The modeled cABCDE trauma treatment process starts with the patient arriving in the resuscitation room and the handover between emergency services and the trauma leader. Initial information about the patient (e.g., gender, accident type, time of accident) must be entered into the system. If necessary, the resuscitation and circulatory shock subprocess can be started directly and at any time during the process via a BPMN non-interrupting call activity, which starts the subprocess parallel to the main process. The main process inherits the variables automatically, and the subprocess can return the documented treatment variables to the main process.

The main process is divided into grouped subprocesses, which are called via BPMN signal events and separate the individual treatment stages of the cABCDE approach. For example, the “A-Airway process” is shown in Fig. 2. For user interaction, the BPMN user task element is used, which creates an editable form during workflow execution. In this way, data can be submitted to the system, or a selection between different recommendations can be made. If a decision is made, an XOR gateway is used whenever only one subsequent sequence flow is selected. AND gateways are used to start parallel processes. For complex decisions and calculations, DMN tables are integrated into the process model via BPMN decision tasks. Also, external Java Code files (e.g., for protocol generation) are executed and triggered via BPMN service tasks.

Example subprocess—Airway assessment in TraumaFlow modeled with BPMN

In this way, the patient is evaluated by going through the cABCDE stages. At the end of each stage, the user is asked to reevaluate the finished stages (e.g., repeat B-stage when the breathing deteriorates). After a successful initial treatment in the resuscitation room, the patient is transferred either to the operating room or to the intensive care unit, and the process can be finished in TraumaFlow. Afterward, it is possible to generate treatment documentation.

In Fig. 3, the TraumaFlow user interface is shown, which is the visualized workflow instantiation of the modeled treatment process. On the left side, the currently active tasks are color-coded by priority. On the right side, detailed information about the current task and decision options is displayed. For every task, a ‘Back-Button’ was modeled to enable a return to the previous task. TraumaFlow provides additional features, such as a timer for the main treatment and the resuscitation process. During treatment, different medical scores (e.g., Injury Severity Score) can be calculated in real-time to improve decision-making. All the system-recommended actions can be traced through the underlying process model for explainability, which is also accessible during runtime. In addition, the WfMS logs every executed task and decision made to the second. Based on the system logs, TraumaFlow enables detailed documentation of the training session. Hence, a prefilled hospital internal handover protocol, the trauma protocol of the DGU trauma register, and a detailed training protocol are generated at the end of every process execution.

TraumaFlow user interface

To evaluate the guideline adherence, every scenario execution was rated by an experienced trauma instructor with up to 12 points for correctly recognized and applied treatment. In Fig. 4, the number of participants that perform essential treatment steps is shown for scenarios with and without TraumaFlow. For example, of all 14 participants, only five performed trauma treatment following the cABCDE scheme without decision support. In addition, tranexamic acid was not given once. With the use of TraumaFlow, all participants followed the cABCDE approach and applied tranexamic acid as proposed in the S3-Guideline.

Essential treatment steps performed by 14 participants. Thereof, nine steps are generic (max. number of participants is 14) and three are scenario-specific (max. number is seven due to altered scenario execution)

In Fig. 5, the number of evaluation points achieved by the differently experienced participants is shown. Participants with low and medium experience benefit most from the decision support (mean improvement for low experience: 5.1 points and medium experience: 5.7 points). Medium- and highly experienced participants performed perfect treatment with decision support. In total, a significant improvement in guideline adherence and completeness of treatment was achieved with the use of TraumaFlow (6.6 ± 1.2 points without vs. 11.6 ± 0.5 points with TraumaFlow, p < 0.00001 (Mann–Whitney-U-test)). A detailed analysis of the clinical evaluation can be found in [27].

Number of evaluation points achieved by differently experienced participants

During scenario execution, the obstacles in the interaction of the participants with TraumaFlow were recorded. In Fig. 6, the mean number of usage errors per session is shown. ‘Unnecessary’ repetitions occurred due to contradictory parameter inputs, which is a security feature to reevaluate the user input. Repetitions were noted 0.6 ± 0.9 times per training scenario. Operating errors occurred due to ambiguous questions of the system 0.6 ± 0.6 times on average. To correct the user input, the ‘Back-Button’ was used 0.6 ± 0.8 times per scenario. In particular, active decisions against (0.2 ± 0.6 times per scenario) and without the recommendations of TraumaFlow (1.9 ± 2.5 times per scenario) were mostly made by medium- and highly experienced participants. Wrong parameter inputs that led to incorrect recommendations were noted in 0.4 ± 0.7 events per scenario.

Usage errors of differently experienced participants

In addition, system errors such as user-unexpected recommendations, wrong recommendations, wrong process paths, and slow responses of the system were noted (Fig. 7). For the user, unexpected recommendations occurred 11 times in total. Thereof, two were wrong recommendations.

System errors during all training sessions

The usability of TraumaFlow was evaluated by the participants with the standardized UEQ (User Experience Questionnaire) [28], which is used to assess the acceptance and users' subjective experience with software systems. The UEQ consists of 26 items related to six categories: Attractiveness (aesthetic and emotional impression), perspicuity (how easy to learn), efficiency (solving tasks without unnecessary efforts, fast usage), dependability (how secure and predictable), stimulation (how exciting and motivating), and novelty (how creative and innovative is the design). Users rate each item on a scale from − 3 to 3. Positive values over one represent a positive evaluation. TraumaFlow was evaluated positively in 5 of 6 categories (Fig. 8). The dependability was rated neutral due to slow system performance during several training sessions and user-unexpected recommendations.

Results of the UEQ Questionnaire

In addition, the participants were asked to answer seven questions about the user experience with TraumaFlow (Fig. 9). The large majority (92%) of the participants felt more confident with computer-aided decision support and considered TraumaFlow useful for the training of polytrauma treatment (92%). In addition, 62% rated the training effect with support higher than without, and 85% believe that TraumaFlow can be helpful in a real treatment situation. Most participants (69%) could use the system with the tablet device without problems, and 85% considered the recommendations of the system plausible. While, 69% think the system behaves as expected, 15% experienced unexpected behavior.

Results of the Likert Scale questions about user experience with TraumaFlow

For all training sessions, the treatment duration was recorded (Fig. 10). The participants were not given a time limit during the training, but the vital signs of the mannequin responded to incorrect or delayed treatment steps. Without TraumaFlow, the treatment was significantly shorter than with the system (mean time without: 5.9 min ± 1.9 min versus 11.7 min ± 2.3 min with TraumaFlow, p < 0.0001 (Mann–Whitney-U-Test)). The initial treatment in the real-world resuscitation room usually takes approximately 20–30 min until the patient is transferred to the operating room or intensive care.

Treatment duration with and without TraumaFlow