Management reasoning scripts: Qualitative exploration using simulated physician-patient encounters

Scripts are defined generally as

high-level, precompiled, conceptual knowledge structures … [that] represent general (stereotyped) event sequences, in which the individual events are interconnected by temporal and often also causal or hierarchical relationships; that can be activated as integral wholes in appropriate contexts [and] contain variables and slots that can be filled with information present in the actual situation, retrieved from memory, or inferred from the context [15].

Additionally, “a script is not a simple list of events but rather a linked causal chain; a script can branch into multiple possible paths that come together at crucial defining points” [16]. Merging these other theoretical concepts with our empirical observations, we defined management scripts as “precompiled conceptual knowledge structures that represent and connect management options and clinician tasks in a temporal or logical sequence to facilitate development of a rational management plan” [14].

The concept of illness scripts dates to 1984 [17], and has been the subject of rich discussion and research [15]. Illness scripts are knowledge structures (mental representations) that reflect how an illness developed (i.e., the sequence of events occurring in a given patient) and figure prominently in theories of diagnostic reasoning [18, 19]. Key features of the illness script include the enabling conditions (risk factors), faults (disease pathology), and consequences (symptoms, physical signs, test abnormalities) [17].

By contrast, management scripts are a new concept—namely, mental representations that guide development of a management plan. A recent empirical study described “therapy scripts” in the selection of an antimicrobial, but did not extend this model beyond that specific task [20]. Another recent conceptual model of management scripts highlighted their temporal evolution [21], but was not based on empirical data and focused on activities (reasoning, decisions) within the clinician. In our qualitative study [14] we independently developed a model for management scripts that includes activities without the clinician (e.g., involving the patient) as well as within. However, given the multiple features requiring discussion in that report we were unable to fully explore our insights regarding management scripts. In short, previous work suggests that management scripts potentially play an important role in management reasoning, but further exploration of this concept is warranted.

Human memory seems to be divided into working memory (of limited capacity) in which computations are performed and links are established between new and old information; and associative (long-term) memory (of limitless capacity) in which information is stored and retrieved based on patterns (direct association between new information and existing exemplar[s]) [3, 22]. It has long been recognized that human thinking (including clinical reasoning) involves dual processes:

Type 1 thinking supports non-analytical clinical reasoning, often manifest in rapid generation of diagnostic hypotheses and pattern matching (association between mental representations [scripts, schemas, propositions, mental models, cognitive maps, chunks] of this case and prior cases) [4].

Type 2 thinking supports analytical clinical reasoning, manifest in slow, deliberate, systematic integration of clinical data, diagnostic possibilities, and additional data gathering (hypothetico-deductive reasoning). It “enables uniquely human facilities, such as … mental simulation and consequential decision making” [23]. Research in diagnostic reasoning suggests that novice trainees rely more on analytical reasoning whereas experts typically use more non-analytical reasoning [4].

In our conceptual exposition, we noted that “[w]e presume that management reasoning reflects a balance of non-analytical processes … and analytical processes …, yet the relative contributions remain unknown. … It seems plausible that management reasoning may be inherently more analytic (deliberate, planned, and systematic) than diagnosis” [2]. Our rationale was that the numerous factors in play (treatment options, patient preferences, shared decision-making, explicit cost-benefit tradeoffs, etc.) and the absence of a single best solution all combine to both inhibit the generation of needed patterns and prevent their application. However, in the absence of empirical data we suggested further research into this hypothesis.

We aimed to conduct an introductory investigation of management scripts and dual process thinking in management reasoning.

We reviewed a convenience sample of 10 videos of staged simulated physician-patient encounters, used in an IRB-approved study of rater training conducted in 2006 [24, 25]. Each video (45 s to 7 min long) portrayed an outpatient follow-up counseling visit between a resident physician and a patient. To enhance the range of observable behaviors, for each medical problem (hypertension, hyperlipidemia, fibromyalgia, diabetes mellitus, and papillary thyroid cancer) we selected 1 poor and 1 superior performance. Six videos (2 pairs) were extemporaneous dialogues between the physician and a standardized patient. Physicians were coached in their level of performance; patients were instructed to respond authentically while portraying the “same” person in both cases. The other 4 videos were scripted dialogues.

All investigators independently reviewed each video at least twice using a coding form (e-Box 2 in ESM) with open-ended prompts, including: “In what ways was reasoning automatic, fast, and reliant on pattern recognition (reflecting System 1)?” and “In what ways was reasoning deliberate, effortful, and slow (reflecting System 2)?” Each reviewer also documented “epiphanies”—novel insights, themes, and connections that extended beyond what was visible in the video.

Our analysis used a social constructivism paradigm and constant comparative approach [26], mirroring the method used in a previous study of diagnostic reasoning [13], proceeding in three stages. In Stage 1, author DAC organized each investigator’s raw observations and epiphanies into a 120-page, single-spaced document. In Stage 2, the investigator team reviewed this document individually and as a group, and engaged in multiple voice and electronically mediated conversations to iteratively reorganize, reconceptualize, elaborate, and refine these observations into a new 21-page list of critical insights, of which approximately 20% referred to management scripts or Type 1/Type 2 thinking (see e‑Box 3 in ESM). From this list we identified several distinguishing features (facets of variation) of management scripts.

We recognized that our empirical data were insufficient to deeply probe these concepts, especially since management scripts were an unanticipated finding. Thus, in Stage 3, we integrated conceptual frameworks, theories, and empirical evidence from the extensive literature on scripts, dual process thinking, and diagnostic reasoning (including [3, 4, 6, 7, 15‐17, 23, 27]) as we continued to iteratively revise our models (including > 12 h of voice communication and multiple written drafts). We reflected, discussed, and collected new data (i.e., re-reviewing videos) until no additional insights were forthcoming from this dataset. All reviewers came to full consensus on all concepts.

Three of us (DAC, SJD, CRS) are practicing internal medicine physicians, and two of us have PhDs in cognitive psychology (LDG) and education/cognition (SJD); these backgrounds surely influenced our perspectives on clinical reasoning and clinician-patient encounters. We also developed this model of management reasoning [1].

In nearly all encounters, the physician followed a coherent, fluent, structured sequence of dialogue and activity that appeared to be largely predetermined or preplanned. We interpreted these as outward (observable) manifestations of cognitive scripts (high-level, precompiled mental representations of interconnected events [i.e., clinician tasks]). Scripts guide clinicians as they prioritize and organize management options such as diagnostic tests, treatments, consultations, patient education, shared decision-making, and monitoring to generate a management plan. We discerned that scripts differed in their scope, automaticity, quality, and generality.

As our understanding evolved, we recognized that management scripts differ substantially from the illness script prevalent in diagnostic reasoning; we summarize these differences in Tab. 1. Most salient, the illness script is predominantly retrospective—it reflects the temporal evolution of the illness up to the point of presentation. Moreover, the illness script is often treated as interchangeable with the diagnosis: multiple scripts (diagnostic hypotheses) are activated early in the case presentation, and these may then be deliberately evaluated to select the script that best matches the available data. In a sense, the illness script is a destination (diagnosis).

By contrast, the management script is decidedly prospective—it guides clinicians as they teach, collect new information, jointly make decisions, and ultimately recommend treatment. Management reasoning involves more than evaluating treatment options against available data to select the best match. It also requires communication and shared decision-making with the patient; dynamic interplay among people, systems, and settings; and ongoing monitoring and adjustment of the plan [2]. As such, we envision the management script as a pathway (not a destination) that leads to a management plan (the destination). Our empirical observations suggest six key features (components) of management scripts (Tab. 2):

Drawing on the above concepts, we elaborated a tentative heuristic framework for how management scripts might operate in practice (Fig. 1).

The number of available scripts and detail of each script will vary. A beginning medical student might have only a general script for “starting a medication”; a primary care physician might have a diagnosis-specific script for “treating new hypertension”; and a hypertension specialist might have separate scripts for patients with diabetes, renal failure, and sleep apnea. As with the illness script [18], management script activation is probably a subconscious cognitive event (Type 1 thinking).

Next comes “selection” of the most relevant script—the script most likely to efficiently lead to an effective management plan. We believe, based on research in diagnostic reasoning (illness scripts) [4, 27], that selection involves pattern-matching with prior cases (see e‑Fig. 1) based on various case features (diagnosis, comorbidities, known or assumed preferences, etc.). Script selection probably varies from largely subconscious to largely deliberate (i.e., varying degrees of Type 1 and Type 2 thinking) depending on the familiarity of the clinician with this particular problem and context.

Finally, the selected script is “instantiated”. In its general form (prior to instantiation), the script consists of multiple possible branching pathways and multiple empty “slots” [16]—placeholders that can be filled with patient- and context-specific information. Instantiation fills the slots and selects a specific pathway. For the illness script, “[t]he script, or generic event sequence, maps onto the general clinical picture of a disease, whereas each individual patient can be considered an instantiated illness script, with both typical (central) or atypical (peripheral) features, which appear in a certain order” [15]. For the management script, slots might include the severity or urgency of diagnosis, comorbid conditions, current medications, allergies, patient preferences, logistic constraints, etc. Some slots likely remain empty at first, and may be filled later with additional information, or they may remain unfilled. Script instantiation explains how management scripts (and management reasoning) can be tailored to the patient and context. More generic scripts apply across a broader spectrum of cases, but require more effort to fill in empty slots; more specific scripts are faster and less effortful (fewer empty slots), but apply to fewer situations. Experience with a given problem (diagnosis, patient comorbidities and preferences, system constraints, etc.) promotes development of more, and more specific, scripts relevant to that problem; and this in turn promotes greater efficiency in developing a tailored management plan. Tab. 2 provides a hypothetical example of management script instantiation.

Once instantiated, the script guides the clinician in the temporal unfolding of the encounter as the management plan is developed. Cognitively, we propose that the script includes the key features described above (i.e., problem, management options, preferences, education needs, interactions, and encounter; see Tab. 2). Behaviorally, the script guides the choice among various management tasks.

If the encounter proceeds in close alignment with the instantiated script, it may be sufficient to result in an acceptable management plan. However, in many cases the initial script as instantiated will be insufficient. Scripts “do [not] provide the apparatus for handling novel situations” [16], such as an infrequent diagnosis or event, novel connections/insights about the disease/patient/system, or most patients seen by novice clinicians. Moreover, scripts can be thrown off course by distraction (interruption by another script [e.g., realization that labs tests suggest hyperthyroidism]), obstacles (an impeding condition is present or a necessary condition is absent [the patient asks an unexpected question, the preferred drug is too expensive, or a diagnostic test will be delayed]), and errors (an action is completed inappropriately [the clinician fails to address a patient’s strong preference]) [16].

Novel situations, obstacles, and errors all require iterative adjustments. Many adjustments can be made to the initial script by consciously substituting new information or filling previously unfilled script slots (i.e., Type 2 thinking). If more extensive adjustments are needed, or in the case of distraction, new candidate scripts are activated, selected, and instantiated, and the process continues. We observed in some videos a failure to adjust (i.e., a flawed or inadequately tailored script was used); we speculate that reasons for this include limited knowledge or experience (i.e., weak or few scripts) and limited time. It is possible for several scripts to be operating at once [16]; for example, scripts for “newly diagnosed hypertension” and “encouragement of lifestyle change” could run concurrently.

Eventually, the management plan is sufficiently well developed that the plan is implemented, the encounter ends, and the clinician moves on to the next patient. Of course, the management plan continues to evolve over time (the ongoing strategy script) in clinician-patient encounters, between such encounters (e.g., clinician ordering additional tests, or patient deciding to stop taking medications), and through interactions with other members of the healthcare team.

Our conceptualization of the management script differs considerably from the illness script [15, 17] (see Tab. 1), aligning instead with the original construct of the cognitive script (“a predetermined, stereotyped sequence of actions that define a well-known situation” [16]). Our conceptualization of the management script also differs from a previous conceptualization in which management script activation leads directly to selection of management options [21]. By contrast, in our model activation is followed by selection and subsequent instantiation of a script, which guides tasks and temporal events that iteratively consider options (together with other script features) to culminate in a management plan.

This study was intended as an introductory exploration of these topics, and has limitations. First, the videos constituted a convenience sample. They were intended to be realistic, but were limited to 5 conditions in adult outpatient primary care, originally designed for a study of rater training [24, 25], and selected to represent extremes of performances. Four videos were scripted; although this limits their utility in supporting novel insights, this limitation is mitigated as the script writers were not part of our study team. Nonetheless, our findings are best viewed as laying a groundwork for future investigation. Second, our findings are based on observable behaviors. We can speculate about but cannot confirm underlying cognitive processes; indeed, some behaviors could reflect more than one reasoning approach. Third, we authors were both the developers of the guiding conceptual framework [1] and the observers and analyzers in this study, so there is some risk of confirmation bias. Fourth, although our analysis method differed somewhat from typical qualitative analyses, we were transparent in our methods, results and inferences; and our appeal to published literature and theories constitutes a type of data triangulation. We believe our approach was rigorous and sufficiently robust to support the proposed (limited) implications. Finally, many of our insights relied in part on theoretical considerations, and some could have been derived from theory alone. However, the fact that no authors have previously identified these insights (including ourselves, despite extensive discussion spread over several years) highlights the added value of our video-stimulated empirical approach.

Although our empirical observations did not allow direct inferences about Type 1 and Type 2 thinking, these observations stimulated deeper insights that, when merged with existing theories of clinical reasoning, resulted in a novel conceptual model for dual process thinking in management reasoning. Both Type 1 and Type 2 thinking appear to interact with both the mental representation (the script) and the management tasks, usually in an iterative process (e-Fig. 2). Type 1 thinking appears to influence the activation, selection, and initial instantiation of the management script (e-Fig. 1); namely, the initial script is selected quickly, unconsciously, and largely based on pattern recognition. This operates best when the clinician possesses a large library of relevant scripts and each script has a well-developed series of slots, which permits more accurate pattern-matching and greater case-specific tailoring. Tailored (instantiated) scripts serve to focus and simplify the encounter by prompting case-specific questions, limiting options, and guiding education (see Tab. 2). If (when) script revision is required, the processes of activation, selection, and instantiation probably become more deliberate (effortful) and more hypothetical (involving mental simulation)—in other words, increasingly Type 2.

Type 2 thinking takes the script selected and instantiated (presumably through Type 1 processes—at least initially), and deliberately and consciously seeks to align the script with the unique case (to refine the instantiation; and, if needed, revise the script selection). These efforts are helpful if they promote tailoring that culminates in a better-optimized management plan; these efforts are wasteful if they simply consume time and energy (slow and effortful) without improving the final plan. Many Type 2 processes are likely part of the script itself: a largely-automated (Type 1) script could contain planned branch points that briefly pass control to Type 2 (i.e., a back-and-forth between Type 1 and Type 2 thinking) [16]. For example, a 40-second “canned” educational message (Type 1) might pause for a planned question to confirm understanding (Type 2) and then resume. Alternatively, Type 2 thinking likely arises spontaneously when the clinician goes “off script” in response to novel situations, obstacles, and errors [16].

We propose, for illustration, that an airplane’s flight plan is analogous to a management script. The flight plan specifies the destination, route, and anticipated potential challenges. Some plans are detailed, others less so. If all goes as planned, most of the flight is on “autopilot” (Type 1 thinking). Familiarity (repeated experience) with the route, aircraft, and crew and a more detailed flight plan allow more automaticity. Many in-flight adjustments (e.g., response to minor turbulence) are also performed automatically, with greater experience (in general, and with the specific route and aircraft) enabling larger adjustments to be performed with minimal effort. However, deviations from the flight plan are almost always required (Type 2 thinking), especially in long flights during which weather, air traffic, and ground conditions could change dramatically. Thus, the pilot must periodically seek new information, re-evaluate, and redirect. Occasionally, unanticipated challenges or a new destination will cause the original plan to be discarded and replaced. This combination of flight plan (script), autopilot (Type 1), and hands-on adjustments (Type 2) guides actions that ultimately lead to the final destination (customized management plan).

As an introductory investigation with nontrivial limitations, immediate applications of our findings are constrained. However, we believe that management scripts, and their interrelationship with cognitive processes as discussed above, have potentially profound implications for education research, theory, and practice. We previously speculated that “[e]xplicit consideration of treatment costs and benefits, use of rubrics to guide management decisions, and thoughtful shared decision-making all suggest a slow, deliberate process” [14]. If true, this would imply that both training and clinical practice of management reasoning are of necessity rather inefficient. However, management scripts offer a transformative perspective. Rather than teaching only long lists of treatment options, educators might encourage activities designed to promote development of management scripts. Pending further specific evidence, we can borrow methods from what is known about development of cognitive schema and illness scripts [7, 18, 27], including deliberate practice with virtual patients [28], teaching general scripts (such as SPIKES [29] for breaking bad news), learning from errors [30‐32], and structured reflection [33, 34].

Our findings suggest avenues for future research. Among these, we need more empirical evidence to test our model, including confirming the distinctions between management scripts and illness scripts (Tab. 1) and reconciling gaps such as how management scripts develop; how activation, selection, and instantiation work; and how Type 1/Type 2 thinking vary over the course of the encounter. Additionally, scripts offer new targets for assessment, including script activation, selection, and instantiation; yet whether and how these processes can be isolated and measured remains to be seen. Future research might again use videos as stimuli (i.e., observable behaviors); such videos might strategically vary according to management script features (Tab. 2) and emphasize the iterative nature of management reasoning (Fig. 1). Lastly, we will need new methods that look beyond observable behaviors to understand deeper cognitive processes.