Percussion Entropy Analysis of Synchronized ECG and PPG Signals as a Prognostic Indicator for Future Peripheral Neuropathy in Type 2 Diabetic Subjects

Diabetic peripheral neuropathy (DPN) is one of the most common chronic complications of diabetes. It has become an essential public health crisis, especially for care in the home. Synchronized electrocardiogram (ECG) and photoplethysmography (PPG) signals were obtained from healthy non-diabetic (n = 37) and diabetic (n = 85) subjects without peripheral neuropathy, recruited from the diabetic outpatient clinic. The conventional parameters, including low-/high-frequency power ratio (LHR), small-scale multiscale entropy index (MEISS), large-scale multiscale entropy index (MEILS), electrocardiogram-based pulse wave velocity (PWVmean), and percussion entropy index (PEI), were computed as baseline and were then followed for six years after the initial PEI measurement. Three new diabetic subgroups with different PEI values were identified for the goodness-of-fit test and Cox proportional Hazards model for relative risks analysis. Finally, Cox regression analysis showed that the PEI value was significantly and independently associated with the risk of developing DPN after adjustment for some traditional risk factors for diabetes (relative risks = 4.77, 95% confidence interval = 1.87 to 6.31, p = 0.015). These findings suggest that the PEI is an important risk parameter for new-onset DPN as a result of a chronic complication of diabetes and, thus, a smaller PEI value can provide valid information that may help identify type 2 diabetic patients at a greater risk of future DPN.


Introduction
The prevention and early detection of diabetes mellitus (DM) in high-risk subjects such as those with metabolic syndrome have become important issues in preventive medicine and public health [1,2]. As for type 2 diabetic patients, the outcomes of microvascular and macrovascular complications can be Values are expressed as mean ± SD. Group 1, healthy elderly subjects; Group 2, diabetic subjects; Group 3, diabetic subjects with peripheral neuropathy 6 years after baseline measurement. The total number of test subjects was 122. WC, waist circumference; BMI, body mass index; SBP, systolic blood pressure; DBP, diastolic blood pressure; PP, pulse pressure; HDL, high-density lipoprotein cholesterol; LDL, low-density lipoprotein cholesterol; HbA1c, glycosylated hemoglobin; FPG, fasting plasma glucose. ** p < 0.001 Group 1 vs. Group 2. p values of the parameter larger than 0.017 are regarded as not statistically significant between two groups. The total number of subjects is 122 in this table.

Ethical Issues, IRB, and Consent Form
The study was approved by the Institutional Review Board (IRB) of Hualien Hospital (Hualien City, Taiwan) [26,27] and Ningxia Medical University (Yinchuan City, Ningxia Province, China) Hospitals (No.2018-229). All subjects gave written informed consent.

Study Protocol
Each subject was required to refrain from caffeine-containing beverages and theophyllinecontaining medications for at least 8 h before the baseline data measurement. Before taking the measurements, all subjects were requested to sign informed consent forms and complete questionnaires on demographics and medical histories, as well as receive blood sampling for serum biochemical analysis.
A detailed explanation of the aim and procedures was provided, as well as the measurement of synchronized ECG and PPG signals to be used for follow-up study. The test subjects received a standardized medical examination by a doctor that consisted of anthropometric, physiological, and biochemical measures at baseline. The indices of atherosclerosis and autonomic nervous function were subsequently computed. All diabetic patients underwent regular clinic treatment and follow-up in an outpatient clinic for at least eight years (i.e., two years for DM identification and six years for the follow-up period). DPN was diagnosed as the presence of symptoms of numbness, tingling, or pain of distal extremities lasting for more than 3 months in the same diabetes outpatient department through neurophysiological study [29].

Follow-up and DPN Status
The DPN status for the subjects in Group 3 at each follow-up stage was ascertained by questionnaire and clinical medical examinations. The screening DPN from type 2 diabetes patients at the baseline and follow-up periods was based on the presence of symptoms of numbness, tingling, or pain of distal extremities lasting for more than 3 months and a confirmed diagnosis by the clinic doctor (i.e., in accordance with neurophysiological study). The study population comprised a sample of 27 type 2 diabetes patients with DPN (aged 62.81 ± 1.71 years) who underwent a synchronized ECG and PPG signals measurement at baseline and then were followed for at least 6 years after the baseline measurement at the same hospital.

Baseline Measurements and Protocol of Measurement of Synchronized Electrocardiogram (ECG) and Photoplethysmography (PPG) Signals
All measurements were performed over a period in the morning (i.e., 08:30-10:30). In addition, to minimize the potential errors in the infrared sensor readings arising from involuntary vibrations of the participants, all subjects were allowed to rest in a supine position for 30 min in a quiet room with a temperature maintained at 26 ± 1 • C. Blood pressure readings were obtained once over the left arm of the supine subjects using an automated oscillometric device (BP 3AG1; Microlife Corporation, Taipei, Taiwan) with an appropriate cuff size. A self-developed six-channel electrocardiography ECG-PWV-based system [26,27] was used to acquire 1000 successive recordings of the RRI signals and digital volume pulses (DVPs) within 30 min. To validate the application of the ECG-PWV system in assessing autonomic function, the RRI series was used for the LHR [14,16], MEI LS , and MEI SS computations. Accordingly, the present study analyzed the RRI signals by dividing the MEI according to a small scale (MEI SS , mean value of sample entropy on a scale from 1 to 5) and large scale (MEI LS , mean value of sample entropy on a scale from 6 to 10) for comparison [20]. The DVPs of PPG with the R wave on RRI as a reference point could be used for the electrocardiogram-based pulse wave velocity (PWV mean ) [26,27] and PEI [21,22] computations for assessing the autonomic function considering the degree of atherosclerotic change and autonomic function, respectively ( Figure 1). In our previous study [21], changes in the BRS caused one to five cardiac cycle delays under the effects of fingertip DVP amplitude variations followed by synchronized RRIs. Accordingly, in obedience with the fluctuation tendency in the PEI computation, the percussion entropy had a length of the fluctuation pattern equal to two (i.e., PEI main contributor), and was expressed as BRS, while the percussion entropy, with a length of the fluctuation pattern equal to three (i.e., PEI major offset), was indicated as the biological complex system. synchronized RRIs. Accordingly, in obedience with the fluctuation tendency in the PEI computation, the percussion entropy had a length of the fluctuation pattern equal to two (i.e., PEI main contributor), and was expressed as BRS, while the percussion entropy, with a length of the fluctuation pattern equal to three (i.e., PEI major offset), was indicated as the biological complex system.

Statistical Analysis
The values are expressed as mean ± SD in Tables 1-3. The comparisons of the continuous valuables were analyzed using a Student's unpaired t test with Bonferroni correction, and the differences between the categorical variables were assessed using a chi-square test. For the goodness-of-fit test and relative risk analysis, the PEI values were arbitrarily divided into three categories by the interquartile range method. The PEI was processed as both continuous and categorical variables and was undertaken in the Cox proportional hazards model to analyze the multivariate parameters according to the expansion of DPN. The relative risks (RR) were predicted with Cox regression analysis with corresponding 95% confidence intervals [30]. The following traditional risk factors for DPN were included as variables in the model: age, BMI, resting systolic and diastolic blood pressure, total cholesterol, triglyceride, waist circumference, pulse pressure, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, glycosylated hemoglobin, and fasting plasma glucose from baseline (i.e., PEI provided) to the end of the follow-up period (no longer than 6 years for each patient). According to the chi-square goodness-of-fit test in SPSS, the null hypothesis was rejected with a computed chi-square value larger than the level of significance. The synchronized ECGs (a) and the right index finger photoplethysmography (PPG) signals (b) were obtained for percussion entropy index (PEI) computation. With (a) the R wave on Lead II as a reference point, the time differences (∆T 2 (c) and ∆T 3 (d)) for the second toe were obtained. The PWV mean was calculated by dividing the distances from different points of reference (L) with ∆T (i.e., PWV = L/∆T). The PWV mean , in the evaluation of the degree of atherosclerosis in the lower extremity of the body, was obtained by averaging the PWV values from both sides of the foot.

Statistical Analysis
The values are expressed as mean ± SD in Tables 1-3. The comparisons of the continuous valuables were analyzed using a Student's unpaired t test with Bonferroni correction, and the differences between the categorical variables were assessed using a chi-square test. For the goodness-of-fit test and relative risk analysis, the PEI values were arbitrarily divided into three categories by the interquartile range method. The PEI was processed as both continuous and categorical variables and was undertaken in the Cox proportional hazards model to analyze the multivariate parameters according to the expansion of DPN. The relative risks (RR) were predicted with Cox regression analysis with corresponding 95% confidence intervals [30]. The following traditional risk factors for DPN were included as variables in the model: age, BMI, resting systolic and diastolic blood pressure, total cholesterol, triglyceride, waist circumference, pulse pressure, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, glycosylated hemoglobin, and fasting plasma glucose from baseline (i.e., PEI provided) to the end of the follow-up period (no longer than 6 years for each patient). According to the chi-square goodness-of-fit test in SPSS, the null hypothesis was rejected with a computed chi-square value larger than the level of significance. The Statistical Package for the Social Sciences (SPSS, version 14.0 for Windows, SPSS Inc., Chicago, IL, USA) was utilized for all statistical analyses.

Results
The results from the fours indices, LHR, MEI SS , MEI LS , PWV mean , and PEI, were first computed for the diabetic subjects with peripheral neuropathy within six years (i.e., Group 3) for comparison with the healthy elderly subjects (i.e., Group 1) and diabetic patients without peripheral neuropathy Diagnostics 2020, 10, 32 6 of 13 (i.e., Group 2). Subsequently, three new diabetic subgroups using different PEI values were identified for the goodness-of-fit test. Finally, the Cox regression analysis of risk factors for the incidence of DPN within six years after the PEI provided for diabetic patients was verified.
3.1. Comparison among LHR, MEI SS , MEI LS , PWV mean , and PEI for Age-Controlled Healthy and Diabetic Subjects with and without DPN After the entire follow-up process had been carried out and the DPN status was confirmed, the results from the comparison of the four previous computational parameters (i.e., LHR, MEI SS , MEI LS , and PWV mean ) with the PEI for DPN identification assessment among the three groups of subjects are shown in Table 2. Although the value of PWV mean was significantly higher in Group 2 compared with the Group 1 subjects (p < 0.017), there was no notable difference between Groups 2 and 3. On the other hand, the PEI showed highly significant differences among the three groups (p < 0.001) ( Table 2).

Three Diabetic Subgroups Using Different Percussion Entropy Index (PEI) Values
The distribution of the PEI exhibited an approximately normal curve with a mild skew toward higher values. The values of the quartile ranges in the distribution were 0.27-0.54, 0.55-0.65, and 0.66-0.82 for the lower, middle two, and upper quartiles, respectively, for the prognostication of subjects with type 2 diabetes who are more prone to develop DPN. The diabetic patients in Group C showed remarkably higher HbA1c levels than those in the diabetic patients in Group B (p < 0.017). On the other hand, no significant differences were noted in the demographic and hemodynamic parameters, as well as the fasting blood glucose and serum lipid profile between any two groups ( Table 3). In summary, a comparison of characteristics among subjects in the three categories revealed no significant differences in age, body mass index, waist circumference, systolic blood pressure, diastolic blood pressure, pulse pressure, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, and fasting plasma glucose.

The Goodness-of-Fit Test
According to the chi-square goodness-of-fit test result in SPSS, the null hypothesis (i.e., no association between the PEI and DPN) was rejected with a chi-square value (i.e., the computed chi-square value, χ 2 = 8.00) larger than the level of significance (χ 2 = 5.99, α = 0.05). That is, diabetic patients with a smaller PEI value were associated with the future development of DPN within six years after the PEI was provided.

Cox Proportional Hazards Model
A total of 27 type 2 diabetic patients developed DPN among 85 study patients (31.8%) within six years after the baseline examinations in this study. The progression to DPN in patients in the three categories within six years and corresponding relative risks for the incidence of DPN assessed by the Cox proportional hazards model are shown in Table 4. The Cox model revealed a graded association, with the diabetic subjects with a small PEI (i.e., Group C) at 2.90× greater risk of developing DPN on follow-up, relative to the diabetic subjects with a large PEI (i.e., Group A) after adjustment for entry age, waist circumference, BMI, systolic and diastolic blood pressure, total cholesterol, triglyceride, pulse pressure, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, glycosylated hemoglobin, and fasting blood sugar. In addition, the Cox model revealed a graded association, with the diabetic subjects with a moderate PEI (i.e., Group B) having almost equal risks for developing DPN on follow-up relative to the diabetic subjects with a large PEI (i.e., Group A).

Cox Regression Analysis
The regression analysis using the Cox proportional hazards regression analysis of risk factors for incidence of DPN is shown in Table 5. The PEI was also significantly associated with the risk of developing DPN when it was treated as a continuous variable. The relative risk of incidence of DPN within six years of follow-up in diabetic patients for the PEI was 4.77 (p = 0.015), whereas the relative risks of incidence of DPN for the value of fasting plasma glucose and glycosylated hemoglobin were 1.01 (p = 0.033) and 0.73 (p = 0.041), respectively. The term of interaction between HbA1c and FPG was not significant (p = 0.205).
Compared with fasting plasma glucose and glycosylated hemoglobin, smaller PEI values can provide valid information that may help identify type 2 diabetic patients at a greater relative risk of future DPN from baseline measurement (i.e., PEI provided) to the end of the follow-up period (i.e., within six years after the PEI was provided).

Discussion
Type 2 diabetes and its related complications are associated with the long-term damage and failure of various organ systems [31]. The overall impact of bad glucose control on vascular complications and major clinical outcomes in type 2 diabetes is still an open problem. While good glucose control has an undoubted benefit in the microvascular system of diabetic patients [6][7][8]32], good blood glucose control also improves microvascular disease and should be implemented early and maintained for the optimum length of time. A previous review study [31] highlighted the need for implementing programs for early detection, screening, and awareness to mitigate the burden of managing the complications. Therefore, diagnosis classifies a patient as having or not having a particular disease. In fact, diagnosis was recognized as the primary guideline for treatment and prognosis (i.e., what is going to happen in the future), and is still considered the key component of clinical practice [33]. However, it is not an easy job to control the appropriate glucose for diabetic patients; this is the reason DPN is still one of the most common chronic complications of diabetes. Recently, a study [23] that determined the exact PPI intervals was reported to provide a prognosis of peripheral neuropathy from diabetic patients. However, real-time computation was not able to obtain immediate index information of the test subjects. Furthermore, it did not provide valid information regarding the degree of risk of developing future DPN that may encourage type 2 diabetic patients to follow a good lifestyle.
This study addressed results from the indices LHR, MEI SS , MEI LS , PWV mean , and PEI, which were first computed for diabetic subjects with peripheral neuropathy six years after baseline measurement (i.e., Group 3) for comparison with diabetic patients without peripheral neuropathy (i.e., Group 2). Although the values of the vascular stiffness indices, including MEI LS and PWV mean , were significantly different in Group 2 compared with the Group 1 subjects (p < 0.017), there were no notable differences between Groups 2 and 3 (p > 0.017). On the other hand, the PEI (i.e., BRS assessment index) showed highly significant differences among the three groups (p < 0.017) ( Table 2). These results are consistent with the major outcomes in the previous study [23]. Significantly smaller values for the PEI were noted for Group 3 compared to the other two groups (e.g., Group 1 vs. Group 2 vs. Group 3: 0.73 ± 0.01 vs. 0.63 ± 0.01 vs. 0.59 ± 0.02), which is consistent with the same findings, where diabetic neuropathy was found to be a more significant crucial factor of spontaneous BRS assessment than carotid elasticity in type 2 diabetics in [34].
Therefore, most diabetic patients with a smaller PEI value were only concerned about the relative risks of the future development of DPN, and were not focused on achieving a smaller PEI value, because DPN, which has a lifetime prevalence of approximately 50%, is the most common diabetic complication. DPN is also the leading cause of disability due to diabetic foot ulceration and amputation, gait disturbance, and fall-related injury [3,35]. Neuropathy not only causes problems such as a decreased quality of life, poor sleep, and depression in diabetic patients, but the quality of life is also greatly affected [36][37][38]. Although PEI has recently been introduced to assess the complexity of BRS [21][22][23], the significance of smaller PEI values concerning the identification of subjects with type 2 diabetes who are more prone to develop diabetic neuropathy is unknown. That is, it would be difficult to predict how many and who will develop DPN in advance. Thus, a model of clinical practice focused on DPN prognosis and predicting the likelihood of future outcomes associated with PEI may be more useful for diabetic patients [33].
According to the results in Table 2, the present study adopted the PEI as the first measurement of all of the recruited diabetic patients to create the basis of quartiles in the diabetic population's distribution of the PEI: the upper 25% (i.e., n = 22, Group A, 6 DPN included), the middle 50% (i.e., n = 42, Group B, 10 DPN included), and the lower 25% (i.e., n = 21, Group C, 11 DPN included). The diabetic patients in Group C showed remarkably higher HbA1c levels than those in the diabetic patients in Group B (p < 0.017) ( Table 3). On the other hand, no significant differences were noted in the demographic and hemodynamic parameters, as well as the fasting blood glucose and serum lipid profile between any two groups (i.e., Group A vs. Group B and Group B vs. Group C) ( Table 3). In the diabetic patients with smaller PEI values in Group 3, almost 50% had developed DPN within six years (Table 4). These results are consistent with statements in the study [31]. According to goodness-of-fit test result in the study, the null hypothesis (i.e., no association between smaller PEI values and developing DPN within six years after the PEI was provided) was rejected, with the chi-square value being larger than the level of significance. An association exists between diabetic patients with smaller PEI values and the development of DPN within six years after baseline measurement.
A total of 27 type 2 diabetic patients developed DPN among 85 study patients (31.8%) in the six years after baseline examinations. This finding is consistent with similar results reported in [37,38]. The progression to DPN of patients in the three categories within six years and the corresponding relative risks for the incidence of DPN assessed by the Cox proportional hazard survival model are shown in Table 4. The Cox model revealed a graded association, with the diabetic subjects with a small PEI (i.e., Group C) at 2.90-times greater risk of developing DPN on follow-up relative to the diabetic subjects with a large PEI (i.e., Group A) after adjustment for entry age, waist circumference, BMI, systolic blood pressure and diastolic blood pressure, total cholesterol, triglyceride, pulse pressure, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, glycosylated hemoglobin, and fasting plasma glucose. Type 2 diabetic patients with smaller PEI values may have a larger relative risk of developing DPN within six years. In addition, the relative risk of incidence of DPN within six years of follow-up in diabetic patients for the PEI was 4.77 (p = 0.015) in Table 5. This study was designed to use synchronized ECG and PPG signals (i.e., PEI) in predicting the development of peripheral neuropathy from type 2 diabetes. The PEI has recently been introduced not only to assess the complexity of BRS but also to show the significance of smaller PEI values concerning the identification of subjects with type 2 diabetes who are more prone to DPN.
The current study has its limitations. Firstly, the number of subjects recruited was relatively small. In addition, it may be difference between DPN attack confirmation time and checkout time in our study for non-fixed follow-up to each diabetic patient. Therefore, Kaplan-Meier survival analysis was not adopted in this study. Nevertheless, highly significant associations between PEIs and relative risks of developing DPN were still significant. Secondly, this study focused on the Cox proportional hazard survival model for diabetic patients, and the optimal BRS delay between the amplitude and RRI series would be set at one to five heartbeat cycles for all test subjects with the same setting. Thirdly, the impact of periodontal therapy on diabetes control was not investigated because of the limited number of diabetic patients. Subsequently, the period of baseline measurement was more than two years, because of the limited number of subjects in each group. Finally, as an observational study, the values of PEI and the proposed parameters could be used to identify the risk factors for a prediction task by using simple machine learning algorithms (such as SVM, LDA, or even deep learning) in the future.

Conclusions
This study represents the first attempt to investigate the clinical prognostic feasibility of applying the PEI, and demonstrates enhanced sensitivity in differentiating between diabetic subjects without DPN and diabetic subjects with DPN within six years after baseline measurement, compared to single-scale indices (i.e., LHR and PWV mean ) and multiple temporal-scale indices (i.e., MEI SS and MEI LS ). Our findings suggest that diabetic patients with smaller PEI values are more prone to developing DPN, which is of potential importance for application in the area of the point-of-care diagnostic devices.  Acknowledgments: The authors are grateful for the support of Texas Instruments, Taiwan, in sponsoring the MSP tools and assisting in developing novel signal-processing techniques as a contribution to preventive medicine in this study. The authors would like to thank the Guest editor and Reviews for their insightful recommendations, which have contributed greatly to the improvement of this work.

Conflicts of Interest:
The authors declare no conflict of interest.