Integrated Basic Science Within the Instructional Design of Pattern Recognition Training

Background:

Ensuring that medical students and novice physicians become good diagnosticians remains at the core of medical education. For more than six decades academics and clinicians have sought an understanding of the most successful diagnostic reasoning strategies and how to educate future physicians in these strategies. Research has revealed that most diagnosticians apply a mix of pattern recognition, deductive backward reasoning, i.e. searching for symptoms or signs that justify a tentative diagnosis, and inductive forwards reasoning strategies, i.e. analyzing the most probable diagnosis based on signs and symptoms. There is some evidence in support of novice clinicians being more prone towards deliberate deductive and inductive diagnostic reasoning strategies while experienced clinicians generally rely more on pattern recognition. Unlike novices, experts generally identify the most likely diagnosis based on pattern recognition, followed by deliberate and often unconscious deductive reasoning, aimed at justifying or ruling out the identified tentative diagnosis/-es. The dual-process theory offers an explanation for this two-step reasoning strategy, dividing human cognition into two systems; the intuitive system 1 (pattern recognition) and the deliberate and analytic deductive/inductive system 2. Pattern recognition (system 1) is immediate, often very accurate, and requires minuscule resources although it has been criticized widely for being prone to unconscious heuristic biases. The deliberate system 2 processes are slower and require significantly greater cognitive resources than system 1 processes. System 2 processes are generally considered less prone to unconscious biases, although several findings suggest otherwise. Most people prefer to apply the efficient and non-strenuous system 1 whenever possible, and only active system 2 processes when absolutely necessary. Unfortunately, pattern recognition (system 1) in a diagnostic setting relies on domain-specific experience, which is generally unavailable for novice physicians that therefore rely on the significantly more demanding system 2 operations during diagnostics. Several authors argue that educators should consider teaching or promoting pattern recognition if possible as it enables accurate, efficient, and less demanding diagnostics. Realistic and extensive pattern recognition training is possible within medical specialties that rely mainly on visual processing such as pathology, radiology, and dermatology. Large case libraries with annotated x-rays, pathology slides, or dermoscopic images can be made readily available for practice, enabling students to attain strong mental representations of the relevant differential diagnoses before initiating their clinical careers. There is extensive literature on how to facilitate the development of strong mental representations through pattern recognition training. Although identification of bone fracture types, cellular abnormalities, and skin conditions is important, additional knowledge retrieval and processing is necessary in order to provide patients with the correct treatment regime. Artificial intelligence and trained pigeons (yes, pigeons) handle image diagnostics impressively well, with several authors reporting accuracies on par with expert clinicians. However, unlike expert clinicians, machines and pigeons are currently unable to retrieve one or more complex precompiled diagnosis-specific scripts or schemas from long-term memory based on subtle diagnostic cues, followed by a deliberate analysis of the most likely diagnosis and appropriate treatment action. Illness script theory attempts to explain this incredible memory retrieval and processing operation through a simple framework. Illness scripts are defined as mental representations or schemas of an illness or disease that contains enabling conditions, i.e. demographics and medical history, consequences, i.e. the symptoms and presentations of the illness/disease, and faults, i.e. the biomedical explanation for consequences and enabling conditions. When an experienced physician recognizes a certain pattern of enabling conditions, consequences, and faults it activates and retrieves one or more illness scripts from long term memory, unlocking all of the knowledge stored within that script, including the physiology of the disease and interrelated scripts (differential diagnoses). Little is known concerning the formation and consolidation of illness scripts during medical and clinical training but there is increasing evidence in favor of combining the underlying physiology and anatomy (faults) with demographics (enabling conditions) and symptoms (consequences) within the instructional material of educational interventions in medicine. An underlying conceptual understanding of the medical condition's causal mechanisms seems to help trainees consolidate and inter-link their illness scripts, resulting in faster and more accurate script activation and utilization. When causal mechanisms or "the biomedical science" is included in standard instructional designs for internal medicine, containing clinical manifestation of the disease, students perform significantly better, especially on delayed tests and transfer tests. Similar positive effects on diagnostic performance have been observed when "extended" basic science descriptions that explain the underlying sociological and behavioral causal mechanisms are integrated within the instructional design for complex medical conditions that include a combination of social, somatic, and psychological problems. Student's immediate and retained diagnostic performance and transfer of knowledge improves if biomechanical visual analogies of the causal mechanisms are introduced within the description of the causal mechanisms. The positive effect of integrating biomedical science is best harnessed when it is intermixed with classic curricula describing clinical manifestations of diseases. The positive effects from classic textbook diagnostics on written cases also translate towards visual diagnostics, e.g. dental radiology, and the underlying conceptual knowledge of procedural skills, e.g. lumbar puncture. Integrating the causal explanations for visual criteria used in visual diagnostics increases both the immediate and long-term diagnostic performance of students. The improvement in diagnostic performance associated with integrating biomedical science has been found to be resilient against speed-accuracy tradeoffs, which indicates that a strong representation of "faults" within an illness script increases diagnosticians' ability to rapidly and accurately activate the script. Although basic science explanations seem to translate across the various medical modalities it is important to note that visual diagnostics varies significantly from remaining diagnostic modalities. When experienced dermatologists examine a skin lesion they immediately form a global impression generating one or more tentative diagnoses that are usually very accurate. Subsequently, they engage in a backward reasoning strategy attempting to find features that justify or reject their tentative diagnosis. The global impression is a result of intuitive system 1 operations while the deliberate analysis of differential diagnoses and feature search is deliberate system 2 operations. Although experts from remaining clinical specialties such as internists, neurologists, and cardiologists rely heavily on pattern recognition their assessments are based on input (symptoms, medical history, laboratory tests, etc.) that are gathered deliberately, requiring a larger degree of system 2 processes early in the diagnostic reasoning process. Investigators theorize that this difference in the "point in time" where pattern recognition is used by the various diagnostic specialties should be reflected in the education of the various specialties. Based on these reflections, the optimal educational intervention for clinicians that rely mainly on visual processing (dermatology, pathology, and radiology) ought to be case-based training with direct visual feedback coupled to a curriculum with a concise and relevant instructional design. The formerly mentioned studies show that integrating basic science within such an instructional design improves the diagnostic accuracy and transfer of knowledge to similar disease categories. However, former studies in visual diagnostics have failed to establish whether improved performance following the integration of basic sciences improves clinicians' pattern recognition, deliberate feature search strategy, or both. The educational interventions within these studies have been short and included less than 3 training cases per diagnosis, which investigators of this trial consider low in regards to the formation of mental representations for the visual classification of diseases (pattern recognition). Finally, former studies have not examined the effect of integrating basic science within the instructional design of training interventions for visual diagnostics and its effect on learning behavior (total duration and the number of times accessing instructional material), learning curves (formation of mental representations), and skill/knowledge retrieval. Investigatorsors hypothesize that the introduction of basic science explanations within the instructional design of case-based training in visual diagnostics will improve students' learning curves, retention, and retrieval of knowledge/skill following a washout period. To our knowledge, there are no previous studies that elaborate the effect of integrating biomedical explanations for visual criteria in a prolonged case-based educational intervention aimed at training students' pattern recognition skills. Acknowledging that learning may differ when focusing on novice and more advanced learners and that immediate performances do not always predict long-term learning outcomes, this is an important research gap to fill.

Research question:

In a group of medical students with limited dermatological training, what is the effect of integrating biomedical causal explanations of visual criteria during a prolonged case-based skin cancer training program in visual pattern recognition when compared with an identical instructional design without biomedical explanations? How will the displacement of students' cognitive resources from practicing pattern recognition towards understanding the pattern, affect their learning behavior, learning curve (accuracy and time per diagnosis), and retrieval of pattern recognition skills following a washout period?

Method:

The study will be conducted as a randomized controlled trial with an allocation ratio of 1:1. Enrolled participants must be actively enrolled at the Faculty of Health and Medical Sciences, University of Copenhagen, and have passed exams in basic histopathology and cellular physiology. Exclusion criteria are former training in dermoscopy or skin cancer diagnostics in general. A minimum of 60 students will be recruited. All participants are required to download a mobile application (Dermloop) containing the educational material. During sign-up participants will be asked to fill out information concerning their demographics, enter a six-digit trial code and sign a digital consent. Upon registration, participants will automatically be randomly allocated to the basic science or feature group. During the study all participants will complete four steps (see figure):

1. Pre-test
2. Digital training session in skin cancer diagnostics concentrated on seven differential diagnoses (nevi, melanoma, seborrheic keratoses/solar lentigo, basal cell carcinoma, dermatofibromas, and vascular lesions) that span 7 days. Instructional material differs between the two study groups. The basic science group will have access to modules that describe the characteristic visual criteria for each diagnosis and their underlying histopathological cause. The feature group will read identical descriptions of the visual criteria without an explanation of the underlying causal mechanisms.
3. Retraining session
4. Retention test

Pre-test:

The pre-test consists of 12 randomly sampled items (generalizability coefficient of 0.7) from a test item library (n= 25 items) with established validity evidence for skin cancer diagnostics.

Digital training session:

The session consists of an introduction and presentation of six diagnoses (nevi, melanoma, seborrheic keratoses/solar lentigo, basal cell carcinoma, dermatofibromas, and vascular lesions).

Introduction The introduction includes a short written introduction about skin cancer diagnostics and the six diagnoses that will be included in the educational intervention. Participants will be asked to briefly read through the various dermoscopic criteria that are characteristic of the seven diagnoses.

Case-based practice

Participants will be asked to practice on 500 skin lesions within 7 days. The training consists of quizzes with direct feedback. Participants will be asked to diagnose skin lesions based on the age and gender of the patient, a clinical image, a dermoscopic image, and the location of the skin lesion. Participants will receive immediate feedback following their choice of diagnosis. The feedback consists of their diagnosis, the correct diagnosis, access to the instructional design of both diagnoses, and an ability to toggle between the images and location. Each quiz is 10 cases long and the distribution of diagnoses is random across each quiz, but with an overall distribution across 100 cases of:

Diagnosis distribution:

Melanoma 20% Nevi 20% Seb. K./ Lentigo Solaris 20% Dermatofibroma 10% Basal cell carcinoma 10% Hemangioma 10% Squamous cell carcinoma. 10%

Retraining session:

Following the 14 day wash-out period, all participants will be asked to access the application and practice on another 100 cases within two days.

Retention test:

Seven days after the retraining session has been concluded participants will be asked to answer a retention test. The retention test consists of 12 randomly sampled items (generalizability coefficient of 0.7) from the same test item library (n= 25 items) used for the pre-test.

Primary outcomes:

Slope and plateau (if reached) of the initial learning curve (accuracy and time per diagnosis) Slope and plateau (if reached) of a secondary learning curve (accuracy and time per diagnosis) following a washout period

Secondary outcome:

Time spent reading instructional material Number of times that the instructional material has been accessed Change in performance from pre-test to retention test.