A systematic review identifying effective teaching methods and their combinations for increasing empathy in physicians: pairwise and network meta-analysis | BMC Medical Education

Previous empathy interventions for healthcare professionals have been shown to work with a moderately large effect size [16,17,18] and this was confirmed in this study. However, empathy interventions for physicians are often designed without knowing the most effective and efficient “ingredients” – what works, and what works well together?

Which teaching methods are effective for teaching empathy to physicians?

The PMA found that including the didactic teaching method was associated with greater intervention effectiveness compared to control. This finding diverges from a previous meta-analysis on medical students, which found no one teaching method was better than another [18]. One reason may be that our study encompassed a greater number of studies (111 versus 18 studies), based on our inclusion of practicing physicians in addition to medical students. Pedagogically, didactic activities may be especially crucial for supporting skill acquisition in that it provides the theoretical structure to organize and assimilate newly acquired knowledge [48].

The reason the other teaching methods were not significant on their own may reflect the concept of intentional alignment from pedagogical research [49]. This is the notion that the teaching goal (i.e., desired outcome of the intervention) should align with the criteria of success (i.e., outcome measure) and with the learning activities (i.e., teaching methods). That is, didactic approaches may be particularly effective for targeting the aspect of empathy that is assessed on standardized measures used in research.

Moreover, it is possible that some teaching methods may be effective in some teaching contexts but not in others. For example, previous meta-analyses show considerable variability of feedback: some types of feedback can be powerful, whereas feedback that lacks specificity, timing alignment, and clarity is ineffective [50]. In fact, a seminal meta-analysis found approximately one-third of feedback interventions decreased performance [51]. From a practice perspective, this further highlights the importance of detailed reporting in studies. By providing in-depth description of the training itself, studies can be grouped more effectively; thus, with heterogeneity is reduced, meta-analytic conclusions become more robust.

What intervention characteristics are important for empathy training?

In terms of number of teaching approaches, the PMA revealed that interventions using two teaching methods were more effective in increasing empathy than those with three. Although, it should be noted that there were only a small number of studies representing these combinations as (see relative line widths in Fig. 1; and Supplemental Appendix 12 for frequency of intervention combinations), and wide confidence intervals (Fig. 3), therefore replication is needed to confirm this finding. These results are aligned with other meta-analyses on empathy interventions for mental health professionals and parents [32, 52], which shows that focused interventions can be very impactful for teaching empathy skills. However, these results are somewhat surprising in light of previous literature for medical training, which shows the effectiveness of simulation-based education of using at least three teaching methods [53] (i.e., didactic, rehearsal, and feedback). Instead, as we describe in the NMA section below, it seems that certain combinations of teaching methods can be especially effective when paired together.

The PMA also showed which intervention characteristics may be more or less important for teaching empathy: format was statistically significant but not the number of sessions or facilitator type. In-person group formats were more effective than online and/or independent formats, which are in line with research showing that peer collaboration may be particularly beneficial for acquiring an interpersonal skill such as empathy [54] as group settings may be uniquely conducive for acquiring social and self knowledge [55]. The non-finding of session number is in line with previous studies which did not find a significant effect of duration as a moderator of intervention effect [18, 24, 25]. Again, underreporting intervention details makes it hard to study the impact of specific intervention characteristics. For example, while the duration of the empathy interventions overall was not associated with effectiveness in our study, duration of specific teaching methods could be, though this is rarely reported by studies.

The type of facilitator did not impact intervention effectiveness, suggesting that empathy skills need not be taught only by professionals, as those with comparatively fewer credentials can be as effective educators for empathy skills [56]. Additionally, learning outcomes may be boosted with a focus on learner experience, such as motivation, engagement, and satisfaction [57, 58]. In pedagogical research, there is strong meta-analytic evidence showing that positive teacher behaviours (e.g., warmth, non-judgement, and genuineness) are significantly associated with improved learning outcomes [59]. In medical education, recent systematic reviews also highlight how clinical supervisors are a key influence of student empathy development, where exposure to emotionally distant or “efficiency driven” role models may contribute to the overall decline of empathy during medical school [60, 61]. Student self-report (although not a blinded outcome in trial design) may serve an important function for understanding educational outcomes [49].

Which study characteristics impact the effectiveness of empathy training?

Study characteristics which were associated with intervention effectiveness included measurement type and control group type. In line with previous meta-analyses [25], larger effects were seen on objective measures as compared to patient-report and self-reports of empathy. It may be that self-reports have less reliable psychometric properties and more prone to bias, which could negatively impact its measure of empathy [62]. Moreover, it is possible that interventions demonstrated stronger effects on objective measures of empathy because these assessments closely mirror the structure and content of the training itself. This may reflect a form of “teaching to the test” [63], where participants are prepared to perform well on behaviors they know will be evaluated. As such, objective measures may potentially overestimate the real-world impact of the intervention compared to patient-reported outcomes, which are more sensitive to the nuanced and relational aspects of empathy [64, 65].

We also found a greater effect size when comparing studies which used a pure control group over those which used education-as-usual. While this was not found in previous meta-analyses on empathy interventions [18, 25], this inconsistency may be in part due to a lack of precision when it comes to defining control groups; this is especially the case when it comes to behavioural interventions [66]. Overall, effect sizes for moderator analyses were small in magnitude, a finding that is common in PMAs and indicative of high heterogeneity between studies. This may reflect the diversity of behavioural interventions that exist to target a skill as complex as empathy; it will be important for future studies to further explore this heterogeneity.

What combinations of teaching methods are effective?

NMA rankings revealed the most effective empathy intervention was comprised of didactic and rehearsal methods, followed by didactic and reflection. The third, fourth and fifth ranks were 3) didactic, rehearsal, reflection and observation; 4) didactic, rehearsal, observation, and feedback; and 5) all five teaching methods together. The most common teaching methods in these combinations included didactic (5/5) followed by rehearsal (4/5), reflection and observation (3/5). These findings align with results from previous meta-analyses that suggest a mix of teaching methods is effective for teaching empathy, the importance of didactic teaching, and a trend towards the benefit of rehearsal methods [18, 23].

These results also show that certain combinations of teaching methods are more effective than combining all methods together (i.e., DHROF). Consistent with the PMA, the two top combinations of the NMA involve only two teaching methods. Notably, while the highest performing interventions comprised only two teaching methods, the effects of interventions with more methods were more robust (i.e., yielded smaller confidence intervals). By examining the SUCRA rankings, it is possible to see that the combinations of teaching methods that show statistically significant effect sizes (see Fig. 3) include at least one “surface learning” component (mastery of content) and at least one “deep learning” component (connections between, and extension of, the content) [21], which generally achieved through a component that is experiential (see Table 1). Specifically, the combination of didactic and “experiential” (rehearsal and reflection) teaching methods were the most effective intervention combinations.

Didactic teaching may be necessary to give physicians the cognitive framework upon which they can build the concepts as they learn; however, this teaching method may not be the most effective alone. Instead, didactic may work especially well when combined with rehearsal or reflection activities; this provides for the opportunity for application and consolidation of empathy skills, which is key to learning [21, 67]. Acquiring a high level of competency of procedural skills in medicine requires a similar pedagogical approach, which first requires acquiring cognitive knowledge of a skill, following adequate exposure and practice, before being able to expertly performing the skill [68].

Furthermore, these NMA results also highlight practice gaps between what is commonly done (i.e., frequency of intervention combinations) compared to what is most impactful (i.e., using effective intervention combinations). For example, feedback is often identified as an active ingredient in medical education literature [53, 69], so educators may feel that it is an essential component for training all types of patient physician communication behaviors; that may explain why four out of five of the most frequent interventions contain feedback. What our study suggests, however, is that it may not be as “essential” as presumed; in fact, three of the five significant intervention combinations do not include feedback (i.e., DH, DR, DHRO). It is possible that what is effective may vary for the type of communication behavior being taught. Notably, a teaching method like feedback is an expensive component for teaching empathy, and if it is possible to achieve the same empathy outcome without it, this may be more cost-effective. Therefore, our study shows that NMA can be an insightful tool for identifying effective interventions, thus providing some direction of what key ingredients program developers and educators should focus on, rather than solely relying on what is common practice.

Limitations & future directions

Results should be interpreted with the following limitations in mind. First, publication bias was detected, which is not uncommon in PMA and NMA [70], and could inflate the magnitude of the observed findings [71]. Nonetheless, this speaks to the need for future work to emphasize pre-registration and publication of null findings. There is especially need for additional research on how best to address publication bias, especially in large NMAs [72]. Second, heterogeneity is a known problem in NMAs involving behavioral and educational interventions [43, 73, 74]. This could be addressed by standardizing interventions, methods and reporting [75, 76]. Following the CONSORT-SPI tool [77] and providing better documentation in original studies (including details of sampling, intervention content and delivery method) are crucial to improving aggregations of evidence.

Third, when studies reported multiple types of empathy measures, we prioritized objective measures over patient-reported or self-report outcomes because objective measures allowed for greater cross-study comparability and are less susceptible to expectancy effects [78]. We recognize, however, patient-perceived empathy is a critical dimension, given its association to patient outcomes [79]. Moreover, excluding patient-reported measures may limit insights into how these interventions impact patients’ actual experiences. For example, one study has demonstrated that the impact of physician empathy on patients’ emotional quality of life depends upon the type of consultation and the patient’s emotional skills [65]. Patient-perceived empathy can provide deeper understanding of the impact of physician empathy. Therefore, it will be crucial next step to assess the ecological validity of empathy interventions with real patients as well.

Fourth, a great majority of the studies reviewed were from either two regions, Europe and Central Asia or North America (80%, 87/109), which may not capture the diverse ways in which empathy training is approached. It would be ideal for future work to investigate how different cultural and linguistic contexts may change the teaching and measurement of empathy [80]. Fifth, the literature search for this review was conducted approximately three years prior to manuscript submission (October 2022) and therefore does not include the most recent studies on empathy interventions. Updating the search to incorporate newer evidence would be a valuable direction. Lastly, while the current study did not find an impact of intervention duration, most studies did not examine long term follow-up to see how long intervention effects last. Future studies should focus on booster sessions to determine whether this could mitigate empathy decline [68, 81].