapy(CRRT)is a new blood purification technology developed with hemodialysis.At present,different recommendationsof heparin dose in CRRT are not the same and the treatment projectis still based on the experience of clinicians,so the dose of heparin and individualized programs in CRRT is worth exploring.Methods: 543 patients were selected from January 2014 to January 2017 who received continuous veno-venous hemofiltration(CVVH)by using heparin as anticoagulant in department of Nephrology in the Second Affiliated Hospital of Dalian Medical University.Exclusion criteria:(1)Presence of hemoptysis,gastrointestinal bleeding,cerebral hemorrhage,fundus,conjunctival hemorrhage,oral mucosal bleeding,skin ecchymosis,menstruation increase bleeding tendency;(2)The existence of cerebral infarction,pulmonary embolism,myocardial infarction,thrombosis of deep vein thrombosis and other diseases;(3)Have been applied to anticoagulation therapy patients.The participants general clinical data were collected that including gender,age and weight.Laboratory items were including Hemoglobin,hematocrit,blood platelet,platelet distribution width,prothrombin time,prothrombin time activity,prothrombin time ratio,international normalized ratio,fibrinogen,activated partial thromboplastin time,thrombin time,d-dimer,triglyceride,total cholesterol,low densitity lipoprotein,high densitity lipoprotein,apolipoprotein A,apolipoprotein B.Recording the first dose and additional dose of heparin.Using the German Fresenius bedside blood filter,filter is Fresenius AV600 S polysulfone membrane.The transmembrane pressure,pressure drop,venous pressure,cardiopulmonary bypass line,filter and small pot of blood coagulation and machine alarm were recorded.543 patients were divided into well heparin anticoagulation group(489 cases)and failure heparin anticoagulation group(54 cases);the well anticoagulant group were randomly divided into a training set(409 cases)and test set(80 cases)to establish artificial neural network model,and established multiple linear regression model.SPSS 22.0 software was used for data analysis,P < 0.05 with statistical significance.The application of metlab related variables mapping.Using Anaconda software Python3.4 to construct artificial neural network model.Various methods were compared with 5%,10%,15% coincidence rate and the coefficient of determination R2 value.Result: 1.Artificial neural network model was established in well heparin anticoagulation group to analysis respectively of the first dose and additional dose of heparin with other variables.Compared to the real value of first dose of heparin in well heparin anticoagulation group,the predictive value of the first dose of real value in 5%,10%,15%,coincidence rate were 41%,68%,83%;compared to the real value of addition dose of heparin in well group,the predictive value of the addition dose of real value in 5%,10%,15% coincidence rate were 38%,60%,81%.2.We use bivariate correlation to analysis of first dose heparin and additional dose with 22 variables in well group,results show that the first dose of heparin and gender,weight,Hb,HCT,PLT,PT%,APTT and ALB were positively correlated,but negatively correlated with age,PDW,D-d,PT,PT-R PT-INR(P<0.05).First dose of heparin and weight were significantly correlated(correlation index r=0.707,P < 0.01).Additional dose of heparin associated with sex,weight,PLT was negatively correlated with age(P<0.05).3.A linear regression(P < 0.01)was established between the dose of heparin and weight.The determination coefficient R2 was 0.499 in first dose of heparin and weight relation,and was 0.082 in addition dose of heparin and weight relation.Two equations as follows(1)(2): Y1 0.277X1 2.044(1)(Y1 was the first dose of heparin,X1 was weight)Y2 0.023X1 3.783(2)(Y2 was the addition dose of heparin,X1 was weight)4.Multiplelinear regression(P < 0.01)was established between the dose of heparin and weight.The determination coefficient R2 was 0.560 in first dose of heparin and weight relation,and was0.106 in addition dose of heparin and weight relation.Two equations as follows(3)(4): Y1 0.273X1 0.021X2 0.372X3 0.064X4 3.121(3)(Y1 was first dose,X1 was weight,X2 was Hb,X3 was D-di,and X4 was Alb)Y2 0.023X1 0.005X2 0.143X3 0.172X4 3.269(4)(Y2 was additional dose,X1 was weight,X2 was Hb,X3 was TC,X4 was APO-B)5.Comparison of artificial neural network model,multiple linear regression model and the prediction of the linear regression based on weight.Each model was compared with the real value of heparin dose of 5% coincidence rate,10% coincidence rate,15%tcoincidence rate and the determination coefficient R2.The coincidence rate and R2 of the three models of artificial neural network were higher than that of multiple linear regression model,and the coincidence rate and R2 of the multiple linear regression model were higher than that of linear regression.6.Compared with two groups,the level offirst dose of heparin in failure group wassignificantly lower than that inwell group(33.27±3.43 U/kg VS 26.62±5.61 U/kg)(P<0.0l).The level ofadditiondose of heparin in failure group waslower than that inwell group(10.01±1.87 U/(kg·h)VS 8.02±2.39 U/(kg·h))(P<0.05).Conclusion: 1.Compared to traditional method,Artificial neural network model and multiple linear regression model has better predictability in the formulation of heparin dose in CRRT,and the artificial neural network model is best.2.In the formulation of CRRT heparin anticoagulant program,divided by weight as a reference,but also to consider the gender,age,hemoglobin,platelets,D-dinner,serum albumin and other factors on the impact of a reasonable dose.Multiple linear can be used to predict the formulation of heparin dose.