Article, Emergency Medicine

Kinect-based real-time audiovisual feedback device improves CPR quality of lower-body-weight rescuers

a b s t r a c t

Background: chest compression quality is associated with rescuer posture and body weight. We designed a Kinect module-based real-time audiovisual feedback device to investigate the relationship between rescu- er posture, body weight, and CC quality.

Methods: A total of 100 healthcare professionals were enrolled as participants in this randomized trial. A Kinect- based sensor system was used to monitor the depth and rate of CC and provide further Real-time feedback. All participants were asked to perform continuous CC on a manikin with and without feedback for 2 min individually in either a kneeling or standing position.

Results: A kneeling posture can provide higher rate of CC than a standing posture can (111.4 +- 22.6 per minute vs. 99.1 +- 18.9 per minute, p value = 0.005). Real-time AVF feedback can provide a better compression depth, rate, and effective compression ratio (6.16 +- 1.88 cm vs. 5.54 +- 1.89 cm, p value = 0.02; 103.2 +- 21.0/min vs. 96.7 +- 25.8/min, p value = 0.03; 62.6 +- 28.0% vs. 51.0 +- 33.2%, p value = 0.004). Regardless of the effect of real-time feedback, the CC depth correlated to the rescuers’ body weight. Rescuers who weighed below 71 kg benefited from the Kinect module-based real-time AVF device in terms of improved CC quality.

Conclusion: The Kinect-based AVF device can significantly improve CC quality in manikin training in rescuers with their body weight b 71 kg.

(C) 2017

Introduction

Adequate compression rate and depth are emphasized by the 2015 American Heart Association and European Resuscitation Council guide- lines [1-4]. The 2015 International Consensus on Cardiopulmonary Resuscitation and Emergency Cardiovascular Care Science With Treat- ment Recommendations also suggest the use of real-time audiovisual feedback and prompt devices during cardiopulmonary resuscitation (CPR) in clinical practice as part of a comprehensive system for care for cardiac arrest [5].

Several technologies have been used to provide real-time monitor- ing or feedback for CPR, including pressure sensors, accelerometers, force sensors, impedance signalling, and motion detectors [6,7].

? The authors have no potential conflict of interests to disclose.

* Corresponding authors at: Department of Emergency Medicine, Tri-Service General Hospital, National Defense Medical Center, No. 325, Sec. 2, Cheng-Kung Road, Taipei, Taiwan.

E-mail addresses: [email protected] (S.-H. Tsai), [email protected] (W.-I. Liao).

Commercial audiovisual feedback (AVF) devices are usually not readily available or cost-effective. Kinect-based motion sensing devices are a newly developed technology for detecting and recording chest compression quality. Kinect is a motion-sensing input device manufactured by Microsoft for the Xbox 360 gaming console. Kinect provides three-dimensional motion capture of objects and recognition capabilities. A self-developed software is required for driving the Kinect camera. It is increasingly applied in medical fields such as balance train- ing and rehabilitation tools [8]. Kinect-based AVF devices have been combined with self-developed software to improve the depth and rate of CC in CPR training [7,9]. However, previous studies regarding Kinect-assisted CC have only concerned standing posture use and single scenarios [7].

Several studies have demonstrated that rescuer posture and body weight can influence CC quality. A previous study demonstrated signif- icant differences in compression rate and depth between CPR per- formed on manikins placed on the floor and those placed on a stretcher at rescuers’ knee height [10]. When performing CC, rescuers developed fatigue more easily in the standing posture than they did in the kneeling posture [11]. When performing CC on a manikin on the

http://dx.doi.org/10.1016/j.ajem.2017.09.022

0735-6757/(C) 2017

floor, rescuers with lower body weight tended to become fatigued, which resulted in gradually declining CC quality compared with those with higher body weight [12]. Rescuers with a body mass index (BMI) b 26 tended to perform CC with inadequate depth and finished CC faster than those with a BMI N 26 [13]. Although numerous feedback systems for CPR exist, studies investigating the effect of Kinect-based AVF on CPR quality are limited. In addition, no study has evaluated the effect of Kinect-based real-time AVF in terms of rescuer body weight and position.

The present study examined the effectiveness of Kinect-based real-time AVF devices and the association between the rescuer pos- ture and body weight, CC quality, and real-time feedback. We hy- pothesized that CC quality might be influenced by the rescuer posture and body weight and that a Kinect-based real-time Feedback device might produce improved CC quality. We concluded that healthcare providers with lower body weight tended to provide subop- timal quality CPR and would benefit from Kinect-based real-time AVF devices.

Methods

Data collection

This study is a randomized trial with 100 participants who are em- ployees of Tri-Service General Hospital and ambulance service, includ- ing doctors, nurses, and emergency medical technicians. All had received CPR training and certification within 2 years of the study. This prospective collection of CC data was approved by the Institutional Review Board of the Tri-Service General Hospital, National Defense Medical Center, Taipei, Taiwan.

Study protocol

All participants underwent full BLS or ACLS training programs accord- ing to the 2010 International Liaison Committee on Resuscitation guide- lines within 2 years of the start of this study. CPR was administered in one-rescuer and compression-only mode to focus the efficiency of chest compressions. Participants were divided into two randomized groups: those kneeling beside the manikin on the floor and those standing beside the manikin lying on a bed with a hard backboard (bed height = 60 cm). Participants performed continuous CCs on a manikin without feedback for 2 min and shifted to with feedback groups (Fig. 1). All participants had at least a 1-h rest period between two CPR settings.

A self-developed software was paired with the Kinect (Microsoft, Redmond, WA, USA) motion-sensing module to monitor the number and Depth of CCs and provide real-time AVF (Fig. 2). The horizontal field of the Kinect sensor was approximately 50-70 cm, and the vertical field was 70 cm, resulting in a resolution of slightly N 0.6 mm per pixel. The participants attached a hand marker (4 x 4 cm) made of paper to the top surface of their left hand. Kinect sensors detect and record the position and motion of participants’ hands every 1/20 s. The feedback group performed CC with real-time AVF during the entire 2-min test. Participants could adjust their CCs according to the information pre- sented on the monitor, including the depth and rate of compressions. The non-feedback group performed CC without any feedback and the Kinect device was used to record only the depth and rate.

Data analysis

The CC quality was analysed according to the CC rate and depth and the effective compression ratio (compressions with depth N 5 cm/total compressions). To determine the validity of each CC, at the beginning

Fig. 1. Study flow chart. Participants performed continuous CCs on a manikin without feedback for 2 min and shifted to with feedback groups. All participants had at leasta 1-h rest period between two CPR settings.

Fig. 2. Kinect-based real-time audiovisual feedback system for chest compression. (A) Experimental device set-up. (B) A marker was placed in participants’ hands and the device was calibrated before the test. (C, D) Audiovisual feedback device displaced the depth and rate during a real-time test.

of the session, the system acquired the height of the rescuer’s hand in contact with the chest wall at the point of maximum release. A Kinect camera detected the marker location at the beginning of the CC. The po- sition of the hand marker was recorded throughout the 2-min CC peri- od. Each compression depth was calculated as the difference between the minimum and maximum heights of the hand during each compres- sion phase. The period between two consecutive maxima was consid- ered a full compression phase. An adequate CC was defined as that with a depth of N 5 cm.

Statistical analysis

All continuous variables were presented as mean +- standard devia- tion. Analyses were performed on an intention-to-treat basis. The data were compared across groups by using the chi-squared test for categor- ical variables and the two-tailed paired t-test for continuous variables. The correlation between compression depth and body weight was eval- uated using Pearson product-moment correlation. A receiver operating characteristic curve was plotted to classify the CC effect on the basis of body weight. For the two-tailed tests, p b 0.05 was considered statisti-

without real-time feedback (depth: 6.16 +- 1.88 vs. 5.54 +- 1.89, p =

0.016; rate: 103.2 +- 21.0 vs. 96.7 +- 25.8, p = 0.026; effective ratio:

62.6 +- 28.0 vs. 51.0 +- 33.2, p = 0.004). CC in a kneeling posture on the floor resulted in a higher CC quality compared with a standing pos- ture beside a bed, in terms of higher compression rate (111.4 +- 22.6 per minute vs. 99.1 +- 18.9 per minute, p = 0.005). There was larger varia- tion around interrupted time in the kneeling position with feedback compared to other groups, but it did not reach statistical significance.

As shown in Fig. 3A, irrespective of the effect of real-time feedback, the CC depth correlated to the rescuer body weight (r = 0.227, p b 0.05). We determined that the optimal cut-off point of body weight for adequate CC quality is 71 kg, with a sensitivity of 0.328 and specific- ity of 0.883 (Fig. 3B). Participants with lower body weight (b 71 kg) per- formed CC with a lower compression depth and effective compression ratio than those with higher body weight did, despite being assisted

Table 1

Quality of chest compression is evaluated by using the Kinect-based real-time audiovisual feedback.

cally significant. The analyses were performed using SPSS Ver.18.

Sample size determination

Based on a study design for repeated measures, an alpha of 0.05, a power of 90% and an estimated effect size (primary outcome compression depth, 4.88 +- 0.519 cm versus 5.32 +- 0.88 cm, calculation based on pre- vious data [7]) resulted ina sample size of n = 17 at least of each group.

Results

With feedback (n = 100)

Without feedback (n = 100)

p value

The setting of the Kinect-based real-time AVF is presented in Fig. 2. In the current setting, the module could monitor the CC depth with an accuracy of b 1 mm error. A total of 100 volunteers were enrolled and all participants completed the study. The distribution of the continuous variables is normal and parametric. Table 1 presents the performance of the participants with or without real-time feedback. The compression depth, compression rate, and effective compression ratio of the real- time feedback group were significantly higher than those of the group

* p values b 0.05 were considered statistically significant (with feedback vs. without feedback).

Compression death (cm) all Kneeling

6.16 +- 1.88

6.37 +- 1.92

5.54 +- 1.89

5.38 +- 1.68

0.02?

0.03?

Standing

6.05 +- 1.87

5.61 +- 1.99

0.15

p value

0.42

0.56

Compression rate all

103.2 +- 21.0

96.7 +- 25.8

0.03?

Kneeling

111.4 +- 22.6

103.6 +- 29.2

0.16

Standing

99.1 +- 18.9

93.3 +- 23.4

0.09

p value

0.005#

0.06

Effective compression (%) all

62.6 +- 28.0

51.0 +- 33.2

0.004?

Kneeling

64.7 +- 28.3

50.0 +- 31.6

0.02?

Standing

61.5 +- 28.0

51.4 +- 34.1

0.05

p value

0.60

0.84

Interrupted time (s) all

3.5 +- 9.1

2.6 +- 3.9

0.30

Kneeling

4.6 +- 14.8

2.5 +- 5.3

0.39

Standing

2.9 +- 4.1

2.6 +- 3.0

0.56

p value

0.40

0.88

# p values b 0.05 were considered statistically significant (kneeling position vs. standing position).

Fig. 3. (A) Correlation between rescuer body weight and depth of chest compressions. (B) Receiver operating characteristic curve for the depth of chest compression and body weight. The cut-off point of the ideal body weight of participants to perform adequate quality chest compression is 71 kg.

by the Kinect AVF device (Table 2). Seventy-four rescuers weighing b 71 kg maintained a significantly higher CC quality with the Kinect Real-time feedback device than did those without feedback (depth:

5.93 +- 1.71 vs. 5.34 +- 1.79, p = 0.031; rate: 102.3 +- 21.7 vs. 95.1 +-

26.8, p = 0.03; effective ratio: 58.8 +- 27.1 vs. 47.1 +- 32.4, p = 0.013). However, in the remaining participants weighing N 71 kg (n = 26), no statistically significant difference was observed between the compres- sion depth, rate, and effective compression ratio with or without the use of a feedback device (depth: 6.81 +- 2.19 vs. 6.11 +- 2.09, p = 0.259; rate: 105.4 +- 18.8 vs. 101.3 +- 22.5, p = 0.478; effective ratio:

73.2 +- 28.2 vs. 61.8 +- 33.4, p = 0.164; Fig. 4).

Discussion

According to our review of the relevant literature, this was the first CPR simulation study to compare the differences in CC quality in terms of body weight and body posture by using a Kinect module- based real-time AVF device. Our study demonstrated that the Kinect module can provide real-time AVF. Healthcare providers with a lower body weight and healthcare providers who performed CC in the

Table 2

Comparison between rescuers weight over and under 71 kg.

standing posture tend to provide suboptimal CC quality and would ben- efit from Kinect-based real-time AVF.

Several technologies have been used to provide real-time monitor-

ing or feedback for CPR, including pressure sensors, accelerometers, force sensors, impedance signalling, and motion detection [6,7]. Accelerometer Feedback devices are commonly used to detect rescuer performance during CC training and actual CPR. However, they can overestimate the compression depth when CPR is performed on a pa- tient on a mattress, particularly in the case of low-body-weight rescuers [14,15]. Unlike other CPR feedback devices, the Kinect sensor is com- mercially available to non-professionals outside hospitals or healthcare facilities. The benefits of the Kinect device include its cost-effectiveness and its ability to merge with video games to further enrich training sce- narios. We believe that Kinect-based real-time AVF can effectively mon- itor CC quality, as previously described [7,9].

Regarding CC posture, our study revealed that performing CC in a kneeling posture provides a significantly higher compression rate than does performing CC on a bed. CC force is generated by gravity. Performing CC in a kneeling posture enables a high compression force to be exerted by positioning the rescuer’s shoulders directly above the patient’s chest [16]. Kneeling on a bed or standing on a footstool is the common rescuer position during hospital CPR [16]. Fewer CCs were de- livered effectively in the standing posture because it demands more power than using the kneeling posture does. Thus, devices have been developed to mimic the kneeling posture and promote CC quality [17,

Compression death (cm)

Body weight b 71 kg (n = 74)

Body weight ? 71 kg (n = 26)

p value

18]. In our current study, we also observed that participants in kneeling

postures can provide a higher CC quality through a significantly in- creased effective compression ratio.

Non-AVF 5.34 +- 1.79 6.11 +- 2.09 0.07

AVF 5.93 +- 1.71 6.81 +- 2.19 0.03?

p value 0.03# 0.25

Compression rate (/min)

Non-AVF 95.1 +- 26.8 101.3 +- 22.5 0.29

AVF 102.3 +- 21.7 105.4 +- 18.8 0.52

p value 0.03# 0.47

Effective compression (%)

0.05

Non-AVF

47.1 +- 32.4

61.8 +- 33.4

AVF

58.8 +- 27.1

73.2 +- 28.3

0.02

p value Interrupted time (s)

0.01#

0.16

Non-AVF

2.9 +- 4.3

1.9 +- 2.6

0.25

AVF

3.8 +- 10.1

2.8 +- 5.3

0.64

p value

0.41

0.42

*

AVF: audiovisual feedback.

* p value b 0.05 is considered significant (body weight b 71 kg vs. body weight ? 71 kg).

# p values b 0.05 were considered statistically significant (with feedback vs. without feedback).

Previous studies have shown that CC quality can be influenced by factors such as gender and BMI, and individual rescuer and CPR training programs should consider these factors to maximize beneficial out- comes for victims [13]. Previous studies have demonstrated that a great- er BMI is associated with adequacy of CC [19,20]. Moreover, rescuer body weight has been shown to be a major determinant of adequate CC in novice rescuers [21]. Our study produced that heavy-weight res- cuers did not performed higher quality of CC compared to lightweight rescuers. Moreover, we found that the cut-off value of the body weight of rescuers that required the Kinect-based AVF device was 71 kg. Res- cuers with body weights higher than 71 kg could deliver adequate CC even without the Kinect-based real-time AVF device, whereas those with body weights b 71 kg could benefit from the Kinect-based real- time AVF device. This finding indicates that participants can adjust their posture and strength to fit the ideal depth and rate of high quality CC and overcome the disadvantage of low body weight.

Fig. 4. Real-time audiovisual feedback can improve the quality of chest compression in rescuers with body weight b71 kg. In participants with body weights b 71 kg, Kinect-based real-time feedback device can significantly improve the chest Compression quality performed in terms of compression depth, compression rate, and effective compression ratio. *p b 0.05.

Previous studies have shown that real-time feedback devices can as- sist participants in delivering longer, more effective, and steadier CC over time. An extrapolation of these simulation results may allow the interval between Rescuer switches to be prolonged beyond the currently recommended 2 min when a feedback device is used [12,22]. Feedback from the automated training device was sufficient to produce a signifi- cant improvement in both CC rate and depth [23]. Resuscitation training implementation combined with real-time AVF was reported to be inde- pendently associated with improved CPR quality, an increase in surviv- al, and favourable functional outcomes after out-of-hospital cardiac arrest [24]. In the present study, lower-body-weight rescuers can im- prove their CC quality by using Kinect-based feedback devices. We rec- ommend that rescuers with lower body weight (b 71 kg) benefit from using feedback devices such as Kinect to maintain high-quality CC. However, the advantages of using feedback devices for heavy rescuers (71 kg and more) are limited.

Our study had several limitations. First, the Kinect-based feedback system is newly developed and its validity is limited to the training en- vironment. The actual effect of application of these systems to humans warrants further in-field investigations. Second, the study was designed to involve 2-min continuous CC. The effect of time disruption caused by the hands-off time during rescuer switches, breathing, or pulse checking was not considered in this study. Third, we did not perform direct com- parison of the Kinect-based feedback device and other CPR feedback de- vices. Finally, all participants enrolled in our study were medical staff. Additional studies are necessary to compare the Kinect-based feedback device and other CPR feedback devices and determine the effect of im- provement in CC and the benefit of Kinect-based AVF for non-profes- sional providers.

Conclusion

The Kinect-based AVF device can significantly improve CC quality in manikin training in rescuers with their body weight b 71 kg.

Acknowledgements

We thank Prof. Li-Chen Fu, Shih-Huan Tseng, Yen-Pin Hsu, and Kuan-Yu Chen from the Department of Electrical Engineering, National Taiwan University for the research and development of the software used in this study. This study was supported in part by Research Grants TSGH-C104-059 and TSGH-C105-058 from the Tri-service General Hos- pital, National Defense Medical Center, Taipei, Taiwan; MAB-105-069 and MAB-105-089 from National Defense Medical Center, Taipei, Taiwan.

Authors’ contributions

J.C.W. had full access to all of the data in the study and takes respon- sibility for the integrity of the data and the accuracy of the data analysis.

W.I.L. and S.H.T. contributed to the design of the study. Y.L.C.; Y.H.C. and

S.J.C. contributed to the literature review and writing of the manuscript.

J.C.W. contributed to the data collection. W.I.L. contributed to the statis- tical analysis. All authors contributed to the review and approval of the final manuscript.

Additional information

The authors have no competing financial interests to declare.

References

  1. Berg RA, Hemphill R, Abella BS, Aufderheide TP, Cave DM, Hazinski MF, et al. Part 5: adult basic life support: 2010 American Heart Association guidelines for cardiopul- monary resuscitation and emergency cardiovascular care. Circulation 2010;122: S685-705.
  2. Deakin CD, Nolan JP, Soar J, Sunde K, Koster RW, Smith GB, et al. European resusci- tation council guidelines for resuscitation 2010 section 4. Adult advanced life sup- port. Resuscitation 2010;81:1305-52.
  3. Kleinman ME, Brennan EE, Goldberger ZD, Swor RA, Terry M, Bobrow BJ, et al. Part 5: adult basic life support and cardiopulmonary Resuscitation quality: 2015 American Heart Association guidelines update for cardiopulmonary resuscitation and emer- gency cardiovascular care. Circulation 2015;132:S414-35.
  4. Soar J, Nolan JP, Bottiger BW, Perkins GD, Lott C, Carli P, et al. European resuscitation council guidelines for resuscitation 2015: section 3. Adult advanced life support. Re- suscitation 2015;95:100-47.
  5. Perkins GD, Travers AH, Berg RA, Castren M, Considine J, Escalante R, et al. Part 3: adult basic life support and Automated external defibrillation: 2015 international consensus on cardiopulmonary resuscitation and emergency cardiovascular care sci- ence with treatment recommendations. Resuscitation 2015;95:e43-9.
  6. Yeung J, Davies R, Gao F, Perkins GD. A randomised control trial of prompt and feed- back devices and their impact on Quality of chest compressions–a simulation study. Resuscitation 2014;85:553-9.
  7. Semeraro F, Frisoli A, Loconsole C, Banno F, Tammaro G, Imbriaco G, et al. Motion de- tection technology as a tool for cardiopulmonary resuscitation (CPR) quality train- ing: a randomised crossover mannequin pilot study. Resuscitation 2013;84:501-7.
  8. Ruff J, Wang TL, Quatman-Yates CC, Phieffer LS, Quatman CE. Commercially available gaming systems as clinical assessment tools to improve value in the orthopaedic set- ting: a systematic review. Injury 2015;46:178-83.
  9. Semeraro F, Marchetti L, Frisoli A, Cerchiari EL, Perkins GD. Motion detection tech- nology as a tool for cardiopulmonary resuscitation (CPR) quality improvement. Re- suscitation 2012;83:e11-.
  10. Sebbane M, Hayter M, Romero J, Lefebvre S, Chabrot C, Mercier G, et al. Chest com- pressions performed by ED staff: a Randomized cross-over simulation study on the floor and on a stretcher. Am J Emerg Med 2012;30:1928-34.
  11. Foo NP, Chang JH, Lin HJ, Guo HR. rescuer fatigue and cardiopulmonary resuscitation positions: a Randomized controlled crossover trial. Resuscitation 2010;81:579-84.
  12. Hasegawa T, Daikoku R, Saito S, Saito Y. Relationship between weight of rescuer and quality of chest compression during cardiopulmonary resuscitation. J Physiol Anthropol 2014;33:16.
  13. Jaafar A, Abdulwahab M, Al-Hashemi E. Influence of rescuers’ gender and body mass index on cardiopulmonary resuscitation according to the American Heart Associa- tion 2010 Resuscitation guidelines. Int Sch Res Notices 2015;2015:246398.
  14. Perkins GD, Kocierz L, Smith SC, McCulloch RA, Davies RP. Compression feedback de- vices over estimate chest compression depth when performed on a bed. Resuscita- tion 2009;80:79-82.
  15. Jung MH, Oh JH, Kim CW, Kim SE, Lee DH, Chang WJ. Does accelerometer feedback on high-quality chest compression improve survival rate? An in-hospital cardiac ar- rest simulation. Am J Emerg Med 2015;33:993-7.
  16. Hong CK, Park SO, Jeong HH, Kim JH, Lee NK, Lee KY, et al. The most effective rescuer’s position for cardiopulmonary resuscitation provided to patients on beds: a randomized, controlled, crossover Mannequin study. J Emerg Med 2014;46:643-9.
  17. Jones AY, Lee RY. Rescuer’s position and energy consumption, spinal kinetics, and ef- fectiveness of simulated cardiac compression. Am J Crit Care 2008;17:417-25 [dis- cussion 26-7].
  18. Oh J, Chee Y, Lim T, Cho Y, Kim IY. Chest compression with kneeling posture in hos- pital cardiopulmonary resuscitation: a randomised crossover simulation study. Emerg Med Australas 2014;26:585-90.
  19. Krikscionaitiene A, Stasaitis K, Dambrauskiene M, Dambrauskas Z, Vaitkaitiene E, Dobozinskas P, et al. Can lightweight rescuers adequately perform CPR according to 2010 resuscitation guideline requirements? Emerg Med J 2013;30:159-60.
  20. Ashton A, McCluskey A, Gwinnutt CL, Keenan AM. Effect of rescuer fatigue on perfor- mance of continuous external chest compressions over 3 min. Resuscitation 2002; 55:151-5.
  21. Oh JH, Kim CW. Relationship between chest compression depth and novice rescuer body weight during cardiopulmonary resuscitation. Am J Emerg Med 2016;34: 2411-3.
  22. Buleon C, Delaunay J, Parienti JJ, Halbout L, Arrot X, Gerard JL, et al. Impact of a feed- back device on chest compression quality during extended manikin CPR: a random- ized crossover study. Am J Emerg Med 2016;34:1754-60.
  23. Johnson M, Peat A, Boyd L, Warren T, Eastwood K, Smith G. The impact of quantita- tive feedback on the performance of chest compression by basic life support trained clinical staff. Nurse Educ Today 2016;45:163-6.
  24. Bobrow BJ, Vadeboncoeur TF, Stolz U, Silver AE, Tobin JM, Crawford SA, et al. The in- fluence of scenario-based training and real-time audiovisual feedback on out-of- hospital cardiopulmonary resuscitation quality and survival from out-of-hospital cardiac arrest. Ann Emerg Med 2013;62:47-56.e1.

Leave a Reply

Your email address will not be published. Required fields are marked *