| Introduction: Diabetes mellitus(DM)is one of the most serious public health problems in the world,and its prevalence continues to rise at an alarming rate.According to the latest data,it is predicted that there will be about 783.2 million people suffering from DM in the world by 2045.China has the largest number of DM patients,and the number is expected to increase to 174 million by 2045.Diabetic foot(DF)is one of the most common and serious complications of diabetes.Its treatment cycle is long and expensive,which poses a huge burden on patients,families and society.The rate of new diabetic foot ulcer(DFU)in Chinese diabetic patients within 1 year is about 8.1%.The recurrent ulcer rate within 1 year is about 31.6%.About 4 million diabetic patients worldwide develop foot ulcers every year,and an average of DF patient is amputated every 20 seconds.The prognosis of DF is very poor,even exceeding the mortality and disability rate of most cancers.Diabetic foot osteomyelitis(DFO)is one of the manifestations of severe stage of DF,with a high recurrence rate and amputation rate.Clinical prediction model(CPMs),also known as clinical prediction rules and risk prediction model,is a model based on big data and machine learning(ML).Using a variety of mathematical formulas to estimate the probability of an individual suffering from a certain disease or a certain outcome in the future,it can provide more and more important reference information for clinicians,and has important clinical significance and application value.This study is divided into two parts.in the first part,the clinical data of patients with diabetic foot in the first affiliated Hospital of Dalian Medical University are collected and integrated.Through single factor analysis,multi-factor analysis,least absolute shrinkage and selection operator(LASSO),and combined with clinical significance,the relevant risk factors affecting amputation are screened out,and the risk prediction model of amputation is constructed.And through the random split method to verify the model internally.The second part of the study also uses the above clinical data to increase the follow-up of short-term death events after discharge,combined with univariate analysis,multi-factor analysis,LASSO machine learning algorithm and clinical significance,to screen the death-related risk factors and establish a risk prediction model,which is verified internally by K-fold cross-validation method.Part 1 Development and validation of predictive model for the risk of amputation in diabetic foot osteomyelitisObjective:Based on the analysis of the clinical data of inpatients with diabetic foot in the first affiliated Hospital of Dalian Medical University,the risk factors related to amputation were screened,and the risk prediction model of amputation was constructed and verified and evaluated internally.Materials and methods:The clinical data of patients with diabetic foot who met the inclusion and exclusion criteria in the first affiliated Hospital of Dalian Medical University from June 2012 to June 2022 were collected,including demographic data,previous disease history,local clinical manifestations of the foot,chronic complications of DM,laboratory data,imaging data,lower limb surgery and so on.First of all,the missing data value,abnormal value check and variable conversion,random separation of training set and verification set are carried out.Then R software is used as the data processing tool,with the training set data as the main body,through t-test,chi-square test,single factor analysis,multi-factor Logistic regression analysis,LASSO machine learning algorithm and clinical significance to determine the risk factors closely related to amputation;the above risk factors are analyzed again by multi-factor Logistic regression analysis to establish the amputation risk prediction model and draw the line diagram.Finally,based on the training set data,the performance of the prediction model is internally tested through the area under the curve(AUC)of the receiver operating characteristic(ROC),calibration curve analysis and decision curve analysis(DCA).Based on the verification set data,the reliability and stability of the prediction model are internally verified.Results:1.A total of 357 patients met the inclusion and exclusion criteria of this study.Univariate analysis showed that hypertension,heart failure,Wagner grade,diabetic nephropathy,diabetic peripheral neuropathy,white blood cell count,neutrophil ratio,neutrophil count,lymphocyte ratio,platelet count,hemoglobin,glycosylated hemoglobin,wound secretion bacteria culture,lower limb artery stenosis and lower limb artery occlusion may be the risk factors of amputation.2.Multivariate Logistic regression analysis showed that hypertension,heart failure,Wagner grade,neutrophil count,lymphocyte ratio,hemoglobin,glycosylated hemoglobin,wound secretion bacteria culture and lower limb arterial occlusion were closely related to amputation.3.Calculating the area under the ROC curve of risk factors screened by multi-factor Logistic analysis indicates that the discrimination degree of single factor prediction is not high,so it is necessary to further reconstruct the multi-factor risk prediction model.4.The screening variables of LASSO machine learning algorithm showed that sex,age,smoking,marriage,hypertension,heart failure,Wagner grade,diabetic nephropathy,white blood cell count,hemoglobin,serum albumin,glycosylated hemoglobin,wound secretion bacteria culture and multiple plaques of lower extremities were risk factors for amputation.5.Based on the above research results,combined with clinical significance,hypertension,heart failure,Wagner grade,diabetic nephropathy,neutrophil count,lymphocyte ratio,hemoglobin,glycosylated hemoglobin,wound secretion bacteria culture and lower limb arterial occlusion were selected for the construction of amputation risk prediction model.6.The effectiveness of the model is evaluated by the training set data,the AUC value(0.963)shows that the model has good distinguishing performance,the calibration curve shows that the prediction probability is in good agreement with the actual probability,and the DCA curve shows that the model has good clinical applicability.Using the verification set data to verify the model,the AUC value(0.980)shows that the model has good discrimination performance,the calibration curve shows that the prediction probability is in good agreement with the actual probability,and the DCA curve shows that the model has good clinical applicability.Conclusion:1.Univariate and multivariate analysis showed that the risk factors of DF amputation included hypertension,heart failure,Wagner grade,neutrophil count,lymphocyte ratio,hemoglobin,glycosylated hemoglobin,wound secretion bacteria culture and lower limb arterial occlusion.However,the discrimination of single factor for the risk prediction of DF amputation is not high,so it is necessary to further build a multi-factor risk prediction model to improve its risk prediction ability.2.Comprehensive multi-factor Logistic regression,LASSO machine learning algorithm and clinical significance.Ten variables were selected to construct the prediction model=-9.562-1.394*hypertension+1.078*heart failure+2.130*Wagner grade-1.297* diabetic nephropathy+0.161*neutrophil count+0.095*lymphocyte percentage-0.028* hemoglobin+0.260*glycosylated hemoglobin-0.971*wound secretion bacteria culture+ 1.594*lower extremity arterial occlusion,and draw a diagram to facilitate clinical application.It is proved that the model has good calibration and differentiation.Part 2 Development and validation of predictive model for the risk of death in diabetic foot osteomyelitisObjective:Through the analysis of the clinical data of inpatients with diabetic foot in the first affiliated Hospital of Dalian Medical University,the risk factors related to death were screened,and the death risk prediction model was constructed and verified and evaluated internally.Materials and methods:The case data of the study were the same as those in the first part,and the deaths were followed up at 6 and 12 months after discharge by telephone.After data preprocessing,using R software as a data processing tool,through t-test,chi-square test,univariate analysis,multi-factor Logistic regression analysis,LASSO machine learning algorithm and clinical significance,to determine the risk factors closely related to shortterm death(6 months,12 months);the above risk factors were analyzed again by multifactor Logistic regression analysis,and the death risk prediction model was established and the line chart was drawn.Finally,the performance of the prediction model is evaluated and internally verified by AUC,calibration curve analysis and DCA.Results:1.A total of 357 patients were included in the construction of death risk prediction model within 6 months,and a total of 328 patients were included in the construction of death risk prediction model within 12 months(29 patients were discharged less than one year and were not included in the study).2.Univariate and multivariate analysis of 6-month death showed that heart failure,cerebrovascular disease,Wagner grade,albumin,glycosylated hemoglobin and amputation might be closely related,while univariate and multivariate analysis of 12-month death showed that heart failure,Wagner grade,glycosylated hemoglobin,lower limb arterial occlusion and amputation might be closely related.3.The area under the ROC curve of death risk factors for 6 months and 12 months is calculated respectively,suggesting that the degree of differentiation of single factor prediction is not high,so it is necessary to further reconstruct the multi-factor risk prediction model to improve its prediction performance.4.Based on multivariate Logistic regression analysis,LASSO machine learning algorithm and clinical significance,six risk factors including heart failure,Wagner grade,cerebrovascular disease,serum albumin,glycosylated hemoglobin and amputation were selected to establish a 6-month death prediction model,and five risk factors including heart failure,Wagner grade,glycosylated hemoglobin,lower limb arterial occlusion and amputation were selected to establish a 12-month death risk prediction model.5.The AUC value(0.920)of the 6-month death risk prediction model showed that the model had good distinguishing performance,the prediction probability matched well with the ideal line,and had good clinical practical efficiency,while the 12-month death risk prediction model AUC value(0.749)showed that the model had good distinguishing performance,the prediction probability matched well with the ideal line,and had good clinical practical efficiency.Conclusions:1.Univariate analysis and multivariate analysis showed that the death within 6 months may be related to heart failure,cerebrovascular disease,Wagner grade,albumin,glycosylated hemoglobin and amputation,while the death within 12 months may be related to heart failure,Wagner grade,glycosylated hemoglobin,lower limb arterial occlusion and amputation.The above single factors do not have a high degree of discrimination for short-term death risk prediction,so it is necessary to further build a multi-factor risk prediction model to improve its risk prediction ability.2.Combining multivariate Logistic regression,LASSO machine learning algorithm and clinical significance,six variables were selected to construct a 6-month death risk prediction model=0.038-1.395*heart failure+1.208*cerebrovascular disease+0.628* Wagner grade-0.687*serum albumin-0.461*glycosylated hemoglobin+0.693*amputation.The model is proved to have good accuracy,reliability and clinical applicability.Five variables were selected to construct a 12-month death risk prediction model=-2.309+0.669*Wagner grade+1.015*heart failure-0.143*glycosylated hemoglobin-0.925 *lower limb arterial occlusion+1.046*amputation.It is proved that the model has good accuracy,reliability and clinical applicability. |
PDF Full Text Request