Empathy is considered a crucial component of healing relationships, but is not a predictable outcome of medical education (1, 2). Numerous recent investigations have identified a gradual and often persistent deterioration in student and resident empathy as training proceeds (3–8). Although empathy has also been shown to increase following educational experiences that emphasize relational learning objectives, medical students’ views on social, ethical, and interpersonal issues tend to become more negative and cynical over time (9–11). At all stages of training, however, high personal well-being and greater quality of life appear to protect students and residents from declining empathy (12–14).

One shortcoming of studies on empathy in medical education is that students’ viewpoints on what affects empathy have not been included. Commonly used empathy scales measure levels of empathy, but do not assess students’ opinions on what affects their ability to be empathic, an area that may yield useful information in understanding and promoting the development of empathy in physicians. The structured conceptualization process known as concept mapping offers a method of assessing hundreds of individual viewpoints on empathy without using a preconceived questionnaire (15, 16). Integrating qualitative and quantitative statistical methods, concept mapping encompasses brainstorming, idea synthesis, sorting of ideas, nonmetric multidimensional scaling, and hierarchical cluster analysis. The end product is a visual depiction of group-generated ideas, showing all the major viewpoints and their interrelationships as determined by the participants themselves.

Concept mapping has been successful in numerous settings using Concept Systems software (2004, Ithaca, N.Y.) (17–19). First, participants generate a list of factors using a structured brainstorming process. Next, participants identify relationships among these factors by sorting the list into categories of related factors. Multivariate statistical techniques are applied to this information using Concept Systems software, which then depicts the sorted factors as points on a graph, creating a “point map.” On a point map, factors frequently sorted together are positioned more closely, while those sorted separately are located farther apart. Intercoder reliability and semantic validity are built into the method’s capacity for revealing contextual meaning while unitizing and preserving the original texts simultaneously (20). Concept mapping thus maximizes conceptual accuracy, facilitating a succinct yet verbatim presentation of participants’ unique and common points of view.

This study posed three questions: What are the perceptions of students regarding factors affecting empathy in medical education? What types of perceived empathy factors emerge during medical education? Do students’ perceptions differ based on their varying levels of experience? Using a new construct obtained through concept mapping, we present a comprehensive, student-centered response.

Each class and all residents in the first postgraduate year (PGY-1) at Albany Medical College (N=637) were invited to participate in an anonymous brainstorming survey approved by the college’s institutional review board; 293 participants (45.9%) responded. The survey, sent electronically, asked participants to respond to a focus statement and to list their year in medical school or residency, gender, ethnicity, and age. The same population was then invited via e-mail to sort and rate the group’s responses. Sample sizes of 20 to 40 are considered normative for sorting and rating; in this study, 34 respondents participated in the sorting and rating tasks (16). Participant demographics are compared with the larger population in Table 1.

The survey defined empathy as a “complex internal response involving compassion, understanding, and momentary sharing of feelings between oneself and another. It is the capacity to experience the feelings of another, while simultaneously maintaining a sense of separateness.” The survey then asked participants to respond to a focus prompt: “During the course of becoming a doctor, a variety of things occur that affect one’s ability to be empathic. Please list all the factors you can think of.” Using standard rules of brainstorming, the prompt was kept as broad as possible so as not to presage the types of clusters that might emerge, and participants were encouraged to generate as many statements as they could (21, 22). Because of the resulting oversaturated list of responses, standard principles of text analysis known as “idea synthesis” were utilized to minimize redundancy and maximize clarity of statements (16, 20). During this process, each statement listed was examined based on clarity of language, number of ideas per statement, and redundancy of language or meaning. Statements were edited for spelling and grammar, multiple ideas contained in one statement were parsed into separate statements, and repetitive statements describing one idea were collapsed into one representative statement.

Following completion of the final list of perceived empathy factors, each factor was written on an index card. Each participant was given a complete set of index cards with the instruction to “sort the cards into piles that make sense.” The only restrictions given were “factors may not be placed in two piles simultaneously, there may not be a pile consisting of all the factors, and any factor thought to be unique to the rest must be placed into its own pile.” Each participant then rated the relative importance of each empathy factor using a 5-point scale (1=not at all important, 5=most important).

Multidimensional Scaling Analysis

Using Concept Systems Software Program (CSSP) Version 4b (2004, Ithaca, N.Y.), each participant’s sorted pile of factors was put into a binary symmetric similarity matrix consisting of as many rows and columns as there were factors. CSSP then added the matrices, forming a group similarity matrix. The results were represented as a two-dimensional (x, y) configuration, or “point map,” on which factors frequently sorted together were positioned more closely and those sorted together less often were further apart (23).

CSSP used Ward’s algorithm to group the multidimensional scaling results into clusters of related coordinates, initially providing as many clusters as there were factors (24). To select a solution that most closely reflected the perceptions of participants, configurations were first examined by successively combining two or more clusters until all factors were located in a single cluster. Solutions excluding overlaps in content achieved the truest partitioning of the conceptual similarities and dissimilarities described in the raw data (25–27). Factor bridging values were also examined to guide decision making about the internal consistency of each potential cluster (16). Low bridging values indicated stronger conceptual similarity. Titles were chosen by the research team when consensus was reached on acceptable boundaries for the various groupings of factors.

Mean importance ratings were calculated for each factor and for each cluster. Raters were grouped by level of experience, gender, and ethnicity. Group rating patterns (“pattern matches”) were compared using a ladder graph (28). Strong agreement between groups of raters on any mean cluster importance rating was indicated by a nearly horizontal line. Concordance/discordance between ratings was compared using Pearson’s r. Student’s two-tailed t test assessed potential associations between mean cluster ratings and differing levels of experience, gender, and ethnicity.

Two hundred ninety-three brainstorming participants generated 1,683 statements, which were reduced to a final listing of 160 perceived empathy factors (Table 2). A smaller sample of medical students (n=32) and PGY-1 residents (n=2) sorted and rated the relative importance of each factor. Multidimensional scaling results are shown in Figure 1. The point map generated a low stress value (0.25), indicating a stable pattern of relationships among the sorted factors (16). On the map, each point symbolizes a factor, referenced by an identifying number found in Table 2. Points closer together on the map represent similar concepts. For example, in the lower right quadrant of the map, factors 55 (“mentors”) and 56 (“examples set by senior teachers and doctors”) are located close together, while factors less related, such as factors 55 and 135 (“prejudice,” upper left quadrant) are found farther apart.

The four-cluster solution (Figure 2) best minimized average cluster bridging values while providing the most useful solution. The clusters broadly describe four dimensions of medical education perceived by students as affecting empathy. The “personal connections” cluster represented the personal experiences, relationships, and values that students perceived as impacting empathy. Factors such as “too draining to empathize,” “impatience,” and “cynicism” comprised a “negative feelings and attitudes” cluster. Facilitating experiences, such as “directly and significantly changing a patient’s life,” “remembering why I wanted to become a doctor,” and “clinical experience with doctors who demonstrate empathy” comprised a “mentoring and clinical experiences” cluster. Factors such as “assembly line medicine,” “sleep deprivation,” and “being mentally, spiritually, and physically drained by school, stress, and work overload” comprised an “undermining school and work experiences” cluster.

Importance ratings and bridging values are itemized in Table 2. “Listening” was rated the most important factor. “Mentoring and clinical experiences” was considered the most important cluster (Figure 2). Factors central to the integrity of each cluster had lower bridging values and were emphasized in cluster titles. Additionally, because factors such as “clinical experience with doctors who demonstrate empathy” and “mentors” were given higher importance ratings and considered integral student experiences, they were reflected in the title of the “mentoring and clinical experiences” cluster even though their bridging values were higher.

Cluster importance ratings are compared by level of experience in Figure 3. The PGY-1 group was excluded due to its very small size (n=2). Despite a trend toward lower importance ratings by third-year students on the “negative feelings and attitudes” and “undermining school and work experiences” clusters, each group tended to rate factors similarly, resulting in little variation in the hierarchy of ratings (r≥0.86). The lowest correlation was found between second- and third-year students for the “negative feelings and attitudes” cluster (r=0.50), but when all levels of experience were compared by cluster, no significant relationships between mean cluster importance ratings and level of experience were observed (Student’s t test, p>0.05). Strong agreement was found between genders for all cluster importance ratings (r=0.95); no associations were found between importance ratings and ethnicity.

This investigation of medical student empathy is the first to examine students’ opinions on what influences the development of empathy during medical education and offers an immediate, student-centered context for interventions designed to enhance physician empathy. According to the views of students in this study, both personal and educational factors impact empathy during medical education. Factors such as mentoring, role modeling, clinical experiences, personal wellness and connections with others, negative attitudes toward patients, and inhibiting school/work experiences were perceived by students as continuously active from the earliest stages of medical education. Supporting the emerging association between quality of life and empathy during training, these viewpoints highlight the increasing relevance of the connection between personal experiential factors and the attainment of empathy during medical education (12–14, 29).

How can medical schools structure empathy education in a way that builds upon what students and residents perceive as most important? According to students in this and other studies, mentoring may be an important resource for processing experiences in and reactions to the educational environment, as well as for modeling empathic care (30–32). Physicians who have themselves negotiated the sometimes negative context of training and remained empathic are ideally positioned to help students and younger physicians to process difficult interactions with patients and the sometimes empathy-eroding effects of school, work, and life experiences.

Students gave similar factor importance ratings regardless of year in medical school (Figure 3). Based on previous research demonstrating a decline in empathy by the end of the third year of medical school, we expected factors that devalue empathy to be seen as more important by third- and fourth-year students (3, 7, 8). Although the sample size limited our ability to make generalizations, this finding suggests that as students gained experience, a significant shift in the importance of negative empathy factors may have been prevented by the educational culture of the study population, or that students’ preexisting attitudes and beliefs were not significantly changed by medical education. If the latter is correct, early education directed at challenging assumptions about patients and patient care may be crucial to promoting empathy in medical education.

For psychiatry educators, the teaching relationship offers a natural conduit for empathy education. The four-cluster model may be utilized at each stage of education. In the first 2 years of medical school, efforts to define and explore the concept of personal well-being in both self and other, the impact of context and environment on empathy, and the possible clinical effects of preconceived attitudes and beliefs toward and about patients may be emphasized. The third, fourth, and postgraduate years offer multiple opportunities for mentoring and identifying problems of negative countertransference and burnout. Resident supervision sessions may further nurture empathy by facilitating self-awareness and personal growth. Student ratings of perceived empathy factors suggest that the supervisor’s ability to model empathic concern and listen for breaches in empathy may be crucial to empathy development for young physicians.

Empathy education for an entire student body will be more effective, and easier to implement, if the issues targeted are important to the group as a whole. For example, when factor importance ratings within the “mentoring and clinical experiences” cluster were examined year by year, factors 57 (“clinical experience with doctors who demonstrate empathy”) and 35 (“listening”) were the most highly rated, suggesting that mentoring and teaching relationships that model empathic behavior are extremely important to students at all levels of training. Similarly, in the “school and work experiences” cluster, factors 98 (“being mentally, physically, and spiritually drained by school, stress, and work overload”) and 21 (“amount of time spent with patients”) were most highly rated, suggesting that the connection between burnout, well-being, and the ability to be empathic is perceived by students as important at each stage of training. This finding is consistent with other recent studies on well-being and empathy during medical school and residency (13, 33, 34).

Limitations and Future Directions

Constructing the concept map at other institutions will establish a broader based model of students’ perceptions of what affects empathy in medical education. Although the size of the sorting group was normative for concept mapping and representative of the larger population of medical students (Table 1), it may have been too small to allow significant differences in rating patterns among groups to emerge (16). Although this study attempted to maintain the integrity of the student responses by retaining a large number of perceived factors, future concept mapping studies could produce a more manageable factor list by using a smaller number of brainstorming participants and proceed from brainstorming to sorting in one sitting, utilizing a web-based design (16). The four-cluster model, as opposed to a larger number of clusters, has implicit limitations but was chosen to provide medical educators with an easily utilized map of students’ perceptions while preserving the most cohesive thematic meaning. Although other studies may further delineate the identified themes, it is likely that because of the large number of factors identified in this study, the general outline described by the four clusters will remain.

This initial map of students’ views on factors affecting empathy offers a new perspective on the development of empathy during medical training. Future studies might objectively measure and compare students’ empathy levels with their perceptions on what influences empathy during medical education. Importance ratings of factors within one or more clusters could also be compared with students’ empathy levels before and after interventions designed to explore, amplify, or promote empathy. Additional studies comparing teacher and student ratings of student-perceived factors as well as educators’ perceptions of what influences empathy will further add to our understanding of the relationship between empathy and medical education.

Adjacent points depict empathy factors most often sorted together; points further apart represent factors less frequently sorted together. Factor titles are listed in Table 2.

Shaded polygons depict the most distinct regions of the empathy factor point map identified during hierarchical cluster analysis. Cluster thicknesses illustrate each cluster's mean importance rating relative to the importance ratings of all other clusters. Cluster content and mean importance ratings are itemized in Table 2.

Importance ratings: 1=not important; 5=most important

At the time of submission, the authors declared no competing interests.