基于机器学习构建髋部骨折患者术后谵妄的预测模型

打开文本图片集

中图分类号 R683.42 TP181 文献标识码A 文章编号：2096-7721（2025）12-2133-07

AbstractObjective：Toconstructandvalidateanomogrampredictionmodelforriskof postoperativedelirium （POD）inhipfracture patientsbasedonmachinelearing.Methods：18OhipfracturepatientswhoweretreatedattheFirstAfliatedwithNanjing Medical fromJanuarytoDecember2O24 were enroled.Theleastabsolute shrinkageandselectionoperator（LASSO） algorithm wasused toselect predictivevariablesforPOD.Four models wre developed： Logisticregresion（LR），randomforest （RF）， extreme gradientbostingtree（XGB），andsupportvectormachine（SVM）.Theperformanceofeachmodelwasvaluatedusingrceiver operating characteristic （ROC）curves，withthemodelshowingthehighestareaundercurve（AUC）valueselectedasthefinalmodel. Calibrationcurvesanddcsionceaalysis（CA）wereeploedtoessteaccuracyadliicalutilityofteodel，spctiely. Results：Among the 180 patients，55 （30.56%） developed POD.LASSO identified seven predictive variables： age，intraoperative blood lo，anesthsiaduratio，tubatioti，etalstateeamiation（E）sore，arlsoncomorbidityindex（CC）soed postoperative neutrophil-to-lymphocyteratio（NLR）.TheAUCvaluesforRF，XGB，LR，andSVMwere0.93，0.89，0.79，and0.87， respectively，withRFdemonstrating thesuperiorperformance.ThenomogrambasedonRFwassubsequentlydeveloped.DCAindicated thatthenomogamprovidedamoderatenetenfitinpredictingPODincdenceinelderlypfracturepants.TheHosmer-Lemesho（HL） test confirmed good calibration of the prediction model （χ2=2.231 P=0.758 ）.Conclusion：The RF-based nomogram enables healthcare professionaltuatelysssOiskinfracturepatints，facilitigtielyprevetiemeasresptiingsouotio improving healthcare quality，and providing recommendations for patient recovery and outcomes.

KeyWordsHipFracture;PostoperativeDelirium; Machine Learning;PredictionModel; Nomogram

谵妄是髋部骨折患者术后常见且严重的并发症，它不仅会损害患者的认知功能，还会延长住院时间，增加医疗费用[1]。（剩余9982字）