Abstract
Introduction
Long communication latencies in exploration spaceflight will necessitate in situ resolution to medical problems. Integrating automation into the care paradigm can address challenges posed by resource gaps inherent to spaceflight operations. However, it is not clear what aspects of exploration care are most well suited for automation integration.
Methods
To probe the potential role of automation in spaceflight medicine, we began by decomposing the human-automation system to first characterize the work domain(s) of the human tasks. Using the lens of point-of-care ultrasound, we leveraged existing analogous Earth medical domains to conduct in situ observations in a hospital emergency department to understand how clinicians process contextual information to provide urgent care using ultrasound and semistructured interviews with specialists to identify key procedural information components for automation.
Results
This investigation allowed us to characterize the dynamic system surrounding a task that does not exist in its intended—currently inaccessible—use case (ie, point-of-care ultrasound on Mars) to guide future human-automation systems development.
Conclusion
We conclude that specific aspects of the care environment that influence the result of a task or process (“mediating factors”) from candidate work domains call for distinct, targeted guidance for automation support and are valuable in providing system developers with tunable automation level and implementation guidelines within and/or between those work domains. Such evidence-based design practice is directly translatable to automation assistance for medical providers in resource-limited environments as well as to any situation where a person's sensory processing, perception, decision making, or response selection could be aided by automation to accomplish a task.
Introduction
Future long-duration spaceflight missions will involve limited access to resources, limited potential for rapid return to Earth, and significant communication delays. All these aspects will require self-sufficient medical care provided by the crew—they will need to make quick and accurate diagnostic and treatment decisions. Automating elements of the diagnostic and clinical decision-making process can mitigate the challenges of exploration environments. However, there has not been a systematic assessment of the diagnostic and decision-making process in emergent medical situations, so it is unclear which parts of this process could be automated and where this automation could contribute most meaningfully to self-sufficient medical care. Performing qualitative assessments of healthcare settings on the ground offers an avenue for a systematic characterization of these environments that can be leveraged for future automated system design.
Relevant ground-based domains with emergent, time-critical medical operations include hospital emergency departments (EDs), rural hospitals, and austere/wilderness settings. Although not resource limited, urban hospital EDs illustrate standard best practices in ideal conditions to contrast with rural hospitals and austere/wilderness settings that must contend with increasingly limited resources. While not perfectly reflective of an exploration spaceflight environment, some shared resource limitations include isolated/remote sites of practice, limited access to real-time consultation, delays in access to definitive medical care, limited access to specialized care (eg, facilities and specialist practitioners), unpredictable weather/terrain, and restricted supplies, equipment, and number of providers (Figure 1).
Examples of unique characteristics and similarities between wilderness medicine and long-duration exploration missions.
This systematic assessment required observation of a diagnostic procedure that is used regularly in emergent medical situations on the ground and leverages technology that can be translated to the space exploration environment. Trauma point-of-care ultrasound (POCUS) fit these requirements for several reasons: 1) POCUS is often used in wilderness medicine settings due to its portability, affordability, and image storage and transmission capabilities1,2; 2) in Earth medical care settings, POCUS is used for very specific patient management decisions—the extended focused assessment with sonography in trauma (eFAST) can inform whether to expedite transfer to the operating room to manage internal hemorrhage or place a chest tube to treat a pneumothorax; and 3) ultrasound technology has been used extensively for human spaceflight research on the International Space Station.
This work leverages qualitative data collection and analysis, including structured observations in the naturalistic environment and semistructured interviews with subject matter experts (SMEs) to create a systematic understanding of POCUS that enables the future automation of elements of this process. From the structured observations, we extracted situation-awareness cues to track information processing through the care team. The semistructured interviews with SMEs were used to validate the data collected from the observations and examine transferability between those observations and the space exploration environment. With this work, we establish a systematic understanding of a medical procedure likely usable during future exploration missions and, more broadly, set the precedent for qualitative assessments of operational tasks to better inform technology design.
Methods
This study consisted of two data-collection elements: 1) structured observations to characterize task flow and decision making in the clinical environment and 2) semistructured interviews with SMEs to characterize the transition between high- and low-resource environments. Analysis of these 2 datasets resulted in decision-resource map models (DRMMs) that map task flow and decision making to environments with varied resources. These DRMMs then were validated and assessed for applicability to the spaceflight domain through SME focus group interviews. The 2 data-collection elements were informed by cognitive systems engineering methods, including a domain accessibility decomposition and a human–machine teaming systems engineering guide. This work was reviewed, approved, and overseen by the Massachusetts Institute of Technology institutional review board, the Committee on the Use of Humans as Experimental Subjects. Each element of the methodology is shown in Figure 2.
Elements of the methodologic process for this study.
Systematic Data-Collection Approach Development
A domain accessibility decomposition was conducted to guide systematic data collection in the clinical environment. To structure our observations and interviews, we first decomposed the key spaceflight characteristics of interest by goal (ie, goal of care in that environment), environmental parameters (ie, external factors imposed by the physical environment onto the user/operations), user (ie, training/expertise/skills expected of the user), tools (ie, resources/tools accessible and limitations in the domain), and team (ie, parameters surrounding the composition and coordination of the team within the domain). We then mapped those characteristics to sources from accessible domains as well as the mechanism of data capture (Table 1).
Domain-accessibility decomposition.
Structured Observations: POCUS in the ED
Structured observations were conducted in a hospital ED (Acute Care of Massachusetts General Hospital, which is also a teaching hospital for Harvard Medical School). This setting provided numerous opportunities to observe everyday users of POCUS from expert, intermediate, and learner perspectives and develop a baseline understanding of how clinicians use POCUS for decision making at the bedside. The eFAST is particularly interesting as a case study for investigating areas with the potential for automation in closed-loop medical decision making and is employed regularly in the ED to identify free fluid and/or pneumothorax, among other applications.
POCUS exams are often used for management of urgent and emergent cases, making the ED a suitable domain for understanding the trauma POCUS use case. The diversity in agents, expertise, applications, information sources, resources, and criticality surrounding POCUS also result in a highly dynamic observation environment that requires structured observations to gain understanding of the diagnostic process. We developed a structured observation template to ensure that meaningful and relevant data could be collected rigorously and consistently across observations. The template was developed through pilot observations in the ED (accompanied by a physician) to identify patterns in clinician behavior when performing POCUS and to format the template to be easily navigable and facilitate quick data recording during the exam. Over several sessions, the observer accompanied a physician in various areas of the ED (including fast track, urgent, and acute); many of the patient interactions involved the use of POCUS. Throughout these sessions, the observer took written notes to record consistencies and areas of variability in POCUS scans between patient interactions that would be valuable in the formal observations (eg, how did the practitioners communicate? How did they troubleshoot? What tools did they use to assist in decision making? How did they document outcomes?).
From the unstructured written notes recorded during the pilot sessions, we identified 5 consolidated focus areas that encompassed commonalities in the data: communication (ie, verbal call-outs between individuals), resources (ie, hardware, software, consult, and other), documentation (ie, recording of exam outcomes), information flow (ie, relevant sources of information to the procedure, such as vital signs/symptoms, patient history, symptoms/presentation, and other), and operations (ie, open-response prompts involving practitioner expertise, exam prioritization among competing needs, and other roles that practitioners are responsible for before/during/after the exam). These 5 categories, as well as an unstructured general event notes field for notable data that did not fit within the structured categories, comprised the final observation template.
Formal observations followed the same general structure as the pilot sessions. The observer would arrange to be present in the ED for all or part of a shift (shifts lasting 6–8 h) with an identified practitioner; then the observer would accompany the practitioner throughout their patient interactions and record POCUS observations as scans occurred.
We also noted when the data were recorded relative to the scan (ie, pre, during, and post) as well as attributing the data to a particular phase in POCUS imaging (ie, acquiring the image, interpreting the image, diagnosis using the image, and care planning following the diagnosis) to characterize the temporal aspect of information flow. Given the plurality of practitioners in this process, the template includes to and from fields to attribute data to a particular individual (eg, a verbal call-out was expressed by a supervising resident [From: “Sen. Res.”] directed to the intern conducting the POCUS [To: “Intern”]). The template is provided in the online Appendix A.1.
Participants in the study were required to be adult hospital clinical personnel authorized to provide medical care. Prospective participants were identified via email announcement through relevant email lists and by word of mouth. The population under investigation during these procedures was the clinicians performing the scan, and no identifiable patient information was recorded.
Interviews: POCUS in High- and Low-Resource Domains
Given the resource-limited nature of exploration spaceflight in contrast to the observations of the urban ED, we interviewed physicians/practitioners from rural and austere environments to map the changes between high- and low-resource domains. We engaged specialists with POCUS expertise in virtual semistructured interviews lasting up to 90 min that were designed based on the MITRE Human–Machine Teaming Systems Engineering Guide, shown in Figure 2, as a guiding principle for this work. 3
Interviews focused on the potential for automation in the POCUS process and differences in high- and low-resource settings. Interviewees were recruited from urban hospitals, rural/remote hospitals, austere settings, and spaceflight medicine domains as well as from varying levels of POCUS expertise ranging from first-year residents whose ultrasound knowledge was still being developed, attending-level physicians, physicians who were also ultrasound educators, and wilderness medicine/space medicine experts to identify where informational needs differ based on a user's level of expertise. Interviews were structured to focus on extracting clinical perspectives on predictability, observability, anomaly detection, attention/information presentation, self-monitoring, improvisation, and “job smarts” (ie, unwritten information that is usually learned through exposure/experience 3 ) during the POCUS task to better understand the potential role of automation.
Participants were eligible for the interview portion of the study if they were adult medical personnel who had performed a bedside ultrasound in the previous 5 y. All participants consented to a recorded virtual interview. A prescreening survey (provided in online Appendix A.4) was used to verify eligibility criteria. Interviews were conducted virtually, recorded, and transcribed. Prospective participants were identified via email announcement through relevant email lists and by word of mouth.
Coding (ie, descriptive labels attached to data) was performed on observation data using a software tool (ATLAS.ti 4 ). Digital copies of the observation templates were uploaded into the software, where recorded data were tagged with descriptive codes; ATLAS.ti allows a user to review all data that have been tagged with a particular code. Open coding involves identifying themes and interpretations of text without using a predetermined coding scheme. 5
We used an affinity research approach to elicit patterns and themes arising from interview data. 6 In this method, affinity maps are constructed after data collection by first breaking down the data into small elements and transferring those elements into clusters (ie, groupings of elements that share an idea), identifiable by color coding. For this work, we synthesized general content impressions for each interview question. These content impressions took the form of direct quotes and/or short representative summaries of responses for the participants. These content impressions were copied onto color-coded notes and grouped into shared themes among the data.
Decision-Resource Map Models
Observations and interviews with providers illuminated the roles and responsibilities of different agents for information gathering, processing, and task execution that occurred upstream, during, and downstream of the POCUS exam. We developed DRMMs as high-level visualizations of these major operations surrounding POCUS and to highlight fundamental differences in those processes between the high- and low-resource domains. DRMMs were developed for both ends of the resource spectrum studied to enable structured comparisons between the two. The DRMMs capture the providers (gray boxes), tools/resources that they access (green bubbles), and the diversity in information that they gathered through this process (blue bubbles), which influence the choice of tools and outcomes (brown bubbles). Additional icons include footnotes (purple circles) and identified key decision points (star symbol).
Validation: SME Focus Group Interviews
Once findings from the observation and interview portions of the study were compiled and condensed into clear outcomes, we conducted a study validation step that solicited feedback on our findings from clinical SMEs. Clinical SMEs were engaged to 1) evaluate domain-specific operations visualizations and key environmental care factors synthesized from the observation and interview portions of the study (via a survey [Qualtrics]) and 2) provide feedback on prospective automation integration characterization for translation between high- and low-resource domains extrapolated by the research team throughout the study. This consisted of semistructured virtual discussion interviews lasting up to 60 min in groups of 2 to 3 participants and included both written survey and group discussion portions.
Prospective participants were identified via email announcement through relevant email lists and by word of mouth. Participants were eligible for the study if they were adult medical professionals and had experience with bedside ultrasound. Participants must have performed a bedside ultrasound in the previous 5 y.
Results
Structured Observations: POCUS in the ED
The researcher accompanied 20 participant clinicians as they attended to patients, including attending physicians, residents (including senior residents, interns, specialized procedure and trauma surgery residents), fellows, physician assistants, and medical students. Observed exams included focused eFAST, FAST, cardiac, rapid ultrasound for shock (RUSH), thoracic/lung exams, and applications in guiding procedures and more unstructured investigations of various points of interest on the patient.
Data resulting from the observations, recorded using the observation template, were tagged in ATLAS.ti with descriptive terms that summarized key concepts identified by the researcher. 7 For example, a call-out (communication) from a provider to a patient that instructed them to “take a couple of normal breaths” was tagged as “Patient positioning/action.” A provider documenting that they need to conference with cardiology to see what they needed regarding a scan was tagged as “Follow-on action.” Data could be logged with more than one appropriate tag. High-level codes of perception, comprehension, and projection were extracted to encompass and relate these higher-granularity tags and characterize the phases of the workflow of clinical POCUS (Table 2). Definitions for tags are provided in online Appendix A.5.
Observation coding results.
Interviews: POCUS in High- And Low-Resource Domains
We interviewed 10 (8 male and 2 female) participants (ie, medical doctors from internal medicine, radiology, family medicine, emergency medicine, and aerospace medicine specialties), each of whom had experience in 1 or more of our domains of interest: 3 participants were primarily hospital based, 4 participants had or were currently practicing in rural/remote medicine, 5 participants had backgrounds in austere environments, and 3 participants were experienced in spaceflight medicine/operations (including both government and commercial industries). The mean age of participants was 33.6 y with a standard deviation of 4.3 y. Self-reported medical care expertise ranged from 1 to 12 y, and POCUS experience ranged from 1 to 11 y.
Decision-Resource Map Models
DRMMs for both Level I trauma care hospital operations (Figure 3) and for rural/austere settings (Figure 4) can be used to track the operational differences between resource-varying domains. For both DRMMs, the process structure begins with the patient encounter (top left of Figure 3), visualizes prehospital/care facility providers, the tasks for which they are responsible, and the resulting information outputs. This prehospital track may be bypassed if the patient arrives directly at the hospital rather than by ambulance. In this case, they would enter a triage process to characterize criticality before entering the ED floor.
Urban hospital decision-resource map model.
Austere decision-resource map model.
The next block in the process structure following the patient encounter is titled “Upon ED arrival” and refers to the operations that occur within the destination care facility. The hospital DRMM includes higher- and lower-criticality conditions: For the Level I trauma center, trauma activation criteria define the most emergent case (“STAT”) vs the less emergent trauma case (“Alert”) shown here by red and yellow lines, respectively. The items in this phase are grouped in clusters, where many of the processes happen in parallel.
The cyclic element in the center of the DRMM illustrates the constant reevaluation of patient status. When trend shifts in symptoms, vital signs, or presentation occur, practitioners carry out continuous monitoring, filtering, and reevaluation of the appropriateness of the care paradigm. The small arrows branching from the cyclic element indicate the option to shift from the current cluster/paradigm to a more appropriate one at any time. Additionally, within the circular element, there are standardized classification systems that are practiced in this setting to manage hospital logistics due to the vast number of patients seen each day.
The third block is “Outcomes/management,” where providers use these information sources to intervene in many ways that are accessible to them.
Within each block of this DRMM, there are roles associated with specific providers. For example, the “Lower criticality” and “Highest criticality” clusters include paramedics, ED technicians/nurses, physicians and physician assistants, specialized trauma and ED teams, lab technicians, scribes, and administrative staff. Each of these roles is responsible for structured tasks within the hospital ecosystem. Because of this dispersal of responsibilities across a large team, many tasks happen in parallel. For example, nurses/ED technician providers can be applying vital sign monitoring devices and drawing blood, whereas physicians can be collecting history, characterizing the physical appearance, and conducting a physical exam, all while a scribe documents. At the same time, although the “Highest criticality” condition includes many of the same resources present in the “Lower criticality” condition, due to the quick initiation of a trauma team protocol, these resources can be put into effect expeditiously; that is to say, the biggest difference between the “high” and “low” criticality tracks is not the types of resources available but rather the speed with which they are initiated.
Due to the nature of this type of hospital environment (a Level I trauma center), the “Outcomes/management” cluster contains extensive options for management, including direct procedures/interventions, operating rooms or other floors/specialties, hospital admission, and transfer to another facility if needed. The result is that this DRMM includes access to “definitive care” or access to the best and comprehensive route of care for a given disease/disorder. 8
Key takeaways resulting from the hospital (urban) DRMM include 1) due to the breadth and depth of the team with clearly allocated and familiar roles and responsibilities, many processes can happen in parallel, leading to quick timelines; 2) due to the vast amount of resources, “definitive care” is always within reach for these providers; and 3) there is strong usage of standardized classification systems (eg, trauma activation criteria and emergency severity index) to manage hospital logistics and patient throughput, and these standardization systems predictively regulate operations.
The general structure is consistent for the austere DRMM: There is a “Precare facility” block (rather than “Prehospital” because the care facility is not limited to a hospital in the austere setting), followed by an “Upon care facility arrival” block (where interventions may take place) and an “Outcome/management” block.
These 3 blocks (and their associated providers/tools/information sources) may be present in some types of austere conditions, such as rural hospitals. However, the goal of this DRMM is to capture a spectrum of austere environments, from rural hospitals to more remote/wilderness conditions, with spaceflight lying on the most austere end of the spectrum (as illustrated in the austerity spectrum in Figure 4). In the most austere conditions, there may be no care facility available, and most (if not all) of these resources may be inaccessible. Instead, there may be only 1 responder, who would be responsible for evaluation, intervention, and management, all at 1 site. Further, they may be limited to their own skills/expertise, tools on hand, and the potential for remote consult and evacuation/transport. For this most austere case, a “Minimum provider available” block is listed as the first item in sequence (top left of Figure 4).
In this DRMM, there are fewer gray provider boxes to reflect the smaller number of specialized providers available in more austere settings. These boxes include more general task/role language, with purple text to indicate when the team is likely to be even smaller for the more austere cases (ie, when a provider/resource may not be available). Therefore, a key process takeaway that is illustrated in the austere DRMM is that in the case of smaller teams, the scope of responsibility (eg, physical tasks and decision making) can be much broader than in a hospital to account for missing team members.
Within the “Upon care facility arrival” block, there is a sequence of events following triage that begins with the provider(s) determining a differential diagnosis and appropriate management course of action, establishing a pretest probability of injury, and initiation of transfer and/or remote consult. The following elements are “Primary survey” and “Secondary survey,” leading to continuous provider monitoring through symptom/vital signs/presentation vigilance. The conclusion of this element returns to the first block in the post-triage sequence, where again the provider evaluates the differential diagnosis/probability of injury and determines whether a care paradigm shift is warranted (in the context of current patient status). This central patient care/monitoring loop departs from the parallel process seen in the hospital (urban) DRMM as a result of fewer providers who are able to conduct many processes concurrently. Therefore, this DRMM illustrates a distinct cycle of events that can and is likely iterated as needed but happen relatively sequentially rather than in parallel. A resulting key takeaway from this DRMM is that smaller teams also precipitate a shift from parallel processes to sequential tasking, iterating as needed, resulting in a slower timeline of care.
The primary survey element DRMM also includes a distinct awareness of “actionable information” regarding patient management (ie, what information can lead to actions that are accessible to caregiver vs information that requires resources and/or expertise that is not available for management) that must be discerned by the provider(s) when initiating care. In the first element following triage, there is also an illustrated use of patient population and injury risk/heuristics to guide care. Within the outcomes/management block, the outcomes include the understanding of the “optimal intervention” for the situational care level (ie, the best care that can be provided given the resources at hand). Finally, in austere conditions, there are often mission operations and/or defined priorities that impact the boundaries of care provision. Mission operations implications may include events where considerations need to be made outside the patient, such as the health/survival of the team, mission, or vehicle. In these events, differentiation between critical care (eg, resulting in evacuation) and care that would benefit from early care but may be delayed to complete the mission may be required.
Results: Mediating Factors
Through coding and the affinity map method to generate themes in the data, we recognized distinct variables that influenced care practices in each domain. Each domain (ie, Hospital [Urban], Austere, Rural, and Spaceflight) reflected unique aspects that impact POCUS operations. Five shared categories related to direct domain impacts on POCUS operations arose from the affinity mapping: “Treatment/differential diagnosis/planning,” “What you see,” “Equipment you have,” “Who you are with,” and “What your options are.” By analyzing these patterns in operations relative to the different domains, an emergent set of factors arose in the data; these are key system attributes that mediate how a user executes their task (ie, POCUS and broader medical operations). Attributes that arose from the analysis are “Mediating factors” (MFs)—key system attributes that mediate how a user executes the task at hand. They include comfort zone, filtering/information relay, actionable information, technology interface, timeline, mission operations, cognitive burden, gestalt builders/intuition, team composition, and training. Each is described in Table 3.
Mediating factors: key system attributes that mediate how a user executes the task.
SME Focus Group Interviews
SME focus group interviews were used to validate that the contents of the DRMMs identified through observations and interviews were representative of the POCUS care process. Eight participants (6 males and 2 females) included medical doctors (internal medicine, emergency medicine, pediatrics, toxicology/addiction medicine, aerospace medicine, POCUS subspecialty), a physician assistant (emergency medicine), and a registered diagnostic sonographer. The mean age of participants was 41 y with a standard deviation of 13.0 y. Self-reported medical care expertise ranged from 6 to 37 y, and POCUS experience ranged from 2 mo to 14 y. Participants included government and commercial spaceflight medical professions. One participant completed the survey only and did not participate in the discussion portion of the interview.
The survey was completed via the Qualtrics survey platform. In individual virtual breakout rooms, participants watched a prerecorded orientation video to introduce and describe the study goals, DRMMs, and mediating factors content. Participants then were asked to provide ratings and comments regarding the extent to which the DRMMs and MFs compared with their experience/understanding. The survey questions can be found in online Appendix A.6.
The hospital (urban) DRMM was identified as “accurate without significant adjustments” or “some adjustments helpful but not critical” for 7 of 8 ratings. Small, noncritical adjustments were suggested to improve interpretation. The critical adjustment comment reflected the inability of the hospital (urban) DRMM to translate to low-resource settings; the purpose of the austere DRMM was to encompass the low-resource setting (rather than the hospital [urban] model). This comment was concluded to be a result of miscommunication in the orientation materials (discussed further under “Conclusions”). These adjustments were integrated into the final DRMM.
The austere DRMM was evaluated to be “accurate without significant adjustments” or “some adjustments helpful but not critical” in 5 of 8 ratings. Suggested adjustments, both noncritical and critical, were largely framed around the difficulty in capturing such a wide spectrum of care in a single diagram. In response to these comments, the DRMM (Figure 4) was updated to include a pathway for the most austere (barest resource) scenario (“Minimum provider availability”) as well as more effectively communicate the main pathway that includes resources that may be available in other austere settings (identified by purple text).
The relevance of the MFs was rated as “accurate without significant adjustments” in all 7 recorded responses (1 participant did not provide a rating for this question). SMEs provided open responses on the importance of patient factors (eg, relevant patient history/role in team), team familiarity, ultrasound fluency, and gravity impacts on imaging; these items were reflected in mission operations, training, comfort zone, and technology interface, respectively. These results are summarized in Tables 4 and 5.
Subject matter expert group interview decision-resource map model survey results summary.
Subject matter expert group interview mediating factors survey results summary.
a One participant did not provide a rating for the question regarding mediating factors’ relevance.
These suggested adjustments were integrated into either the DRMM and/or into the content description. The final versions of the DRMMs and MFs with SME feedback incorporated are included in Figures 2 and 3, respectively.
The second portion of the SME focus group interviews included a discussion-based format, where we asked for open-ended feedback regarding the following prompts:
How might automation play a role with respect to these mediating factors as you transition from high- to low-resource domains (Hospital [Urban], Austere [Spectrum], Spaceflight)? Can you provide an example of automation or artificial intelligence applicable or suggest a type/application of automation that would be beneficial?
The comments were recorded and integrated into an automation support for mediating factors table (see Table 6).
Discussion
Structured Observations and Interviews
The interview process required active adaptation to gain the deepest understanding of the problem space. While we were conducting the first sets of interviews, we noticed preliminary themes and concepts in the data that we wanted to investigate in further detail in subsequent interviews. We identified factors associated with decision making, teaming, training, and domain translation that were not interrogated in detail in the current interview structure. We developed an alternative script to elicit more details on these factors from the spaceflight medicine subpopulation of the participants. The alternative script (provided in online Appendix A.3) focused on factors relating to the translation of clinical care between these settings into exploration spaceflight, containing clusters of questions themed around: decision making and planning (perception, comprehension, and projection components of clinical decision making); procedures/task execution (task execution and reassessment/reiteration components of clinical decision making); team training and crew composition; and a discussion of how a shifting domain may impact aspects of care identified in early interview analysis. Seven interviews followed the framework-derived structure, and 3 interviews used the alternative/targeted script.
DRMMs and Validation with SMEs
DRMMs illustrate the constraints in hospital and austere settings resulting from access to tools/resources, informational inputs, and team composition. These models emphasize how a provider’s scope of responsibility (eg, for decision making and hands-on tasks) widens as resources become more limited. Further, in comparison with each other, these DRMMs reveal the deviation from parallel, simultaneous tasking in high-resource care to sequential, iterative operations in low-resource care, which lengthens the timeline of care. In low-resource scenarios with longer care timelines, some variables become more impactful in constraining the operations, such as the processing of actionable and prioritized information and the presence of established mission operations protocols. When considering how automation may support care operations before, during, and after a task, it is critical to understand how the constraints of the environments impact those operations. This understanding enables system developers to target areas where automation implementation is most appropriate for resource and information management within the team.
Mediating Factors Automation Support Table
Table 6 summarizes three main components for each MF: 1) high-/low-resource transition: how automation can support a general transition from high to low resources; 2) domain implementation concept: domain-specific concepts for automation implementation with illustrative examples; and 3) concept automation level (range): conceptual ranges for levels of appropriate automation corresponding with those domain implementation concepts. Automation levels follow definitions described by Parasuraman, Sheridan, and Wickens. 9
Mediating factor automation support table.
Although the table describes each of these items for every MF, some general trends emerge from the table. In well-controlled settings where individuals are working within their familiar roles and responsibilities (eg, as an urban hospital), the concepts typically include lower levels of automation, where the user retains more decision-making authority as well as being made more aware of how the information is being used to complete the task. Also, some MFs (eg, comfort zone) include higher transparency of the automation in this domain because hospitals function as a training setting where users need to learn and develop their skills to build proficiency in performing the task independently in the future. In comparison, in the spaceflight setting, we are not training for independence from automation but instead are using automation permanently to augment crew capabilities. Therefore, more austere conditions often involve higher-concept automation levels.
There are also concept implementations where the automation level is higher across all settings, such as in cognitive burden, training, and timeline. Cognitive burden reflects the universal goal of reducing the risk of task saturation; this implementation can be accomplished by predicting when task saturation may occur and offloading noncritical responsibilities from the decision maker. Higher automation levels for task training were seen as beneficial across settings, with implementation concepts including use as a practice tool, monitoring/tracking training deficiencies, and customized training that reflects the unique needs of the user. Maintaining an updated timeline of care (eg, how the choice of certain exams, procedures, and logistics of time-delayed ground consult will impact the timeline) was an area with medium to high levels of automation among all care settings.
The effects of the MFs are contingent on the work domain. For example, actionable information is less impactful on high-resource hospital users, where definitive care is always within reach.
Limitations
Due to constraints in accessing observation sites, POCUS observations were limited to high-resource care, namely a Level I trauma center. This site was excellent to establish a baseline understanding of POCUS operations for high-resource settings, although observations in low-resource sites to supplement interviews would have been insightful for developing DRMMs and MF analysis. Further, the urban hospital DRMM that resulted from these observations depicts the most well-resourced hospital condition.
Although observations and SME focus group interviews included a range of participant disciplines (eg, MD, PA, sonographer), individual interview participants who volunteered for this study were limited to physicians only. Diversity in interview participants likely would further expose automation support factors (potentially impacting MFs) related to occupation and skill.
Conclusions
Through this analysis, we exposed gaps that arise in care when shifting from high- to low-resource environments. Identifying these gaps is invaluable in human–machine teaming system design because it allows developers of automated systems to target support systems for points in the care process when domain effects more strongly influence practitioners. This process is necessary when making automation design extrapolations onto a prospective workflow from an analogous site to a novel domain, such as when designing for the envisioned world problem. 10 To our knowledge, this visualization of care operations and resulting synthesis of direct domain effects on a user for emergency/exploration care have not been performed in previous work. A review of low adoption rates of existing automated systems for decision support in clinical applications included that such systems do not seem well matched to the reality of the clinical workflow. 11 This work contributes to reducing this weakness in system design by incorporating real-world observations and insights from users who are experienced with POCUS and with operations in each of our domains of interest.
To further identify how automation can close the gaps created by MFs when transitioning domains, we implemented clinician feedback into generated concept implementations and levels of automation in response to each of these MFs in hospital (urban), austere (spectrum: rural–wilderness), and space domains (mediating factors automation table). This table can be referenced by system designers to illustrate how the function and magnitude of automation shift as resources become scarcer and when comparing between MFs. This work supplements previous specialized applications in supporting ultrasound with automation/autonomous algorithms12–14 by providing more general POCUS operations concerns and guidance for automation, including in resource- and expertise-limited domains and users.
We have developed guidelines for integrating automation for crew health and performance in Earth-independent missions 15 and incorporated terrestrial medical care practices modified for the novel planetary exploration environment. 16 We also demonstrated human-in-the-loop design (participatory design 17 ) to secure continuous, multisource domain expertise input throughout our process to directly inform our design conclusions. This work contributes to the efforts of the National Aeronautics and Space Administration space medicine experts to inform components of Earth-independent medical operation systems. 18
Supplemental Material
sj-pdf-1-wem-10.1177_10806032251351589 - Supplemental material for Qualitative Assessment of Terrestrial Care Settings to Inform Self-sufficient Spaceflight Medical Care
Supplemental material, sj-pdf-1-wem-10.1177_10806032251351589 for Qualitative Assessment of Terrestrial Care Settings to Inform Self-sufficient Spaceflight Medical Care by Allison Porter, Katya Arquilla and Aleksandra Stankovic in Wilderness & Environmental Medicine
Footnotes
Acknowledgments
We thank Prof. Dava Newman, Dr. Andrew Liu, and Dr. N. Stuart Harris for their support of this work, their invaluable input on the design of this investigation, and their feedback on the preparation of this manuscript.
Author Contribution(s)
Financial/Material Support
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by NASA HRP Grant 80NSSC20K0844 and by student fellowship support from the MIT Morningside Academy for Design.
Consent to Participate
Consent to participate in this study was obtained through a combination of verbal (observations) and written (observations, interviews, and subject matter expert focus group interviews) consent.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
