|Home | About | Journals | Submit | Contact Us | Français|
As most countries face the ageing of their population, understanding successful and pathologic ageing is a research priority. Longitudinal studies examining the ageing process from middle-age are required to establish causal and valid relationships. This systematic review of the literature aimed at identifying large community-based longitudinal studies either including exclusively elderly people or following people from middle-age (50+ years at enrolment) to death, and resulted in a selection of 72 cohort studies. Design features of selected studies show that most were conducted in North America or Northern Europe, most included both genders, and follow-up period was often less than 10 years. Many cohorts focused on cardiovascular health, cognitive decline or osteoporosis. Usually collected variables comprise of self-reported data on socio-demographics, chronic diseases and functional status, as well as measures of cognition, anthropometrics and physical performances. Biological samples were taken in about 60% of the studies, and a third also undertook genetic analyses. This review summarises information on design and content of large population-based cohorts of older persons, and represents a valuable background from which additional data may be retrieved.
Parallel to the ageing of populations, the number of dependent older persons has increased considerably, as well as the need for costly long-term care services. In this context, geriatric research has been interested in the determinants of health in later life, and, more recently, the concept of frailty has emerged as a key issue (Rockwood et al. 1994). There is still much debate around the nature and measurement of frailty (Hogan et al. 2003), and further results from population-based longitudinal studies are needed to investigate the determinants of successful ageing and frailty (Fried et al. 2001; Peel et al. 2005). Numerous cohort projects have already been undertaken, or are still ongoing, focusing on the different populations and different health domains. An insight into the existing studies might be useful for the researcher before undertaking a new project and for the clinician interested in a particular health domain. Overviews of the design and content of the previous projects are not so easily available, even though most large cohort studies did publish articles on methodology or have a website providing this type of information. Summary information allowing to compare similar cohort projects is particularly difficult to find. This review aimed at identifying large population-based cohort studies including either middle-aged adults (aged 50 and over) or focusing on older persons. A literature search did not retrieve any comparable work: previous reviews focused on the disease-related morbidity (Pryer et al. 1995; Bosworth and Siegler 2002), or on a specific theme such as functional decline (Stuck et al. 2002). A similar, yet less systematic work, was undertaken in Canada. Its selection of cohorts was slightly different and results may be considered as complementary (Health Canada, Review of longitudinal studies on ageing, available on the Internet).
Our review provides a summary of design characteristics and domains of investigation of selected studies examining age-related health events in older adults and represents a useful source of references for research on health in ageing.
Inclusion criteria for selecting cohort studies were a lower age limit at enrolment set at 50 years or more and a minimal sample size of 500 subjects at baseline.
By setting this arbitrary lower sample size limit, we intended to select cohort studies large enough to provide sufficient statistical power given attrition over time. Then, to select cohorts representative of the general population, the search was limited to the population-based studies.
We first identified keywords or MesH terms used in several publications related to cohort studies of elderly people. The search strategy combined the terms “ageing” or “aged” with terms related to the study design: “cohort study or studies” or “longitudinal study or studies” and “population-based”. The research was then limited to middle-aged (45–65 years) or aged subjects (65 years or older) and to publications in English, French or German. Medline (Ovid 1966–2002), PsycInfo (1967–2002) and SocioFile (1974–2002) were searched on 31.10.2002, and 1,500, 24 and 1 publications, respectively, were retrieved. The same search strategy was rerun periodically from November 2002 to July 2004 with an extension to the publications of the years 2003 and 2004.
The Cochrane database was searched from October 2002 to January 2003, as well as in November 2003, and in July 2004 and did not yield any further references.
During October to November 2002, a search was conducted on the Web, using the words “cohort” or “longitudinal study”, combined with “elderly” and “health” and identified known or additional cohorts by their Internet site. Internet sites of geriatrics societies were consulted, as well as those of centres and institutes of ageing. Finally, bibliographies of relevant articles or handbooks were hand-searched.
Retrospective or historical cohorts were not considered for inclusion. Studies not focusing on health (e.g. psychological studies) and those focusing on a narrow clinical theme were excluded, as well as studies in which data were collected from administrative or medical databases only, without contacting the participants either face-to-face or by questionnaire. Finally, studies among twins, among very specific ethnics, among people who had been exposed to nuclear irradiation, and studies selecting subjects with a specific disease at inclusion were not selected.
First, titles and abstracts of all 1,525 publications were screened by the first author, using the exclusion criteria. Abstracts of the 732 remaining publications were reviewed independently by the two authors. In case of disagreement, more information was searched on the study, and inclusion of the study was discussed.
To gather maximum information on each cohort, abstracts of publications related to the same cohort study were then searched in Medline using the entire name or acronym of the cohort as keyword. Whenever this strategy did not identify any publication, or only a few, we retrieved the publications of the principal investigators and identified articles related to the cohort study by reading the abstracts.
The following data were extracted from abstracts and articles: name/acronym of the study, number of subjects, age at inclusion, gender of participants, year of beginning and duration of follow-up, setting (country/town), domains assessed, number of publications in Medline and availability on the Internet. For each cohort, abstracted data were sent to one of the original investigators for review. We also requested information on the data we could not find in the publications. Finally, the investigators were asked for references of publications describing the study methodology and its main findings.
As our aim was not to evaluate the validity of the outcomes but to examine populations and themes under study, we did not assess the methodological quality of the selected studies.
Review of the 732 abstracts by two authors resulted in the selection of 67 cohorts. Then, reading the articles led to an exclusion of 28 studies because age or number of subjects at baseline did not match our inclusion criteria. Nineteen cohorts were retrieved from the Internet search, among which 7 were excluded and 28 were retrieved from other sources, among which 7 were excluded, leaving a sample of 72 cohort studies corresponding to our criteria (Fig. 1).
Table 1 shows the selected studies by world region in alphabetical order and summaries their design: gender, age and number of subjects at inclusion, country, year of baseline assessment, duration of follow-up and major themes under study are displayed. It also indicates how many publications were retrieved using the name or acronym of the cohort as keyword in Medline 1966–2003 and whether the data abstracted from publications were validated by the investigators of the study (about 30% of original investigators did not respond to that request, and data for those studies are displayed as found in publications).
Finally, one reference per cohort study was chosen upon the indication of the investigators. Three studies had no indexed publications retrieved in our review process, so that their website is cited as a source of information.
Longitudinal studies of older people have been undertaken in all continents, including South America and Asia. In Europe, half of the selected cohorts originated from the United Kingdom, the Netherlands, Sweden and Germany. Most cohorts were strictly community-based, only seven projects also included participants living in institutions and one was conducted in convents.
Many studies include men and women, except for certain themes, such as osteoporosis, breast cancer and prostate cancer. Longitudinal studies focusing on cardiovascular diseases initially enrolled mainly men, and similar studies were later undertaken in women.
Age at inclusion varies with the disease under study; studies on cardiovascular events tend to recruit people in their 50s, while most studies on dementia recruit people aged 70 years or older. Studies including only younger elderly people were quite rare, as those recruiting persons older than 80 years at baseline.
If we look at cohorts focusing on specific health problems, we observe that cardiovascular disease and dementia have been the objects of many studies. On the other hand, there are only five longitudinal studies focusing on cancer. Our domain of interest, age-related fragilisation, has been examined in 28 studies, among which 23 were interested in the functional decline, while 5 focused on successful ageing.
Duration of follow-up is highly variable, from 2 years to more than 30 years. However, the majority of selected studies followed their participants for less than 10 years.
Table 2 shows a summary of variables collected in each study: although a high proportion of data are self-reported, most studies also collected data derived from observation (measure of physical or mental performances, clinical assessment), at least in a sub-sample of participants. Besides socio-demographic data, current health status, chronic conditions and medical history, a majority of studies did assess mental health using standardised tests such as Folstein’s mini-mental evaluation, Geriatric Depression Scale or Centre for Epidemiological Studies–Depression Scale. Evaluation of cognitive impairment was more frequent if the study included people aged 60 years or more at baseline. Other risk factors for functional decline such as visual and auditive impairments, incontinence, and falls were not routinely collected. These characteristics were more frequently assessed in studies of successful ageing, while neglected in studies focusing on a specific disease. A few studies described health care services utilisation.
A third of the cohorts undertook DNA sampling and analyses. Contrary to our hypothesis, DNA sampling was not undertaken more frequently in the recent studies. In fact, some longitudinal projects, that began before genetic testing was easily available, did collect and store blood samples and are now undertaking these analyses. Other studies collected genetic material during a recent follow-up wave.
While comparing data collected and publication themes, we observed that many cohorts tend to underuse information (data not shown): most papers report results on cardiovascular diseases and dementia, while data on functional status, physical performances or health care utilisation tend to be less frequently reported.
Even though we focused on large cohort studies of older persons, the number of studies retrieved was large and populations studied were heterogeneous. The number of publications related to each study was also highly variable and seemed independent of design features. In particular, larger projects did not always lead to a higher number of scientific publications.
It is interesting to observe that some investigators have taken the unique opportunity of longitudinal design in order to collect additional data during follow-up, such as new assessment tools or new technologies like DNA sampling that were often not part of the baseline study protocol.
On the other hand, most studies do collect a great deal of information that is underused in further analyses and publications. Ideally, reasons for assessment of a variable and further use in analytic planning should be decided before the study begins and have valid justification. However, data collection is also influenced by the legitimate concern of collecting data that might be useful later, according to scientific developments and also to compete with other studies (Deeg and Van der Zanden 1991). The gap between variables collected and those used in published analyses might also result from the inability to answer research questions due to insufficient statistical power. Most studies did not recruit participants in a homogeneous age category, but only set a lower age limit at baseline, despite the fact that the health picture is very different at the age of 60 and 80. Studying age-related events might be difficult if the sample size of each age category is small, in particular when attrition over time is taken into consideration. Following a large and homogeneous sample during many years seems necessary to come to valid results.
Finally, there may be insufficient resources available for data analysis, which requests a high level of scientific competence. Underuse of available databases is a very frequent problem in medical studies and more attention should be given to solutions that may overcome this. In particular, allocating time and resources for data analysis and paper redaction is a necessity that might be underestimated by funding sources. Part of the solution is in collaboration and sharing of data among researchers, respecting the huge investment consented by researchers to collect cohort data and to find financial resources.
This work also illustrates the difficulty of retrieving accurate and comprehensive information on this type of study. Depending on the cohort, our initial search strategy retrieved only 30–60% of publications identified by the name or acronym of the cohort on Medline 1966–2003, thus indicating low sensitivity of that initial strategy. We urge researchers to choose a name or acronym at the beginning of the study, to mention it in each related publication, and to create a website, in order to facilitate access to the information.
Then, we had to contact the principal investigators to get accurate and comprehensive results, because information on study design is not always available in published material. For instance, the number of people included at baseline tends to vary from one publication to another, particularly when some assessments were undertaken in a sub-sample only. We therefore recommend that each publication contains a brief but accurate description of the original study design, including more details on participation rates and characteristics of non-participants, or refers to publications describing study design. It would also enable the reader to estimate the representativeness of the population sample under study.
This work of course has some limitations. First, we limited our search to cohorts of subjects that were middle-aged or aged at enrolment. As cohorts recruiting subjects under 50 years were less likely to include a follow-up long enough to observe problems specific to ageing subjects, we decided to exclude such studies from our search. We are, however, conscious that ageing is a continuum, and that any lower limit of age is arbitrary and therefore limits the extent of the results.
Secondly, we excluded studies in developing countries, although ageing of the population will soon be a prevalent problem in these countries also. However, lifestyle, socio-economic circumstances and health care systems are very different in these countries and we believe that many results of these studies would not be applicable to our developed setting and should be studied separately.
Finally, despite our systematic search combined with manual searching, it is of course possible that we overlooked some important projects in the field. Our search on the Internet retrieved other lists of cohort studies, such as the ones from Health Canada (2004), review of longitudinal studies on ageing and from the National Institute on Ageing (2005). When compared to our results, we found that several studies retrieved in our review were not included in these works. Furthermore, our review encompasses a larger number of cohorts, although these databases included studies recruiting young adults as well. Therefore, we think that our review constitutes a valuable resource for researchers involved in geriatric studies, not only as a background for communication and exchanges, but also to foster the use of available resources, to learn from others’ experiences and to help set priorities for future research in ageing communities.
We wish to thank the investigators of the cohorts for their help in validating our results. Many thanks also to the people who participated in these longitudinal studies.