Desenvolvimento, validação e aplicabilidade clínica da plataforma sênior saúde móvel para análise da variabilidade de frequência cardíaca em idosos

Abstract

INTRODUCTION: Heart Rate Variability (HRV) has been the subject of study within the field of gerontology in recent years, considering that several studies suggest an intimate relationship between changes in HRV, age, and cardiovascular risk. Studies indicate the need to monitor individuals for prolonged periods, highlighting that the identification of cardiovascular risk is dependent on the measurement time. However, several studies use only short-term records and measurements for HRV analysis due to the difficulties encountered in submitting the individual to the use of devices for 24 hours, such as electrocardiogram and Holter. On the other hand, these short records are often unable to identify risks and adverse health outcomes. In this context, wearable devices, such as smartwatches, have been incorporated as a tool within the perspective of remote, continuous, and longitudinal monitoring, and can be used as a new form of data acquisition for HRV analysis. However, as they use photoplethysmography sensors to obtain heart rate (HR) data, the time series obtained by the sensor may present failures, and missing data are often observed, requiring the development of intelligent models to create HRV indices. OBJECTIVES: #Paper 01: To present the strategies adopted for the development of a software solution, entitled Senior Mobile Health (SMH), for HRV analysis, based on a photoplethysmography sensor embedded in a commercial smartwatch. #Paper 02: To validate the Senior Mobile Health software for HRV analysis through a study with 231 participants. In addition, the paper aims to present the architecture and main functionalities of the Senior Mobile Health platform. #Paper 03: To establish HRV reference values in community-dwelling Brazilian elderly through a solution developed using a smartwatch. METHODS: #Paper 01: Seeking to develop a software solution for HRV analysis for a smartwatch, the following techniques were mainly used: (1) interpolation to fill in the missing data in the time series; (2) modeling of the HRV algorithm, to approximate the data obtained by smartwatches with the data obtained by the gold standard (Polar H10) through several smart models; (3) methods for removing artifacts and inconsistencies from the series; (4) statistical tests to analyze the difference between the HRV values obtained by the solution created in this research and the values obtained by the gold standard. #Paper 02: The second paper presents the validation process of the software solution developed in paper 01. For this, 231 individuals over 60 years of age were recruited and asked to use a smartwatch on their non-dominant arm for 24 hours. The time series extracted from the smartwatch was analyzed using the software solution (SMH0 developed and presented in paper 01 and using the Kubios® software. The values obtained through the two tools were compared and through the Wilcoxon test, it was observed the existence or not of statistically significant differences, between the SSM and Kubios, for each of the HRV indices, establishing the significance level of α 0,05. Then, Bland-Altman graphs were constructed to calculate the limits of agreement between the two methods (SSM and Kubios®). Finally, to evaluate the HRV metrics resulting from the SSM and measure how far the predicted values were from the reference value, that is, those generated by Kubios, we calculated the MSE, defined as the mean squared difference between actual values and the estimated values. #Paper 03: It has a study with 220 community-dwelling seniors who used a smartwatch for 24 hours. The time series was extracted and the HRV algorithm developed in this thesis was applied to propose HRV reference values in individuals over 60 years old, divided into age groups, using a smartwatch. RESULTS: #Paper 01: To fill in the missing data of the time series, several techniques were applied, and it was possible to observe that Piecewise Cubic Hermite Interpolating Polynomial (PCHIP) presented a lower error measured by the Root Mean Square Error (RMSE) when compared to the other selected techniques. For the process of modeling the HRV algorithm with the objective of generating the approximation of the data obtained by Fitbit with the data of the gold standard (Polar H10), several intelligent models were applied. The intelligent model that presented the lowest RMSE was the Long Short-Term Memory (LSTM) neural network, showing good long-term learning ability. To remove artifacts and inconsistencies from the created time series, the Kamath method was applied, since this method provided the results most similar to those of Kubios®. Each step of the workflow was validated individually. And, finally, the evaluation of the process was carried out using the paired Student's t-test to determine whether the paired observations (software developed and data obtained by the Kubios® software) are significantly different from each other. As a result, it was possible to observe that there is no statistically significant difference between the HRV indices generated by the solution developed in the research and the indices generated by Kubios®. #Paper 02: The results of the Wilcoxon test showed that there was no statistically significant difference between the data obtained by the SSM software developed in this research and the data generated by the Kubios® software for all HRV indices (SDNN: p = 0.08; RMSSD: p = 0.59; pNN50: p = 0.53). The graphic results of the Bland Altman test, for the three HRV indices, showed the means of the differences close to zero and the narrow limits, thus resulting that the data evaluated by the two methods are essentially equivalent. Finalizing the validation process, we applied the MSE to evaluate the differences between the values predicted by the algorithm and the real values generated by Kubios®. The obtained MSE was 0.256 for SDNN, 0.211 for RMSSD, and 0.191 for pNN50. A small MSE value indicates that our model adequately fits the gold standard data. In short, the results found suggest that there is no significant difference between the SMH platform compared to Kubios®. #Paper 03: The reference values of mean and standard deviation observed for men were: SDNN 127(15), RMSSD 24(7.7), and pNN50 4.1(3.5), while the following values were observed for women: SDNN 111(36.6), RMSSD 19(6.8) and pNN50.4(2.9). In this paper, statistically significant differences were observed in the HRV indices between men and women, with men presenting higher HRV indices when compared to women. Regarding the change in HRV over age, women showed a significant increase in parasympathetic indices in the age group of 70 to 79 years. Through correlation analyses, it was possible to observe a strong correlation between heart rate at rest and all HRV indices in men and women. When observing the correlation between mean heart rate and HRV indices, only women maintain strong correlations. Regarding the physical activity variables, a strong and positive correlation was found between the number of steps and the SDNN among men and a moderate and positive correlation for women. CONCLUSIONS: We conclude that: #Paper 01: The software solution developed for HRV, through the techniques and methods adopted, did not present statistically significant differences with the Polar H10, adopted here as the gold standard and real value to be achieved. Thus, the values predicted by our solution showed a good fit with the target values. #Paper 02: The SSM platform can provide reliable results associated with HRV metrics and therefore healthcare professionals can incorporate this platform during the remote monitoring process of their patients for HRV follow-up. #Paper 03: An increase in HRV values in parasympathetic indices was observed mainly among women aged 70 to 79 years, indicating that, in this period, among females, there is an increase in vagal activity, a cardioprotective mechanism against cardiovascular problems. We also highlight the strong correlation found between HRV and HRV in both men and women, however, when observing the correlation between average HR and HRV, the correlation becomes weak in the men's group, however, it remains strong in the women's group.