Research On The Predictive Value Mining Of Continuous Physiological Data Based On The Digital Six-minute Walk Test

Background:The six-minute walk test(6-MWT)is a method often used to assess exercise capacity in patients with respiratory diseases.The 6-MWT assesses a patient’s exercise capacity by the distance walked in six minutes(6-MWD).Compared to the cardiopulmonary exercise test(CPET),the gold standard for exercise capacity assessment,the index of the 6-MWT is relatively limited,and the 6-MWD is largely influenced by patient’s individual differences and environment.Individual physiological data such as heart rate(HR),respiration rate(RR),oxygen saturation(Sp O2),blood pressure(BP)and pulse rate contain a lot of individualized health and disease information.Collecting these physiological data may provide additional indices for the 6-MWT.The Digital Six-Minute Walk integrates a wearable mobile device,software system,and intelligent algorithm system into the 6-MWT,allowing for lowload,continuous collection of physiological data such as HR,RR,Sp O2,and BP of the patient before,during and after 6-MWT.However,the clinical application value of these collected physiological data remains to be explored further.The aim of this study is to explore the correlation between physiological data and patients’ status,and to explore the clinical application value of physiological data in 6-MWT to provide individualized assessment and prediction for patients.Part 1: Predictive value of continuous physiological data in 6-MWT for lung function Objective:This study aimed to explore the predictive value of data obtained from the digital six-minute walk test system for the classification of pulmonary function in patients with chronic obstructive pulmonary disease(COPD).Methods:A retrospective analysis of data from COPD patients who underwent 6-MWT and pulmonary function tests from June 28,2019 to July 28,2022 in the General Hospital of the Chinese People’s Liberation Army,Hainan Hospital of the Chinese People’s Liberation Army General Hospital,and Xiangya Second Hospital of Central South University.COPD patients were diagnosed and assessed for severity according to the Global Initiative for Chronic Obstructive Lung Disease(GOLD)criteria.The 6-MWT was performed according to the American Thoracic Society(ATS)criteria.The digital six-minute walk system was used to continuously monitor exercise and physiological data during the 6-MWT.Spearman correlation analysis was used to analyze the correlation between 6-MWT data and pulmonary function indices.Data with moderate correlation(r > 0.4)or higher were included in the machine learning model(Logistic Regression,SVC-linear,SVC-poly,Random Forest,Light GBM,XGBoost),and the 5-fold cross-validation was used to assess the performance of the machine learning model.Results:6-MWT data from 501 eligible COPD patients were included.Spearman analysis showed that forced expiratory volume in 1 second(FEV1)was associated with 6-MWD(r=0.575,p<0.01),six minutes walk work(6-MWW)(r=0.531,p<0.01),distancedesaturation product(DSP)(r=0.526,p<0.01),heart rate recovery after one minute(HRR1)(r=0.438,p<0.01),and minimum peripheral oxygen saturation(Sp O2 min)(r=0.438,p<0.01)were moderately correlated.First second as a percentage of the expected value(FEV1%)was moderately correlated with 6-MWD(r=0.466,p<0.01),6-MWW(r=0.426,p<0.01),HRR1(r=0.430,p<0.01)and strongly correlated with DSP(r=0.673,p<0.01).Forced vital capacity(FVC)was moderately correlated with DSP(r=0.464,p<0.01).The Borg respiratory scale and the Borg fatigue scale were weakly or strongly correlated with the lung function indices.5-fold cross-validation results showed that the XGBoost model had the highest prediction accuracy rate(0.854).The model had the best prediction for COPD grade I and grade II(grade I: Precision 94.76%,Recall 92.5%;grade II: Precision 83.46%,Recall 93.89%).Conclusion:There was stronger correlation between 6-MWD,6-MWW,DSP,HRR1,Sp O2 min and pulmonary function indicators than the subjective assessment scale(Borg scale),moreover,the DSP correlates more strongly than the commonly known 6-MWD,which could be used as simple alternative indices for monitoring and assessing pulmonary function in COPD patients.The XGBoost model provides a comprehensive multidimensional approach to predict lung function with the highest accuracy.Part 2: Predictive value of continuous physiological data in 6-MWT for adverse events Objective:This study aimed to analyze the performance of adverse events in 6-MWT and their relationship with physiological data,and to predict adverse events based on physiological data.Methods:A retrospective analysis of data from CRD patients who underwent 6-MWT from June 28,2019 to July 28,2022 at the General Hospital of the Chinese People’s Liberation Army,Hainan Hospital of the Chinese People’s Liberation Army General Hospital,and Xiangya Second Hospital of Central South University.A digital 6-MWT system based on the Sens Echo wearable device was used to continuously monitor physiological data and record adverse events during 6-MWT.Adverse events were defined as chest pain,dyspnea,leg cramps,paleness,or other signs/symptoms requiring cessation according to ATS guidelines.Two groups were divided according to whether adverse events occurred during the 6-MWT.The characteristics of the data collected for anthropometry,HR,RR,and Sp O2 were analyzed.The area under the curve(AUC)of the receiver operating characteristic(ROC)was used to analyze features with statistically significant(P<0.001)differences between the two groups,and features with AUC>0.8 were analyzed to identify thresholds for adverse events to provide a prediction based on one-dimensional data.For multidimensional data prediction models,features were extracted from 2 minutes before the start of 6-MWT and 1 minute after the 6-MWT were incorporated into machine learning models(Logistic Regression,SVC-rbf,SVC-linear,SVC-poly,Random Forest,Light GBM,XGBoost),and the average of AUC after 5-fold cross-validation was used to evaluate the performance of different machine learning models.Results:A total of 2206 6-MWT data was collected,and adverse events occurred in 125(5.7%)cases.Dyspnea was the most common event(reported by 86 patients).About95% of the adverse events occurred 1 minute after the start of 6-MWT.Physiological characteristics that were significantly different between the two groups(P<0.001)were extracted mainly from HR and Sp O2,including: HRR1,DSP,Sp O2 baseline value(Sp O2 base),Sp O2 end value(Sp O2 end),difference between Sp O2 base and Sp O2min(ΔSp O2),Sp O2 min,DSP,Sp O2 recovery value(Sp O2 recovery),time for Sp O2 drop by 4%,and time for Sp O2 drop to 90%.The DSP was 251.75m%(AUC: 0.831,95% CI: 0.769-0.905),the 6-MWW was 23.307(kg·km-1)(AUC: 0.831,95% CI: 0.736-0.876),and the 6-MWD was 301m(AUC: 0.825,95% CI.0.756-0.895)which were the cut-off values for identifying adverse events in this study.The Light GBM model showed the highest AUC with 0.874 ± 0.063 among the machine learning models.Conclusion:Continuous monitoring of physiological parameters in 6-MWT is useful to quantify the physical condition of patients during the test.Compared to the commonly used 6-MWD,the DSP extracted from physiological parameters can better identify adverse events,so we recommend monitoring the DSP besides the continuous monitoring of Sp O2 min recommended by the ATS guidelines.The Light GBM model provides a comprehensive multidimensional approach to predict adverse events with the highest accuracy.The digital 6-MWT based on wearable sensors provides an additional means of safe monitoring for patients,which optimizes the traditional 6-MWT.Part 3: Predictive value of continuous physiological data in 6-MWT for exerciseinduced desaturation Objective:This study aims to provide physiological indices and predictive models that can easily predict exercise-induced desaturation(EID)in patients with chronic respiratory disease,and to explore the value of continuous monitoring based on the digital sixminute walk system for guiding pulmonary rehabilitation.Methods:A retrospective analysis of the data of patients with CRD who underwent 6-MWT using the digital six-minute walk system from June 28,2019 to July 28,2022 at the General Hospital of the Chinese People’s Liberation Army,Hainan Hospital of the Chinese People’s Liberation Army General Hospital,and Xiangya Second Hospital of Central South University.The EID was defined as Sp O2≤88%,and patients with base Sp O2≤88% before the 6-MWT test were excluded.The identification rate of EID was compared between the Sp O2 min obtained by continuous monitoring and the Sp O2 end obtained by intermittent monitoring.According to the calculation method with high recognition rate,the patients were divided into EID group and non-EID group,and the differences of 6-MWT data between the two groups were analyzed.Features with statistically significant differences(P<0.001)between groups were analyzed using the area under the curve(AUC)of the receiver operating characteristic(ROC),and those with AUC > 0.8 were screened and analyzed for critical values to provide onedimensional data-based predictions.For multidimensional data prediction models,features were extracted from 2 minutes before the start of 6-MWT and 1 minute after the 6-MWT were incorporated into machine learning models(Logistic Regression,SVC-rbf,SVC-linear,Random Forest,Light GBM,XGBoost),and the average of AUC after 5-fold cross-validation was used to evaluate the performance of different machine learning models.The occurrence of EID and adverse events was characterized to analyze the role of continuous monitoring of patient status during 6-MWT to guide pulmonary rehabilitation.Results:A total of 2045 6-MWT data was included.There were 448 cases of EID(21.9%)identified by Sp O2 min ≤ 88% and 182 cases of EID(8.9%)identified by Sp O2 end≤ 88%.Approximately 95% of EIDs occurred 1 minute after the start of 6-MWT.Physiological characteristics that were significantly different(P<0.001)between the two groups included HR base,HR end,HRR1,respiratory rate baseline value(RR base),respiratory rate end value(RR end),respiratory rate maximum(RR max),Sp O2 base,Sp O2 end,ΔSp O2,DSP,Sp O2 recovery,desaturation area(DA),desaturation distance ratio(DDR),time for Sp O2 drop by 4%,and time for Sp O2 drop to 90%,and the time for Sp O2 drop to the minimum value.DA of 936min%(AUC: 0.846,95% CI: 0.792-0.900)and DDR of 1.976min%/m(AUC: 0.844,95% CI: 0.792-0.896)were the cutoff values for identifying EID in this study.The Light GBM model showed the highest AUC with 0.958±0.035 among the machine learning models.According to whether EID with adverse events occurred in 6-MWT,patients were classified into four types: "can do,do do"(wearable device does not detect EID,and patient does not complain of subjective symptoms and rests),"can’t do,but do"(wearable device detects EID,but patient does not complain of subjective symptoms and rests),"can’t do,not do"(wearable device detects hypoxia,and patient complains of subjective symptoms and rests),"can do,not do"(wearable device does not detect hypoxia,but patient complains of subjective symptoms and rests)to discussion the implications for personalized pulmonary rehabilitation.Conclusion:The cut-off value of DA identification for EID in this study was 936 min% and DDR was 1.976 min%/m.The Light GBM model provides a comprehensive multidimensional approach to predict EID with the highest accuracy.Continuous monitoring based on wearable devices can identify more EID and classify patients undergoing 6-MWT into four categories,which can guide the implementation of individualized pulmonary rehabilitation.Physiological changes precede the patient’s subjective symptoms,which provides evidence for the prediction of physiological parameters and shows the advantages of continuous monitoring by digital 6-MWT.