A survey of the use and impact of International Journal of Epidemiology's Education Corner

A survey of the use and impact of International Journal of Epidemiology’s Education Corner Ellie Medcalf, Jonathan Y Huang , Onyebuchi A Arah , Michael O Harhay , Stephen R Leeder, and Katy JL Bell * School of Public Health, Faculty of Medicine and Health, University of Sydney, Sydney, NSW, Australia, Editorial Board, International Journal of Epidemiology, Sydney, Australia, Biostatistics and Human Development, Singapore Institute for Clinical Sciences, Agency for Science, Technology and Research, Singapore, Singapore, Centre for Quantitative Medicine, Duke-NUS Medical School, Singapore, Singapore, Department of Epidemiology, Fielding School of Public Health, University of California Los Angeles, Los Angeles, CA, USA, Department of Statistics, UCLA College, Los Angeles, CA, USA, Department of Public Health, Research Unit for Epidemiology, Aarhus University, Aarhus, Denmark, Department of Biostatistics, Epidemiology and Informatics, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, USA, Clinical Trials Methods and Outcomes Lab, Palliative and Advanced Illness Research Center, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA, USA and Editor-in-Chief, International Journal of Epidemiology, Sydney, Australia

from e-mail lists, does not include invitation via twitter; response rate approximately 11%). Of respondents, 56% (120) self-identified as male, 43% (92) as female and 1% (1) as other. Respondents were from 54 countries and their mean age was 46.4 years. Most had a PhD, DPhil, DSc or ScD in public health or epidemiology (64%) and worked in an academic role in teaching and/or research (82%) ( Table 1).
Of the 213 respondents, 43% (91) said that they had used IJE Education Corner articles in their teaching, research or practice: 24% (52) used Education Corner articles specifically in their teaching (45 in epidemiology courses), 33% (70) in their research and 10% (22) in their public health/clinical practice ( Table 2). We also found that Education Corner articles are used in teaching, research and practice across all regions of the world ( Table 2).
Many respondents had used at least two Education Corner articles in their teaching or research, with basic epidemiological study designs the most popular topic. For example, the top two articles selected by respondents for use in teaching and research were 'Classification of   Each of the following responses were provided once: Agriculture industry/ anthropology/qualitative health/ATLAS.ti, clinical research, contract researcher, epidemiology-related surveys, not-for-profit, PhD student, regulatory affairs, social epidemiology.
Eleven respondents answered 'Yes' to both epidemiology and biostatistics courses, two answered 'Yes' to both epidemiology and other courses. b Three respondents answered 'Yes' to using in all of beginners, intermediate and advanced courses; four answered 'Yes' to using in both beginners and intermediate courses; six answered 'Yes' to using in both intermediate and advanced courses. c Data missing for five respondents who used articles in their teaching and 14 respondents who used articles in their research. epidemiological study designs' 5 and 'Case-control studies: basic concepts'. 6 Other popular articles included introductions to epidemiological concepts and methods, such as 'Mediation analysis in epidemiology: methods, interpretation and bias' 7 and 'Competing risks in epidemiology: possibilities and pitfalls'. 8 These and other articles that survey respondents reported using are shown in Supplementary Table S1 (available as Supplementary data at IJE online).
Respondents who did not use IJE Education Corner articles in their teaching, research or practice provided suggestions for increasing their engagement with articles (Supplementary Table S2 Given this lack of familiarity with the Education Corner, respondents also suggested that there was a need for increased promotion and accessibility of Education Corner articles. Respondents also indicated that they would be more likely to use Education Corner articles if topics were relevant to their work. Some indicated that papers covering foundational epidemiology would be most useful to them, whereas others wanted papers covering more advanced epidemiological concepts, or specific topics more applicable to their area of work.
'I think the articles are very good, but I mostly teach undergraduate medical students, so the articles are too advanced.' 'Some more basic concepts would be good for foundation-level students.' 'Wider scope from basic to advanced.' 'Educational topics of comprehensive and state-of-theart subjects.' 'Case studies for teaching epidemiological research from diverse contexts, especially community settings.' These survey data have formed part of an evaluation of Education Corner by a team of IJE editors, and inform the expanded vision for the Education Corner, outlined in the accompanying Editorial. 9 In particular, we have changed the format in the hope of increasing accessibility and encouraging uptake of both basic and advanced methods. We will also be increasing outreach to under-represented voices at all career stages. We plan to continue to grow Education Corner for those learning about and using epidemiology wherever they are in the world, as a contribution to the health of populations.

Ethics approval
This negligible risk study was assessed by the University of Sydney Human Research Ethics Office and exempt from formal ethical review.

Data availability
Available on request.

Supplementary data
Supplementary data are available at IJE online.

Author contributions
E.M. did the data analysis, contributed to data interpretation and wrote the original draft of the manuscript. J.H., O.A., M.H. and S.L. contributed to conceptualization, recruitment and data interpretation, and reviewed and edited the manuscript. K.B. directed the study and was primary supervisor, contriubuted to conceptualization, recruitment and data interpretation, and reviewed and edited the manuscript. Katy Bell is guarantor.

Funding
There was no specific funding for the study.  1 Their argument-aided by directed acyclic graphs (DAGs)-relies on within-person models with a single treatment measure and causally related outcomes. Whereas this framing is conventional for change score analyses, 2 it may not apply to other settings that are commonly represented in panel data: that is, change score analyses with dyadic (two-person) data can yield valid treatment estimates and may be preferable to more general methods. I motivate this claim with a common study design: sibling comparison analysis.
Consider the causal DAG (A) in Figure 1, which is a typical two-sibling comparison model. 3,4 Subscripts i ¼ 1; . . . ; N and j ¼ 0; 1 indicate cluster and sibling, respectively. There are four variables: a treatment, X ij ; an outcome, Y ij ; a change score, DY i ¼ Y i1 À Y i0 ; and an unobserved fixed effect, U i , that confounds treatments and outcomes. Greek letters denote linear causal effects. Following standard fixed effects assumptions, the unobserved confounding effect w (U i ! Y ij ) and the treatment effect d (X ij ! Y ij ) are equivalent for both siblings, and Y ij does not cause Y ij 0 . 3,5 The absence of causally related outcomes in observational data may be justified substantively or evaluated with placebo tests. 6 Mallinson and Elwert (2021) used this model to estimate the effect of birth outcomes on academic performance within sibling clusters, 4 although it readily applies to other dyadic settings, such as studies on health interventions among spousal pairs 7 or friends. 8 Our estimand in DAG (A) is the treatment effect d. Regressing Y ij on X ij yields a biased treatment effect estimate, covariate adjustment notwithstanding, because Y ij and X ij are generically confounded by the unobserved fixed effect U i . However, change score regression, DY ¼b 0 þb 1 X i0 þb 2 X i1 , offsets equi-confounding bias that is transmitted via X ij U i ! Y ij and X ij U i ! Y ij 0 through outcome differencing. 2,4 Additionally, the confounding bias that is transmitted via X ij U i ! X ij 0 is blocked because the regression conditions on both treatments. The resulting partial regression coefficientb 2 ¼ d precisely estimates the treatment effect. Likewise,b 1 ¼ Àd.
Still, a standard conditional maximum likelihood estimator also precisely estimates the treatment effect in DAG (A). 3 We may thus question the value of change score estimation, but its utility is evident in the presence of treatment-to-outcome spillover. The causal DAG (B) in Figure 1 introduces a spillover effect h (X ij ! Y ij 0 ). A conditional maximum likelihood estimator cannot estimate the treatment effect d with this spillover. In contrast, change score regression,DY ¼b 0 þb 1 X i0 þb 2 X i1 , still precisely estimates the treatment effect (b 2 ¼ d), and it also