‘One size does not fit all’: The value of person-centred analysis in health professions education research

Research question

How is empathy in HPE students associated with clinical performance?

How is empathy in HPE students associated with clinical performance?

Method and analysis that can be used

Collect scores on empathy using Jefferson’s Scale of Physician Empathy, collect clinical performance scores, run statistical analysis for association between the independent and dependent variables for the whole sample

Collect scores on empathy using Jefferson’s Scale of Physician Empathy, collect clinical performance scores, divide the sample into sub-groups of similar scoring students using a cluster analysis, run statistical analysis for association between subgroup membership as an independent variable and clinical performance as the dependent variable

Implications of findings for educational practice

Leads to general implications—If the association between empathy and clinical performance is positive, this can lead to an evidenced-based implication that training to foster empathy among students will help in better clinical performance (“One size fits all” approach)

Can lead to nuanced implications per subgroup—If the findings show that the groups with high and moderate empathy scores have good clinical performance and the low empathy score subgroup has poor performance scores, the implication would be that students with low, moderate and high empathy scores could receive different training programmes: modularization of education. (Personalized or “One size does not fit all” approach)

Possible further research

Provide empathy training to all students in the same manner and study the effects

Provide tailor-made empathy training to the students in each subgroup and study the effects of this personalized approach

Jacobs et al. 2014 [5]

Boscardin et al. 2012 [11]

Fokkema et al. 2014 [19]

Generated 5 profiles of teachers on the basis of their conceptions of learning and teaching, which had implications in the form of personalized faculty development activities

Generated 3 profiles based on students’ clinical performance which were used to customize remediation activities for improvement of performance

Generated 5 profiles on the basis of residents’ and physicians’ perceptions of workplace based assessments and made personalized recommendations for introduction of innovations in workplace based assessments

Kusurkar et al. 2013 [2]

Mak-van der Vossen et al. 2016 [12]

Dotters-Katz et al. 2016 [20]

Generated 4 profiles of students on the basis of the combination of their intrinsic and controlled motivation which had implications for academic performance, exhaustion from study and learning approaches

Generated 3 profiles of unprofessional behaviours of undergraduate students which gave an insight into what customized remediation measures could be used for each profile

Generated 3 medical graduate profiles on the basis of their attitude and motivation to teach in order to customize faculty development activities

Orsini et al. 2018 [6]

Lambe and Bristow 2011 [13]

Berkhout et al. 2017 [21]

Generated 4 profiles of students on the basis of the combination of their intrinsic and controlled motivation which had implications for study success and well-being

Generated 3 student profiles based on prior academic achievement and interview rating at the time of medical school admission to identify students who are most likely to need learning support

Generated 5 student profiles on self-regulation of their clinical learning in order to personalize support and supervision

Type of data

Can be used for continuous or categorical data

Can be used for continuous or categorical data

Can be used for a combination of rank-ordered statements and interview data

Required sample size

A good sample size is important for cluster stability. A thumb rule is a sample size of minimum 100

Medium size (at least 70 samples) as well as (very) large sample sizes can be handled, depending of the number of indicators in a given sample

Sample size can be relatively small (65 participants [18]) or larger (152 participants [21]). The quality of the sample is more important than the quantity of the sample. Researchers try to select a varied and diverse sample

Advantages

– Provides more generalizable findings owing to the nature of the data and the ability to handle large sample sizes

– Can lend itself to longitudinal follow-up of profiles to see if they change over time

– Flexibility of the model specification in LCA provides advantage over cluster analysis, which may not yield an optimal representation of the profiles [19]

– Can lend itself to longitudinal follow-up of profiles to see if they change over time

– Statistical and interpretational criteria are used to determine the optimum number of clusters, which means that researchers themselves can determine the number of classes and determine an understandable and practical to use ‘latent factor’ that describes the difference between classes

– This is a robust and systematic way of studying subjectivity. It can be precise and rigorous (depending on the choices made while conducting the analysis), and yet keeps the richness of descriptive data by including post Q‑sort questions or interviews

Disadvantages

– Because of its exploratory nature, it may be random and not generate similar profiles in different samples

– Cannot be used for small sample sizes

– Limited generalizability

– Sample size should not be too small, especially if there is a large number of indicators. This problem might be solved by using a penalty parameter while statistically estimating the latent class model

– Limited generalizability. Q‑studies are not designed for generalizability purposes, they are for uncovering authentic viewpoints within the sample (and sample should be high quality). From there, if desired, prevalence of viewpoints can be tested in larger population through other methods

– Difficult to study change in profiles over time