To address this gap, the present study integrates the DSA framework into a hybrid DES-ABM simulation of patient transport. Unlike prior models that treat communication as idealized or implicit, this approach explicitly simulates stochastic information handoffs and awareness breakdowns among agents. By linking communication success to measurable outcomes such as delay and cancellation, the model offers a novel way to assess coordination performance and inform hospital system design from a cognitive perspective.
This study employed DES and ABM to simulate the current and hypothetical operations of interhospital transportation at a level 1 trauma center for studying and assessing SA distribution and transactions quantitatively. In this study, the DES model simulates all the tasks within the interhospital transportation as a series of discrete events while ABM complements DES by simulating various cognitive states (i.e., SA) and actions of the agents. Taken together, the integration of DES and ABM effectively can represent events as outcome of actions taken (i.e., task elements) by particular human agents (i.e., social elements) possessing specific SA (i.e., knowledge elements) that would comprehensively cover the three elements specified in the DSA theory. Portions of the methods and conceptual framework build upon the author’s earlier work with permission and appropriate adaptation for the present study.
The model in this study adopts a pragmatic approach to represent DSA, treating effective communication between agents — particularly the success or failure of critical message-passing — as a proxy for system-level SA. While formal definitions of DSA emphasize shared understanding, coordination, and the ability to anticipate events, our simulation does not directly track the cognitive state of each agent or their shared knowledge base. Instead, we infer that improved DSA is functionally reflected in better operational outcomes, such as reduced delays and cancellations. This proxy approach aligns with the limitations of our data and the operational focus of the study.
The study protocol was approved by the Virginia Tech Institutional Review Board (#17-383).
This DSA simulation model focused on the processes of transporting patients and equipment between multiple locations, including the critical knowledge or SA communicated between agents as adapted from prior DSA modeling efforts. The simulation model was developed with three major steps: (1) building a conceptual model, identifying processes and SA distribution and transactions; (2) collecting quantitative data through observations, interviews, and queries on a database; and (3) using DES and ABM to develop a quantitative model evaluating patient flow and DSA interventions.
The services and process flows modeled in this simulation were selected based on their direct relevance to intrahospital transportation operations, as defined and overseen by the Carilion Transfer and Communication Center (CTaC; the hospital’s command and control center for patient flow). Inclusion criteria required that each process: (1) be actively coordinated by CTaC and the intrahospital transportation team in the hospital, (2) generate observable data within the patient flow management system (i.e., TeleTracking™), and (3) involve observable and measurable tasks and SA transactions between agents. Processes outside the transportation team’s purview (e.g., triage, discharge planning) were excluded, as they fall outside the operational responsibilities of the transportation team. This focused scope allowed for accurate measurement of transport-related performance and evaluation of DSA within CTaC’s operational boundaries.
The conceptual model was derived from the qualitative DSA model of intrahospital transportation (i.e., combined network for clinical transportation) in Alhaider et al., as shown in Fig. 1, and further refined with additional interviews and observations. The intrahospital transportation team consists of dispatchers, team leaders, and transporters. The combined network in Fig. 1 captures the three elements of DSA (i.e., social, task, and knowledge) of intrahospital transportation when initially managed by CTaC. The network identified unnecessary task loops (dotted line) due to missing or inaccurate SA transactions when transporters arrive to pick up patients but are unprepared because ward nurses fail to provide the SA on patient needs, condition, and/or (hand-off) readiness. The operational data affirming this deficiency were captured by TeleTracking, which records patient transport services (i.e., time. location, completed jobs, cancelation/delay cause; refer to Alhaider et al.. The analysis of SA interactions was informed by the Event Analysis of Systemic Teamwork (EAST) framework, which has been widely used to model distributed cognition/SA in complex sociotechnical system. Observational data were analyzed to map agent interactions, knowledge requirements, and communication links, following procedures used in previous studies.
To verify and refine the qualitative model (Fig. 1) for completeness of intrahospital transportation functions and accuracy, interviews involved CTaC intrahospital dispatchers, intrahospital transportation team leaders in the hospital, and clinical transporters, followed by eighteen hours of observation on the intrahospital dispatchers in CTaC and seventeen hours on transporters in the hospital. Figure 2 presents the conceptual model of processes (depicted by the small grey boxes), types of agents and communication between them (depicted by colored circles and arrows connecting them, respectively), and the SA of each agent (depicted by the dotted box inside each process).
The conceptual model depicts clinical/intrahospital transportation at the Carilion Clinic from transport request to patient drop-off at room/exams/discharge. First, the nurse placing the transport request (Np) starts the process of “created request (arrival)” (top left of Fig. 2) entering relevant SA about the transport request (i.e., patient or equipment information) into TeleTracking (TT). The transport request stays pending until a clinical transporter (CT) becomes available to pick up the request. For equipment, the CT obtains the information from TT (i.e., burgundy colored text in the dotted box) to determine where to go and what equipment is required. Then, the CT travels to the origin of the transport request to pick up the equipment from the technician (Tech) and proceed to the drop-off destination and inform the charge nurse at drop-off (Ncd) upon arrival.
For patient transport request, the CT obtains the SA on the transport from TT (denoted as burgundy colored text in the dotted box) to determine what are the necessary equipment (e.g., wheelchair, monitor, stretcher) and destination before traveling to the patient’s origin. Then, the CT travels to the patient and the process afterwards depends on the purpose of the transport:
In this study, DSA is operationalized through the success or failure of information transactions, specifically whether critical cues reach the appropriate agent at the right time. This approach serves as a pragmatic proxy for system-level DSA, focusing on measurable outcomes in the intrahospital transportation process, such as delay and cancellations, which reflect the effectiveness of SA transactions. The model emphasizes dyadic communication protocols (e.g., nurse-transporter hand-offs) due to their critical role in the transportation workflow, as captured by TeleTracking data and observations. While this approach does not directly measure the distribution of shared knowledge elements across all agents or allow compensatory SA sharing, it aligns with the study’s objective to quantify the impact of SA transactions on operational performance in a complex sociotechnical system.
To translate the conceptual model into a simulation model, operational data were extracted from a hospital historical database via TeleTracking while SA transaction and accuracy data were collected by observations as described in previous data collection protocols.
The TeleTracking data provided the following seven types of operational data on intrahospital transportation from February 2020 to September 2021:
The proportions of transport status and pre-scripted reasons for deficient transport were computed for the simulation based on transport data of patients and equipment across different origins and destinations. The transport data were used to compute discrete probability (or multinomial) distributions for assigning status to the transport job requests in the simulation (Table 2).
Transport requests in TeleTracking system may be canceled due to various operational constraints as shown in Table 1. Each canceled request is logged with a specific status and deficiency reason. If re-requested, the transport is assigned a new trip ID and treated as a separate event. In this study, canceled and repeated requests were not merged but analyzed individually to reflect their distinct impacts on resource use and workflow, consistent with how they are recorded and managed in practice.
In this simulation, destination capacity is indirectly incorporated through the workflow defined by Carilion Clinic’s TeleTracking system. A transport request is only generated once the destination confirms availability and readiness to accept the patient. This reflects actual hospital operations, where only the receiving unit can authorize transport, and sending units, such as ward nurses, cannot initiate requests on their own. As a result, bottlenecks caused by downstream capacity constraints are naturally reflected in transport delays or cancellations recorded in the data.
The TeleTracking patient flow management system does not capture any data on the sub-processes carried out during the dispatch and patient periods in which SA elements were being transacted between agents to minimize the risk of delays and cancelations. To simulate SA distribution during the dispatch and patient periods, clinical transport staff were observed and followed during the daytime shift (10 am-4 pm) at Carilion Roanoke Memorial Hospital for approximately 150 h over four months to collect data on the SA transactions. The observations covered 318 transports from the time of the request until drop-off and recorded on the DSA observation sheet (Fig. 3).
The DSA observation sheet recorded which agents held and transacted the SA necessary for efficient transports and which sub-processes might have deficient SA distribution. The fields on the necessary SA and process name can be pre-filled based on the conceptual model. If SA (e.g., patient vitals, nurse information) is not applicable for a sub-process of the transport, SA ratings are omitted. Note that an “agent” may be an information system. In the sheet, delay or cancelation, related to SA or not (e.g., broken bed, emergency with other patients, complications in patient health), was further detailed with (1) what deficiency reasons (Table 1) could be associated with SA and (2) where in the transport process could the reason occur. The DSA observation sheet thus helps identify which SA transactions fail and the failure percentages with respect to particular reasons of delay or cancelation.
RISK® modelling and analysis software application was used to identify statistical distribution for the quantitative data from TeleTracking and observations. The distribution fitting process began by generating histograms for visual comparison of candidate distributions and their parameters (mean, standard deviation, upper limit). Goodness-of-fit tests then evaluated the optimal distribution match, followed by visual verification comparing fitted distributions with the empirical histogram. Table 3 presents an example of the fitted distributions for completion time of the processes and sub-process for transport requests from the 8 floor in hospital. (Note that “create request” and “await assignment” time were based on TeleTracking data rather than records from the DSA observation sheet.)
TeleTracking data on the transport status were also fitted into distributions and Table 4 presents an example of all the fitted transport status of completed transports, delayed, and canceled for the simulation model of the 8th floor in the hospital.
The DSA observation sheet further identified SA-related transport deficiencies. Table 5 shows the proportion of the reasons for delays and cancelations attributable to deficient SA and the corresponding multinomial distribution.
The conceptual model (Fig. 2) was converted into a simulation by first building a DES of the major processes followed by ABM within several of these major processes.
The clinical transportation simulation started with arrival of the request from each origin and ended at patient/equipment drop-off. The processes and agents were modelled with five simulation objects, which are self-contained modelling constructs defined by distributions and parameters based on TeleTracking and observation data:
Using these five object types, six types of routes prescribing different sequence of sub-processes were simulated:
Figure 4 presents an example of simulating one-way patient pick-up from the 8 floor inpatient ward to a hospital unit performing medical procedures using Simio simulation software. Part 1 includes an arrival object to generate transport requests based on the arrival rate distribution. Part 2 involves assigning the transport request to an available clinical transporter in their shifts. Transporters would then check transport requirements and pick up equipment and perform hand-off with the nurse based on the fitted distributions in Table 3. The handoff may lead to cancelation or preparation of the patient for transport. Patient preparation may also lead to cancelation or continuing the transport that involves calling the virtual care unit to confirm pick up status. Process times for hands off, prepare patients, and call virtual care are based on the fitted distributions in Table 3. In Part 3, the transporter formally picks up the patient for transport and then drops off patient at destination with process time based on the fitted distributions in Table 3. Finally, Part 4 is an artificial simulation object (sink) to denote job completion.
To model SA transactions and their operational impact, agent-based objects (e.g., transporters, nurses) were embedded with a state variable of SA and SA-dependent decision making for performing several sub-processes to complete a job request. In Simio, a state variable can be used to represent the utilized knowledge of the object according to some user-defined logic or statistical distribution for the clinical transportation at Carilion Clinic, the simulation model defined ten knowledge state variables (Table 6).
After defining the states, SA transactions were simulated with (Simio) “add-on steps” that define the logic for transacting SA from one agent to another agent for a change in the SA state variable. Table 7 outlines four generic add-on steps within an object to model SA transactions. These steps are built into a process object that requires a transporter to decide on how to continue with the transport.
Figure 5 shows the connection between the add-on steps in Simio. The state variables are first assigned numerical values with percentage to differentiate deficient from non-deficient transports for every origin based on multinomial distributions (e.g., Tables 4 and 5). Then, a decide step follows to direct agent either to a set node step for proceeding to the next sub-process; or to another decide step for directing the agent to encounter a delay step and then proceed to the next sub-process, or to a set node step for job cancelation. These add-on steps define how the agent objects proceed with sub-processes with and without delays and cancelations. For example, Fig. 6 shows the Simio add-on steps for simulating the “hand-off” process in one-way transports from the 8th floor to show portions of delays and cancelations due to deficient SA about equipment, nurse, and patient. Table 5 specifies that 55% and 5% of equipment-, 35% and 30% of nurse-, and 40% and 25% of patient-related delays and cancelations were due to deficient SA, respectively. Alhaider described the exact implementation in Simio.
Combing the DES and ABM in the Simio software, the full simulation model included six separate types of routes highlighted in different colors in Fig. 7. The model consisted of 28 patient origins, 29 equipment origins, 12 destinations, and more than 200 simulation objects.
The full simulation model was modified (after verification and validation) to test two interventions derived from deficiencies highlighted in the DSA combined network model to enhance communication and coordination (i.e., SA transactions):
The intervention of updating the charge nurse would require the transporter assigned to the request to call and inform the inpatient floor/ward nurse about ETA, name of patient to be picked up, and transport destination of the patient, thereby distributing these SA elements to the nurse for timely preparation of the patient for “hand-off”. To incorporate this intervention into the Simio model, a new simulation object representing the calling process (i.e., short phone call) with a minimum of 10 s and a maximum of 20 s was inserted before “hand-off” object (refer to blue box in Fig. 8). The time range was an estimation based on mock calls performed by the transporters during observations. This inpatient floor intervention process was simulated only for one-way transports from 6th floors to seven destinations. This inpatient floor (i.e., origin) was chosen because the number of transport requests exceeded one thousand.
To reflect the change in SA of the nurse as a result of the call by the transporter, new add-on steps were embedded in the “hand-off” process (i.e., the simulation object immediately after the calling process). Figure 9 shows two generic add-on steps for this intervention: assign (Contact Nurse) step and decide (Reached Nurse?) step. The assign step portioned transports into either good or deficient SA, so the following decide step would apply the logic that charge nurse with good SA would lead to transport proceeding without delay or/and cancelation but charge nurse with deficient SA would lead to transport proceeding with a delay or/and cancelation. As a phone call cannot guarantee perfect SA transmission, we simulated that the transport would lead to 50%, 75%, and 100% of transport without delay or cancelation after the phone for sensitivity analysis.
Figure 10 presents an extension to hand-off process in Fig. 6 by incorporating new add-on steps denoted in blue ovals. Alhaider described the exact implementation in Simio.
The intervention to update the X-ray unit would require the transporter assigned to the request to call the X-ray staff regarding ETA and name of patient, thereby distributing these SA elements to improve timely preparation of the X-ray room. To incorporate this intervention into the Simio model, a new simulation object representing the calling was added after “Pick-up patient”. The estimated duration for this intervention process (i.e., short phone call) followed a uniform distribution with a minimum of 20 s and a maximum of 25 s based on mock calls. This inpatient floor intervention process was simulated only for round-trip transports from fourteen floors to the X-ray unit. The intervention modelling in Simio followed the same method described for the intervention to update the charge nurse.
