Abstract
This study evaluated the performance of a dual-sensor cardiopulmonary resuscitation (CPR) feedback system in estimating chest compression depth and rate over a range of depth and rate combinations during rigid and compliant surface conditions using a computer-controlled motion system to simulate chest compressions. Ten dual-sensor CPR pads were tested using a computer-controlled motion system which simulated chest compressions at target depths of 1.9, 3.8, 4.8, 6.4, and 8.9 cm and target rates of 60, 80, 100, 120, and 140 compressions per minute (cpm). A rigid surface was simulated by applying motion only to the anterior sensor, and a compliant surface was simulated by applying motion to both the anterior and posterior sensor, challenging the algorithm to calculate a net compression depth by subtracting motion of the posterior sensor. For all simulated compressions, including every rate and depth combination and for both rigid and compliant surface simulations, the mean (±sd) depth error was 0.05 (±0.08) cm and the rate error was −0.55 (±1.44) cpm. A dual-sensor CPR system accurately estimates compression depth within ±0.25 cm and compression rate within ±3 cpm over a wide range and combination of clinically relevant chest compression depths and rates during both rigid and compliant surface simulations. Use of a computer-controlled motion system provides a more direct assessment of accuracy than manual compressions performed on instrumented manikins.
Introduction
Cardiac arrest is a life-threating condition that occurs from sudden cessation of cardiac function and a resulting loss of blood circulation. Cardiopulmonary resuscitation (CPR) is a procedure used by rescuers that comprises chest compressions, ventilation, and when indicated, defibrillation shocks, in an effort to restore spontaneous circulation. Chest compression depth and rate are important variables associated with increased shock success and improved patient outcomes following resuscitation from cardiac arrest [1–4]. Chest compression quality is often suboptimal during both in-hospital [5] and out-of-hospital [6] cardiac arrests; however, providing real-time feedback to rescuers improves guideline compliance [1,7–10].
Current guidelines established by the International Liaison Committee on Resuscitation call for chest compressions on adults to be performed at a depth of at least 5 cm, while avoiding excessive depths greater than 6 cm, and at a rate of 100–120 compressions per minute (cpm) [11]. For pediatric patients, the recommended target is a patient-specific depth of at least one third of the anterior–posterior chest diameter, resulting in targets of approximately 4 cm for infants, and 5 cm for children, at rates prescribed for adults [12]. Given the relatively narrow target range of 1 cm for compression depth in adults and the shallow depths recommended for pediatric cases, the ability to accurately estimate depth for each compression in real-time is required for a rescuer to achieve guideline recommended targets.
The ability to accurately estimate compression depth depends on the environment in which compressions are being provided and the type of sensor system being used. A single sensor placed on the sternum, for example, can be accurate within 2–3 mm if chest compressions are performed on a rigid, noncompliant surface, such as the floor [13,14]. Scenarios arise, however, where compressions are performed with concurrent movement of the entire thorax, such as when compressions are performed on a compliant surface such as a mattress, during transport, or in pediatric cases where body movement may differ depending on the technique used by the rescuer (e.g., two-finger versus two-thumb with encircling hands). Such scenarios can introduce significant error in estimating depth when using single-sensor systems, which may lead to rescuers performing compressions outside of target depths [15]. Studies using systems that incorporate dual accelerometers or other technologies have demonstrated the ability to correct for this [13,14,16–18]. However, as these studies used volunteers to perform manual compressions within or near guideline recommendations for adults, a question remains in the literature regarding the accuracy of these sensor systems over a wider range and combination of depths and rates. Further, the variability from compression to compression that is inherent during manual chest compressions provided by a rescuer, makes it difficult to standardize the performance of a CPR sensor system to measure compression depth. This study used a controlled and reproducible method to assess sensor performance over a range of compression depths and rates that are applicable to both pediatric and adult CPR.
The purpose of this study was to evaluate the performance of an accelerometer-based, dual-sensor CPR feedback system in estimating compression depth and rate over a clinically relevant range of depth and rate combinations in a controlled and reproducible manner provided by a mechanical chest compression simulator.
Methods
A monitor/defibrillator (R Series, ZOLL Medical Corporation, Chelmsford, MA) and a sample of ten dual-sensor CPR pads (OneStep™ Pediatric CPR Electrodes, ZOLL Medical Corporation, Chelmsford, MA) were used for the performance testing protocol. A set of CPR pads include two accelerometer sensors. An anterior sensor is housed in an oval-shaped “puck” that is placed over the patient’s sternum during chest compressions to calculate displacement of the chest. A second sensor is housed within the posterior electrode to correct for movements of the patient during compressions.
Each set of sensors was evaluated on a computer-controlled motion system that simulated chest compressions as they would occur on both rigid and compliant surfaces, over a range and combination of depths and rates.
Dual-Sensor Cardiopulmonary Resuscitation Pads.
The ZOLL OneStep™ Pediatric CPR electrodes (Fig. 1) are single-use, disposable, multifunction cardiac electrodes intended to be used with compatible ZOLL defibrillators for ECG monitoring, defibrillation, cardioversion, noninvasive pacing, and CPR assistance/feedback by trained personnel, including physicians, nurses, paramedics, emergency medical technicians, and cardiovascular laboratory technicians. These electrodes are indicated for use on a patient less than 8 years of age or weighing less than 25 kg.

ZOLL OneStep™ Pediatric CPR Resuscitation Electrode. Anterior pad (left) has attached CPR sensor to be placed over the sternum, and posterior pad (right) with embedded CPR sensor to be placed on the back as indicated on the pad labels.
Computer-Controlled Cardiopulmonary Resuscitation Motion System.
Chest compressions were simulated using a custom motion system (FTM Design, Hollis, NH) designed to test the performance of accelerometer-based CPR sensors (Fig. 2). The system comprises two computer-controlled actuator assemblies that apply linear motion independently to two platforms that each sensor can then be attached to (Fig. 2, A and B). Each actuator (Akribis Systems, Singapore) is capable of up to 28 cm of travel and forces up to 1032 N. High-resolution (50 nm) absolute position encoders (Renishaw Inc., West Dundee, IL) are used to measure and control position. The linear encoders generate position data generated by the read head optics and etched pattern on the encoder scale. As such, there is no ability, nor is it required, to calibrate the encoder position measurement. The system was leveled, and its operation was verified before data collection. For testing, the anterior sensors were removed from their foam encasements, and the posterior sensors were removed from the pads so that the sensors could be securely fastened to the motion platforms.

Schematic of computer-controlled CPR motion system. The CPR motion system is used to simulate chest compressions. The anterior sensor was fixed to the platform labeled “A,” and the posterior sensor was fixed to the platform labeled “B.” During rigid surface simulations, linear motion was applied only to platform “A,” while platform “B” remained stationary. During compliant surface simulations, motion was applied to both platforms (gray arrows indicate magnitude of linear motion).

Schematic of computer-controlled CPR motion system. The CPR motion system is used to simulate chest compressions. The anterior sensor was fixed to the platform labeled “A,” and the posterior sensor was fixed to the platform labeled “B.” During rigid surface simulations, linear motion was applied only to platform “A,” while platform “B” remained stationary. During compliant surface simulations, motion was applied to both platforms (gray arrows indicate magnitude of linear motion).
Simulated Chest Compressions.
The motion system was programmed to perform sinusoidal up-down movements that simulate chest compressions on a patient laying on either a rigid surface or a compliant surface, such as a mattress. During the rigid surface simulation, movement was applied only to the anterior sensor, while the posterior sensor remained stationary. During the compliant surface simulation, movement was again applied to the anterior sensor to simulate chest compressions, however the posterior sensor was also moved to simulate displacement of the thorax that might occur due to a patient lying on a mattress. The displacement applied to the posterior sensor during the compliant surface simulation was added to the displacement applied to the anterior sensor so that both rigid and compliant simulations were performed to achieve the same net target depths. All sensors underwent simulated compressions at a total of five target depths (1.9, 3.8, 4.8, 6.4, and 8.9 cm) and five target rates (60, 80, 100, 120, and 140 cpm). Target depths were chosen to cover a range that would be applicable to both pediatric and adult populations. Given that the recommended depth for adults is 5–6 cm, targets just above (6.4 cm) and below (4.8 cm) were selected to ensure performance was tested on both sides of the recommended window. To achieve similar net target depths during the compliant surface simulation, which included movement of the posterior sensor in addition to the anterior sensor, the depths applied to the anterior sensor were 2.4, 4.6, 5.6, 7.4, and 10.9 cm, while depths of 0.5, 0.8, 0.8, 1, and 2 cm were applied to the posterior sensor. For example, to create a target depth of 1.9 cm during dual-sensor movement, the anterior sensor was displaced 2.4 cm, while the posterior was displaced in phase with the anterior sensor at a depth of 0.5 cm. To create more realistic and clinically relevant combinations of rates and depths [19], the greatest target depth of 8.9 cm was performed at rates of 60, 80, and 100 cpm. All other compression depths were performed at 80, 100, 120, and 140 cpm. ∼200 compressions were performed at each combination of depth and rate for each sensor.
Data Analysis and Statistics.
Data were collected on the ZOLL R Series monitor/defibrillator and exported for later offline analysis. The CPR detection algorithm subtracts motion measured by the posterior sensor from the motion measured by the anterior sensor. The net motion is then used to calculate depth and rate for each compression. The analysis first calculated the depth and rate errors (estimated depth − target depth; estimated rate − target rate) for each compression. Each compression was then determined to be within target or not. For depth, a compression was classified as in target if the algorithm calculation was at or within ±0.25 cm of the target depth set by the motion system. For rate, a compression was classified as in target if the calculation was at or within ±3 cpm of the target rate. The depth and rate thresholds for determining whether a compression was in target were chosen based on previous reports of error in the literature [14,17,20] and provide the precision for rescuers to perform guideline compliant chest compressions, particularly for narrow target depths within a 1 cm window. The performance was calculated (number of in-target compressions/total number of compressions), and 99% confidence intervals were calculated for each target depth and rate. Statistical analyses were performed using STATA 15 statistical software (StataCorp LLC, College Station, TX).
Results
A total of 72,578 simulated compressions were performed on ten sets of CPR sensors. Because the CPR detection algorithm uses characteristics of the motion profile, some initial compressions are required to determine if the motion is indeed due to chest compressions, resulting in some small variability in the number of compressions detected for each test condition. For all compressions, including every rate and depth combination and for both rigid and compliant surface simulations, the mean (±sd) depth error was 0.05 (±0.08) cm, and the rate error was −0.55 (±1.44) cpm. The mean depth and rate errors for both rigid and compliant surface simulations were identical to those for all compressions combined (0.05 (±0.08) cm and −0.55 (±1.44) cpm).
Tables 1 and 2 describe the performance of the CPR feedback system in estimating compression depth and rate, respectively. The number of compressions, number of compressions in target, and confidence intervals are shown for compressions performed at each specified target depth and target rate during rigid and compliant surface simulations. In addition, Figs. 1 and 2 show histograms of depth error and rate error for each target depth and rate for both rigid (red) and compliant (blue) surface simulations.
Depth accuracy. Total number of compressions, number of compressions within ±0.25 cm of target depth (n, %), and confidence intervals for sensors tested at each target depth.
Rigid surface simulation | Compliant surface simulation | |||||||
---|---|---|---|---|---|---|---|---|
Compressions (n) | In target (n) | In target (%) | 99% confidence interval | Compressions (n) | In target (n) | In target (%) | 99% confidence interval | |
Depth (cm) | ||||||||
1.9 | 7640 | 7640 | 100 | 99.93–100 | 7640 | 7639 | 99.99 | 99.9–100 |
3.8 | 7638 | 7638 | 100 | 99.93–100 | 7640 | 7640 | 100 | 99.93–100 |
4.8 | 7640 | 7638 | 99.97 | 99.88–100 | 7640 | 7637 | 99.96 | 99.86–100 |
6.4 | 7640 | 7615 | 99.67 | 99.46–99.82 | 7640 | 7621 | 99.75 | 99.56–99.87 |
8.9 | 5730 | 5623 | 98.13 | 97.62–98.56 | 5730 | 5635 | 98.34 | 97.86–98.74 |
Rigid surface simulation | Compliant surface simulation | |||||||
---|---|---|---|---|---|---|---|---|
Compressions (n) | In target (n) | In target (%) | 99% confidence interval | Compressions (n) | In target (n) | In target (%) | 99% confidence interval | |
Depth (cm) | ||||||||
1.9 | 7640 | 7640 | 100 | 99.93–100 | 7640 | 7639 | 99.99 | 99.9–100 |
3.8 | 7638 | 7638 | 100 | 99.93–100 | 7640 | 7640 | 100 | 99.93–100 |
4.8 | 7640 | 7638 | 99.97 | 99.88–100 | 7640 | 7637 | 99.96 | 99.86–100 |
6.4 | 7640 | 7615 | 99.67 | 99.46–99.82 | 7640 | 7621 | 99.75 | 99.56–99.87 |
8.9 | 5730 | 5623 | 98.13 | 97.62–98.56 | 5730 | 5635 | 98.34 | 97.86–98.74 |
Rate accuracy. Total number of compressions, number of compressions within ±3 cpm (n, %), and confidence intervals for sensors tested at each target rate.
Rigid surface simulation | Compliant surface simulation | |||||||
---|---|---|---|---|---|---|---|---|
Compressions (n) | In target (n) | In target (%) | 99% confidence interval | Compressions (n) | In target (n) | In target (%) | 99% confidence interval | |
Rate (cpm) | ||||||||
60 | 1910 | 1910 | 100 | 99.72–100 | 1910 | 1853 | 97.02 | 95.86–97.93 |
80 | 9548 | 9394 | 98.39 | 98.02–98.7 | 9550 | 9531 | 99.8 | 99.65–99.9 |
100 | 9550 | 9379 | 98.21 | 97.83–98.54 | 9550 | 9514 | 99.62 | 99.43–99.77 |
120 | 7640 | 7613 | 99.65 | 99.43–99.8 | 7640 | 7637 | 99.96 | 99.86–100 |
140 | 7640 | 7619 | 99.73 | 99.53–99.86 | 7640 | 7637 | 99.96 | 99.86–100 |
Rigid surface simulation | Compliant surface simulation | |||||||
---|---|---|---|---|---|---|---|---|
Compressions (n) | In target (n) | In target (%) | 99% confidence interval | Compressions (n) | In target (n) | In target (%) | 99% confidence interval | |
Rate (cpm) | ||||||||
60 | 1910 | 1910 | 100 | 99.72–100 | 1910 | 1853 | 97.02 | 95.86–97.93 |
80 | 9548 | 9394 | 98.39 | 98.02–98.7 | 9550 | 9531 | 99.8 | 99.65–99.9 |
100 | 9550 | 9379 | 98.21 | 97.83–98.54 | 9550 | 9514 | 99.62 | 99.43–99.77 |
120 | 7640 | 7613 | 99.65 | 99.43–99.8 | 7640 | 7637 | 99.96 | 99.86–100 |
140 | 7640 | 7619 | 99.73 | 99.53–99.86 | 7640 | 7637 | 99.96 | 99.86–100 |
Discussion
This study demonstrates the ability of a dual-sensor system to accurately estimate compression depth within ±0.25 cm and compression rate within ±3 cpm over a clinically relevant combination of depths and rates. Further, the CPR algorithm performed similarly during the compliant surface simulation as compared to the rigid surface simulation, demonstrating the ability of the algorithm to subtract movement of the posterior sensor from the movement of the anterior sensor to accurately calculate net compression depth. While this study tested the performance of commercially available pediatric CPR pads, the range of target depths selected was appropriate for both pediatric and adult populations. Therefore, the same dual-sensor technology would also be appropriate for use in adult CPR pads. By using a computer-controlled motion system, we were able to systematically control for variables that may otherwise introduce error in studies using manual compressions performed on manikins to test rate and depth accuracy.
There are two main sources of error that contribute to variability in the calculated compression parameters. The first source is from the motion system; the difference between the parameter settings of the motion system and the motion produced. The magnitude of this error is negligible given the 50 nm resolution of the absolute linear encoders to measure position and the actuators used. The second source is from the CPR sensor system; the difference between the motion experienced by the sensor and the calculated depth and rate. Given the relatively small amount of variability introduced by the motion system, we offer that this approach is a more direct assessment of the accuracy provided by the sensor system. Future work should include programming the motion system to perform movements based on displacement signals from manual chest compressions performed in clinical cases.
The reported mean and standard deviation for depth and rate error (0.05 (±0.08) cm and −0.55 (±1.44) cpm) were identical for both the rigid and compliant surface simulations and all depths and rates combined. Figures 3 and 4 show significant overlap of the two simulations (shown in purple), with minimal differences as seen in red (rigid surface simulation) and blue (compliant surface simulation) in each histogram. The similarities in performance between the two simulations highlight the ability for the algorithm to correctly subtract the travel of the posterior sensor from the travel of the anterior sensor, as the net target depths for both the rigid and compliant simulations were identical.

Histograms of depth error for each target depth (left) and for each target rate (right). Distribution of depth error for all compressions is shown in bottom right of both target depth and target rate subplots. Rigid surface simulation data are shown in red and compliant surface simulation data are shown in blue. Purple areas of the histogram indicate overlap of data from both surface simulations.

Histograms of depth error for each target depth (left) and for each target rate (right). Distribution of depth error for all compressions is shown in bottom right of both target depth and target rate subplots. Rigid surface simulation data are shown in red and compliant surface simulation data are shown in blue. Purple areas of the histogram indicate overlap of data from both surface simulations.

Histograms of rate error for each target depth (left) and for each target rate (right). Distributions of rate error for all compressions are shown in bottom right of both target depth and target rate subplots. Rigid surface simulation data are shown in red and compliant surface simulation data are shown in blue. Purple areas of the histogram indicate overlap of data from both surface simulations.

Histograms of rate error for each target depth (left) and for each target rate (right). Distributions of rate error for all compressions are shown in bottom right of both target depth and target rate subplots. Rigid surface simulation data are shown in red and compliant surface simulation data are shown in blue. Purple areas of the histogram indicate overlap of data from both surface simulations.
While there is a small positive bias for depth error for each target depth (Fig. 3, left), the greatest depth of 8.9 cm shows a negative skew. The depth of 8.9 cm was performed only at lower rates (60, 80, and 100 cpm), and was the only depth performed at a rate of 60 cpm. The depth error for compressions performed at 60 cpm (Fig. 3, right) shows a negative bias, which may be due to the lower accelerations causing a small underestimation of depth and driving the negative skew for the histogram corresponding to 8.9 cm compressions (Fig. 3, left). As seen in Fig. 3 (right), the distribution of depth error becomes narrower with higher compression rates, which may be due to the higher accelerations detected by the sensor and the improved ability for the CPR algorithm to estimate depth as compared to lower rates.
During simulated compressions performed at higher rates (120 and 140 cpm), the rate error (Fig. 4, right) shows a ±1 cpm error at a rate of 120 cpm and a −1 and +2 cpm error at a rate of 140 cpm. This is an artifact of the CPR algorithm sampling frequency. The algorithm operates at 125 Hz, corresponding to an 8 ms sample interval. When simulated compressions are performed precisely at 120 cpm, each compression cycle is 500 ms, which comprises 62.5 samples (500 ms/8 ms). Due to the discrete sampling rate of 125 Hz, the period of the cycle is approximated by either 62 samples (which corresponds to 496 ms) or 63 samples (which corresponds to 504 ms). This results in a period corresponding to either 119 CPM (60 s/0.504 s) or 121 CPM (60 s/0.496 s). Therefore, the system cannot accurately measure at a precise rate of 120 cpm, as the hardware limitation forces it to approximate between these two values. This limitation arises from the discrete nature of the sampling rate and the resolution of the system. A similar phenomenon occurs at 140 cpm, causing the algorithm to calculate the rate at either 139 cpm or 142 cpm.
Previous studies [14,16,17,21] investigating the accuracy of sensor technology to estimate compression depth and rate have used manikins instrumented with different types of hardware as a reference for comparison. To our knowledge, the error associated with manikin sensor systems has not been rigorously tested or reported in the literature. Aase and Myklebust, however, have reported an estimated error of ±3% due to factors such as slight changes in supply voltage, the rescuer’s hand position on the chest, and loose fitting mechanical components [16]. Another study by Noordergraaf et al. also acknowledged this limitation and overcame it by instrumenting a manikin with a linear potentiometer to measure compression depth, and a linear voltage differential transducer to measure thorax movement into the mattress [21]. Similarly, Ruiz de Gauna et al. instrumented a manikin with a string potentiometer in the chest of the manikin [17]. Beesems and Koster took additional steps to calibrate the manikin used in their study at a range of depths [14]. While all these methods may provide reasonable reference signals, the error introduced by (1) different types of hardware used to instrument manikins, (2) the inherent variability of depth, rate, and hand position of manual compressions, and (3) the interactions between manikin simulations and different sensor technologies have not been studied.
In addition, the manual chest compressions performed in these studies were within or near guidelines for depth and rate in adults. For pediatric cases, it is recommended to perform compressions at least one third of the anterior–posterior chest diameter, which may call for depths of 4 cm in infants [22], or less. To understand depth accuracy for compression depths relevant to pediatric cases, we performed compressions at 1.9, 3.8, 4.8 cm, all at rates of 80, 100, 120, and 140 cpm.
The dual-sensor system consistently and accurately estimated depth and rate across a wide range of clinically relevant combinations of depth and rate. Further, the performance was similar for both rigid and compliant surface simulations, showing that movement of the posterior sensor can be removed, thereby correcting for movement of the patient’s thorax to correctly report true compression depth.
Limitations.
The main limitation is that a computer-generated waveform to simulate chest compressions is not necessarily representative of manual compressions. For this study, a sinusoidal waveform was used to simulate chest compressions, as clinical data have shown the displacement waveform of manual compressions to approximate a sine wave [6,23,24]. While a computer-generated sine wave does not contain features that may be present in a manual compression waveform, it does provide a controlled, reproducible movement that can be used to compare waveforms of different magnitude and frequency.
Conclusions
This study demonstrates the ability of a dual sensor system to accurately estimate compression depth within ±0.25 cm and compression rate within ±3 cpm over a clinically relevant combination of depths and rates during both a rigid and compliant surface simulation. Use of a computer-controlled motion system provides a more direct assessment of accuracy than manual compressions performed on instrumented manikins.
Funding Data
ZOLL Medical Corporation (Funder ID: 10.13039/100015345).
Conflict of Interest
The authors are employed by ZOLL Medical Corporation.
Data Availability Statement
The datasets generated and supporting the findings of this article are obtainable from the corresponding author upon reasonable request.