Self-regulated learning and the future of diagnostic reasoning education

Introduction

Diagnostic reasoning is a foundational ability of health professionals. Although clinical reasoning has been taught to students for millennia, the National Academies of Sciences, Engineering, and Medicine (NASEM) has recently recommended enhancing clinical reasoning education [1]. The goal of enhancing clinical reasoning education includes improved diagnostic accuracy and reduced diagnostic error (i.e., clinical reasoning performance), as well as better resource utilization and improved patient experience. While perfect diagnostic accuracy is unrealistic, attaining a threshold such that treatment can be delivered without knowing a complete diagnosis (management reasoning) or having the knowledge that more information is needed (diagnostic reasoning) are key skills for the learner. There is an absence of randomized controlled trial (RCT) data that clinical reasoning education improves actual patient outcomes, but observational evidence that health professions learners’ diagnostic ability improves over their undergraduate and graduate medical education careers seems adequate to pursue improved clinical reasoning education.

To operationalize NASEM’s recommendation, health professions educators need not only help learners improve their clinical reasoning, but teach them how to develop expert performance (i.e., high diagnostic accuracy with a low error rate). Traditionally, the primary strategy used to teach clinical reasoning focuses on essential clinical reasoning knowledge acquisition, specifically illness scripts (i.e., diseases’ symptoms, signs, laboratory and radiological findings) and base rates of diseases (i.e., prevalence, incidence), as well as diagnostic frameworks (e.g., pre-renal, intra-renal, post-renal kidney injury), because knowledge is associated with diagnostic success [2]. The Competencies to Improve Diagnosis have described these as well as other curricular objectives for clinical reasoning [3]. However, this approach presupposes that students know the learning and behavioral strategies that lead to expert clinical reasoning performance. In fact, students need to both understand these frameworks, as well as the techniques for how to best learn them. Another approach to teaching diagnostic reasoning is low-fidelity simulation (e.g., paper-based, case-based or problem-based learning), with or without higher-fidelity simulation (e.g., simulated manikins). However, there is no definitive evidence on the most effective way to teach clinical reasoning. If health professions educators want to increase the likelihood of students developing expert performance in clinical reasoning, they need to teach and motivate students to use evidence-based learning strategies, particularly because life-long learning is necessary in the health professions where knowledge continuously expands and changes.

Self-regulated learning to develop expert performance in diagnostic reasoning

An evidence-based learning strategy that is strongly associated with expert performance in other fields (sports, video gaming) is self-regulated learning (SRL) [4, 5]. SRL is defined as the modulation of “self-generated thoughts, feelings, and actions that are planned and cyclically adapted to the attainment of personal goals” (Table 1) [6, 7]. Another related strategy, deliberate practice, refers to a highly structured learning activity where learners set explicit learning goals to improve performance in a skill that they co-create with a coach who carefully observes performance to provide specific, meaningful feedback [7]. Performance improves through an iterative process of effortful, highly-focused practice, followed by assessment and targeted feedback [8]. Within medical education, deliberate practice has been linked to success via simulation training and procedural skills training [9, 10]. It should be noted, though, that studies on deliberate practice are contextually narrow, though elements can inform expertise in clinical reasoning. Regardless, both require some tolerance for repetition and “boredom”. While health professions educators are encouraging students to use SRL, there is no literature regarding its utility in improving diagnostic reasoning [11, 12].

Evidence shows SRL is both associated with expert performance and is a teachable competency that can be applied to medical education [13, 14]. Teaching and encouraging the use of SRL-related techniques appears to help students better utilize them in practice and promotes students’ learning [15], [16], [17]. Prior studies have shown associations between perceived SRL levels and better academic performance of medical students [12, 18]. There is also suggestion that SRL is important in clinical reasoning and life-long learning in medicine, which is especially beneficial to early-stage trainees, although the data are limited [17, 18]. Most have looked at learning outcomes associated with student self-report of SRL behaviors. A single-site study suggested that self-reported use of SRL leads to gains in self-efficacy [19]. A qualitative study of post-graduate trainees suggested that use of certain SRL strategies may aid in trainee success via improved self-confidence and autonomy [20]. It should be noted, though, that SRL’s impact on diagnostic accuracy and efficiency in clinical reasoning is lacking, as is its longitudinal impact on clinical reasoning skills.

Best practices for teaching self-regulated learning

SRL teaching should be done in a developmentally appropriate way based on learner level. For early-stage health professions students, it is particularly important that the educator and learner co-create learning goals due to the Dunning-Kruger phenomenon (i.e., students do not know what they do not know) [21, 22]. Strategies that emphasize learner goal-setting and self-efficacy have the greatest impact [23, 24]. In more advanced learners, metacognitive approaches (e.g., structured reflection on a differential diagnosis) are more ideal, as compared to novice learners, who might benefit from a more socio-cognitive approach (e.g., positive encouragement when a student formulates a proper differential diagnosis) [25], [26], [27], [28]. Using these best practices allows for improved learner ability to use SRL effectively.

We believe that early-stage trainees (medical students and residents/fellows) would benefit greatly from learning how to become “expert learners”. We recognize that such content will be a challenge to fit into already crowded health professions curricula. On the other hand, we believe that teaching SRL to health professions students is as important as any other content area, given the ever-changing landscape of medical knowledge [29]. SRL is a strategy that master diagnosticians use to learn. We believe that there are many opportunities to incorporate SRL into health professions education.

Given studies in other domains highlighting the critical nature of SRL in developing expert performance, we believe that health professions schools should teach students how to apply SRL and deliberate practice to their learning. Educators can provide learners feedback on their use of it to achieve NASEM’s goal of enhancing diagnostic reasoning education [30, 31].

Contexts for developing and practicing SRL

One significant challenge of developing expert performance in clinical reasoning is the lack of iterative practice with multiple cases for a given problem or disease ranging from typical to atypical presentations. Other industries (e.g., aviation) have used simulation to fill in the gaps of inexperience to provide employees opportunities to develop critical knowledge and skills in safe environments [32]. Health professions pre-clinical and clinical curricula routinely uses low-fidelity simulation (e.g., paper-based, case-based or problem-based learning) to provide learners opportunities to practice clinical reasoning, but uptake of higher-fidelity simulation (e.g., simulated manikins) has been limited by the high cost, both time and money [33].

Furthermore, both low-and high-fidelity simulations are typically structured as “one and done” experiences (e.g., one case of myocardial infarction (MI) or one case of pulmonary embolism), rather than longitudinal curricula that help students build and maintain their clinical reasoning. The rationale for this approach is that students and trainees will “see” these problems and diseases in their clinical experiences and continue to develop their clinical reasoning in this way. Some schools require students to fill out patient logs to confirm they have cared for patients with curricularly “required” diseases and if not, they are assigned a case to make up for deficiencies. Again, most of these requirements are “one and done” approaches (e.g., “You didn’t see an MI on your clerkship, so you must complete this single assigned MI case and you will fulfill your requirement.”). It is known that repetition is key in the development of clinical reasoning abilities. It allows for pattern recognition and, with time, the ability of the learner to override thought-processing systems when faced with information that is not in agreement with known patterns [34]. Unfortunately, particularly in undergraduate health professions education, opportunities for repetition are often few.

Given these challenges and the limited educational benefits of “one and done” approaches to developing diagnostic reasoning expert performance, we recommend applying SRL to such exercises. The key to this approach is a large database of cases that cover common and cannot miss diagnoses in the common problems that health professionals will encounter, as well as meaningful assessment data. The former can be created with significant effort from textbooks, clinical problem-solving exercises published in journals, podcasts, and web-based resources/applications. Using the SRL cycle and principles of deliberate practice, learners should co-create appropriate developmental goals with evidence-based strategies for achieving them (e.g., spaced learning, interleaved practice) with a coach, ideally, to provide guidance and motivation (Table 2). Then, both learner and coach should monitor and reflect on performance to determine whether goals were achieved. An action plan should be developed to create new goals if they were achieved or to modify the strategies used if the goals were not achieved. If no coach is available, then feedback can be provided through the resources themselves and students can use this feedback to self-reflect and build their own goals, strategies, and action plans. It should be noted that in order for SRL to be successful, motivation on the part of the learner is a pre-requisite. This is one of the areas that can be regulated by the individual [35]. For this reason, SRL works particularly well in scenarios where the learner is highly motived to succeed, such as remediation. For SRL to work more generally in health professions education, it is vital to establish a learner’s motivation for completing a given clinical reasoning task.

To achieve the goals described in Table 2, we recommend the following elements be included in simulated cases: (1) an anatomical/pathophysiological diagnostic framework or an algorithmic approach to the problem, which provides the student with a useful knowledge organization structure to help build their own frameworks over time (e.g., in a case of syncope and dyspnea, begin with frameworks for each based on pathophysiology), (2) data selection opportunities or a question prior to revealing a patient’s clinical findings that encourages students to list findings that significantly increase or decrease the probability of diseases on the differential diagnosis (i.e., questions that foster a hypothesis-driven approach to data collection), and (3) prioritization and justification of the differential diagnosis (hypotheses) of at least 2–3 diseases, preferably focused on the most likely and cannot miss diseases [36].

We recommend a developmental approach to cases both in terms of fidelity and typicality. Early learners should train with typical presentations of common and cannot miss diseases using low-fidelity simulation (e.g., paper or web-based cases with words only) with faculty or scaffolding built into the case (e.g., algorithms) to aid them given that self-efficacy is a critical determinant of the SRL quality [20]. Low-fidelity simulations also reduce cognitive load, which can negatively impact student performance [37]. Once students master such cases, moderate to high fidelity simulation can be introduced, which require additional diagnostic competencies. For example, in a chest pain case caused by aortic stenosis, an audio file of the murmur, an ECG showing left ventricular hypertrophy, and a chest X-ray showing an enlarged cardiac silhouette can be provided, rather than the description. Alternatively, the entire case can be run in a simulation lab, where the learner elicits the history, performs the physical exam, and requests and interprets labs/studies. Once learners demonstrate competence in prioritizing and justifying the differential diagnosis, scaffolding can be reduced/removed and atypical presentations of common and cannot miss diseases can be introduced.

Regardless of format, meaningful feedback is essential for SRL and deliberate practice. In increasing order of resource utilization, feedback options include: (1) a gold standard note learners read or a video they watch that explains the clinical reasoning clearly, (2) peer feedback from co-students or final year students, (3) feedback from graduates, (4) faculty feedback, and (5) video review with self-assessment followed by faculty review of the video with synchronous or asynchronous feedback to the learner. The last option is likely most effective, as learners discover the gaps between their self-assessment and the observer’s assessment, which has the highest potential to improve their ability to monitor and self-reflect. An example of a case database that gives particularly strong feedback to learners is Healer from the New England Journal of Medicine [38]. We recognize there is no substitute for real clinical experience, but we believe simulation is an important complement that can enhance clinical reasoning knowledge acquisition and may be transferrable to actual patient care.

SRL and clinical experiences

Using SRL in the clinical experiences is essential to developing expert diagnostic reasoning performance as these learning experiences are most likely to transfer to future clinical encounters. We know that the transition to clinical experiences for health professions students leads to dynamic changes in how they learn. External motivation increased and use of metacognitive strategies decreased in one study [17]. Another showed that novice clinical students’ self-regulated learning is influenced greatly by others, including peers, particularly in goal-setting. This contrasts with more experienced clinical students, who are less apt to be influenced by a single individual and can more independently set goals and develop a clear learning trajectory [39]. We believe that this knowledge can be leveraged to better utilize SRL in the clinical experiences of health professions learners. Early emphasis on maintaining metacognitive approaches to learning is a key area that should be focused on, given the decrease in utilization of these strategies early in clinical experiences. Additionally, given the significant influence that peers and individuals at more advanced career/learning stages have on novice clinical students, this is a group that teaching on how to teach SRL techniques may be of great benefit, particularly how to co-create clinical reasoning goals. The aim is to create a more individualized clinical experience for the learner.

SRL and assessment

There is a dearth of information regarding the role of SRL in assessment in health professions education. However, in other fields, the effective use of SRL correlates significantly with performance [4, 5]. If a similar strong correlation can be demonstrated with diagnostic reasoning performance, SRL assessment could be an assessment game-changer because diagnostic reasoning is both content-and context-specific. Content specificity refers to the fact that diagnostic reasoning is highly dependent on clinical knowledge (i.e., a dermatologist may be an expert in diagnosing skin conditions, but a novice in diagnosing cardiac conditions) [40, 41]. Context specificity refers to the fact that clinicians’ diagnostic reasoning performances correlate poorly across different occasions even when the patient’s clinical findings and diagnosis remain the same [42, 43]. These two characteristics of diagnostic reasoning lead to the requirement for large sample sizes to assess the competence of clinical reasoning performance.

An example of an approach focused on performance process is self-regulated learning micro-analytic assessment (SRL-MAT). It has been used to generate information on the strategies learners’ used when faced with a diagnostic reasoning case [44]. In SRL-MAT, learners are asked directly about their strategies for diagnostic reasoning, motivations, and planning process while in the midst of a clinical case (Table 3). Based on this information and the results at the conclusion of the case, the teacher can give targeted feedback that the learner can apply to their learning and diagnostic process.

A final assessment consideration is the nature of longitudinal, aggregated feedback provided to the learner, which is critical for learners to visualize their areas of growth and continued weaknesses. For diagnostic reasoning, this feedback should likely be separated by diagnostic problem and specific disease (e.g., dyspnea, pneumonia; chest pain, GERD), given content specificity. Under each, diagnostic accuracy, differential diagnosis accuracy, knowledge of diagnostic frameworks, data collection thoroughness, and efficiency can be listed, depending on the assessment. In addition, time on task, cognitive load, and confidence could be added to provide evidence of automatization of diagnostic reasoning (i.e., quicker to correct diagnosis with less cognitive effort signifies expert performance), as well as calibration (i.e., correlation of diagnostic accuracy and confidence). A recent development in the graphic depiction of such data is “learning curves”. They provide a visual answer to the question, “How well am I learning?” [44]. Validity evidence for learning curves exist in visual pattern-recognition skills, such as ECGs and radiograph interpretation [45, 46]. By using learning curves, learners and educators can visually identify improvement over time and determine whether competence has been achieved when compared to peers or consensus-based benchmarks. Learners can then modify their learning goals and/or strategies if they are not achieving diagnostic competency benchmarks at the target time intervals.

The use of SRL as an assessment tool has thus far been limited. However, the perceived benefits of its use are exciting and may allow for an improved ability to assess learners’ diagnostic reasoning capabilities. We view diagnostic reasoning as one of the pre-eminent pillars of medical education and life-long learning in medicine. Being able to view the learning trajectories of students and physicians in this realm via assessment is key and can serve as a means for early intervention for learners in any stage of their career who may be falling behind. It can also afford individuals on an accelerated learning curve the opportunity to push themselves further to becoming the “expert learner” that we strongly advocate for. SRL is uniquely suited to diagnostic reasoning assessment.

The future of SRL: novel technology

The technological innovations being implemented within medicine are growing by the day. We envision SRL being able to utilize them as a means of improving learner performance, and not just for their novelty. Virtual reality is being used as a method of instruction in numerous fields and is just now making its entry into medicine. Its primary uses have been in anatomy teaching, but the opportunities to use this in diagnostic reasoning are limitless [47]. Students or residents would have the ability to use programmed cases that place them in the middle of the hospital ward or clinic and not just looking at a computer screen. The immersive experience is more realistic. In addition, it would allow for a true preceptor to insert themselves within the simulation such that SRL can be engaged in with the learner. In this way, learners can talk through goals for a case, thought process, differential diagnosis, and ultimately what strategies worked and did not work.

As this technology becomes more refined, we hope SRL can be used to further individualize medical education. We are in an era where artificial intelligence (AI) and machine learning are en vogue. With further development, SRL-based simulated learning can adapt to the strengths and weakness of the learner. Such a learning environment could also adjust the case based on learner goals at the beginning of the SRL session. As the case advances, if the clinical scenario appears to be too simple for the learner, additional information could be given that would add to its complexity. The use of AI and machine learning is a fantastic opportunity to expose learners to the multitude of diagnostic reasoning strategies out there, and set them up for success as life-long learners.

Conclusions

Becoming an expert learner is no easy feat. It is one thing to say that all physicians are life-long learners, but it is entirely another to practice this on a regular basis. Self-regulated learning, combined with deliberate practice, is a competency that we believe is central to this philosophy as it teaches learners “how to learn”. The future of SRL is bright with opportunities for significant growth in simulation, assessment, and machine learning/artificial intelligence applications. We believe that SRL has the potential to help achieve the NASEM goal of improved diagnostic accuracy and reduced diagnostic error. We all can become expert learners. SRL is one effective way to get there.