Device Description – SimpleSense-BP

The SimpleSense-BP Software Application is a software module within the SimpleSense Platform that accesses the physiological parameters like ECG, heart sounds, and thoracic impedance captured by the FDA-cleared SimpleSense Platform [7] (K212160) for processing into the vital sign outputs of the product. These outputs are returned to the SimpleSense Mobile Application or SimpleSense Web Application user interface (UI) for display and review by medical professionals. No manual operations are required by medical professionals for the device to function. Figure 1 presents the SimpleSense platform, including the hardware product, mobile app, and web interface.

Fig. 1

SimpleSense-BP includes the following four software modules – (1) An input type, format, and data validation module to assess if the inputs supplied to SimpleSense-BP meet a pre-specified requirement by design. (2) A signal quality assessment is performed on the input data from the SimpleSense device. Only signals with adequate quality for further computation are used. (3) A feature extraction module that extracts a predefined set of features from the SimpleSense data and merges the data with the user's demographic information. The device-derived parameters are inclusive, but not limited to, Time between R peak (ECG) and S1 (Heart sounds), time between R peak (ECG) and S2 (Heart sounds), ECG morphological analysis (ECG), S1 and S2 root mean square amplitudes, ratio of S1 and S2 root mean square amplitudes (Heart sounds), R-R intervals (ECG), respiratory rate (thoracic impedance), posture and activity. The demographics data includes – age, gender, height, and weight. Finally, (4) a model for systolic and diastolic blood pressure that takes as input the set of features extracted by the feature extraction module along with the demographics data and returns an estimate of the systolic and diastolic blood pressure associated with that segment of SimpleSense device data which have been shown to have known associations to blood pressure [6].

Study DesignSubject Population

A prospective, multicenter, nonrandomized, observational study (NCT05473702) was conducted in Pennsylvania, New York, and New Jersey in the USA by Nanowear Inc. R&D and ClinCept LLC, Columbus, Georgia, USA in Georgia.

Subjects from the general population in the three geographic regions were screened and enrolled to meet the inclusion and exclusion criteria, as presented in Table 1. The study was conducted with three phases of enrollment with the same enrollment criteria. For the first arm, at least 85 subjects were required from Pennsylvania and Georgia for training and validation and New York and New Jersey for testing.

Table 1 Subject Population enrollment criteria and BP clinical stratification boundaries

For the second arm, to ensure models operate accurately over a broad range of blood pressures, both within the specified range of the original model chosen by baseline reference value and outside the range of the baseline model, Nanowear enrolled an additional cohort of subjects. The enrollment was done until at least 10 subjects had a change in BP, meeting the criteria of at least 15 mmHg systolic or 10 mmHg diastolic for each of the 4 models used by the device to capture real-world examples and occurrences of patients who undergo such changes. The enrollment aimed to capture both induced and physiological changes in BP, ensuring a comprehensive evaluation of the device's performance.

For the third arm, to ensure the device maintains its accuracy for the labeled period of calibration, Nanowear enrolled another cohort of subjects, and performance evaluation was performed with measurements at baseline, at the end, and at an intermediate time point(s). Subjects were enrolled following IEEE 1708 specifications of at least 85 subjects, and at least 21 subjects in each clinical stratification of Normal, Prehypertension, Stage 1 hypertension, and stage 2 hypertension. 4 measurements after the initial calibration measurement shall be made for each enrolled subject, separated by 7 days to cover an evaluation period of 21 days. The final enrollment totals and subject characteristics for all arms are described in Sect. 3.1.

Blood Pressure Measurement Procedure

Two independent nurse practitioner observers performed all measurements using the SimpleSense device and each subject's gold standard sphygmomanometer. Endpoints for BP classification were determined according to the 2014 eighth JNC report [8]. Baseline blood pressure measured at the start of the test was considered for classification. Three measurements were made with subjects seated in a chair with elbows resting on the armrest, back upright against the backrest, and feet flat on the floor. The average of the three measurements was used to establish the clinical stratification of the subjects. The participants were required to wear the SimpleSense device, while the two trained observers used a sphygmomanometer and an Omron blood pressure cuff GUDID 10073796266353 to measure their blood pressure sequentially with no more than 90 s between measurement initiations to limit the effect of natural time-dependent BP variability and no less than 60 s to allow recovery of the brachial artery in the arm. The data collected simultaneously and in sync by SimpleSense was utilized to train models to estimate blood pressure using SimpleSense data and demographics information. Figure 2 provides an illustration of the study procedure for development and validation.

Fig. 2

Diagram of the overall study conducted for the development and validations of SimpleSense-BP

Nanowear presented the details of the procedure in a previously published article [6], where we followed the most recent recommendations available at the time [9]. To summarize, study participants were asked to perform blood pressure-modifying activities to raise the SBP and DBP dynamic range based on suggested methods in the literature [10]. Subjects sat on a chair or stool with their legs up and held a warm heat pad wrapped in an insulating cloth for 10 min to lower their blood pressure. Subjects were then asked to walk briskly to raise their blood pressure (light exercise) for 10 min. Subjects were then given an ice pack to hold in their hand for about 10 min, called cold pressor stimulus, to increase their BP. After each activity, the observers recorded three consecutive blood pressure readings or 24 total sphygmomanometer readings and 12 Omron readings per patient over the 90-min recording period. The results and how the changes were distributed are shown in the supplementary materials. We now expand on this work with additional test populations and sub-group analysis. Further, we have updated the algorithm to use hypertension diagnosis specific models for each class of hypertension to potentially find diagnosis specific relationships between SimpleSense data and blood pressure.

Data Preparation and Analysis

A 60-s-long segment of data from SimpleSense that corresponds with the observer-entered sphygmomanometer blood pressure was extracted for all the subjects. This SimpleSense data was combined with the sphygmomanometer readings of systolic and diastolic blood pressure from the observers to form the data sets for training and testing. Each segment was subjected to a data quality assessment, and segments of data of insufficient quality due to noise were removed from further consideration, a feature already existent from the SimpleSense Platform [7]. From the recorded data that has been deemed to be of acceptable quality, a dataset was prepared with the 60-s-long segments of SimpleSense device data and the associated target systolic and diastolic blood pressure values for training the SimpleSense-BP algorithm.

The Pennsylvania and Georgia datasets were used as training, and the data sets from New York and New Jersey were sequestered as independent test validation data of the AI-based algorithm.

Accuracy, as defined by the Institute of Electrical and Electronics Engineers (IEEE) 1708:2014 (including the 2019 amendment)[11] and the International Standards Organization (ISO) 81,060–2: 2018 standards[12], were used to evaluate the accuracy of estimated blood pressure. Following the recommendations to induce change by the IEEE 1708 standard, study participants performed activities that can modulate blood pressure for up to 90 min.

For the first and third arm, Mean Absolute Difference (MAD) (Equation S1), Mean Difference (MD) (Equation S2), and Standard deviation of differences (SD) (Equation S3) are used to analyze the performance of the models. The statistical aspect of the criteria is discussed in IEEE 1708-2019a and in ISO 81060–2. Additionally, the MAD value for the systolic and diastolic pressures were also computed using bootstrap with resampling. 1000 iterations were performed, taking samples equal to the available test set observations with replacement to arrive at the bootstrap mean MAD and the 95% confidence interval.

where \(_\) is the test device measurement, \(_\) is the average of the adjacent two reference measurements taken before and after device measurement as defined in ISO 81060–2:2018, and \(n\) is the data size.

To report the accuracy of the device, mean absolute difference (MAD), mean difference (MD), and standard deviation of difference (SD) were used. The MAD, MD, and SD are reported with data binned in different patient-specific parameters to evaluate the effect of potential confounders. Bootstrap resampling was conducted to calculate the MAD values for lower 95% and upper 95% confidence intervals of the MAD for the Systolic and Diastolic BP model. A pre-specified limit for MAD of no greater than 6 mm Hg [11], and MD ≤ ± 5 mmHg and SD ≤ 8 mmHg [12] was considered acceptable based on established validation practices for blood pressure measurement devices in the field.

For the second arm, three measurements for each condition were averaged and not averaged for error performance calculation. Both results are presented. We performed two sub-analyses – Sub analyses 1 and 2 as described in Sects. 2.3.1 and 2.3.2. Overall performance as primary analysis and performance in each clinical stratification as secondary analysis were computed for both sub-analyses.

In the supplementary materials, the systolic and diastolic values are graphically plotted as:

Bland–Altman Scatter plots of the measurement differences between the test device and reference measurement vs. the average of them, along with the limits of agreement.

Scatter plots of the test device measurements vs. the reference device measurements with slope and intercept of the fit line and the confidence intervals of the slope and intercept values.

Assessment of Device Performance in Tracking BP Recovery After an Induced Change

This sub-analysis aims to assess the device's performance once the stimulus for inducing BP increase is removed and there is a real-world potential for BP to return to baseline value—this analysis measures the device's accuracy when there is a significant decrease in blood pressure. Specifically, this sub-analysis reflects performance evaluation at the change point (an increase in BP of 15 mmHg systolic or 10 mmHg diastolic) and following change, which is error performance of recalibrated estimates on Variation 2 and rest. Measurements included in the analysis are depicted in Figure S5a. In another analysis, across the entire test population, all significant changes of SBP ≥ ± 15 mm Hg and DBP ≥ ± 10 mm Hg regardless of whether stimulus induced were evaluated with the device recalibrated prior to observed change using sphygmomanometer and performance evaluated on the measurements after the change was observed as depicted in Figure S5b.

Overall Assessment of Device Performance Inclusive of Ability to Track Induced Change and Recovery of BP

Overall performance was evaluated on Static, Variation 1 (BP lowering using warm stimulus and leg raising), Variation 2 (BP increase using brisk walk and cold pressor stimulus), recalibrated estimates on Variation 2, and rest after recalibration. Only those subjects who exhibited an SBP increase of ≥ 15 mm Hg or DBP increase ≥ 10 mmHg due to the application of cold pressor and brisk walk stimulus were included in this performance evaluation. Measurements included in the analysis are depicted in Figure S6a. In another analysis, across the entire test population, all significant changes of SBP ≥ ± 15 mm Hg and DBP ≥ ± 10 mm Hg regardless of whether stimulus induced, nominal changes below the significant change threshold and overall performance across all measurements are compared. Measurements included in the analysis are depicted in Figure S6b.

Blood Pressure Algorithm Validation for the Duration of Validity of an Initial Calibration

To ensure the device maintains its accuracy for the labeled period of calibration, Nanowear enrolled an additional cohort of subjects, and performance evaluation was performed with measurements at baseline, at the end, and at an intermediate time point(s). Subjects were enrolled following IEEE 1708 specifications of at least 85 subjects and at least 21 subjects in each clinical stratification of Normal, Prehypertension, Stage 1 hypertension, and stage 2 hypertension.

4 measurements after the initial calibration measurement were taken for each enrolled subject, separated by 7 days to cover an evaluation period of 21 days.