Cohort Profiles: what are they good for?

The IJE editorial team asked me to write about my views of the Cohort Profiles published in the journal, in light of a pilot study of the longevity and utility of the series.1 In the early phase of co-editing IJE, George Davey Smith and I looked at ways of improving communication, strengthening skills and learning from the history of epidemiology. Our first innovations were Point-Counterpoint discussions of controversial topics and Reprints and Reflections—re-publication of ‘IJE classics’ which soon ran out,2–4 and articles of significance that were not widely known due to being published in non-English languages or obscure journals or just being overlooked. Cohort Profiles were the first among later initiatives aimed at improving access to data and collaboration—Health & Demographic Surveillance System Profiles, Data Resource Profiles, Software Application Profiles—and were the most successful in terms of submissions. We started with cohorts as they are emblematic of epidemiology, due in part to the major causal discoveries made by investigators using cohort methodology and the clear findings and prodigious output of early cohort studies. Cohort studies have been set up for exploring a single hypothesis—for example, the British Regional Heart Study, the first Cohort Profile published in IJE,5 was set up to find out whether variation in drinking water quality (hard/soft) explained the regional distribution of cardiovascular disease6,7 but has made much greater contributions in other areas.8 Many of the most well-known cohort studies published profiles with IJE, although they were not all early adopters. The Framingham Study Cohort Profile was published in 2015 with companion papers, a classic article in our Reprints and Reflections section9 and an editorial documenting the study’s past, present and future.10 The Million Women Study, one of the largest cohorts globally, was only published in 2019.11 Other investigators published their cohort summary papers with other journals12 or before IJE started publishing Cohort Profiles.13,14 IJE’s requirement for an explicit statement on data access may have put some investigators off submitting Cohort Profiles. Data access is of increasing importance in epidemiology, both from the perspective of sharing data to maximize use and to allow replication and application of new analytical methods. This broad consensus among the scientific community and the major funders of epidemiological studies was challenged by the Nurses Health Study investigators.15 Providing the data and code as an annex to articles was suggested over a decade ago,16 and research of data-sharing practices, their evolution and their consequences for the field was proposed.17 The proposed IJE Cohort Profile survey might explore trends in data access arrangements. Our original aims were to: (i) provide readers with honest experiences of running cohort studies so others could learn from them; (ii) improve access to cohort data; (iii) increase collaboration among cohort investigators; and (iv) cover cohorts established from across the world.18 Furthermore, none of the other epidemiology journals were publishing Cohort Profiles. We were surprised by the number of submissions, having expected to publish one or two in each issue. By 2012, we had a backlog of 50 Cohort Profiles published online which had not yet appeared in print. It was clear that ‘something had to be done’. Our solution was to ask authors to provide a ‘Pocket Profile’ summarizing the key features of the cohort, which would appear in the print journal, with the full Cohort Profiles to be published online only. This made good sense, as the printed copy of the journal is distributed to libraries and to a minority of members of the International Epidemiology Association who want it. Most readers of IJE are not members of the International Epidemiology Association and get articles online and not from a printed version. There is a page limit for the printed journal, but such limits are irrelevant for online content, so making changes to the submission criteria for Cohort Profiles does not seem necessary. Since the Cohort Profile series was started in 2004, IJE has published just over 5000 articles, of which 449 (9%) are Cohort Profiles (Web of Science). They are twice as likely to be cited as other IJE articles and contribute nearly 20% of the overall citations to IJE articles, with an average of 62 (median 26, range 0 to 1165) citations each. The most highly cited paper published by IJE since 2004 is a Cohort Profile.19 It became apparent that some cohorts had developed a new focus or had collected new data and it would be worth recording reasons for these changes. Cohort Profile Updates were started in 2014, with 20 published to date. Given our success in publishing citable articles, it is not surprising to find that 30 other medical journals are now publishing cohort profiles. European Journal of Epidemiology was an early adopter, publishing its first cohort profile in 2006,20 but only a handful followed. BMJ Open heads the field, with 145 cohort profiles published since 2012, although some of these are cohorts of patients with disease sometimes enriched with healthy participants, rather than epidemiologically defined cohorts. Having other options for cohort investigators to publish their existence has to be a good thing. But citations do not tell us whether publishing cohort profiles improves access to data or collaboration. Our first objective, to learn lessons for setting up cohort studies, is difficult to evaluate. Cohort studies are complicated to establish and maintain, needing leadership and teamwork. However, the most important requirement is stable funding over many years to gain value. In the UK, there have been initiatives from funding bodies to support cohorts both in the UK and in low- and middle-income countries.21–24 Public health initiatives for COVID-19 in low- and middle-income countries might be used to build infrastructure for long-term cohort studies, initially focusing on consequences of the pandemic and then extended to health and social topics. It would be useful to catalogue the funding bodies of cohort studies globally as part of the IJE’s planned Cohort Profiles survey. Did we help to improve access to cohort data? Alongside the planned survey, a bibliographical analysis of citations to IJE Cohort Profiles could be conducted to determine how many citations relate to de novo hypotheses and analyses involving other investigators outside the original cohort investigators. For example, the Newcastle Thousand Families Study25 had 32 citations to 2020, of which five involved other investigators. However, it would require additional information to know whether the Cohort Profile publication in IJE had prompted data access, rather than the demands of funding agencies to make data widely available.26–29 Teasing out the ways in which de novo approaches to other research groups are made may be difficult to do in a survey, so interviews might be worthwhile. Collaboration between cohort investigators has increased over the past few decades, partly because of the need for replication of findings in areas such as diet and health,30 and in gaining adequate power for evaluating small effects of potential risk factors.31 The coronary heart disease—work stress hypothesis gained traction from findings in the Whitehall II study,32 but a recent collaboration of seven European cohorts found no strong evidence of association between job strain and mortality among all women (hazard ratio, 0.96; 95% confidence interval, 0.82–1.12) and in men (1.01; 0.86–1.19) without cardiometabolic disease.33 The opportunities afforded by the Human Genome Project have also been critical in accelerating collaborations initiated by geneticists, rather than epidemiologists, in obesity,34 diabetes35,36 and lung function,37 among many other areas. Publishing IJE Cohort Profiles cannot claim any influence in building these collaborations. The final aim was to ensure that cohorts from low- and middle-income countries were represented. Here we did quite well, with 113 profiles (25% of all cohort profiles, Web of Science) from low- and middle-income countries: Africa (25), China (29), India (13), other Asian countries (20), Latin America (22) and Central America (4). Maintaining Cohort Profiles in low- and middle-income countries is a priority, and exploring the productivity, utility in improving health, and future opportunities for data linkage for these cohorts would be useful and could form part of the planned IJE Cohort Profile survey. A previous survey of IJE Cohort Profiles was conducted on articles published in 2015–17, focusing on use of cohort data in the evaluation of public health and clinical interventions.38 Few Cohort Profiles presented information about their potential and use for evaluation of interventions, and it was proposed that Cohort Profiles should explicitly include such information. This seems a suggestion worth considering and incorporating into the planned IJE Cohort Profile survey. In summary, cohorts are among the major tools of epidemiology, and modern technology for data capture and linkage is providing more efficient means of running them. It is critical that cohorts in low- and middle-income countries are supported so that they can improve the volume and quality of data that are used to fuel the Global Burden of Disease study estimates of disease occurrence and risk factors using best available evidence.39,40IJE leads the field in publishing Cohort Profiles and it is hoped it continues to do so. My thanks to George Davey Smith for comments on an earlier draft of this article. Shah Ebrahim was Co- Editor in Chief of the International Journal of Epidemiology from 2000 to 2016. He has received grant funding for cohort studies. He directed the British Regional Heart Study from 1992 to 1998 and initiated and directed the British Women's Heart & Health Study from 2005 to 2012 and has supported other cohort studies in the UK and India.