Learning styles and the myths surrounding it: a systematic scoping review of the literature.

Background: Learning style (LS) is the theoretical assumption that each individual has a better form for cognitive processing throughout learning. In medical education, LS has been studied as a tool to optimize medical learning. Teaching in the postgraduate medical environment embraces specic methodological aspects for mastering medical abilities and LS inventories have been widely used for enhance learning. However, no review has been done on this subject until this date. Therefore, a scoping review was performed to explore the extent of evidence on LS and postgraduate medical education. Methods: a systematic scoping review was performed according to PRISMA - ScR and JBI guidelines. We searched MEDLINE, ERIC, LILACS and SCIELO virtual library on February 2020. A peer review was performed with blinding of both investigators and any divergence was resolved by consensus. Searching strategy, search terms, exclusion and inclusion criteria and data charting were structured prior to the beginning of the study. Data was summarized and collated. Analysis of the quality of the evidence was also performed using specic tools. Results: 211 studies were obtained with the search engine after duplicates were removed. Of these, 40 were selected after applying exclusion and inclusion criteria. Two other studies were excluded post initial screening. The majority of studies were from United States. General surgery, internal medicine and family medicine were the specialties that had most studies on LS. Kolb LSI was the most used LS inventory. The majority of studies were observational with a cross sectional design (34 out of 38). Only four studies were RCTs with a low quality of evidence and a high risk of bias. It was also seen that LS may change through training, with work-hours and areas of specialty training. Conclusion: There is a lack of high quality studies to provide reliable evidence for the utilization of LS in postgraduate medical education and it is desirable for more Cohort or Randomized Control Trials in this area for a more robust evidence.


Introduction
Learning is the basic process of human adaptation (1,2). The comforting assumption that each individual has a speci c cognitive, affective and physiological trait that may enhance the learning process has been studied under the name of learning styles (3). Though this theory has been used for nearly 40 years, its validity has been controversial and despite evidence, it has shown some downsides (4,5), learning styles questionnaires are still widely used in learning environments(6,7). There are over 70 different learning styles inventories (VARK, Kolb LSI, Honey and Mumford, MBTI…) that focus on a diversity of psychological theories and its application in higher education is broad, despite not evidence based (7).
Medical education comprehends the complex task of delivering a well-quali ed physician to society. It´s desired abilities are described by the six dimensions of CanMEDS framework for a good medical expert (Scholar, Professional, Communicator, Collaborator, Leader and Health advocate). To optimize learning of these complex dimensions, tailoring learning to individual learning style is a theory explored in the literature.
After graduating from medical schools, students are enrolled in many learning formats in order to initiate their rst steps in a medical career(8). Though formats vary from internships, residency programs and other specialty trainings in postgraduate medical education, these phases are when doctors develop competencies under supervision towards independent practice after completion of their basic medical quali cation. Knowledge translation is a key ability to be developed in this phase and learning styles has been explored as a construct for optimizing the learning process of these professionals (9).
Knowing if there is any advantage on the use of LS in medical specialty training may be of great interest to orient learning strategies and rethink its applicability in medical education. Mapping what and how studies have been using this tool in postgraduate medical education may reveal patterns that could be used to aid medical trainees in their path to becoming independent practitioners and maximize learning throughout residency. Moreover, mapping studies on the eld may give bearings for further research on LS.
We sought to know if medical trainees and residents' LS may be a situational diagnostic tool to orient and organize postgraduation programs and if there is any association between training in each medical specialty and medical trainees learning styles. Despite this being widely used, no study has compiled information regarding postgraduate medical education and learning styles to our knowledge. This paper aims to map and summarize data available in international literature and review its current use.

OBJECTIVES
To deliver what has been studied on the use of learning styles in postgraduate medical education. We aimed for patterns of utilization of learning style inventories and medical specialty training as well as the impact of learning styles as a situational diagnostic tool for postgraduate medical programs. We also wanted to know how this subject is studied, by uncovering the studies designs, tools used to measure the LS, number of participants on average, the specialties that more commonly study these tools and the quality of current evidence available.

Methods
Study design: a peer-controlled scoping review was performed according to JBI and PRISMA -ScR statements (10). Differently from a systematic review, the study question was broad and a scoping review was preferred, thus registration on PROSPERO was not necessary. Descriptive synthesis of data was registered and analyzed.
Review process: we performed the review in a systematic way after de ning our study questions, search method and database, study selection, charting data and summarizing results (11,12).
Research question: What is known in the literature about learning styles in postgraduate medical education? How do different medical specialties differ on their learning styles during postgraduate training? How is LS applicable in postgraduate medical education?
Searching strategy: We opted to search MEDLINE, ERIC, LILACS and SCIELO virtual databases. Two investigators (LM and MD) used the same Mesh terms and proceeded independently on study selection after de nition of inclusion and exclusion criteria. Initially, title screening and abstract was performed and on selected articles exclusion and inclusion criteria were applied. Any discordance was resolved after deliberation and consensus. Data was collected until February 2020.
As for search terms the following Mesh were used: "Internship and Residency"; "Education, Medical, Graduate" and "learning". Also, we looked for all elds containing preferences and styles since there were no Mesh terms in the literature. With Boolean terms we built our search construct: (Residency or internship or graduate medical education) AND "Learning styles" and ltered by Medical Education.
Inclusion criteria were de ned as: 1) Language (Portuguese, English and Spanish); 2) Clear reference of type of learning style inventory used; 3) Traditional postgraduate medical programs.
As for exclusion criteria:1) Undergraduate medical students; 2) Students that were not registered in postgraduate medical training programs; 3) Studies involving medical specialists; 4) Other healthcare students; 5) Postgraduate curricula as sample; 6) Studies that did not specify a medical specialty; 7) Learning style is not an outcome of the study; 8) Not original data.
Data extraction was peer reviewed and disagreements were resolved by consensus. Data charting was done on a purpose built MS Excel Spreadsheet and the following variables were de ned for obtaining information: author (year), country; sampling population; medical specialty; study design; LS tool used; and a summary of evidence brought by each study individually.
The response rate was de ned as the amount of total included participants divided by the total number of invitations or eligible students at screening. Studies that didn´t specify this information were labeled as "not available" in supplementary table 1. The number of participants was de ned as the number of residents that responded to the questionnaires, while the number of evaluations was de ned as the amount of evaluations that were used in the study analysis.
Data analysis: Learning styles of different medical specialties postgraduate programs were highlighted and discussed according to the type of learning style analysis and its utility in each program. We sought to see de ciencies in current evidence regarding our study questions. Studies that brought more than one population of interest (i.e. college or undergraduate study) and medical residents were also included, however, we have focused on the data primarily from these last ones to build our table.
To de ne study design, we have used a previously published article (13). We have also explored the quality and biases in the included studies using known tools to analyze the risk of bias such as ROBINS I for clinical non-randomized and quasirandomized trials (14). This was performed in a blind manner by two of the authors (LM and RV). Disagreements were resolved by a third party (MD).

Results
Two studies were excluded after title and abstract screening because one of them was about junior doctors (UK) training (15) Other fourteen studies (20, 22, 24, 26, 28-32, 34, 42, 46, 48, 52 e 54) haven't described a response rate and one RCT(42) had its number of participants pre-de ned and excluded part of the eligible participants during randomization.
Randomized clinical trials (RCT) represented four studies with a total of 285 LS evaluated, (average of 71,25 per study -range 18 to 121). Analysis of the risk of bias using ROBINS I showed that all the four studies had a high risk of bias. Three out of the four RCT used the Felder and Solomon's Index of Learning Style (ILS) and one used the VARK (Visual, Aural, Read/write, and Kinesthetic) questionnaire.
The outcomes used, vary among studies, but usually the LS were evaluated as secondary outcomes. The most used tool to analyze LS was the Kolb Learning style inventory with a total of 20 studies (52%), followed by the VARK questionnaire and the Felder and Solomon (three studies each). The Honey Alonso and the Vermunt Inventory of Learning Styles (ILS) was used in two studies each and eight other studies used different LS inventories (see table 1 for all LS inventory used).
The specialties that most studied LS were general surgery (n=10), followed by internal medicine (n=6) and family medicine (n=5). In a subdivision by clinical and surgical specialties, we found that 23 (56%) were in clinical areas and 17 (41,5%) in surgical areas.
How do different medical specialties differ on their learning styles during postgraduate training? Table 1 summarizes results of individual studies. Overall, residents from surgical specialties, (general surgery, ophthalmology, orthopedics, head and neck and neurosurgery) are more activists and show a tendency to prefer a converging learning style on Kolb LSI. Only one study directly compared learning styles of surgical residents and internal medicine residents with similar results despite differences amongst the groups were not signi cant. Family medicine residents showed a tendency to prefer more abstract forms of learning, whereas internal medicine residents had a tendency to prefer more sensing and re ective learning styles. As the eld of specialties narrowed, there was an inclination towards active learning styles with a more concrete component than an abstract one.
What was the quality of evidence found?
Using previously described tools (55) for cross sectional studies, we can conclude that the majority of studies were of low quality with a high risk of bias. Most of the studies haven't set up inclusion and exclusion criteria and even when it did, usually the residents that left their residency program were excluded from analysis. Also, we have seen that some of the studies used more than one population, such as medical students and faculties, which can under power studies results for post-graduation.
There was also an important structural methodological difference among the studies starting with the huge variance seen in the tools used to measure LS (see table 1).
How is it applicable in postgraduate medical education?
A summary of evidence brought by each study individually can be seen in table 1.
As for a general appraisal of the studies, we can divide studies as cross sectional and RCT. Further information regarding individual aspects of studies, were explored bellow.

Discussion
The purpose of this study was to summarize and gather the evidence regarding learning styles in medical postgraduate programs and medical residency. In our review, we found that most of the studies were performed in the USA, Canada and UK which makes it unrealistic to extrapolate this data to other countries.
Regarding specialties, LS were more studied in the family medicine/GP, internal medicine and general surgery so far. This may be explained partially because these specialties are more common, which by consequence, have more trainees than the others. Another possibility might be that these areas produce more in medical education research.
In surgery and internal medicine, since these specialties in general are a prerequisite for further subspecialties, these postgraduate students form a wider and more heterogeneous population cluster of LS. For example, we found that most general surgery trainees had a multimodal learning style when VARK was surveyed differently from the general population.
This nding may be obscured by the study population heterogeneity.
On the other hand, despite being a heterogeneous population, the only LS dimension that was preferred from general surgery trainees was the activist dimension. Therefore utilizing this information may provide ways to improve residents' learning curve in order to have less impact in patients morbidity and mortality undergoing surgical procedures by residents(56,57). About LS inventories used, we have found a great variety of tools among the studies. In total we have found thirteen different tools used. The Kolb LSI was the most used overall and especially in the cross sectional studies. This is one of the most popular tools to analyze one's LS. Kolb  Differently, the tool most used in the RCTs was the Felder and Solomon's Index of Learning Style (n=3, 75%), which might be partially explained by the fact that these three studies were from the same study group (40)(41)(42).
Regardless of study design, methods of standardization and data should be pursued in future research in this eld to increase external validity and to make feasible for a systematic review to compile studies and increase robustness of evidence. So far and with the studies that we have found, it seems not possible to do that because of strong methodological differences amongst the studies.
As for the results found, seven studies have tried to correlate LS with scores in summative assessments (17,23,29,30,32,34,49), however only four have found a positive result (17,23,34,49)  As we can see, the studies hardly can be comparable due to powerful methodological differences. Not to mention the fact that the four studies that had found a positive result had a cross sectional design. A natural consequence of that is the fact that these results may have been in uenced by confounder's bias. For example, the fact that a dominant aural learner scored lower in a written exam than his fellow who learns better by reading or writing, may represent a strong confounder since the written exam itself may favor the last kind of student, hampering ndings previously described (34). Further studies in this eld may take that in account and perform studies with stronger power such as cohort studies and RCT. Finally, the fact that studies with positive result are more likely to be published(59) brings another issue, since we may have not been able to nd more studies with negative results that may not have been published.
One other study have tried to correlate which are the LS at more risk to fail or quit surgery residency (30) and showed that residents that had transferred themselves to a non-surgical program were more likely to learn by observation in Kolb LSI, concluding that this residents were at more risk of leaving surgical specialties and should be prioritized. However, this result is not de nitive since it happened in a subgroup analysis and might be heavily biased and underpowered.
Two studies (30,31) found a positive correlation of the activists LS of general surgery residents with number of cases performed during residency, however once again this maybe partially explained by the fact that if the standard approach to surgical residents are focused exclusively on active surgery, residents with different LS may feel discouraged to enroll in academic activities.
In our opinion, trying simply to correlate learning styles with different parameters like test scores, number of cases operated or risk to fail may be a deterministic thought that does not help to improve the learning process. On the contrary, the variety of LS amongst the residents may improve one's process of learning by increasing satisfaction and showing representativeness among the different cognitive processes. We believe that studies should focus on strategies to adapt the learning process to speci c LS clarifying this process to the students, which by itself may improve learning.
Although it is hard to produce high quality evidence studies in the medical education eld due to speci c methodological issues such as di culty to de ne and measure outcomes and blinding of subjects and researchers, a shift towards more robust studies is urgently needed. For that objective be ful lled, more randomized control trials should be pursued, since purely descriptive cross-sectional studies will not improve current evidence.

Limitations
As for limitation, some countries may not have been found in our study since we limited language in our inclusion criteria. We opted to restrict the language to English, Spanish and Portuguese due to researcher's familiarity.
Our review encountered di culties in obtaining data from studies especially concerning type of specialty training, learning style distribution among studied population and obtaining articles prior to 1980. We did not nd studies in other European countries well known for research production in medical education such as the Netherlands(60).
The quality of evidence was affected mainly due to study design and methodology. There is an urgent need for better performed randomized control trials in education and surrounding learning styles that is not different. Even in the four RCT encountered, a more robust methodology could have aided in a more secure result.
Another issue that may have hindered our research was the fact that some studies have more than one population as sample. We chose to dissect post-graduation data from the rest of the sample, which could have underpowered these studies individually. However, this method may have showed a more representative picture of what is known regarding postgraduation students and residents´ LS.
Lastly, since our study was designed as a scope review, our study delimitation was wide, with three objective questions, which could have made it more di cult to summarize and compare the data we have found. However, as we can see from our results, our study design has showed itself adequate, due to the huge heterogeneity found, amongst the studies. With the data we have gathered, seems unrealistic for a systematic revision to be made in the LS eld.

Conclusion
High quality evidence surrounding the applicability of LS on medical residency and post-graduation is scarce. Further studies are urgently needed on this matter, especially randomized control trials and cohort studies. We are aware of the challenges on medical education research, but a conclusion surrounding applicability of LS on post-graduation seems implausible with the research done so far. The quality of all of the studies found was low with a high risk of bias due to methodological problems More robust research on this eld is possible with studies using stronger methodological features.