The Hexoskin study was performed between October 2018 and February 2019, as part of a multicenter, cross-sectional study. The study was registered with the Netherlands Trial Register (NL6934) and was approved by the Twente Medical Ethics Committee. A total of 15 subjects without known pulmonary or cardiac disease and with normal pulmonary function (ie, FEV₁>80% of predicted), between the ages of 18 and 80 years, were recruited at the University of Twente, Enschede, the Netherlands. Subjects were excluded when they were (1) not able to perform the tests, (2) not able to fit into one of the available Hexoskin sizes, (3) had a pacemaker or implantable cardioverter defibrillator, or (4) were unable to read and understand the Dutch language. All subjects gave written informed consent prior to participation.

Baseline characteristics are described by mean (SD) or median (IQR), depending on normality. Normality was assessed by a Shapiro-Wilk test.

A total of 15 healthy subjects were included in this study; no subjects were excluded. Baseline characteristics are presented in Table 1.

In both sessions, out of 15 participants, individuals had to be excluded from further analysis due to artefacts in Hexoskin data regarding the following tasks: bend-over sitting (n=1, 7%), vacuuming (n=3, 20%), walking with weights (n=3, 20%), and stair climbing (n=4, 27%). Data from vacuuming and walking with weights tasks had to be removed from measurements for both subjects 1 and 2. Moreover, subject 2 data also showed large artefacts in the stair climbing task. In total, 50% of the removed data were explained by the removal of data from measurements for two individuals. Regarding the Hexoskin data, TV could not be distinguished from the artefacts and, therefore, we were unable to match TV between the two methods. If more than 50% of the data within a task were unmatchable, they were removed from further analysis. Figure 2 shows an example of the excluded data.

TV was normally distributed. We listed the tasks with the lowest and highest mean TV. In session 1, the mean TV_SPIRO ranged from 0.67 (SD 0.14) L for bend-over sitting to 1.57 (SD 0.48) L for stair climbing. The mean TV_HX ranged from 0.67 (SD 0.13) L for bend-over sitting to 1.57 (SD 0.46) L for stair climbing. In session 2, the mean TV_SPIRO ranged from 0.61 (SD 0.12) L for bend-over sitting to 1.63 (SD 0.49) L for stair climbing. The mean TV_HX ranged from 0.61 (SD 0.21) L for lying to 1.69 (SD 0.44) L for stair climbing.

With Bland-Altman analysis, the bias and LOA were calculated comparing TV_SPIRO and TV_HX in each session; see Figure 3 for an example. The biases found in subjects were nonnormally distributed. The median bias (median LOA) for all subjects per task are given in Table 2. In session 1, all tasks, except for vacuuming, met the criteria for the bias and LOA. In session 2, all tasks, except for vacuuming, met the criteria for LOA. Two tasks met the criteria for the bias.

Due to excessive movement artefacts, as a result of breathing maneuvers, data from 5 subjects in the Hexoskin group had to be excluded from further analysis in both sessions. All lung parameters were normally distributed. A box plot of all functional lung parameters can be found in Figure 4.

With Bland-Altman analysis, the bias and LOA were calculated by comparing the various lung function parameters measured with the spirometer and the Hexoskin. The bias (LOA) of the SVC, FVC, and FEV₁ for each session can be found in Table 3. All Bland-Altman plots have similar features; see Figure 5 for Bland-Altman plots of SVC. Trends were found in all Bland-Altman plots. None of the parameters met the criteria for accuracy, and Hexoskin data showed large deviations from spirometry data. The coefficient of variation was smaller than 10% for SVC and FVC measured with both methods.

All subjects answered both questions. The mean rating for the fit of the shirt was 6 (SD 1). Comfort had a mean score of 8 (SD 2).

In this study, we determined the accuracy of the calibrated Hexoskin to measure TV in comparison to the spirometer in various tasks of daily living. In all tasks of session 1, accuracy of the Hexoskin was good, with biases within the criteria and acceptable LOA. Moreover, reapplicability of the calibration functions was tested in a repeated session. The high median bias in the second session showed that a recalibration was necessary after removal of the Hexoskin. However, LOA were comparable with the first session, which indicates that there was only an accuracy issue and not a precision problem. The Hexoskin is not yet a valid measurement tool to measure SVC, FVC, and FEV₁, based on bias and LOA. Lastly, the comfort level of subjects was high while wearing the Hexoskin.

A priori, we determined that for good accuracy, the bias had to be between –5% and 5%, and the LOA had to be between –15% and 15%. Good accuracy of the Hexoskin in measuring TV was found in the first session. The median bias was found to be within ±0.6%, comparable to a bias of 3 mL on a TV of 0.500 L. Moreover, all LOA were within the ±15% that we determined a priori, with one exception for vacuuming (15.6%). The median biases were all negative, indicating that the Hexoskin overestimated TV slightly, on average.

The LOA found in the various tasks were slightly lower than the LOA in the previously mentioned studies [12,13], with the exception of the study by Clarenbach et al [11]. We hypothesize that the slight variations in outcome were the result of the performed analysis, in following three ways:

More demanding tasks (ie, bend-over sitting, vacuuming, walking with weights, and stair climbing) showed a slightly decreased agreement relative to the tasks at rest. The decrease in agreement was also reflected in the data we had to remove due to excessive artefacts; they were all part of a demanding task. Similarly, in the study by Brüllmann et al [12], agreement was decreased more highly in demanding tasks compared to agreement in a resting position. The increase in LOA from inactivity to more demanding tasks can be explained by the increase (1) in variation in breathing patterns [18,19] and (2) of movement artefacts.

In session 2, we investigated the possibility of reusing calibration factors. Although the median biases were relatively small, in five out of seven tasks they did not meet the criteria. The LOA in the second session were comparable to the LOA found in session 1 in all tasks, with vacuuming as the exception, within 15%. This is in contrast with earlier studies. Heyde et al [20] found that the bias between RIP and the spirometer was similar in repeated measurements. Moreover, Brüllmann et al [12] compared calibration parameters between two measurements and found that the mean difference was zero.

Various reasons for poor accuracy in session 2 can be distinguished, based on poor calibration. The position and fit of the RIP belts on the subjects affect the calibration needed to get accurate results [12,14,15]. Various subjects were in between Hexoskin sizes; because of this, during measurements with these subjects, the Hexoskin tended to crawl up. Therefore, the chance is high that the Hexoskin and, thus, the thoracic and abdominal bands slightly changed positions (1) during measurements and (2) in between measurements. A possible solution to this problem is presented in the study by Brüllmann et al [12]. The LifeShirt had a snug fit but was also affixed to the trousers of the subject with pins to prevent the upward movement of the shirt. Another explanation for not being able to reapply the calibration factors is magnetic coupling of the two belts in the Hexoskin. This appears to occur due to the mutual inductance between chest and abdominal bands modulating the output of the shirt. This is a known problem of RIP sensors [21]. The chance of poor reproducibility is high, as the influence of the bands on each other is not predictable when reapplying the Hexoskin.

A priori, we determined that all lung volumes should be measured with a bias close to zero and LOA equal to or smaller than 0.150 L. None of the lung volumes were measured within these criteria. To the knowledge of the authors, there have been no studies published investigating the feasibility of RIP technology to measure functional spirometry. Poor accuracy in this study can be explained by (1) movement of the shirt, (2) assumption of a linear relationship between circumference and lung volumes, and (3) the use of calibration factors determined from the “sitting” task. The first explanation has already been addressed in the previous section. The second and third argument are related. Based on earlier studies [22], the assumption was made that there is a linear relationship between circumference and lung volumes. However, this assumption is probably true for normal or resting breathing and not for forced maneuvers. We used the calibration of the “sitting” task. In this task, the volumes exhaled were, on average, 0.7 L, while in SVC and FVC maneuvers, volumes above 5 L, on average, were exhaled. This could explain the large deviation between volumes measured by RIP and the spirometer. The linear relationship between the mean and the difference in the Bland-Altman plots (Figure 5) can be explained by the difference in distribution of volumes (Figure 4) measured with both the Hexoskin and the spirometer.

Although we cannot determine the bias and LOA per subject, this would be interesting. Per subject, all SVC and FVC values measured with the Hexoskin seemed to be in very close range to each other (Figure 5). The mean coefficient of variation was comparable in both techniques. For SVC and FVC, the coefficient of variation was <10% for both techniques. This suggests that it is possible to measure functional lung volumes accurately; however, a proper calibration procedure should be defined containing a set with variations in lung volumes (eg, 200 mL to 2000 mL).

In this study, we identified some limitations. In session 1, the same data were used for determination and validation of the calibration factors. We used this construction to minimize the measurement time for the subjects, and to make the calibration procedure realistic and applicable in a clinical setting for patients with pulmonary diseases. However, if the calibration varies over time within a task without removal of the equipment, we should have found larger LOA. Therefore, it is unlikely that the bias found in this study would change to a significant extent if extra validation data were to be included.

In measurement B, the use of the calibration from the “sitting” task could be the reason for poor accuracy of the measured lung volumes. Accuracy could have been decreased if calibration factors were derived from a sample of breaths larger than 70% of SVC or FVC.

We conclude that the Hexoskin accurately measures TV compared to spirometry values if the calibration procedure is used prior to measurement, as was the case in session 1. Based on our analysis, we conclude that 95% of the measured TV_HX values were accurately measured compared to TV_SPIRO. The other 5% of the measurements will not affect outcomes in home monitoring. During telemonitoring, it is common to measure over hours or days. In such a time span, the outliers will not have a significant impact on the average measurement of TV.

However, the implementation of the shirt as a telemonitoring device is not yet feasible. Currently, based on the results in session 2, calibrations have to be performed prior to each measurement. Due to the long calibration time (approximately 30 minutes), this would not be a practical procedure. Modifications to the shirt to deal with the problems of magnetic coupling will probably solve problems with the nonreproducibility of the calibration. Other solutions are to (1) investigate other calibration procedures that are simpler or shorter in nature or (2) look at uncalibrated data. We suggest, for example, a fixed-volume calibration [15], in which a subject breathes from a known volume for several minutes to obtain the calibration factors. In this case, spirometry measurements become unnecessary.

The shirt can be worn with a high level of comfort during measurements. This is a clinically relevant conclusion, as use of the shirt would not be feasible for patient monitoring if the comfort level was rated lower. Moreover, the manufacturer introduced a zipper into the shirt. This will make use of the shirt even more feasible for patient monitoring, as it will make putting on the shirt easier.

The spirometer-calibrated Hexoskin was able to accurately measure TV in healthy subjects during various tasks; however, calibration factors could not be used after removal and reapplication of the Hexoskin. The Hexoskin cannot be used to assess calibrated lung volumes reliably. In the near future, we will evaluate other calibration options to improve the reproducibility of the calibration. Moreover, we will determine the accuracy in patients and investigate the usability of the Hexoskin for home monitoring of patients with pulmonary disease.