Study Of Sudden Cardiac Arrest Risk Probability Prediction Model Based On Adjusted Heart Rate Variability Metrics

Sudden Cardiac Arrest(SCA)is the sudden termination of cardiac ejection function for various reasons.If effective measures are not taken,it will progress rapidly to sudden cardiac death.There is a saying of golden 4 min in the rescue of SCA patients.For each minute that rescue is delayed,there is a 10～15% decrease in the success rate of rescue.If the delay is more than 10 min,brain death occurs in most patients and it is impossible to recover the consciousness of patients.Prompt defibrillation and early cardiopulmonary that are the core links in the survival chain of SCA can increase survival rate of SCA patients by 3-to 4-fold.The estimation of SCA patients is around 1.03 million in China and most cases occur in out-hospital settings(about 77% of them occur at home and 16% of them occur in public place).Due to geographical location and traffic condition,the average arrival time of medical personnel and ambulance is 16 min in China.It is difficult for SCA patients to receive timely and professional treatment due to randomness and time sensitivity of SCA.It is reported that the survival rate for SCA patients is extremely low and studies demonstrate that the overall survival rate for SCA patients is approximately 1～2% in China.Therefore,SCA risk prediction is import that can help the patients seek professional medical assistance in advance and reserve sufficient time for treatment.It will bring more survival chances for SCA patients.Left ventricular ejection fraction is primary indicator utilized to clinically screen patients with high SCA risk although both of its sensitivity and specificity are low.It is measured by echocardiography and medical personnel that is suitable for in-hospital settings and not suitable for out-hospital settings.Heart rate variability(HRV)based on electrocardiogram reflects the dynamic balance in the autonomic regulation of the heart.It is also utilized in assessment of autonomic disorders and quantification of risk for various cardiac diseases and noncardiac conditions,which has been widely used in the early prediction of SCA risk.But there still have been some deficiencies in the existing prediction model based on HRV metrics.Firstly,the effect of heart rate(HR)on HRV metrics is not adjusted.The HR is determinant factor of HRV metrics,but previous researches use HRV metrics without HR adjustment,which could affect the stability and reliability of results due to different HR level.Secondly,all the prediction models established in previous researches are classification model that could produce only binary results to predict relative SCA risk,and could not predict absolute SCA risk probability in various subjects which is unfavorable to early screening and warning for SCA risk.Thirdly,the predictive performance still needs to be improved.All the prediction models based on HRV metrics use traditional machine learning method,and prediction time is shorter than the deep learning prediction model based on the raw electrocardiogram signal.Most researches include only SCA patients and normal controls,the research needs to be extend to more categories of cardiovascular diseases,for example the most common hereditary heart disease in clinical practice--hypertrophic cardiomyopathy(HCM),to improve the accuracy and reliability of prediction results.In order to solve these problems,we proposed a refined HRV adjustment model based on HR and established a SCA risk probability prediction model with multilayer perceptron of traditional machine learning method and adjusted HRV metrics.The HCM patients were included to validate the prediction results.The main contents of this paper were as follows:1.A refined HRV adjustment model was proposed.First,the exponential decay relationship between HR and HRV was validated in both long-term HRV and short-term HRV.The determination coefficient for exponential fitting of long-term and short-term HRV results with HR was 0.23±0.15 and 0.81±0.29(P<0.05).Therefore,the HR had a greater effect on short-term HRV results and the exponential relationship was more pronounced.Then,the difference of adjustment coefficient between normal controls,HCM patients and SCA patients was validated.Among the 11 HRV metrics,there were significant differences in the adjustment coefficients of 10 metrics except for sample entropy(P<0.05).Lastly,a refined HRV adjustment model was proposed to overcome the drawback of fixed adjustment coefficient in existing exponential adjustment model.The exponential adjustment model with individualized adjustment coefficient was established using exponential relationship between HRV metrics and mean HR,which could better eliminate the effect of HR on HRV results.2.The effect of different HRV analysis method on performance of SCA risk stratification was investigated.The clinically commonly used 11 HRV metrics was used for SCA risk stratification between normal controls,HCM patients and SCA patients.The adjustment HRV method proposed in this paper had the best performance for SCA risk stratification with the area under the receiver operating characteristic curve of 0.73±0.08,which was significantly greater than long-term HRV analysis(0.62±0.05),short-term HRV analysis(0.58±0.04),and fixed coefficient adjusted HRV analysis(0.67±0.05).Compared with unadjusted HRV results,coefficient of variation of the HRV results was significantly decreased from 0.22±0.10 to 0.09 ± 0.02 when adjusted by the method proposed in this paper.The adjustment method of HRV proposed in this paper increased the accuracy and stability of HRV results for SCA risk stratification.3.The effect of different signal length on accuracy of adjusted HRV results was investigated.Compared with the results of 24 h signal length,the Pearson correlation coefficient was 0.37±0.20,0.73±0.14 and 0.97±0.04(P<0.05),relative error was 26.74±21.83%,7.82 ± 5.26% and 3.25 ± 1.77%(P<0.05),the area under the receiver operating characteristic curve was 0.72±0.08,0.83±0.12 and 0.85±0.12(P<0.05)when the signal length was 5min,30 min and 60 min relatively.The adjusted HRV results could reliably characterize the pathological changes of autonomic function of HCM patients in 3 signal lengths,but the accuracy and reliability of adjusted HRV results of 60 min signal length was the highest.4.A SCA risk probability prediction model was established.The selection of optimal feature subset was performed using recursive feature elimination method,and then the multilayer perceptron model was trained and evaluated.The model proposed in this paper could early predict SCA high risk event within 24 h before the onset of its occurrence.The SCA risk probability of normal controls,HCM patients and SCA patients was 0.04±0.14,0.29 ± 0.31 and 0.93 ± 0.22 respectively(P<0.05).The predictive accuracy was 95.05%,sensitivity was 93.10%,specificity was 95.56%,and the predictive performance was better than the existing methods.Additionally,The model proposed in this paper had a simple structure and low computational complexity,making it more suitable for continuous monitoring in portable and mobile devices.Through the above research,a refined HRV adjustment model with individualized adjustment coefficient was proposed in this paper.Secondly,the performance of different HRV analysis method for SCA risk stratification was investigated.Then,the accuracy of adjusted HRV results with different signal lengths was investigated.Finally,a SCA risk prediction model was established that could predict absolute risk probability of SCA.The results show that the exponential adjustment model with individualized adjustment coefficient improves the accuracy and stability of HRV results for SCA risk stratification,and the SCA risk probability prediction model based on adjusted HRV metrics and multilayer perceptron not only has better predictive performance than existing method but also has simple model structure.In the next step,the prediction model proposed in this paper can be integrated with wearable devices for continuously monitoring and early screening of SCA risk.This can help patients with high SCA risk seek professional and effective treatment timely and reduce the mortality of SCA patients.