Article, Emergency Medicine

Potential pros and cons of the kinect-based real-time audiovisual feedback device during in-hospital cardiopulmonary resuscitation


Contents lists available at ScienceDirect

American Journal of Emergency Medicine

journal homepage:

American Journal of Emergency Medicine 36 (2018) 319-338

Potential pros and cons of the kinect-based Real-time audiovisual feedback device during In-hospital CPR

Dear Editor:

I read the article by Wang et al. entitled “Kinect-based real-time audio- visual Feedback device improves cardiopulmonary Resuscitation quality of lower-body-weight rescuers” with great interest [1]. I believe that kinect- based Real-time feedback technology could be an easy solution to improve the quality of in-hospital cardiopulmonary resuscitation (CPR). However, there are possible limitations of this technology. I would like to discuss the following important issues with the authors and the readers.

First, the recent guidelines recommend using audiovisual Feedback devices during CPR for real-time optimization of CPR performance [2]. Many different technologies have been developed to achieve this goal as indicated by the authors. Among them, feedback devices using accel- erometer technology are applied most popularly. However, these devices could not be used during in-hospital CPR because they exaggerate chest compression depth when used on a patient lying on a soft surface, such as a mattress [3]. Recently, several new technologies have been de- veloped to overcome this limitation [4-7]. Although these new technolo- gies can measure CCD correctly on Soft surfaces in Experimental models, they require dual sensors on the front and back of the patient’s thorax. The need for dual sensors might cause another problem, such as increas- ing hands-off time, changing patients‘ position, and obstacles to chest compression, when used in an actual clinical setting.

Kinect-based real-time audiovisual feedback technology could be an- other solution to overcome this defect because the sensor of this technol- ogy is not in contact with the patient’s thorax. It requires only a single marker in front of the thorax or the rescuer’s hand. The key problem is its accuracy. The authors described that the error range of the kinect sen- sor was less than 1 mm [1]. I wonder how the authors calculated this error range. The authors also indicated that the kinect sensor detected the movement of the marker every 0.05 s. If the average chest compres- sion rate was 110 per minute, 0.54 s would be needed for each chest compression. Therefore, the kinect-based sensor could detect the move- ment of the marker 10 times in each chest compression: 5 times in the compression phase and 5 times in the decompression phase. Consid- ering that the average CCD was 5 to 6 cm, the error range could not be less than 1 mm. In addition, there would be a possibility of over-estima- tion when used on a soft surface because it could measure actual CCD by detecting the movement of marker. The authors only described that they used a bed with a hard backboard. If they used a soft mattress, mattress compression depth might be added to the CCD in case of standing pos- ture. The possibility of over-estimation should be verified by another ex- periment. If they presented data measured simultaneously with the kinect sensor and a gold standard device, such as Resusci Anne QCPR (Laerdal Medical, Stavanger, Norway), the accuracy of the kinect sensor would be clearer.

Second, the authors used a fixed bed height of 60 cm. Considering that the average knee height of the Korean healthcare provider was

42.6 +- 1.8 cm, the median gap between the rescuer’s knee height and the bed height was about 17.4 cm [8]. It was already reported that the CCD decreased significantly when the chest compression was conduct- ed beside a bed 20 cm higher than the rescuer’s knee height [9]. There- fore, a high-quality chest compression in a Kneeling posture was easily predictable. However, there were inconsistent results in Wang et al.’s study. Although the CCD of the kneeling posture was deeper than that of the Standing posture in the case of using audiovisual feedback (6.37

+- 1.92 vs. 6.05 +- 1.87, respectively, p = 0.03), there were no significant differences in cases of no audiovisual feedback (5.38 +- 1.68 vs. 5.61 +- 1.99, respectively, p = 0.15). I do not understand why these inconsis- tent results were obtained. They might be caused by different types of CCD measurements or over-estimation of the CCD on a mattress. If the authors measured the CCD simultaneously with gold standard methods, we could determine the reasons. In addition, although the CCD in- creased with feedback, it did not increase significantly if the condition was confined to the standing posture. These results indicate that the kinect-based real-time feedback system would not be effective during in-hospital environments. The authors also analyzed the results accord- ing to the rescuer’s body weight. However, I cannot find the results con- sidering both the rescuer’s body weight and posture. Additional data are needed to verify if this feedback system would be effective during in- hospital environments in cases of low-body-weight rescuers.

Third, the study results explain why accurate feedback is required for underweight rescuers. However, overweight rescuers also deterio- rate the performance of chest compressions. Contri et al. reported the converse effects of the rescuer’s body weight on the CCD and chest wall recoil [10]. Although overweight rescuers could achieve adequate CCD easily, the chance of incomplete chest wall recoil would increase. Therefore, we should monitor the ratio of complete chest wall recoil during CPR. To report chest wall recoil, another algorithm would be needed because the kinect-based sensor might only measure actual CCD apart from the compression rate and the CCD.

The kinect-based real-time audiovisual feedback technology has many possibilities. However, it should be verified by additional experi- ments whether it might improve the qualities of CPR during in-hospital environments.

Je Hyeok Oh

Department of Emergency Medicine, College of Medicine, Chung-Ang

University, Seoul, Republic of Korea Department of Emergency Medicine, College of Medicine, Chung-Ang University, 84 Heukseok-ro, Dongjak-gu, Seoul 06974,

Republic of Korea.

E-mail address: [email protected]

21 September 2017

0735-6757/(C) 2017

320 Correspondence / American Journal of Emergency Medicine 36 (2018) 319338


Wang J, Tsai S, Chen Y, Chen Y, SJ C, Liao W. Kinect-based real-time audiovisual feed- back device improves cardiopulmonary resuscitation quality of lower-body-weight rescuers. Am J Emerg Med 2018;36:319-20.
  • 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.
  • 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.

    BG Yu, Oh JH, Kim Y, Kim TW. Accurate measurement of chest compression depth using impulse-radio ultra-wideband sensor on a mattress. PLoS One 2017; 12:e0183971.
  • Minami K, Kokubo Y, Maeda I, Hibino S. A flexible pressure sensor could correctly measure the depth of chest compression on a mattress. Am J Emerg Med 2016;34: 899-902.
  • Beesems SG, Koster RW. Accurate feedback of chest compression depth on a mani- kin on a soft surface with correction for total body displacement. Resuscitation 2014;85:1439-43.
  • Oh J, Song Y, Kang B, Kang H, Lim T, Suh Y, et al. The use of dual accelerometers im-
  • proves measurement of chest compression depth. Resuscitation 2012;83:500-4.

    Cheong SA, Oh JH, Kim CW, Kim SE, Lee DH. Effects of alternating hands during in- hospital one-handed chest compression: a randomised cross-over manikin trial. Emerg Med Australas 2015;27:567-72.
  • Cho J, Oh JH, Park YS, Park IC, Chung SP. Effects of bed height on the performance of
  • chest compressions. Emerg Med J 2009;26:807-10.

    Contri E, Cornara S, Somaschini A, Dossena C, Tonani M, Epis F, et al. Complete chest recoil during laypersons’ CPR: is it a matter of weight? Am J Emerg Med 2017;35: 1266-8.

    Reply to letter, “Potential pros and cons of the Kinect-based real-time audiovisual feedback device during in-hospital CPR”

    To the Editor:

    Thank you for your interest and comments regarding our study in- vestigating “Kinect-based real-time audiovisual feedback device im- proves cardiopulmonary resuscitation quality of lower-body-weight rescuers” [1]. We appreciate the time you took to comment on our re- search and the important insights you have provided. We have some in- formation to elaborate in the following sections.

    First, the application of Kinect to chest compression had been reported to Resuscitation in 2013 [2]. Our study provided an idea to quantify the CC quality by using Kinect modules. Our study supports that the Kinect module could be used in monitoring CC quality in differ- ent postures and body weight of the trainees. A previous study regard- ing mobile robot has shown that Kinect can provide a precise data for resolution around 0.6 mm within a distance of 0.6 m [3]. Second, we apologized that there are a few typing errors in Table 1 of the present ar- ticle that could be misleading. A request of correction had been submit- ted to AJEM simultaneously. In the correct version, our study revealed that there was no statistically significant difference in the compression depth between the kneeling and standing position with or without feedback. The correct data should be “With feedback, CC in a kneeling posture on the floor resulted in a higher CC quality compared with a standing posture 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.” Pre- vious study had indicated that CC depth decreased when the bed height was 20 cm more than knee height [4]. Nonetheless, the difference be- tween the bed height and average knee height of participants was less than 20 cm in our current study. Therefore, our results are in line with the study mentioned above and the idea of the reader. Third, over-esti- mation is one of the limitations of many CC feedback device when it is performing on a soft surface such as mattress. We believe that the Kinect device cannot overcome the limitation yet. Ideally, CPR should be performed in a hard surface in the clinical settings. Therefore, we conducted our study on an iron stretcher with hard surface, rather

    than soft mattress to minimize the possibility of over-estimation and detection error. We emphasize that the study was a manikin study. Fur- ther evaluation is deemed necessary to verify the benefit of Kinect in real-world resuscitation scene. Finally, adequate chest wall recoil is a part of high-quality CC. We agree with the readers’ comments regarding to the efficiency of chest wall recoil should be incorporated in further studies. Simultaneously measurements of the Kinect sensor with a well-calibrated device could further clarify several important issues.

    In summary, we thank the authors of the letter for their attention to detail. Nonetheless, those mistakes did not change the major results and conclusion of the study. We apologize for the inconvenience to the readers, the editors and the publisher.

    Jen-Chun Wang, MD Wen-I Liao, MD

    Shih-Hung Tsai, MD, PhD*

    Department of Emergency Medicine, Tri-Service General Hospital, National

    Defense Medical Center, Taipei, Taiwan

    *Corresponding author at: Department of Emergency Medicine, Tri-Service General Hospital, National Defense Medical Center, No. 325,

    Sec. 2, Cheng-Kung Road, Taipei, Taiwan.

    E-mail address: [email protected] (S.-H. Tsai)

    8 November 2017


    1. Wang JC, Tsai SH, Chen YH, Chen YL, Chu SJ, Liao WI. Kinect-based real-time audiovi- sual feedback device improves cardiopulmonary resuscitation quality of lower-body- weight rescuers. Am J Emerg Med 2017, Sep. 15 pii: S0735-6757(17)30747-7. https://
    2. 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 training: a randomised crossover mannequin pilot study. Resuscitation 2013;84:501-7.
    3. Viager M. Analysis of Kinect for Mobile Robots. Available at: doc/56470872.
    4. Cho J, JH Oh, Park YS, Park IC, Chung SP. Effects of bed height on the performance of chest compressions. Emerg Med J 2009;26:807-10.

      Predicting 72-hour emergency department revisits: Methodological issues

      Dear Editor,

      We read with great interest the article by Pellerin G and colleagues [1]. After reading this article carefully and critically, we think that some methodological and statistical issues should be considered.

      As reported in the study [1] in the second model using patient demo- graphics and prior year utilization the c-statistics were 0.70 for both de- velopment and validation samples, while in the final model using comorbidities and previous predictors, the c-statistics were 0.74 and

      0.73 for both development and validation samples, respectively. It is crucial to emphasize, difference between the c-statistics in second model and final model is nothing and even clinically negligible. In other words, the overall performances of the two model seems are the same for identifying patients at high risk of revisiting the ED within 72 h. For application in clinical practice, we suggest the authors to test the equality of two c-statistics in second model and final model by using introduced statistical methods [1-3].

    Leave a Reply

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