Astudyofprognosticpredictionmodelfordiabeticfootpatients

Abstract:

Objective：Tocomparetheresultsofthebackpropagation（BP）neuralnetworkmodel，BPneuralnetworkmodeltrainedwithgeneticalgorithm（GA-BP），andCoxproportionalhazardsmodelinthepredictionofamputationandsurvivalprognosisofdiabeticfoot（DF）patients，andtochoosetheoptimalpredictionmodel.Methods：Hospitalizationdataof273patientswithDFwerecollectedwhowereadmittedtoTheFirstAffiliatedHospitalofChongqingMedicalUniversityfromJanuary2014toJanuary2016.ThesepatientswerealsofollowedupbytelephoneuntilDecember2016.Thenthreemodelswereestablished，namelytheBPneuralnetworkmodel，BPneuralnetworkmodeltrainedwithgeneticalgorithm，andCoxproportionalhazardsmodel.Theareaunderthereceiveroperatingcharacteristiccurve，sensitivity，andspecificitywereusedtojudgetheperformanceofthethreemodelsinthepredictionofamputationandsurvivalprognosisofDFpatients.Results：Whentheoutcomeswereamputationanddeath，theBPneuralnetworkmodel（amputa－tion：χ2=7.692，P=0.005；death：χ2=12.071，P=0.000）andBPneuralnetworkmodeltrainedwithgeneticalgorithm（amputation：χ2=10.083，P=0.001；death：χ2=12.071，P=0.000）werebetterthantheCoxproportionalhazardsmodel.However，therewasnosignificantdifferencebetweentheBPneuralnetworkmodelandtheBPneuralnetworkmodeltrainedwithgeneticalgorithm（amputation：χ2=0.200，P=1.000；death：χ2=0.000，P=1.000）.Conclusion：BoththeBPneuralnetworkmodelandtheBPneuralnetworkmodeltrainedwithgeneticalgorithmcanbeappliedtotheanalysisofsurvivalandprognosisofchronicdiseasessuchasDF.