A cross-sectional study was conducted at Medical School of Nantes University (France). Ethical approval was obtained from the Ethics committee of Nantes University. Project staff was not involved in the diploma allocation. All general practice residents (n=380) were eligible to the study. The return rate was ~40% (n=150). Thirteen incomplete answers or with outlying scores (Dixon’s test) were discarded. The final sample (n=137, mean age =26.5, SD=1.3) corresponded to 94 women (mean age =26.6, SD=1.2) and 43 men (mean age =26.5, SD =1.6). The proportions of residents registered in years 1, 2 and 3 of the formation corresponded to ~30% (n = 41), ~39% (n=53) and ~31% (n=43). The gender ratio (F/M) differed significantly between years 1 (25/16), 2 (43/10) and 3 (26/17) (χ²(2, N=137) = 6.293, p=.043). Becoming a general practice resident was a deliberate choice for 123 respondents (~90%).

The questionnaire included items related to socio-demographic information (gender, age, year of the curriculum). Items from the French versions of the Jefferson Scale of Physicians Empathy (f-JSPE; 25 items), the Connor-Davidson Resilience Scale (f-CDRISC; 21 items) and the 20-item Toronto Alexithymia Scale (f-TAS20) was also included in our survey. These scales have demonstrated satisfactory reliability for the assessment of resilience, empathy, and alexithymia.^43-45

The scoring of f-JSPE items is based on a 7-point Likert scale with one corresponding to “full disagreement” and seven corresponding to “full agreement”.¹⁰ items needed reverse scoring because of their formulation. Total f-JSPE score varies from 20 to 140; a high score indicates high empathy. F-CDRISC and f-TAS20 scales use a 5-point Likert scale for item scoring. For f-CDRISC, the item score varies from 0 (“full disagreement”) to 4 (for “full agreement”), and the total score varies from 0 to 84. For f-TAS20, the item score varies from 1 (for “full disagreement”) to 5 (for “full agreement”). Five items needed reverse scoring. Total f-TAS20 score varies from 20 to 100. High scores for f-CDRISC and f-TAS20 indicate high resilience and alexithymia traits.

All general practice residents were contacted by e-mail. They were invited to answer an electronic version of the questionnaire. The access to the questionnaire was granted after the validation of an electronic informed consent in which goals, means, and methods have been described. To ensure resident’s anonymity, each resident encoded an identifier by using the first two letters of his/her first name, a number corresponding to the day of birth (between 01 to 31), a number corresponding to the year of birth (between 00 and 99), a number corresponding to the place of birth (between 00 and 101) and the first two letters of his/her mother’s given name.

Data were collected at the end of the survey. They were analysed with SPSS 21, Sigma Plot 12 and R (3.2.5) software. F-CDRISC scale has recently been validated for resilience measurement in health students.⁴³ The reliability of f-JSPE and f-TAS20 scales was re-assessed in our conditions. The item-score correlation coefficient (rIS) was calculated. As Cronbach’s α coefficient is not a good estimator of internal consistency for multifactorial scales,⁴⁶ greatest lower bound (GLB) and McDonald’s w coefficient were calculated as recommended elsewhere.⁴⁷ The scale appropriateness was deduced from the Kaiser-Meyer-Olkin coefficient (KMO, optimal value above .8) and the Bartlett’s test (optimal p < .001). The item-sampling adequacy was deduced from the anti-image correlation coefficient value calculated for each item (AIC, optimal value above .5).

To confirm the 3-factor structure of the f-JSPE and f-TAS20 scales, a confirmatory factor analysis (CFA) was performed by using the maximum likelihood method. The goodness of fit of a factor model was estimated according to Byrne’s recommendations.⁴⁸ The following indices were calculated: i) the normed χ² (χ² /df, optimal value below 2.0), ii) the standardized root mean square of residuals (sRMR, optimal value below .05), iii) the goodness of fit index (GFI) and the adjusted goodness of fit index (AGFI), both being optimally higher than .90), iv) the root mean square error of approximation (RMSEA, optimal value below .08) and its relative p(close) for which a value above .05 indicates a good fit, and v) the comparative fit index (CFI) (optimal value above .90).

The normality of data distribution was verified by using the Shapiro-Wilk’s test (significance criterion p>.05). Mean (M), standard deviation (SD), 95% confidence interval (95% CI) and the average score (i.e., the ratio between the total score and the number of items of a scale) were calculated for the different scales and different subgroups. A correlation between two variables was deduced from the value of the Pearson correlation coefficient. The differences were estimated by Student’s t-test (for two-group comparison) and by analysis of variance (ANOVA, for multiple-group comparison) in considering a risk α = .05. The threshold of significance of a difference was set at p<.05. When a difference was significant, the effect size was estimated by Cohen’s d coefficient (for two-group comparison) with the correction of Rosnow and Rosenthal⁴⁹ or by η² coefficient (for multiple-group comparison). Small, medium and large effect sizes were respectively characterized by d, η² ≤ .2, .2 < d, η² ≤ .5 and .5 < d, η².⁵⁰

Multiple linear regression analyses were performed to test whether alexithymia and empathy can predict resilience. Gender (male = 1, female = 2) and year of formation (year 1 = 1, year 2 = 2 and year 3 = 3) were also considered as potential predicting factors of resilience. Standardized regression coefficient (b) and p values were calculated to estimate the relationships between the different variables.

The heterogeneity within a dataset can be ascertained by cluster analysis (CA).51 A non-hierarchical CA (K-means) was run by using standardized scores (z scores) for f-JSPE, f-CDRISC and f-TAS20 as clustering variables. The validity of different models (from 2 to 4 clusters) was assessed. The significance of between-cluster differences was calculated by unpaired Student’s t-test (for a 2-cluster model) or by ANOVA and post-hoc Bonferroni correction (for 3- and 4-cluster models). The most likely cluster model contained the highest number of clusters for which all z scores were significantly different. The validity of the retained model was assessed by discriminant analysis (DA), in which z scores were considered as independent variables, whereas the number of clusters corresponds to the dependent variable. For cluster and discriminant analyses, the significance criterion was set at p < .001.

As shown in Table 2, f-CDRISC was significantly best-scored by male residents (low effect size). A significant difference in resilience (low effect size) was observed between the residents of the different years of formation. However, our analysis indicated that such a difference originated more likely from an interaction between gender and year of formation. Gender-related or curriculum-related differences for f-JSPE or f-TAS20 scores were not significant.

Resilience and empathy were positively correlated, r₍₁₃₅₎ = .36, p<.001. Negative correlations were observed between alexithymia and resilience, r₍₁₃₅₎ = -.40, p<.001, and between alexithymia and empathy, r₍₁₃₅₎ = -.38, p<.001. A regression model describing the contribution of the different variables to the resilience was elaborated. It was supported by a significant regression equation, R²_adjusted = .27, F_(4,132) = 13.39, p< .001, Durbin-Watson coefficient = 1.625. The resilience was positively predicted by empathy, b = .255, t₍₁₃₂₎ = 3.19, p=.002, and by year of formation, b = .157, t₍₁₃₂₎ = 2.12, p=.036. Meanwhile, gender (female < male, b = -.231, t₍₁₃₂₎ = -3.14, p= .002) and alexithymia, b =-.284, t₍₁₃₂₎=-3.57, p < .001, corresponded to negative predicting factors. Alexithymia negatively influenced empathy, b =-.270, t₍₁₃₂₎ = -3.21, p=.002. Neither alexithymia nor empathy was affected by gender or by year of formation. This model was validated by CFA as demonstrated by the goodness-of-fit indices, χ²(5, N = 137)/df = .55, GFI = .99, AGFI = .98, sRMR = .042, RMSEA < .001, p(close) = .846.

Different convergent solutions resulting from CA were observed. However, ANOVA shows that the differences produced by 3-and 4-cluster models were not significant (data not shown). For this reason, the 2-cluster model was considered as the most reliable solution. This was validated by DA. The cluster effective calculated by DA were found to be identical to those determined by CA. A very strong correlation between the allocations determined by CA and by DA was observed, r(135) =.98, p< .001.

The gender repartition determined for each cluster was similar (Table 3). By contrast, between-cluster differences calculated for resilience, empathy, and alexithymia were significant. The residents of cluster 1 were more resilient and more empathetic than those of cluster 2 (strong effect size). The residents allocated to cluster 2 exhibited a higher alexithymia profile (strong effect size). The between-cluster difference was also observed when the year of formation was considered as comparison criterion. In particular, residents from 3rd year were more abundant in cluster 1, whereas cluster 2 was mainly constituted by 1st and 2nd-year residents.

To explore our first working hypothesis, f-JSPE and f-TAS20 scales are used for empathy and alexithymia measurements. These scales have been described as reliable tools in different French-speaking samples.^44,45,53-55 However, a psychometric scale needs de novo validation when study conditions are changed.⁵² Our work confirms that f-JSPE and f-TAS20 correspond to reliable tools for empathy and alexithymia measurements in French medical residents.

The average score of alexithymia calculated for French medical residents is similar to those calculated for French asymptomatic adults,⁴⁵ and for German and Japan general populations.^56,57 There are contradictory results concerning a gender-related difference for alexithymia in the literature. Indeed, previous observations highlight some gender-related differences in German and Jordanian general populations.^56,58 However, other reports show that alexithymia is unaffected by the gender of French or Japan individuals.^45,59,60 Our study shows that the gender of French medical residents does not influence their alexithymia trait.

The average score of empathy calculated in the present work is in good agreement with those calculated for medical students⁶¹ and for French medical practitioners.^54,62 The fact that empathy level is affected by gender or not, is currently debated. On the one hand, a higher empathy is reported for females in samples of undergraduate students of Portugal and USA.^63,64 On the other hand, North American and Brazilian male residents have higher empathy levels than their female equivalents.^65,66 The present study does not report any gender-related empathy difference for French medical residents. This suggests that medical formation at Nantes University minimizes or abolishes the empathy difference between male and female medical students during their early training.

The resilience measured for French medical residents is higher than that determined for Chinese, Korean and US general populations.^67-69 By contrast, it is close to that measured for Australian nurses or Brazilian athletes daily living in stressful conditions^70,71 and for Chinese and Turkish eartquake survivors.^72,73 This indicates that medical formation provides a training environment susceptible to support the resilience of medical students.

The relationships between resilience, empathy, and alexithymia have never been fully described, because of a lack of concomitant measurements. It is shown in the present work that empathy and alexithymia correspond to significant predicting factors of resilience. Indeed, resilience is negatively influenced by alexithymia and positively affected by empathy which is also negatively influenced by alexithymia. A decline of medical students’ empathy occurs as students progress in their training.⁷⁴ Our findings suggest that such a decline can be accompanied by the decrease of the students’ resilience, thus rendering the students more vulnerable to PBO. Apart from training programs suggested for resilience enhancement,^75,76 education programs devoted to the reinforcement of students’ empathy could represent an efficient strategy aiming to support resilience. Also, helping students in the understanding of their own emotions could also figure a protection factor against PBO.

Cluster analysis has been used to describe heterogeneous distributions of empathy,⁶⁴ alexithymia^57,77 and resilience³⁷ among pathological and non-pathological populations. However, these results have been obtained with independent measurements of empathy, alexithymia or resilience. In the present study, empathy, resilience, and alexithymia are concomitantly measured, and cluster analysis considers the three variables simultaneously. Two clusters of residents are identified in very good conditions of confidence. The former is mainly composed of 3rd-year residents with low alexithymia trait and high empathy and resilience. The latter is predominantly constituted by 1st and 2nd-year residents with a high alexithymia propensity and low empathy and resilience. This suggests that the medical residency training increases the emotional understanding and the resilience of the residents.

Considering normative data,^61,69 the residents of both clusters exhibit non-pathological levels of empathy and resilience. However, a TAS20 score higher than 2.65 is reported to indicate a strong alexithymia propensity.⁵⁶ Consequently, our work suggests that the residents of the second cluster have a high alexithymic profile. Therefore, this work corroborates Shapiro’s comments concerning the promotion of students’ alexithymia during preclinical and clinical medical formation.⁷⁸ Predictive determinants of alexithymia during early stages of medical training shall be investigated in future studies.

Although significant datasets support our work, it presents several limitations. It corresponds to a monocentric study (Faculty of Medicine of Nantes). The results are based on a limited number of responses (49% of the population of residents). A selection bias may be present in the survey, as collected responses can originate from students with a positive a priori for this study. Furthermore, medical residents are potentially able to identify socially acceptable answers concerning the level of empathy associated with medical practice. Fisher and Katz describe this social desirability-bias on self-report assessment.⁷⁹ It may be responsible for some overestimated responses and the present survey. The last limitation concerns the design of the study which corresponds to a cross-sectional survey. This prevents the observation of time-related changes in empathy, resilience, and alexithymia. Consequently, further longitudinal investigations based on larger samples obtained from different medical schools are necessary.

The authors declare that they have no conflict of interest.