Sequence matters: patterns in task-based clinical reasoning

Background

Clinical reasoning is a complex phenomenon that is associated with taking a history, performing a physical exam, ordering and interpreting laboratory and/or radiographic tests (at times) as well as designing a management plan that is appropriate for a patient’s circumstances and preferences [1], [2], [3]. It comprises processes that allow the clinician to properly diagnose and manage an illness, requiring wisdom, insight, and experience [4]. Deconstructing this inherent complexity somewhat, Juma and Goldschmidt delineated 26 tasks that physicians may implement when encountering patients [1]. The first three tasks focus on framing the patient encounter, the next eight tasks [4], [5], [6], [7], [8], [9], [10], [11] emphasize the diagnostic process, the following 13 tasks [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24] address the management of care, and the remaining two tasks [25], [26] entail self-reflection. Through these stages of framing, diagnosis, management, and reflection, they identified that physicians assess priorities, differentiate most likely diagnosis, and establish management plans [1]. Similarly, McBee and colleagues studied diagnostic and therapeutic reasoning comprised of the aforementioned framework using carefully constructed videotapes [5]. McBee et al. highlight the role of context and expert performance in driving variability across clinical encounters and how physicians employ a variety of clinical reasoning tasks [5]. Additionally, their study findings suggested that resident physicians used tasks in a varied and non-sequential manner [5]. Both of these sets of authors speculate about patterns that might emerge in these tasks to help us better understand the process of clinical reasoning; in this study, we begin this process.

Situated cognition and context specificity

Our investigation was informed by situated cognition [6]. Situated cognition is grounded in the notion that thinking (in this case clinical reasoning) is situated (or located) in experience, dividing the clinical experience into physician, patient, and environmental facets (factors) and their interactions [7]. Clinical reasoning is believed to emerge from these various interactions and thus, modification of or differences in these situation-specific elements would be expected to impact clinical reasoning. As such, there may be more than one path to diagnosis and this non-linearity can impact a physician’s diagnostic and management decisions.

Evidence for this situated cognition approach includes the finding of context specificity [8]. Context specificity is the phenomenon whereby a physician arrives at two different diagnoses for two patients with the same symptoms, findings, and underlying diagnosis, due to the patients’ differing situations. Durning et al. explored the interactions among the physician, patient, and environment in their study of contextual factors (alterations of context such as a patient suggesting a diagnosis, the electronic medical record improperly functioning, or the physician being sleep deprived – patient, environment, and physician contextual factors, respectively) and how they impact diagnostic reasoning of board-certified internists [7]. They found that experts’ performance was impacted by contextual factors and this observed impact was consistent with the tenets of situated cognition and cognitive load theories. Cognitive load theory pertains to our limited cognitive architecture and how we can only hold or process a limited number of pieces of information in our short-term (or working) memory [9]. High cognitive load is believed to be a potential mechanism underpinning context specificity [10].

Here, we investigate the presence of emerging patterns – both sequential (from case beginning to end across all participants) and comparative (comparing the aforementioned sequences in cases with and without contextual factors) – that we believe could yield valuable insights into context specificity and situated cognition. This investigation can reveal how clinical reasoning tasks unfold for physicians and how contextual factors may affect this process. First, we sought to determine if a discernable pattern emerges in physicians’ clinical reasoning. Second, we explored whether specific contextual factors influenced the observed patterns. And finally, we sought to delineate whether these situation-specific (contextual factors present or not) patterns provide additional evidence of the impact of contextual factors on clinical reasoning performance.

Building upon the work of McBee et al. [5], we provide a visual, sequential map that elucidates the patterns extant in clinical reasoning tasks and how they might be altered by contextual factors. Adding to earlier work on patterns in clinical reasoning [11], we examine the clinical reasoning process, or the tasks that physicians complete during a patient encounter. This process-oriented, sequence-based approach expands on previous work that examined the prevalence of epistemic activities in clinical reasoning [12]. Consistent with situated cognition, we hypothesize that this is not a linear process, rather one that is emergent, as clinical reasoning is iterative in nature. More specifically, we predicted that the sequence of task categories would unfold in a way similar to a typical clinical encounter. In other words, broadly, the clinical reasoning process begins with taking a patient’s history (framing), performing a physical and generating a diagnosis (diagnosis), developing a management plan (management), and finally, reflecting on their overall reasoning process (reflection). Within this broad pattern, we predicted smaller iterative cycles of some of these task types. Such findings could help inform our understanding of clinical reasoning and context specificity and provide additional evidence for situated cognition as an appropriate theory for exploring clinical reasoning. Exploring seemingly varied and non-sequential patterns through the lens of context specificity may not only build on what script theory predicts but also allows us to better understand how clinical reasoning unfolds as contexts shift. Moreover, the emergence of a lucid pattern may provide a useful framework for teaching and assessing clinical reasoning processes.

Sequence visualization of task categories

The sample included for this analysis was derived from a larger study focused on clinical reasoning performance [13]. Physicians either viewed a videotape or participated in a standardized patient encounter and then they completed a post-encounter form (PEF) [14]. Immediately following PEF completion, they were asked to think aloud about how they reached their diagnosis and treatment plan while either re-watching the videotape or watching a video of the standardized patient encounter. Physicians viewed two videotapes or participated in two standardized patient encounters. In short, 60 think-aloud transcripts from 30 physicians who participated in two separate cases – one with a contextual factor and one without – were coded for 26 clinical reasoning tasks [1]. Contrary to earlier work [15], rather than focusing solely on initial diagnostic hypotheses, we introduced contextual factors as distractors throughout the clinical reasoning process. For the purpose of this paper, we refer to the organic order in which tasks were demonstrated during a think-aloud as “steps”. For example, a participant may exhibit task 3 as their first step, task 8 as their fourth step, and task 24 as their final step and may repeat some tasks multiple times.

Frequencies of each of the 26 tasks were aggregated and categorized by framing the encounter (tasks 1–3), diagnosis (tasks 4–11), management (tasks 12–24), and reflection (tasks 25 and 26). These categorical counts were visualized using stacked area charting via Microsoft Excel to better understand the emerging, sequential nature of clinical reasoning tasks. Additionally, based on the frequencies, percentages were also calculated in order to numerically visualize the task category sequences.

“Crests” in clinical reasoning task categories

In addition to visualizing said data, based on percentages of task categories, we sought to identify “crests” in the demonstration of clinical reasoning tasks. For the purposes of this study, the highest crest, or surge in the percentage of tasks in a particular category, was identified in each task category. For example, the largest crest constitutes the highest average percentage of tasks during the think-alouds. We employed this crest visualization technique in an effort to better understand when particular task categories peaked in usage. We highlight each crest in our aggregate sample and compare crests between cases with and without contextual factors and participant clinical experience.

Additionally, because each think-aloud accounts for one task count per step, a range of 0–60 task counts (representing 0–60 think-alouds) could be observed for each step. As such, any steps that amounted to less than 11 task counts (or 11 think-alouds) were removed from the dataset. For the purposes of our study, this ensured that all four task categories were consistently and adequately represented and prevented skewed percentages as a result of extended think-alouds that tapered off into a repetitive, single task count. This resulted in different ranges, but allowed for a cleaner and more accurate representation of the cresting within the task categories.

General sequence and pattern

Our first visualization represents all 60 transcripts (or 30 physicians), i.e. think-alouds from all physicians regardless of contextual factor presence (Figure 1). Two outliers (think-alouds that ended with repeated identical tasks) were removed from our visual analysis, resulting in a maximum of 60 steps. This allowed us to view the temporal nature of clinical reasoning tasks as it unfolded during the think-alouds, in aggregate.

Figure 1:

Task-based category by step (n=60).

During the first 15 steps, regardless of contextual factor presence or not, all participants are engaged in a clinical reasoning task. At step 15, our first participant concludes their think-aloud and as a result, our total task count begins its decline. At step 1, 80% of participants are engaged in a framing task, 18% in a diagnosis task, 0% in a management task, and 2% in a reflection task. As the think-aloud progresses into steps 8–12, framing tasks decline sharply (~23%) and diagnosis tasks increase (~75%). Moving further along the think-aloud process, we see management tasks begin to pick up at step 36 and continue until the end of the think-aloud. Reflection seems to occur throughout the think-aloud process; though, the two reflection tasks seem to peak at step 35.

In order to further illustrate the sequential nature of the task-based categories, Table 1 showcases the percentage calculations as bands of intensity. In other words, the darker the band, the higher the frequency of a given task category (framing, diagnosis, management, or reflection). The steps within a crest are highlighted in red.

Table 1:

Crests in clinical reasoning task categories (n=60).

As Table 1 indicates, framing tasks peak during steps 1–3, diagnosis tasks peak during steps 8–12, management tasks peak during steps 36–44, and reflection tasks peak during steps 27–36.

Contextual factor patterns

The difference between cases with and without contextual factors is shown in Figure 2. In the absence of contextual factors, the lines demonstrate larger increases and decreases of various task categories, primarily when framing the encounter. For instance, the “peaks” and “valleys” observed in the frequencies are much steeper as physicians progress through their early tasks (steps 1–16). This variability in amplitude continues throughout the think-aloud but is much more prominent through step 25. Conversely, in the presence of contextual factors, the lines appear less varied. This trend is consistent across the four clinical reasoning task categories. The visual representations depicted in Figure 2 capture this phenomenon.

Figure 2:

Task-based category by step, without and with a contextual factor (n=30).

In the absence of a contextual factor, not only does there seem to be more variability while framing the encounter, but framing tasks also seem to be occurring at a higher rate. Also, in the absence of a contextual factor, management tasks do not occur until much later in the think-aloud process (step 12 vs. step 3 in the presence of a contextual factor). And finally, as documented by the thicker black band across the top of both charts, reflection seems to occur most consistently and at a much higher rate in the absence of a contextual factor.

Contextual factors and experience

The differences in patterns in the absence and presence of contextual factors seem to resemble the differences in patterns between more and less experienced physicians [over a decade (practicing physicians) compared to less than a decade (residents)]. The fluid up and down movements are present in the reasoning tasks displayed by experienced physicians in the absence of contextual factors. Figures 3 and 4 and Tables 2 and 3 show how experienced physicians become more like less experienced physicians in the presence of contextual factors, which serve as cognitive “distractors”.

Figure 3:

Task-based category by step, no contextual factor (n=30) vs. experienced physicians (n=18).

Table 2:

Comparison of crests in clinical reasoning task categories: no contextual factors vs. experienced physicians [all −CF think-alouds, n=30 vs. experienced think-alouds (+CF and −CF), n=18].

Table 2 further delineates the relationship between clinical reasoning task sequencing in the absence of a contextual factor and that of an experienced physician. The crests – highlighted in red and also indicated by darker shaded bands in Tables 2 and 3 – are occurring at nearly identical sequence points for experienced physicians (regardless of whether a contextual factor was present or not) and the whole group of physicians reasoning in the absence of a contextual factor. In other words, we are seeing the highest crest forming at steps 1–3 (framing), 9–11 (diagnosis), 32–38 (management), and 26–28 (reflection) in both think-aloud samples.

Table 3:

Comparison of crests in clinical reasoning task categories: contextual factors vs. intermediate physicians [all +CF think-alouds, n=30 vs. intermediate think-alouds (+CF and −CF), n=42].

Figure 4 offers a different glimpse into the patterns and sequences of intermediate physicians (in the presence and absence of a contextual factor) and all physicians performing in the presence of a contextual factor. The light gray band representing diagnosing tasks is much thicker for both samples and management and reflection tasks are exercised much less frequently (as noted by the thinner gray and black bands).

Figure 4:

Task-based category by step, contextual factor (n=30) vs. intermediate physicians (n=42).

As shown in Table 3, the crests observed in contextual factor think-alouds and intermediate physician think-alouds are synchronous. The highest crest observed in framing occurs during steps 1–3, in diagnosis during steps 8–10, in management during steps 41–43, and in reflection during steps 34–36.