In 2018, we launched a pre-clinical reasoning course at the University of California, San Francisco (UCSF) School of Medicine.15 Many frameworks for understanding CR have been described, with some scholars emphasizing knowledge structures, while others focus on cognitive processes. We find both aspects of reasoning relevant and integrated aspects of both into our curricular design. The CR course emphasizes knowledge architecture (e.g., illness scripts and diagnostic schema), provides case-based opportunities to practice reasoning skills like developing problem representations and articulating prioritized differential diagnoses, and invites consideration of other domains (e.g., communication, interprofessional collaboration, bias) that impact reasoning.15,16 Reasoning concepts are introduced periodically during the first 18 months of medical school and consolidated in an immersive 3-week CR course before the start of clerkships using a combination of lectures, panels, and interactive small groups (see Appendix 1 for details). The latter are led by faculty in internal medicine, pediatrics, neurology, emergency medicine, and family medicine. The goals of the course are to teach medical students how to develop (1) a structured approach to analyzing and discussing a diagnostic problem and (2) a comprehensive understanding of the domains that impact the diagnostic reasoning process and influence the risk for diagnostic error. The CR course emphasizes diagnostic rather than management reasoning because diagnosis is foundational to clinical practice, aligns with pathophysiology and disease-oriented content taught during the pre-clerkship curriculum, and precedes management in common assessment scales (e.g., RIME17).

We invited all students who completed clerkships in medicine, neurology, or pediatrics during the first half of their clerkship year to participate. Medicine, neurology, and pediatrics clerkships were chosen as inclusion criteria given their emphasis on diagnosis. However, students were able to reflect on diagnosis-focused experiences from any of their rotations during interviews. No students had completed their year-long, longitudinal family medicine clerkship nor the fourth-year emergency medicine clerkship; therefore, these diagnostically focused rotations were not part of our inclusion criteria. All students had completed the CR course before clerkships. Invitations were sent by email with several reminder emails sent during the first few months of clerkships. In person reminders were shared during clerkship didactic sessions. Participants received a $20 gift card.

We opted for a qualitative approach as this methodology was well suited for an exploratory study of a less well-understood phenomenon, and enabled us to delve into students’ experiences with and perceptions of reasoning in the clerkships. As there was no pre-existing theory specific to the CR skills of learners transitioning from the classroom to the clerkships, we selected constructivist grounded theory to analyze qualitative data and develop a theoretical model to illustrate our findings.18,19,20 A grounded theory approach provided an opportunity to develop a deeper understanding of students’ metacognitive approaches to and influences on their reasoning. A constructivist approach enabled us to use our own understanding of reasoning to interpret and analyze the data, and to identify themes and perspectives that might otherwise have been unnoticed. Our constructivist research paradigm also afforded exploration of CR as a complex process influenced by individuals’ unique learning experiences in different contexts.20 The study received exempt status from the UCSF IRB.

Two investigators (DC and GD) designed a structured interview guide to explore how students thought through and communicated about the diagnostic process in clerkships. Questions were revised with input from medical education faculty: one non-clinical medical education scientist, two physician leaders of a longitudinal clinical skills course, and an interprofessional education scholarship group, who recommended that we consider sensitizing concepts as we revised our interview guide. The interview guide was pilot tested with two clerkship students and adjusted for clarity. Two senior medical students conducted interviews to reduce the power differential between participants and interviewers with the aims of enhancing participants’ openness to sharing experiences, reducing bias that could be introduced if faculty invested in the curriculum were to interject leading questions, and decreasing the potential for influencing participants’ responses. Student interviewers were trained to follow the interview guide and avoid adding prompting questions outside of general extenders such as “tell me more.” The interview guide (including notes about sensitizing concepts) is provided in Appendix 2.

To prime their memory of a specific example of a diagnostically challenging case, participants were asked to review an admission note from a recent clerkship, but interviews were not limited to that patient encounter. The interviewer asked students to describe their thought processes before, during, and after seeing a patient, and how they shared their reasoning with others. Students were also asked to reflect on the clinical learning environment (including interactions with other team members such as supervisors) and how it influenced their reasoning. Finally, students were asked to describe clinical situations where they noted a risk for diagnostic error.

Interviewers received training in qualitative interviewing from an education scientist (AT). Interviews were conducted by phone in the spring of 2019 and 2020, lasted 30–60 min, and were recorded and transcribed. Recordings were de-identified and stored on a secure research drive. Only general data about participants (clerkship rotations and clinical sites) were retained.

We used a constant comparative, iterative process to analyze interview data and develop a codebook. Each investigator (DC, GD, SK, AT) independently coded several transcripts and met to discuss codes and create a codebook. Three investigators (DC, GD, and SK) then further refined the codebook. Two investigators then independently and sequentially coded each transcript, and one investigator (DC) compared the two independent codes. These three investigators met to resolve any differences in coding and to decide whether any new codes should be added. After ten transcripts were coded, no new major experiences or perspectives appeared in the data.19

All transcripts were coded and organized using Dedoose analytic software (Sociocultural Research Consultants, LLC, Manhattan Beach, California). Three investigators then utilized the coding scheme to undertake an in-depth analytic process, dividing the major code categories between them and reviewing all quotes for each code, identifying over-arching themes and concepts, and looking for connections between codes. The investigators developed a model that described the major themes and relationships between themes, and noted connections between themes in the data and concepts taught in the pre-clinical CR course.

Our research team included two clinician-educators (DC, GD) with experience in CR curriculum design and teaching, one clinician-educator (SK) with experience as an instructor in the CR course, and one education scientist (AT) with experience in qualitative methodology. DC was the main curriculum designer and director of UCSF’s pre-clinical CR course during the study.

DC emphasized the situated, contextual nature of reasoning; GD brought a cognitivist lens to the reasoning process; SK provided the perspective of an early-career clinician educator who recently encountered concepts of CR throughout medical school and residency; AT offered the perspective of a non-clinician scholar with a focus on learning theories. Our team believes that both knowledge architecture and cognitive processes are relevant to reasoning, as opposed to considering these to be mutually exclusive frameworks, and brought an integrated perspective to our analysis.