Avatar-based versus conventional vital sign display in a central monitor for monitoring multiple patients: a multicenter computer-based laboratory study

We developed the Visual Patient Technology in analogy to the synthetic vision technology used in aviation. Synthetic vision renders a virtual image of the flight situation from satellite position data, aircraft attitude (orientation in space), altitude, heading, terrain and traffic data, and other data available in an aircraft. The resulting image looks to the pilots as if they looked out of the window in clear weather. A lake looks like a lake, a mountain like a mountain, etc. Therefore, the flight situation can be interpreted more intuitively than when the pilots have to compile this “image” from lower-level data in their minds, as is the case with conventional instruments. According to user-centered, situation awareness-based design principles and standard works of logic, a model is then ideal if its design has a logical relation to the reality it tries to represent, also termed direct display of information [4, 22]. Just like synthetic vision technology, Visual Patient Technology displays the monitoring information in a way that corresponds to the expected phenomena in the real patient it represents. For example, the Visual Patient avatar becomes cyanotic (skin turns purple) if oxygen saturation is low, or it emits thermal radiation from its body if the body temperature becomes too high. The realistic depiction of reality is also the reason why the patient avatar is upside down in the anesthesia view tested in this study: because this corresponds to the common viewing position of the anesthesiologist on the patient. The three main characteristics of the avatar are the pre-processing of the data, the direct presentation of the information and the integration of the vital signs in several visualizations, e.g., care providers can judge the respiratory rate on the basis of the respiratory rate of the lung as well as the formation rate of the carbon dioxide cloud exhaled by the patient model. These characteristics translate a large number of numerical values and waveforms into an animated model of the patient’s situation, which the care provider can evaluate and store at a glance without first having to read and derive meaning from data values. The translation of the vital signs from the numerical data into the avatar model takes place in real-time as with regular monitoring. Since we started developing and studying Visual Patient Technology in 2012, we found that the technology, compared to conventional number and waveform-based monitors, improved the perception of vital sign information in anesthesia providers, reduced the perceived workload, and received positive user feedback [23, 24]. Additionally, the technology improved perceptive performance in a setting with a distraction [25]. In eye-tracking studies, we found that the technology works because users do not need to read numbers sequentially but can receive information from looking at colorful moving objects, which allows for monitoring using peripheral vision and enables parallel perception of multiple vital signs at the same time [26, 27].

This study was an investigator-initiated, within-subject, prospective, multicenter trial comparing two different multiple patient monitoring interfaces. We conducted this multicenter study in Switzerland with anesthesia providers from the University Hospital Zurich (USZ) and the Cantonal Hospital Winterthur (KSW).

In this study, we included resident and staff anesthesiologists and certified anesthesia nurses who worked in the two study centers at the time the study was conducted. These anesthesia providers were invited to participate and relieved from their duties for the duration of their participation on a regular working day. The participants were included according to an inclusion plan, which served to ensure that the sample included equal numbers of participants of all professional groups and genders in both centers. Eight participants in Zurich and 8 participants in Winterthur assessed the 10-s scenario and 8 participants in Zurich and 14 participants in Winterthur evaluated the 30-s scenario. The last group included more than the minimum of 8 participants required in each group because 6 additional participants were institutionally planned to participate in the study, and accordingly available on the days we tested the 30-s scenario in Winterthur. Participation in this study was voluntary and there was no monetary compensation for the participants. All participants signed a written informed consent form agreeing to the use of their data in anonymous form for scientific purposes.

Before the actual study, we conducted a pilot study with five participants in order to estimate the expected effect size as the basis for sample size planning. These five participants evaluated three 10-s scenarios. In this pilot study, the mean of differences between the participants’ performance with avatar and conventional monitoring was two vital signs with a standard deviation of the differences of 0.8. Using the standard deviation of the differences obtained in the pilot study and a mean of differences of one vital sign, which we considered to be the minimum clinically significant difference, an effect size of 1.25 resulted. For an alpha-error probability of 0.05 and a power of 80%, the required sample size was 8 participants. To detect a difference of one vital sign with the specified power at the 5% significance level in each center, for both scenarios, and both viewing durations, we had to include at least 16 participants per center, or 32 participants total, who each evaluated two scenarios. The computer program used for the sample size calculations was G*Power 3.1 [28].

The participants sat with a data collector in a room where they could observe and evaluate the patient monitoring scenarios undisturbed. Before each session, participants completed a demographic survey, watched a 6-min educational video about Visual Patient Technology (Additional file 2: Video S1), and familiarized themselves with the layout of the conventional patient monitoring interface used in this study.

For the study, each participant evaluated a total of four scenarios of either only 10-s or only 30-s duration. Figure 1 shows a flowchart of the study procedure. We showed participants the 10-s scenarios to simulate at-a-glance monitoring. At-a-glance, or observing the monitor in short glances, corresponds to real-life monitoring behavior of care providers as reported by several studies [7, 8]. We used 30-s scenarios to evaluate how the two monitoring technologies compared when users would be given significantly more time to observe and memorize a scenario. The four scenarios consisted of two identical scenarios, which were shown to each participant once in avatar-based and once in conventional form, hence the total of four scenarios shown. Participants were not aware of the two identical scenarios. The first scenario was picked at random and the following scenarios were presented in interchanging order between scenarios so that the same scenario never appeared twice in a row. The study design allowed us to compare the performance of each participant with the two different monitoring technologies, which allowed for intra-individual comparisons. After each scenario, participants had to name aloud the status of the vital signs they remembered of the critical patients. If a participant could not spontaneously name the status of all vital signs, the data collector queried the vital signs not yet actively mentioned by the participant. For example: “You have not yet named the status of respiratory rate for critical patient number 2, do you remember it? The participants had to choose between the response options “too high”, “normal”, “too low“ or “not perceived”. Also, after each scenario, participants completed the National Aeronautics and Space Administration (NASA) Task Load Index (TLX) questionnaire, a tool to measure perceived workload. It was originally developed for use in the aerospace industry, but has since been extensively used and validated also in human factors and healthcare research [29‐37]. We found validation for the use of NASA TLX to measure subjective workload during patient monitoring tasks in previous studies. In one study, a standardized distraction by a calculation task increased the workload measured by NASA TLX in both avatar-based and conventional patient monitoring [25]. Also, in a previous study with a comparable setting, we found shorter observation times to be associated with higher NASA TLX scores [23]. We collected data for this study using an iPad-based data collection tool [38].

To attain equal conditions for all participants, we recorded and played back the scenarios as videos. Figure 2 and Additional file 3: Video S2 shows the original scenarios used in this study with multiple patients represented at the same time. The conventional monitoring scenarios showed only a conventional number- and waveform-based monitor modelled after a GE Datex Ohmeda monitor (General Electric Company, Boston, MA, USA) using the simulator app SimMon (Castle 2 Andersen ApS, Hillerød, Denmark). The avatar-based scenarios showed the animated avatar in addition to the conventional display. To attain optimum realism, we highlighted the abnormal vital signs in all scenarios, as it is commonly used in modern day conventional patient monitors.

Per scenario, we showed six patients, two of which were critical with multiple vital sign abnormalities, and four of which did not show any vital sign abnormalities. For each patient, the following 11 vital signs were displayed: pulse rate, arterial blood pressure, oxygen saturation, central venous pressure, electrocardiogram ST-segment, respiratory rate, tidal volume, expiratory CO2 concentration, body temperature, brain activity, and neuromuscular conduction.

The critical patients in scenario 1 were a patient in septic shock and a patient with endotracheal tube obstruction. We simulated septic shock as pulse rate, respiratory rate and temperature too high, blood pressure, oxygen saturation and end-expiratory carbon dioxide concentration too low, other vital signs normal. We portrayed endotracheal tube obstruction as oxygen saturation, end-expiratory carbon dioxide concentration and tidal volume too low, other vitals normal.

The critical patients in scenario 2 were a patient in cardiopulmonary arrest and a patient with malignant hyperthermia. We simulated cardiopulmonary arrest as pulse rate, blood pressure, oxygen saturation, end-expiratory carbon dioxide concentration too low, ST-segment too high, other vital signs normal. We presented malignant hyperthermia as pulse rate, end-expiratory carbon dioxide concentration and body temperature too high, blood pressure too low, other vitals normal.

The primary objective of this study was to compare the perceptual performance of anesthesia providers observing central station patient monitor scenarios with both technologies. Therefore, we compared the number of vital signs participants were able to correctly recall with both monitoring technologies. A secondary objective was to compare the perceived workload during the task with both technologies. We evaluated perceived workload because high workload is a psychological stress factor. When workload demands exceed maximum human capacity, it interferes with our ability to remain aware of the situation at hand and will ultimately negatively affect decision making and task performance [3‐7, 18, 35, 39]. In high workload situations, humans are susceptible to attentional tunneling and premature closure, the tendency to rush a decision without considering all aspects first. The goal of successful situation awareness design has been described as transferring the relevant information to the user with the least effort [4]. Another secondary endpoint of this study was the evaluation of the frequencies with which individual vital signs were correctly recalled by the participants.

Figure 3 compares the anesthesia providers’ perceptual performance between avatar-based and conventional monitoring in the 10-s scenarios. Using avatar-based monitoring, anesthesia providers perceived a median of 11 (IQR 8 to 15) vital signs, compared to only 7 (IQR 4 to 9), using conventional monitoring, p < 0.001, mean of differences 4 vital signs, 95% confidence interval (CI) 2 to 6, effect size d = 0.67. Figure 4 compares the perceived workload in the 10-s scenarios. Using avatar-based monitoring, anesthesia providers had a lower median NASA TLX score of 67 (IQR 51 to 75), compared to 77 (IQR 51 to 84), using conventional monitoring, p = 0.034, mean of differences 6 points, 95% confidence interval 0.5 to 11, effect size d = 0.43. In the 30-s scenarios, the differences between the avatar and conventional monitoring were not statistically significant. Anesthesia providers perceived a median of 16 (IQR 13 to 18) vital signs with the avatar, vs. 15 (IQR 13 to 17), using conventional patient monitoring, p = 0.055, effect size d = 0.33. Likewise, NASA TLX scores did not differ in the more extended 30-s scenarios with a median score of 60 for both technologies, p = 0.59, effect size d = 0.13. Additional file 1: Figures S1 and S2 show these results. Additional file 1: Table S1 outlines the evaluation of the frequencies with which the individual vital signs were correctly recalled. In the 10-s scenarios, only two of 11 total vital signs, blood pressure and expiratory CO2 concentration, were perceived more frequently with conventional monitoring than with avatar-based monitoring. The other nine vital signs were correctly recalled more frequently or equally frequently with Visual Patient Technology.

This study was conducted as a computer-based laboratory study and as such has some specific limitations. Most importantly, this study did not test any patient outcome measures and only tested the technology in an undisturbed and simulated environment, which excluded potential effects of distractions that could be present in a real environment. It is, therefore, unclear how the results would translate into a real operating room or intensive care unit environment. Also, workload was measured with the NASA Task-Load-Index. Although this is a commonly used tool and, in our previous studies, we found indications of its validity in patient monitoring tasks, there are other, more direct, measures of stress, which we did not assess in this study, e.g., participants’ heart rate variability [42], pupil dilation [43], and blink frequency [44].

This multicenter study found a clinically and statistically significant performance improvement in the 10-s scenarios with Visual Patient Technology. The variety of anesthesia providers included into this study, with different professional experience and positions, as well as age and sex, increases external validity. The intra-participant design decreases potential effects of random noise.