Eleonora Uphoff, Neil Small, Rosie McEachan and Kate Pickett
For some years, our research has been based in the city of Bradford in northern England. We are often asked to justify our research setting. There seems to be a concern that a cohort population that is not representative of the nation as a whole or of the ‘average person’ cannot produce valuable insights beyond its local setting.
While such concerns are not new, they now seem more present, perhaps due to the rise of Big Data or the increased sharing of and access to data from national surveys and cohorts. Do these reservations represent a push for representativeness and generalisability in epidemiology? If so, this might come at the expense of research aiming to paint a more detailed picture of population health.
It’s worth noting that findings may be generalisable without the sample being representative. The British Doctors Study provided convincing evidence of the risks of tobacco smoking with a sample highly unrepresentative of the general population.
Cohort studies rarely represent the wider population. The most disadvantaged groups, and those most in need in terms of health and wellbeing, are usually underrepresented. This is true for ethnic minorities as well as participants with a lower socioeconomic position. Whereas the dangers of overreliance on those people easiest to recruit are well documented for clinical trials, epidemiologists are not as quick to realise that a failure to recruit those groups labelled as ‘hard to reach’ has important implications for their work. We illustrate this with two examples.
The ethnic density hypothesis proposes that ethnic minorities fare better in terms of health when they live in areas with a high proportion of their own ethnic group. This may be due to area-based social networks and a buffer from discrimination. Ironically, most research on this topic is based on data from areas with very low levels of minority ethnic density. In some studies, a proportion of minority ethnic residents as low as 5% is classified as ‘high ethnic density’. Partly due to the lack of a relevant, rather than representative, population to test this theory, the evidence on this topic is still weak after decades of research.
The well-established, and often considered universal, ‘social gradients in health’ offer another example. While conventional measures of socioeconomic position, such as occupation, income and education, show an unequal distribution for many health outcomes in the general population, these measures do not always display the same social gradients in health for ethnic minorities (see figure below). Socially diverse datasets with a sufficiently large sample of ethnic minorities are required to pick up on such nuances.
To provide a more accurate picture of a diverse society, and to better understand and respond to inequalities in health, the study of diverse populations is vital. Some cohorts — such as the Australian ‘Watch Me Grow’ study, the UK Millennium Cohort Study and the Growing Up in New Zealand study — have made extra efforts to reach disadvantaged groups. Others have shown that reasons for not reaching some groups of the population extend beyond being ‘hard to reach’, including poor community engagement.
As researchers who rely on community participation, we make continuous efforts to engage with study participants and the wider community. This has allowed us to recruit a sample that is largely representative of the ethnic and deprivation area profile.
Demographics and representativeness of the Born in Bradford cohort
|Cohort births*||Non-cohort births*||England†|
|Mother’s age (years)|
|1 (most deprived)||68%||70%||27%|
|5 (least deprived)||1%||1%||15%|
* In Bradford Teaching Hospitals Trust during recruitment period, 2007–2011.
More importantly, knowing your community, and having your community know you, promotes long-lasting accountability. Although achieving this relationship takes considerable effort, it has brought immense benefits for the quality and relevance of our research.
An added bonus is that policy changes which seem very difficult to achieve on a large scale can be achieved locally through relationships with policy makers, and these changes may act as a catalyst for bigger change. After research from our team demonstrated links between air pollution and low birthweight in Bradford, the council acted to replace and update the most polluting buses in the city. The same findings have fed into European multi-cohort studies.
In the interest of our community and society’s most disadvantaged members, we prioritise generalisability over representativeness. Coming from a geographically specific yet demographically diverse cohort, we are confident that our findings benefit those far beyond Bradford’s boundaries.
Eleonora Uphoff is a Research Fellow employed by the University of York and works for the Better Start Bradford Innovation Hub. Neil Small is a Professor of Health Research at the University of Bradford. Rosie McEachan is the Programme Director for Born in Bradford and co-director of the Better Start Bradford Innovation Hub. Kate Pickett is a Professor of Epidemiology in the Department of Health Sciences at the University of York, and Co-Director of the Better Start Bradford Innovation Hub.
Follow us on Twitter: @NoortjeUphoff, @drrosiemc, @ProfKEPickett, @BiBresearch
2 thoughts on “The relevance of a local cohort: the case for generalisability over representativeness”
You have addressed an issue the INDEPTH Network of Health and Demographic Surveillance System (HDSS) field sites in low- and middle-income countries have been grappling with for decades.
We have described our dynamic longitudinal population cohorts ((Sankoh O, Byass P. (2012). The INDEPTH Network: filling vital gaps in global epidemiology. Int J Epidemiol 41(3):579-88.))
We have presented lessons from history in terms of representatives and the value of the cohorts. ((Byass P, Sankoh O, Tollman SM, Högberg U, Wall S. (2011). Lessons from history for designing and validating epidemiological surveillance in uncounted populations. PLoS One 6(8):e22897))
Recently, we looked at how generalisable are the estimates from our cohorts. ((Bocquier P, Sankoh O and Byass P (2017). Are health and demographic surveillance system estimates sufficiently generalisable? Global Health Action, 2017 Vol. 10, 1356621
We work closely with the communities.
We should continue to do our work and do it with scientific rigour.
Hi thanks ffor sharing this