Dec 05, 2023
ウェアラブル活動量センサーを活用し、日々の歩数を超えた身体の健康状態を評価
npj Digital Medicine volume 5、記事番号: 164 (2022) この記事を引用する 5201 アクセス 1 引用 136 Altmetric メトリクスの詳細 身体的健康状態は、個人の実行能力を定義します
npj デジタルメディスン 第 5 巻、記事番号: 164 (2022) この記事を引用
5201 アクセス
1 引用
136 オルトメトリック
メトリクスの詳細
身体的健康状態は、日常生活の通常の活動を実行する個人の能力を定義し、通常、臨床現場でアンケートおよび/または時間制限歩行テストなどの検証されたテストによって評価されます。 これらの測定の情報内容は比較的低く、通常は頻度が制限されています。 活動量モニターなどのウェアラブル センサーを使用すると、身体活動に関連するパラメーターの遠隔測定が可能になりますが、毎日の歩数の測定以外には広く研究されていません。 ここでは、2 回の来院(18.4 ± 12.2 週間)の間に Fitbit 活動量モニター (Fitbit Charge HR®) を提供された肺動脈性高血圧症 (PAH) 患者 22 人のコホートからの結果を報告します。 各臨床訪問で、最大 26 件の測定値が記録されました (カテゴリ 19 件、連続 7 件)。 分ごとの歩数と心拍数の分析から、身体活動と心血管機能に関連するいくつかの指標を導き出します。 これらの指標は、コホート内のサブグループを特定し、臨床パラメーターと比較するために使用されます。 いくつかの Fitbit 指標は、継続的な臨床パラメーターと強く相関しています。 閾値処理アプローチを使用して、多くの Fitbit メトリックがサブグループ間の臨床パラメーター (身体状態、心血管機能、肺機能、血液検査のバイオマーカーに関連するパラメーターを含む) に統計的に有意な差が生じることを示します。 これらの結果は、毎日の歩数がアクティビティ モニターから導き出せる多くの指標のうちの 1 つにすぎないという事実を強調しています。
ウェアラブル活動センサーを使用すると、個人の身体活動を遠隔監視できますが、これまでは 1 日の平均歩数の評価にほとんど限定されていました。 歩行または歩行は日常生活の基本的な動作であり、人間の健康を促進するための重要な指標となっています1。 たとえば、毎日の歩数の増加(4,000 歩未満から 12,000 歩以上)は、全死因死亡率の減少と関連しています 2,3。 入院患者の場合、1 日の歩数の閾値(通常、1 日あたり 1000 歩未満)は、再入院などの不良転帰と関連しています4、5、6。 歩行速度 7、8、9 や時間制限歩行テスト 10、11 などの関連する歩行パラメータも、臨床的に関連する結果を予測することがわかっています。
歴史的に、個人の身体状態を遠隔監視することは困難でしたが、ウェアラブル技術の進歩により、手術後、または慢性疾患患者の通院の合間に継続的に評価できるようになりました。 Fitbit デバイスなどのウェアラブル慣性測定ユニット (IMU) は、関連するスマートフォン アプリで表示できる、IMU 信号から得られる他の指標 (睡眠など) とともに歩数を記録します。 さらに、Fitbit などの多くのウェアラブル デバイスは、光電脈波記録法を使用して心拍数を測定します。
歩数、特に毎日の歩数は、依然として身体活動のリモート評価の最も一般的な指標ですが、分ごとの歩数と心拍数のデータは、アプリケーション プログラミング インターフェイス (API) を使用して Fitbit サーバーからダウンロードできます。 したがって、Fibit を継続的に装着している人の場合、1 週間で 10,080 件の歩数 (単位: 1 分あたりの歩数、SPM) と心拍数 (単位: 1 分あたりの拍数、BPM) の値を取得でき、各ポイントは平均値を表します。その分間の歩数と心拍数。 自由生活環境や非定型的な歩行パターンを持つ患者集団における歩数測定の精度には依然として懸念が残っていますが 12,13、がん、心血管疾患、肺動脈性肺高血圧症、多発性硬化症患者を対象とした研究では、これらのデバイスが正確かつ正確な歩数測定を提供できることが示唆されています。臨床関連データ14、15、16、17。 同様に、比較研究では、Fitbit デバイスからの心拍数測定値は一般に、安静時または活動レベルが低い個人の心電図とよく一致していることが示されています 18,19。 ただし、皮膚の色素沈着などの他の要因も測定精度に影響を与える可能性があります20。
0. Red line shows a normal fit. d Weekly activity map: scatter plot showing heart rate versus step rate. Each point represents one minute where a physiological heart rate was recorded. The grey lines show the upper and lower envelopes of the activity map. The blue line shows a linear least squares fit to the data./p>5000 steps (14/22) to those with <5000 steps (8/22). This arbitrary threshold resulted in 6 statistically significant clinical parameters (Supplementary Table 2 and Supplementary Fig. 2). Subjects with <5000 steps per day had lower 6MWD at visit 1, lower hemoglobin levels at visit 2, poorer pulmonary health (higher physician-assessed WHO FC) at visit 1, and experienced more pedal edema (Pedal Edema score) at visit 2. Two subjects had average daily step counts >10,000 steps per day (PAH 1, 19), but had no other similarities. Sensitivity analysis of threshold values and the number of statistically significant clinical parameters for all Fitbit metrics are provided in Supplementary Figs. 3 and 4./p> 0 (HR(SR = 0), i.e. active). Histograms for HR(SR = 0) (Fig. 1b) and HR(SR > 0) (Fig. 1c) were described by normal distributions, from which we obtained the mean, standard deviation, and skewness. The range of mean HR(SR = 0) was 66.2–111.8 BPM, with standard deviations of 6.4–13.7 BPM (Supplementary Fig. 5). The skewness varied from −0.75 to 2.30, highlighting a broad range of behavior with relatively large tails to the left and right of the peak (Supplementary Figs. 6 and 7)./p>82 BPM (8/22). This resulted in 8 statistically significant clinical parameters (Supplementary Table 2 and Supplementary Fig. 8). Subjects with lower mean values of HR(SR = 0) had lower RHR at visits 1 and 2, and lower peak heart rate at visit 2, but experienced more pedal edema (Pedal Edema score) and more palpitations (Palpitation score) at visit 1, were less able to perform usual activities (lower EQ-5D Usual Activity scores) at visit 1, and experienced more pain/discomfort (lower EQ-5D pain/discomfort scores) at visit 1./p>100 BPM. Both subjects had low fitness slopes (see below), suggesting that they did not access a wide range of heart rate during daily activities. However, PAH 1 had the highest average daily step count in the dataset. We note that 3 subjects (PAH 4, 20, 27) removed the device overnight (see below), which may have resulted in higher mean HR(SR = 0) values since heart rate values during sleeping were likely not included./p>90,000 individuals over 35 weeks, reported that the RHR (assumed to be the true RHR) was dependent on age, BMI and sleep duration, with daily values of RHR from 40–108 BPM25, although 95% of men and women had RHR values between 50–80 BPM, similar to the range found here./p>1. This resulted in 4 statistically significant clinical parameters (Supplementary Table 2 and Supplementary Fig. 9). Subjects with lower skewness values were more likely to have higher resting heart rate at visits 1 and 2, experienced less pain/discomfort (lower EQ-5D pain/discomfort scores) at visit 1 and were more likely to be in better health (higher EQ-5D Index) at visit 1. Two subjects had skewness of HR(SR = 0) values >1.9 (PAH 27, 28): both subjects also had relatively low resting heart rates, longer free-living 6MWD, and higher fitness plot slopes./p> 0 represents HR values while subjects were active. The mean values of HR(SR > 0) were 78.6–121.0 BPM (mean 94.4 BPM), and the standard deviation was 6.5–14.0 BPM (Supplementary Fig. 10). The mean values were only slightly higher than the mean values of HR(SR = 0), although the standard deviations were similar. The mean skewness values for HR(SR > 0) were from −0.57 to 1.35, similar to the range for HR(SR = 0). We compared individuals with mean values of HR(SR > 0) <95 BPM (12/22) to those with >95 BPM, resulting in 4 statistically significant clinical parameters (Supplementary Table 2 and Supplementary Fig. 11). Subjects with lower mean values of HR(SR > 0) had lower RHR at visits 1 and 2, lower albumin levels at visit 1, and experienced more palpitations (lower Palpitation score) at visit 1./p> 0, the mean HR at SR = 0, and the fraction of time inactive (Fig. 2a). The data points for each week for most subjects were tightly clustered in distinct regions. From the loading plot (Fig. 2b), PC1 is dominated by the step rate parameters (+PC1) and the fraction of time inactive (−PC1). PC2 is dominated by the mean heart rate at SR = 0 (+PC2) and the standard deviation of the heart rate for SR > 0 (−PC2). The group of subjects in the fourth quadrant (PAH 3, 9, 12, 19, 23, 27) are characterized by high mean and standard deviation of the step rate, and a high value of the standard deviation of the heart rate at SR > 0. This implies that these individuals exhibit a wide range of step rates and a wide range of heart rates during normal activities of daily life. The group of subjects along the positive y-axis (PAH 1, 10, 14, 17) are characterized by high mean heart rate at SR = 0. High values of HR(SR = 0) imply that these individuals have a high resting heart rate and are unlikely to access a wide range of heart rates during normal activities, even if they have the capacity for moderate or high step rates. The group of subjects along the negative x-axis (PAH 2, 7, 11, 13, 20, 21, 30) are characterized by a large fraction of time inactive. Three subjects (PAH 15, 26, 28) are clustered around the origin. The PCA plot suggests a range of behavior with distinct combinations of metrics associated with heart rate and step rate. To explore these relationships in more detail, we assessed several derived parameters. Distinct groupings of subjects were found for mean HR(SR = 0) >82 BPM, skewness of HR(SR = 0) <1, ambulation product, P > 1000, and fitness slope >0.15 (Supplementary Fig. 12)./p> 0):SD is the standard deviation of the heart rate at SR > 0; SR(SR > 0):mean is the mean step count for SR>0; SR(SR>0):SD is the standard deviation of the step rate for SR > 0; time active is the fraction of minutes with SR = 0./p>0.15 (11/22) to those with slope <0.15, resulted in 3 statistically significant clinical parameters (Supplementary Table 2 and Supplementary Fig. 13). Notably, subjects with slopes >0.15 had lower NT-proBNP levels at visits 1 and 2. B-type natriuretic peptide (BNP) and N-terminal pro b-type natriuretic peptide (NT-proBNP) are biomarkers for cardiac stress, and PAH patients with NT-proBNP levels below about 300 pg L−1 are considered low risk for heart failure26. The mean levels for subjects with slope >0.15 at visits 1 and 2 were 188 ± 180 and 145 ± 165 pg mL−1, respectively. These results suggest that the fitness slope may be a useful indicator of NT-proBNP levels and risk for heart failure. Comparison of subjects with fitness intercepts above (10/22) and below (12/22) the mean (91 BPM) were similar to results for subgroups with HR(SR = 0) above and below 95 BPM./p> 1000 (12/22) to those with P < 1000, resulted in 7 statistically significant clinical parameters (Supplementary Table 2 and Supplementary Fig. 14). An ambulation product value of 1000 was selected as it was close to the median value (1079), and represented a well-defined separation between the two groups (Fig. 4d). Subjects with P < 1000 had lower 6MWD at visits 1 and 2, and experienced more pedal edema (Pedal Edema score) at visit 1. Two subjects had ambulation product values > 5000 (PAH 9, 19). Both subjects had a high ambulation frequency and walked more than 5000 steps per day on average. Both subjects also had relatively lower resting heart rates, longer free-living 6MWD (see below), and higher fitness plot slopes. PAH 1, despite having the highest step count, ranked fourth in ambulation product value as a result of having relatively lower endurance and intensity values./p> 0) for analysis. In this study the average weekly usage was 0.44–0.97. Note that charging the device overnight (e.g. 8 h) once a week results in a weekly usage of 0.95. We also defined the maximum off-time as the longest continuous time during the week that the device was not worn, which varied from less than 1 h to more than 12 h. From heat maps of usage and the maximum off-times for all subjects (Supplementary Figs. 15 and 16) we can further infer how the device was used./p> 0. Yellow cells indicate that the device was worn continuously for the full hour. White cells indicate that the device was not worn (no HR recorded) for the full hour. a Heat map for PAH27 (13 weeks of data), showing low usage (average = 0.49) with the device not worn overnight. b The maximum off time for each week for PAH27 is consistently around 12 h overnight. Each point represents the maximum off-time for each week in the trial. c Heat map for PAH30 (22 weeks of data), showing relatively high usage (0.90), with the device removed for several hours every few days. d The maximum off time for PAH30 is typically 8–20 h and includes overnight hours. e Heat map for PAH10 (13 weeks of data), showing high usage (0.97). For the first 10 weeks the maximum off-time is less than 1 h. f The maximum off time for PAH10 is usually less than 1 h./p>0.94, which corresponds approximately to the 75th percentile. Comparison of usage, resulted in 4 statistically significant clinical parameters (Supplementary Table 2 and Supplementary Fig. 18). Subjects with average weekly usage < 0.94 (15/22) were more likely to have more severe PAH (higher EQ VAS score) at visit 1, worse pulmonary health (higher physician assessed WHO FC score) at visit 1, and experienced more difficulty breathing (modified Borg dyspnea score) at visit 2. Two subjects had average usage < 0.5 (PAH 4, 27), however, both of these subjects removed the device overnight. The third subject who removed the device overnight (PAH 20) also had low average usage (0.60). (Changes in device usage over time are summarized in Supplementary Figs. 19 and 20)./p>320 m (PAH1, 3, 9, 10, 11, 12, 14, 17, 19, 22, 23, 26, 27, 28). Comparison of FL6MWD resulted in 6 statistically significant clinical parameters (Supplementary Table 2 and Supplementary Fig. 23). Notably, subjects with average FL6MWD < 320 m had lower 6MWD at visit 1 and visit 2, experienced more pedal edema (Pedal Edema score) at visit 2, had worse pulmonary health (higher physician-assessed WHO FC) at visit 1, and had lower hemoglobin at visit 2./p> 480 m (PAH3, 23). These subjects were in the fourth quadrant of the PCA plot, which implies that they had a wide range of step rates and heart rates during normal weekly activity, and had ambulation product P values > 1000./p> 400 m (12/22) had higher 6MWD at visit 2, lower NTpro-BNP at visit 2, experienced less chest pain (Angina score) at visit 1, and had better pulmonary health (lower physician-assessed WHO FC) at visit 2 (Supplementary Table 2 and Supplementary Fig. 24)./p>4.0 m/week) (PAH3, 10, 20), and four subjects had a large negative slope (<4.0 m/week) (PAH1, 13, 21, 23)./p> 1 but, as described previously, this subject recorded high FL6MWD values during the first 13 weeks, but then maintained a much lower value in subsequent weeks. It is evident that there is no correlation between the FL6MWD in week 1 and the predicted 6MWD (H6MWD) for an equivalent healthy individual (Fig. 7a)./p> 0. Three subjects (PAH 30, 2, 20, 11) had health state values below 0.52 in their first and last weeks. These subjects were located along the negative x-axis of the PCA plot, characterized by a large fraction of time inactive./p> 0), ambulation P value, fitness slope. Based on the maximum Bayesian Information Criterion (BIC) (Supplementary Table 3), the subjects were categorized into three groups (Supplementary Fig. 28). Group 1 had high ambulation metrics (steps/day, ambulation product P, and FL6MWD), high HR(SR > 0), and high fitness slope (Supplementary Table 4). Group 2 were characterized by the lowest ambulation metrics (steps/day, ambulation product P, FL6MWD), the lowest HR(SR = 0) and HR(SR > 0), and the highest HR(SR = 0)sk. Group 3 had the highest HR(SR = 0) and HR(SR > 0), the lowest HR(SR = 0)sk and fitness slope. The three groups identified from LPA analysis occupied distinct regions of the PCA plot, with the exception of PAH 10 who was in Group 2 (Supplementary Fig. 29)./p> ±0.5). Albumin was correlated with HR(SR = 0) and HR(SR > 0) at visit 1 (r = 0.565 and 0.627, respectively). NT-proBNP was also correlated with HR(SR = 0) at visit 1 (r = 0.585), and was inversely correlated with fitness slope at visit 1 (r = −0.585). RHR at visits 1 and 2 were correlated with HR(SR = 0), HR(SR = 0)sk, and HR(SR > 0). 6MWD at visits 1 and 2 were correlated with FL6MWD. RVSP at visit 1 was inversely correlated with fitness slope. Notably, steps/day and ambulation P did not have strong correlations to the continuous clinical parameters./p> 0.15 had lower NT-proBNP levels, an important biomarker of cardiac stress, at visits 1 and 2. In addition, this approach may contribute to identification of individuals who would benefit from more frequent clinic visits or specific medications./p> 0), i.e. active). From the distributions of these three metrics we obtained the mean, standard deviation, and the skewness. The heart rates were fit to a normal distribution, and the step rate was fit to a log normal distribution. A scatter plot of step rate versus heart rate provided a weekly signature of cardiovascular activity for each individual. From a linear least-squares fit to the data we obtained the slope (heart rate per step rate (BPM/SPM)). The effective area of the heart rate versus step rate (HR vs. SR) plot was determined by first calculating the upper (lower) envelopes. Each point in the upper and lower envelopes represents the average of the maximum (or minimum) HR values at each value of step count in a bin width of 10 SPM. The envelope point is located at the average step rate for all values with HR values. Step rates with no HR values are omitted from the calculation. Bins with no HR values do not have an envelope point. We then performed a linear least-squares fit to the envelopes to determine the area of the HR-SC plot./p> \,1.0\) is considered large./p> 0):SD, SR(SR > 0):mean, SR(SR > 0):SD, time inactive (fraction of minutes with SR = 0). These parameters were selected to represent heart rate and ambulation metrics and to avoid redundancy. For each parameter we used the average weekly value. The variance for the first two principal components were 48.6% and 30.0%, respectively. For 100 independent runs where we randomly selected different weeks, the mean variance of PC1 and PC2 was 77.5 ± 0.58%./p> 0), ambulation P value, fitness slope, FL6MWD, and usage. LPA was performed through package ‘mclust’ (version 5.4.10) in R (version 4.2.1). The optimal number of clusters was determined based on the maximum Bayesian Information Criterion (BIC) through the function ‘mclustBIC’./p>