Our study obtained data on empathy from students in all 6 years of medical training. Between 55% and 78% of each cohort in the Core Science component of the course participated at some point in the study. The comparable figures for the Clinical component were 50% to 82%. These figures, allied with missing data analyses indicating that initial scores did not predict later non-response, would support the view that the findings can reasonably be generalized to the population of medical students at the University of Cambridge.
We found statistically significant gender differences in affective empathy at all 6 years of medical training and in cognitive empathy for 4 years. These findings support those of other studies among different populations, using a range of instruments [24, 25, 27, 32].
Differences in mean scores between men and women were larger than any of the changes in mean scores between different stages of the course. Gender differences in IRI-EC ranged from 1.64 to 3.11 and for IRI-PT from 0.79 to 2.04. Differences in mean scores at different stages of the course were generally less than 1.
We found that with time, affective empathy declined on average for men and sensitivity analysis (removal of outliers) revealed that women's affective empathy declined in the Clinical component of the course. Amongst women in the Core science component of the course affective empathy remained constant on average. There were no significant changes in cognitive empathy amongst women or amongst Clinical men. Sensitivity analysis (removal of outliers) revealed that in the Core Science component of the course men's cognitive empathy increased. However, although these changes were statistically significant, regression coefficients indicated that they were extremely small and therefore of questionable practical significance.
By using the generic IRI, we were able to differentiate between affective and cognitive dimensions of empathy. This is at variance with more recent studies of empathy using the Jefferson Scale of Physician Empathy (JSPE) [9, 22, 24, 27, 30–32, 38–40]. However our approach enables us to set in context empathy scores recorded by our medical students. The mean scores recorded by students in our study for both IRI-EC and IRI-PT (at Year 1, males IRI-EC 19.43, IRI-PT 18.05, females IRI-EC 21.07, IRI-PT 19.37) are similar to those recorded by medical students in other studies [21, 28, 38, 39]. The scores also resemble those obtained from studies of other undergraduate student populations [16, 21, 41, 42].
However, apart from general notions of more empathy being better for patient care there are no benchmarks for medical student empathy. Further, comparisons of scores for a generic instrument such as the IRI recorded by medical students with other populations do little to inform medical education. Given its widespread use in a medical context, perhaps it is now time for benchmarks for medical students to be established for the JSPE, which also take account of possible differences in age and culture [39, 40].
Our findings would suggest that any changes observed in either affective or cognitive empathy amongst Cambridge medical students were small and of limited practical significance. This supports the view expressed by Colliver  in the recent debate about decline in medical student empathy [43–45].
The investigation reported here is limited by being based on one UK medical school, providing a "traditional" course. The voluntary nature of the survey meant that initial response rates were variable (Table 1). Students entering in 2007 and 2008 have had the opportunity to participate on 3 occasions. Of these, 29% of Core Science students and 45% of Clinical students have done so. Nevertheless, the missing value analysis supports the view that those continuing to participate could be considered representative of all student entrants in their year group and that continued participation was not influenced by levels of either affective or cognitive empathy recorded at the beginning of participation. However, since the missing value analysis is based only on initial values for affective and cognitive empathy, we cannot completely exclude the possibility of an association between an unobserved change in either affective or cognitive empathy and the missing values.
Although the IRI enables measurement of different dimensions of empathy it is nonetheless a self-report instrument and we cannot predict the extent to which reported levels of empathy are reflected in the actual behaviours of our students or influenced by socially desirable responses.
The results of this study highlight further research questions. For example how far does the gender difference in affective empathy persist after qualification? Is it a reflection of some innate difference which may have implications for selection?