In for the long haulBMJ 2011; 342 doi: http://dx.doi.org/10.1136/bmj.d942 (Published 15 February 2011) Cite this as: BMJ 2011;342:d942
Of all the methods used for research, the long term cohort study is the most seductive. You identify a group of people who share a date or place of birth or an experience of some kind, then study them over a period of time. Simple—at least in principle. In practice, because the best known cohort studies have also been very large, the logistical effort required to keep the show on the road is impressive in itself. So impressive that you can almost find yourself viewing any useful insights that emerge more as a bonus than the point of the exercise.
Next month sees the 65th birthday of the granddaddy of all cohort studies, the National Survey of Health and Development.1 Set up by James Douglas less than a year after the end of the second world war, it began with interviews of more than 13 000 mothers who had given birth in the United Kingdom during one week of March 1946. Concern over the low birth weight of babies born to less well-off mothers prompted a follow-up survey of more than 5000 of the original offspring. The project just kept on going. When the latest assessment began a few years ago its organisers were still in touch with around 3000 of the cohort.
The success of the project has inspired comparable work in several other countries from Finland to New Zealand, and also further cohort studies in the UK. These include the 1958 National Child Development Study, the 1970 British Cohort Study, and the Millennium Cohort started in 2000.2 But although birth studies of this kind are the most publicised use of the cohort approach, it can be applied to any large group being investigated for all manner of reasons. The European Prospective Investigation of Cancer and Nutrition (EPIC), for example, is studying 500 000 people in 10 European countries to investigate whether cancer is related to diet.3
And further studies keep emerging. The French government has recently funded a study led by epidemiologist Tobias Kurth, a director of research at Inserm (the country’s health and medical research institute) and BMJ consulting clinical epidemiology editor. It will follow 30 000 students from the universities of Bordeaux and Versailles for at least 10 years. “We’ll look at disorders that are most frequent in this age group, especially mental health disorders such as depression,” says Dr Kurth. “We’ll also look at migraine, which often starts between 20 and 30. We’ll look for risk factors which might explain the disease onset. And since we’re planning a long follow-up, we’ll also look for risk factors for diseases that develop later in life.”
Strength of time
The potential of these studies as research tools is clear. Although randomised controlled trials are usually regarded as the best method for tackling research questions, there are circumstances in which they are impossible or simply unethical. Testing the effects of asbestos or tobacco smoke on health are obvious examples. Moreover, following people for many years, or even a lifetime, makes it possible to explore their development, health, or ageing in relation to changes in their personal circumstances or the wider economic and social environment. But that said, do the findings really justify the considerable time and resources that cohort studies absorb?
That many of the biggest are so familiar is an indication of their impact. Think, for example, of Richard Doll’s study of smoking in 35 000 British doctors.4 From 1951 it tracked their mortality for 50 years and showed the increased risk of vascular and respiratory diseases and cancer associated with cigarettes. Think of the Whitehall studies of British civil servants5 and their telling illumination of the differing prevalence of ischaemic heart disease at different levels in the hierarchy. And think of the “natural” experiments at Hiroshima and Nagasaki that generated two cohorts exposed to intense bursts of radiation.6
Two notable US cohort studies
Framingham Heart Study
The study began in 1948 when researchers recruited more than 5000 people from the small Massachusetts town of Framingham (www.framinghamheartstudy.org/index.html). Every two years they have a physical examination and laboratory tests. A second generation was recruited in 1971 from among the original participants’ adult children and their spouses. And more recently some of the grandchildren of the original cohort were enrolled. The issues that the Framingham data have clarified or brought to light include the influence of smoking, cholesterol, blood pressure, physical activity, and obesity on heart disease and stroke.
Nurses’ Health Study
Established in 1976 to investigate the potential long term consequences of oral contraceptives, the study’s 122 000 nurses answered questions about smoking, hormone use, and diet (www.channing.harvard.edu/nhs/). A second cohort was added in 1989 and a third in 2008. These studies have explored the influence of smoking, oral contraceptives, alcohol, obesity, and physical activity on heart disease, breast and colon cancer, hip fracture, and cognitive function.
More specifically, the National Survey of Health and Development claims to have informed all manner of official reports including, in the health field, the 1998 Acheson report on inequalities in health,7 and the more recent review of the topic by Michael Marmot.8 Diana Kuh, director of the national survey, believes that it has been particularly influential in reinforcing the view that what happens in your early years affects your adult life. “In the middle period of the study there was a raft of papers showing associations that seem to be important. This really influenced popular thinking that investing in children is important for later life.”
The survey has now reached a point at which its cohort will soon start to become elderly. “We’re a study that can tell the government about the likely impact of ageing on health and social services,” says Professor Kuh. The data will reveal the extent to which a poor start in life is still affecting the health of 60 year olds. In due course it will also find out whether the influence persists into peoples’ 70s—or if by then it’s faded or been swamped by other factors.
The Millennium Cohort Study, having been going only a decade, isn’t yet in the same league of proved achievement. But it has already carried out four surveys of its 19 000 children. Published analyses include the factors that predispose to the emergence of obesity (children who don’t have breakfast are more likely to be overweight) and the influence of breast feeding on health, says its director, Lucinda Platt. “It can be difficult to point to any particular finding and say that this changed that policy. But the policy community is very much aware of what comes out of these surveys.”
Although cohort studies provide data that may, in some cases, be difficult to obtain by experimental methods, they do have their limitations. In its 2007 report on observational methods in medical research,9 the Academy of Medical Sciences offered the example of folic acid intake during early pregnancy to prevent neural tube defects. In 1989 a cohort study of 23 000 births pointed to a large and specific benefit of the maternal use of folic acid supplements during the first six weeks of pregnancy. The prevalence of neural tube defects was four times higher in children born to women who had not taken folic acid or who had taken it later in pregnancy than among those born to women who had taken it in the first six weeks. But while the study showed a clear association, it couldn’t prove causation. As the academy report points out, it could have been that “high social class women at low risk were more likely to take vitamins.” It took a randomised controlled trial to establish the causal connection.10
Advantages and drawbacks of cohort studies
An Academy of Medical Sciences working party listed some of the pros and cons of cohort studies.9 They included:
The sequence and timing of associations are readily determined
There is no need to rely on long term retrospective recall
They provide a ready estimation of the size of an effect
There is a good opportunity to examine a wide range of both expected and unexpected outcomes
Very large samples are required if the disease outcomes to be examined are uncommon
A long time frame is needed to study most associations with disease
Nor do all new cohort studies receive an unqualified welcome. The 500 000 strong UK Biobank (www.ukbiobank.ac.uk) has been set up to investigate peoples’ illnesses in relation to their genes and their environment. But critics have accused its organisers of lacking clear objectives and of collecting social information that won’t be as robust as the molecular data.
In spite of occasional criticism, cohort studies clearly have a future. While some can provide virtually definitive evidence (no point, for example, in rerunning the doctors and smoking study), the conclusions of others will always be provisional. The birth cohorts, for example. “We know from previous cohorts how early life experiences have influenced adults who are now in their 50s,” says Professor Platt. “But the world’s changing, gender attitudes are changing, the education system and the economic climate have changed.” You don’t have to be a sociologist to know that the life of a child born in the years after the second world war is vastly different from that of a child born this century. So enthusiasts for birth cohort studies can make a powerful case that their projects have to be reinvented and repeated again and again. Judgments of what matters evolve, as do the issues themselves. You can’t look to a study that began even as recently as 20 years ago to find out how hours spent peering at a computer screen instead of kicking a football might be affecting physical and mental health.
Britain has accumulated a wealth of longitudinal data. A project called HALCyon (Healthy Ageing across the Life Course, www.halcyon.ac.uk) is now linking the National Survey of Health and Development with eight other UK cohort studies. The aim is to understand more about how ageing affects physical and cognitive abilities and psychological and social wellbeing, as well as the biology of ageing itself. Four of these studies use information collected during the 1920s and 1930s: a remarkable re-exploitation of data that might otherwise be seen as having only historical interest.
One other thing that is clear about large cohort studies is that you don’t embark lightly on them. Reflecting on his forthcoming study of university students, Dr Kurth commented, “Getting people of this age group and following them up is quite challenging.” That’s putting it mildly. And there’s the matter of funding. If your study is really ambitious you can find yourself moving into the realms of “big science.” Biobank, for example, is costing around £60m (€70m; $97m). And if you aim to keep going for a seriously long time, you must learn the arts of adaptability. Having started in the 1940s the National Survey of Health and Development has witnessed every advance in data storage and handling from index cards through to desktop computing. But it’s kept going.
Cite this as: BMJ 2011;342:d942
Competing interests: The author has completed the unified competing interest form at www.icmje.org/coi_disclosure.pdf (available on request from him) and declares no support from any organisation for the submitted work; no financial relationships with any organisation that might have an interest in the submitted work in the previous three years; and no other relationships or activities that could appear to have influenced the submitted work.
Provenance and peer review: Commissioned; not externally peer reviewed.