Research On Constructing Electronic Frailty Index And Identifying Clinical Subtypes Of Frailty Based On Machine Learning

Development and Validation of Machine Learning-based Electronic Health Record Frailty Index Focusing on Hospitalized Older AdultsObjectives:Promoting "healthy aging" is a key action to actively respond to the aging population,and frailty is the first obstacle to maintain healthy life for older adults,which highlights the significance of early identification and intervention on frailty.Using electronic health records to identify frailty of elderly inpatients is an promising approach to solve this issue.This study aims to use machine learning combined with electronic health records to construct an electronic frailty index for hospitalized older adults.Patients and Methods:This study is a retrospective cohort study,enrolling 47,349 firstadmission patients aged>65 from July 2013 to September 2019 as a training/validation set.A prospective geriatric comprehensive evaluation cohort of 993 cases was constructed as a test set.In this study,frailty-related hospital adverse events(long hospitalization,decreased activities of daily living and inhospital death)were used as outcome variables.The XGBoost model was used to select key variables from 175 variables extracted from electronic health records,and the Electronic Frailty Index(eFI)was constructed according to the accumulative deficits model.The test set was used to evaluate the eFI's predictive capacity for the risk of frailty-related adverse events(all-cause death,all-cause readmission,accidental falls),and verify the correlation between eFI and the classic frailty indices.Results:According to the importance of the variables in the XGBoost model,66 variables were selected to construct eFI.eFI was significantly associated to frailty-related inhospital adverse events(Long hospitalization OR=1.923,95%CI:1.886-1.961,p<0.001;ADL score reduction OR=2.19,95%CI:2.124-2.259,p<0.001;Inhospital death OR=2.255,95%CI:2.050-2.363,p<0.001;Composite endpoint OR=2.217,95%CI:2.171-2.261,p<0.001).eFI was significantly correlated with the risk of adverse events in 1 year after discharge(All-cause death HR=2.830,95%CI:1.935-4.138,p<0.001;All-cause readmission HR=1.431,95%CI:1.263-1.621,p<0.001;Accidental falls HR=1.417,95%CI:1.137-1.766,p=0.002;Composite endpoint HR=1.492,95%CI:1.332-1.672,p<0.001).eFI was significantly correlated with the Comprehensive Assessment of Aged Frailty Index(CGA-FI)(Correlation coefficient=0.679,95%CI:0.612-0.684,p<0.001).According to the comprehensive analysis of the maximum Youden index,the best cut point for eFI to diagnose frailty was 0.258,the area under the ROC curve for frailty diagnosis was 0.821(95%CI:0.794-0.848),and the sensitivity was 0.794(95%CI:0.746-0.836)),the specificity was 0.718(95%CI:0.682-0.752).The proportion of frail patients in the electronic medical record cohort diagnosed by eFI was 27.0%(n=12774).Compared with non-frail patients,frail patients had a significantly higher risk of frailty-related in-hospital adverse events(Long-term hospitalization OR=3.61,95%CI:3.44-3.78,p<0.001;ADL score decline OR=7.64,95%CI:6.99-8.35,p<0.001;Inhospital death OR=5.01,95%CI:4.36-5.75,p<0.001;Composite endpoint OR=4.69,95%CI:4.48-4.91,p<0.001),and frailty was the independent predictors of 1-year adverse events for senile inpatients(All-cause death HR=6.72,95%CI:2.84-15.94,p<0.001;All-cause readmission HR=2.16,95%CI:1.69-2.76,p<0.001;Accidental fall HR=2.01,95%CI:1.30-3.11,p=0.002;Composite endpoint HR=2.33,95%CI:1.86-2.91,p<0.001).Conclusion:The electronic frailty index constructed based on electronic health records and machine learning can predict the frailty-related inhospital and prognostic adverse events of elderly inpatients.This provides technical support for medical institutions to identify frail inpatients.Systematic Identification of Frailty Subtype in Hospitalized Senile Patient Using Machine Learning Algorithm Combined with Structural Equation ModelObjectives:Frailty is a vital manifestation of unhealthy aging in older adults.Effective intervention can improve or reverse the frailty status,and reduce the risk of adverse events in turn.By and large,the existing approaches for frailty intervention are generic and show no difference in improving the prognosis.The identification of frailty clinical subtype can make interventions with more pertinency.This study aims to identify clinical subtypes of frailty and clarify the characteristics of the subtypes.Patients and methods:In this study,the frail population(n=12774)in the electronic health record cohort was used as the training set,and the frailty comprehensive assessment cohort(n=245)was used as the test set.Principal component analysis and cluster analysis were used to analyze the relationship among the variables of the frail older population.Partial least squares structural equation modeling(PLS-SEM)was used as a method for estimating frailty path models with latent variables and their relationships,and the test set was used to verify the correlation between the latent variables and related clinical indicators.The K-means cluster analysis was used to identify the clinical subtypes of frailty.In the test set,was used to classify the prognostic characteristics of different subtypes through Kaplan-Meier survival curve analysis,multivariate logistic regression analysis,and multivariate COX analysis.Results:The frailty PLS-SEM model included six latent variables:malnutrition,metabolic imbalance,comorbidity,inflammation,polypharmacy and functional degeneration.The path coefficient between frailty and functional degeneration was more evident than others,approximately 0.49(95%CI:0.48-0.50,p<0.0001).In the test set,the latent variables of malnutrition,metabolic imbalance,inflammation and functional decline were correlated with MNA-SF score,HbA1c,hs-CRP and SPPB score(p<0.0001).The optimal number of cluster analysis was 3,and the three clinical subtypes of frailty were identified through K-means clustering methods,such as "Inflammation-metabolic","Inflammation-nutrition-functional degeneration" and "Comorbidity".Compared with other subtypes,multivariate logistic regression analysis showed that after being adjusted for age,gender,education level and type of ward,"inflammation-nutrition-functional degeneration"was significantly correlated with the higher risk of inhospital adverse events(Inhospital death OR=2.992,95%CI:2.512-3.564,p<0.0001;Long hospitalization OR=2.846,95%CI:2.557-3.170,p<0.0001;ADL score decline OR=1.330,95%CI:1.155-1.532,p<0.0001;Composite endpoint OR=2.622,95%CI:2.349-3.924,p<0.0001).Multivariate COX regression analysis suggested that after adjusting for age,gender,education level,and type of ward,subtype "inflammation-nutrition-function degeneration"was characterized with a higher risk of adverse event in 1 year after discharge compared to other subtypes(All-cause death HR=2.296,95%CI:0.814-6.475,p=0.116;All-cause readmission HR=1.831,95%CI:1.121-2.989,p=0.015;Accidental accidental falls HR=1.452,95%CI:1.196-3.166,p=0.001;Composite endpoint HR=1.706,95%CI:1.083-2.688,p=0.021).Conclusion:Functional degeneration,inflammatory response,comorbidity,metabolic imbalance,malnutrition,and polypharmacy are the potential mechanisms leading to frailty.Three clinical subtypes of the frail hospitalized population are identified as"inflammation-metabolism," "inflammation-nutrition-functional degeneration" and"comorbidity." The clinical features and prognostic characteristics of these subtypes were significantly different.Further researches could implement more targeted interventions aimed at different clinical subtypes of frailty,expecting to improve frailty status and prognosis for the frail population.Trimethylamine N-oxide,a Gut Microbiota-dependent Metabolite,is Associated with Frailty in Older Adults with Cardiovascular DiseaseObjectives:As a gut microbiota metabolite,trimethylamine N-oxide(TMAO)plays an important role in the aging process.Frailty is significantly related to the poor prognosis of elderly cardiovascular patients.Many research results suggested that the imbalance of intestinal microbiota may induce or accelerate frailty,but the relevant mechanism or effector has not yet been clearly explained.Our study aimed to explore the association between TMAO and frailty in older adults with cardiovascular disease.Patients and methods:This cross-sectional study analyzed a total of 451 people aged 65 years or older who underwent comprehensive geriatric assessments.Frailty status was determined using a frailty index constructed with 48 variables according to the cumulative deficits model.Physical frailty and cognitive frailty were also assessed in detail.Multivariate COX regression was used to evaluate the prognostic value of serum TMAO concentration in elderly patients with cardiovascular disease.The endpoint of this study was all-cause death,all-cause readmission,and accidental falls in 1-year follow-up.Fasting plasma TMAO was measured by mass spectrometry.Results:The proportion of frail subjects was 29.9%(135/451).Plasma TMAO levels were significantly higher in frail patients than in nonfrail individuals(4.04[2.84-7.01]vs.3.21[2.13-5.03]?M;p<0.001).Elevated plasma TMAO levels were independently associated with the likelihood of frailty(OR 2.12,95%CI:1.01-4.38,p=0.046).Dose-response analysis revealed a linear association between the TMAO concentration and the OR for frailty.A 2-unit increase in TMAO was independently correlated with physical frailty(OR 1.23,95%CI:1.08-1.41,p for trend 0.002)and cognitive frailty(OR 1.21,95%CI:1.01-1.45,p for trend 0.04).In the multivariate COX regression analysis,high serum TMAO concentration(TMAO?6.91?M)was an independent risk factor for the composite endpoints(all-cause death,all-cause readmission,and accidental falls)in elderly cardiovascular patients(HR 1.63,95%CI:1.02-2.6,p=0.04).In the frail group,the risk of the composite endpoint increased with the increase of serum TMAO concentration.In the non-frail group,the risk of the composite endpoint did not increase significantly with the increase of serum TMAO concentration.Conclusion:In elderly patients with cardiovascular diseases,the serum TMAO concentration is positively correlated with frailty,the increase of serum TMAO concentration is positively correlated with the increased risk of frailty,and the increase of serum TMAO concentration is positively correlated with the risk of physical frailty and cognitive frailty.High serum TMAO concentration is an independent risk factor for poor prognosis in elderly patients with cardiovascular disease.