Overall, the mean age of our 103 participants at the time of the survey was 57.9 years (standard deviation = 12.2), 55% of participants were male, and 68% had received training in the U.S. (Table 1). With three exceptions, there were no statistically significant bivariate associations (p < 0.05) among journal type, publication year, gender, age at the time of publication, training in the U.S., training in epidemiology, training in biostatistics, recency of epidemiology training, or recency of biostatistics training. The exceptions were that: (1) participants were more likely to be female if their article was published in 2020 than in 2000 (odds ratio (OR) = 4.2, 95% confidence interval (CI) = 1.3–13.2); (2) individuals with formal epidemiological training were four years younger, on average, than individuals without formal epidemiological training (95% CI = 0-8 years younger); and (3) the mean age of individuals that had last received training in epidemiology or biostatistics within five years of publication was 11 years younger than individuals who received training at least five years before publication (95% CI for epidemiology = 7–15 years younger; 95% CI for biostatistics = 7–16 years younger). Additionally, while not statistically significant, we observed that the proportion of female corresponding authors increased over time (36% in 2000, 43% in 2010, and 70% in 2020).
Nearly 43% of the 103 participants (n = 44) had a clinical degree and 37% (n = 38) had a Doctor of Medicine (MD). Most participants had interdisciplinary training; for example, 73% of the 44 participants with clinical training also had formal epidemiological or biostatistical training and 64% of clinicians had formal training in both epidemiology and biostatistics. Although none of publication year, gender, or recency of training was significantly associated with interdisciplinary training (p > 0.05 for each comparison; Table S1), participants who published in general clinical journals and specialty clinical journals were more likely than participants who published in epidemiological journals to have interdisciplinary training (95% CI = 1.5-16.5 for general clinical journals; 95% CI = 0.99-14.3 for specialty clinical journals).
Interdisciplinary training seemed to provide options to clinicians for journal outlets; five of the 32 (16%) clinicians with formal training in epidemiology or biostatistics published their articles in epidemiological journals compared to only one of the 11 clinicians (9%) without this training. Additionally, journals tended to have corresponding authors with formal training in fields corresponding to the journal type. Of the 67 articles published in a clinical journal, 57% had a corresponding author with clinical training. Similarly, of the 36 articles published in an epidemiological journal, 81% had a corresponding author with formal epidemiological or biostatistical training. Nevertheless, of participants who reported that they did not have epidemiological training (n = 35) or biostatistical training (n = 20), only 11 out of 25 respondents (44%) and 3 out of 11 respondents (27%), respectively, indicated that formal training in these fields would have benefitted them.
Participants frequently reported working within interdisciplinary research teams (Table 2). Over 70% of our participants reported that at least one co-author (other than the respondent) had formal training in each of epidemiology (n = 74), biostatistics (n = 69), and clinical medicine (n = 76). Approximately 70% (n = 72) of participants reported that they had teams that collectively had formal training in each of the three areas (Figure 2; including the respondents’ training). Interdisciplinary team composition was not significantly associated with year of publication, type of journal, gender, or recency of training in either epidemiology or biostatistics (p > 0.05 for each bivariate comparison, Table S1).
Interdisciplinary teams typically involved co-authors at multiple stages of the research process. Of participants who reported the research stages to which co-authors contributed, over half indicated that co-authors with epidemiological (n = 57), biostatistical (n = 42), and clinical training (n = 40) were involved in all four research stages (design, analysis, interpretation, and reporting). The year of publication, type of journal, and gender of the participant did not significantly affect whether a co-author with epidemiological or biostatistical training was involved in any of the four research stages (p > 0.05 for each bivariate comparison). Participants who reported that they had completed their epidemiological training <5 years from the date of publication all reported that they had a co-author (other than themselves) with epidemiological training who contributed to the design and interpretation of the research. The recency of participants’ biostatistical training was not significantly associated with whether a co-author trained in biostatistics contributed to any of the research stages (p > 0.05 for each comparison).
Figures S1 and S2 show how journal type, publication year, and gender of the participant relate to participants’ confidence in their own and their co-authors’ ability to apply epidemiological and biostatistical concepts related to their article. In general, there were no significant bivariate associations among these variables; however, participants who published in 2020 were significantly more likely to say that they were extremely confident in their epidemiological abilities whereas participants who published in 2000 or 2010 were significantly more likely to say that they were very confident (p = 0.012 and p = 0.030, respectively).
All responses to the open-ended question (n = 38) are given in the Supplemental Digital Content (Table S2). Responses fell into three major themes. The first theme was more detailed explanation of the respondents’ training or of their co-authors’ training (e.g., “this paper was published while I was in the process of defending my PhD dissertation in Epidemiology”). Some responses in this theme mentioned specific experts and academic departments with whom the participants trained (often in lieu of formal training), some mentioned specific analytic techniques, and some addressed the sectors in which their co-authors had training.
The second theme addressed the challenges in defining and reporting on academic training. Respondents mentioned that the term “biostatistics” may not be inclusive of some statistical mathematical expertise, especially for individuals who trained outside of the U.S. or in earlier time periods. Similarly, some participants commented that training in the social sciences may be an adequate replacement for training in epidemiology, depending on the specific focus of the analysis (e.g., “my coauthors had training in health services research and pharmaceutical health outcomes research. I think training in these fields is also very applicable for applied epidemiology/policy analysis”).
The third theme involved reflective responses. Some participants reflected on the importance of interdisciplinary training within teams (e.g., “I don't think it is possible for any one individual to hold all the expertise needed for this sort of project”). Others reflected on how over-confidence in ones’ abilities can be problematic to the rigor of science (e.g., “Training is essential. Too many people assume they know how to interpret data and as a result, the quality of many papers suffer.”). One participant who gave a response in this category was particularly self-critical: “At the time, I was very young and overconfident. As it was twenty years ago and I have learned a lot since then, I would not be surprised to find errors in the paper, if I were to read it again today.” Finally, some responses within the third theme reflected on the administration of training programs: “All students at schools of public health should be required to take a course or seminar on the importance of skepticism in the interpretation of one's research.”