老年衰弱前期/衰弱的红外热成像特征分析及应用价值探究OA

Objective This study aims to identify the infrared thermographic characteristics of pre-frailty and frailty in elderly populations,and to construct a risk prediction model for the onset of pre-frailty and frailty using machine learning(ML)techniques,thereby providing objective evidence for clinical prevention and intervention strategies targeting pre-frailty and frailty in the elderly.Methods This study conducted a retrospective analysis of health examination data from 489 elderly individuals collected from communities in Chengdu and surrounding areas between October 2021 and October 2023.After applying 1∶1 propensity score matching(PSM)to balance baseline characteristics between groups,infrared thermal imaging features and independent influencing factors of pre-frailty/frailty in older adults were analyzed.The PSM dataset was randomly split into a training set and a validation set at a ratio of 7∶3.Five machine learning algorithms,namely Logistic regression(LR),support vector machine(SVM),gradient boosting machine(GBM),neural network(NNET)and CatBoost,were used to construct risk prediction models for pre-frailty/frailty in the elderly.Model performance was evaluated using the area under the receiver operating characteristic curve(Area under the curve,AUC),accuracy,sensitivity,specificity,and F1-score.Decision curve analysis(DCA)was conducted to assess clinical utility.The optimal model was subsequently interpreted using the SHAP(Shapley additive explanations)method.Results Infrared thermography of pre-frail and frail elderly individuals exhibits characteristic cool deviations(ΔT<0)in the Middle Jiao and Du Meridian.All five machine learning models demonstrated favorable discriminative ability and clinical utility,with the Logistic regression model exhibiting the best overall performance—achieving AUC values of 0.790(95%CI:0.730-0.851)in the training set and 0.793(95%CI:0.700-0.887)in the validation set.Based on SHAP analysis,the feature contribution ranking for the Logistic regression model was as follows:number of chronic diseases,PSQI(Pittsburgh sleep quality index)score,Du Meridian ΔT,Middle Jiao ΔT,and PHQ-9 score.The number of chronic diseases(OR=1.467,95%CI:1.253-1.719),PSQI score(OR=1.105,95%CI:1.033-1.182),and PHQ-9 score(OR=1.183,95%CI:1.045-1.339)were identified as positive predictors of pre-frailty/frailty in the elderly,whereas Du Meridian ΔT(OR=0.387,95%CI:0.199-0.755)and Middle Jiao ΔT(OR=0.186,95%CI:0.055-0.625)served as negative predictors.Conclusion Machine learning models developed by combining Du Meridian ΔT and Middle Jiao ΔT demonstrate robust predictive performance and clinical utility,showing potential value in early screening of frailty among the elderly and supporting clinical decision-making.