Impact of video quality when evaluating video-assisted cardiopulmonary resuscitation: a randomized, controlled simulation trial

This prospective, randomized, controlled, parallel group and double-blind simulation trial was conducted in the University Hospital of Cologne and the EMS Department of the Fire Brigade of Kerpen, Germany. The study was approved by the Ethics Committee of the University Hospital Cologne (Approval number 18–130, 12th of February 2019) and was registered at the German Clinical Trial Register (Registration number DRKS00015297). Data was collected from September 2019 to February 2020. Written and informed consent was obtained from each participant prior to inclusion.

CPR was simulated using a standardized CPR training manikin (Ambu Man basic, Ambu GmbH, Bad Nauheim, Germany). Video sequences were recorded with a Tamron objective (Tamron SP 24-70 mm F/2.8, Tamron Corporation, Saitama, Japan) on a Nikon D750 (Nikon Corporation, Tokyo, Japan) with a resolution of 1920 × 1080 pixels per inch and 30 frames per second, to obtain a video with maximum quality. The camera was placed laterally of the manikin on a tripod of 1.80 m height, facing the manikin and the helper performing CPR from approximately 2.5 m of distance. A metronome (metronom beats app, Stonekick, London, UK; downloaded from Google Play Store) was set to 110 bpm (or the corresponding lower/higher frequency) to ensure guideline-conformant compression frequency, and the internal compression depth indicator of the manikin was used to ensure sufficient compression depth. In the final video sequences, neither metronome nor compression depth indicator was visible respectively audible. Seven high quality video clips of 12 s each were recorded, showing either guideline-conformant optimum performance compression-only CPR, or a typical error often made during CPR (Table 1).

Video resolution was artificially reduced by using a software tool (XMedia Recode, V. 3.4.3.6, Eschenbergen, Germany) to obtain three quality levels: high (resolution 1920x1080ppi), medium (resolution 320x240ppi) and low (resolution 128x96ppi), all with a frame rate of 30fps (Fig. 1). High quality served as standard, while the lower resolutions are typically used on mobile devices, when data is transferred via wireless networks and compression is necessary. Video sequences and quality levels were randomized with a randomization algorithm provided by the Institute of Medical Statistics and Computational Biology (IMSB) of the University of Cologne. During the evaluation phase, video sequences were analyzed on an ASUS X541UA laptop with a 15.6″ screen. A standardized questionnaire with multiple-choice answers was used to evaluate CPR quality.

Foty-six paramedics and 47 emergency physicians were randomly recruited on the campus of the University Hospital of Cologne, Germany and the EMS Department of the Fire Brigade of Kerpen, Germany. There were no primary exclusion criteria.

Each evaluator had to judge seven video sequences (Fig. 2). Six of those included the six different error types and one did not comprise any error. Only one error was included per video sequence (see Table 1). Participants were asked to evaluate CPR performance after each video sequence, using a standardized questionnaire with multiple-choice answers about compression rate, compression depth, compression point and release after compression. Additionally, participants were not informed about the number of errors in one video sequence and, thus, indication of more than one error was possible. During the evaluation of every video sequence, participantshad the possibility to indicate each category as “not evaluable”. There was no time limit for answering the questionnaire.

The primary outcome parameter was correct identification of the error presented in the video sequence depending on the video quality. Secondary outcome parameters consisted of classification of CPR performance depending on the presented error, false classification of correct CPR performance, frequencies of “CPR performance not evaluable”, indication of further, allegedly errors and finally differences in classification depending on different professions.

Errors presented in the video sequences were evaluated correctly in 71.5% at low quality, in 76.8% at medium quality and in 77.3% at high quality (Fig. 3). Overall, 75.2% of all presented video sequences were correctly evaluated. There was no significant difference in correct error classification depending on quality levels of the presented video sequences (p = 0.306).

Regarding the classification of the specific errors depending on the different video quality levels, results are shown in Fig. 4. Wrong hand position was recognized correctly in 100% at low and medium quality levels and in 97% at high quality level, revealing no significant differences in dependence of video quality (p = 1.0). Likewise, correct CPR performance (p = 0.363), low and high compression rate (p = 0.617; p = 0.847), superficial compression depth (p = 0.619) and incomplete release (p = 0.198) did not show significant differences depending on video quality. In contrast, 13.8% of increased chest compressions at low quality, 50% at medium quality and 42.9% at high quality level were classified correctly, revealing a significant difference in different video quality stages (p = 0.006).

Further analyses focussed on the frequencies of correct error classification depending on the presented error. Thereby, we found a significant difference in correct classification depending on the presented error (p < 0.001; Fig. 5). In the following subgroup analysis, we evaluated differences in correct classification between two errors, as shown in Table 2.

Correct CPR was falsely evaluated as “incorrect” in 41.3% of all videos and thus evaluators attributed errors to the correct CPR performance. Correct CPR performance was falsely classified as low compression rate (21.6%, [n = 8]), superficial compression depth and incomplete thorax release (each 27%, [n = 10]), high compression rate, increased compression depth and wrong hand position (each 8.1%, [n = 3]).

As shown in Table 1, there was only one error per video sequence (except for the video with correct CPR, which contained no error). However, in 184 of 650 video sequences (28.3%), 262 additional errors, that were not shown in the video sequence, were indicated by the evaluators (Fig. 6b). While allegedly errors did not significantly depend on video quality levels (p = 0.422), the presented error was of significance for the indication of additional errors (p < 0.001). Lowest frequency of additional allegedly errors was found in video sequences of correct CPR performance (7 of 92 = 7.6%, Fig. 6a). Highest number of indicated additional allegedly errors was seen in video sequences showing wrong hand position during CPR (43 of 93 = 46.2%), revealing a significant difference compared to correct CPR (p < 0.001).

This study was not powered to find differences in the evaluation of different professions. However, our results reveal some interesting findings regarding interprofessional differences between paramedics and emergency physicians. Summarizing all quality stages, correct classification of the presented errors was achieved in 73.9% for paramedics and in 76.5% for emergency physicians, showing no significant interprofessional difference (p = 0.468). Significant interprofessional differences were only seen in low quality video sequences, where 78.7% (n = 85) of the presented errors were correctly evaluated by emergency physicians, but only 64.2% (n = 68) by paramedics (p = 0.023). In medium and high quality, interprofessional evaluation showed no significant differences (p = 0.873 and p = 0.419 respectively). However, by analysing the results of emergency physicians and paramedics separately, there was a significant influence of the video quality level on correct errors classification by paramedics (p = 0.021) but not by emergency physicians (p = 0.8). For paramedics evaluating a video sequence, increased chest compressions in a low video quality were significantly less frequently recognized, compared to medium and high quality levels (12.5% vs. 60% vs. 60%; p = 0.007). Further, video quality had a significant influence on detection of incomplete thorax release after compression evaluated by paramedics (p = 0.045) but not by emergency physicians (p = 0.393). It was most frequently detected at a high quality level (92.3%), followed by a medium quality level (73.3%) and low quality level (50%). In both professional groups, the presented error in a video sequence had a significant influence on correct error classification (p_paramedics < 0.001; p_{emergency physicians} < 0.001). Finally, there was no interprofessional difference regarding indicating “CPR performance not evaluable” (p = 0.415) and indicating further errors (p = 0.728).

This study was conducted with a resuscitation mannequin made for education and evaluation of medical personal, which is only a substitute for real patients. In accordance with the current European Resuscitation Council (ERC) Guidelines [28], we confined to a compression-only CPR approach. Future studies might incorporate ventilation in their CPR scenario. This study only investigated a maximum resolution of 1920 × 1080 pixels and framerate of 30fps. It is quite possible that a higher resolution and framerate can lead to different results as 4 k resolution (4096 × 2160 pixels) and framerates till 60fps are available. Videotaping was performed in a closed environment with sufficient light, constant weather, no disturbances and perfect technical facilities (e.g. tripod). Further evaluation in real-life situations is recommended. It must also be considered that we only used one video position. Different positions may positively or negatively influence the ability to recognize and correct certain mistakes.