Cohort Profile: The Ageing Trajectories of Health - Longitudinal Opportunities and Synergies (ATHLOS) project.

Why was the cohort set up?

The number of people aged 60 years or older is projected to significantly increase in the coming decades worldwide. According to United Nations estimates, this figure is expected to more than double by 2050 and to more than triple by 2100. Population ageing poses major challenges for the traditional social welfare state due to the greater needs for health and social care of older people.

This project, Ageing Trajectories of Health – Longitudinal Opportunities and Synergies (ATHLOS), funded by the European Union’s Horizon 2020 Research and Innovation Program, aims to achieve a better understanding of the impact of ageing on health by developing a new single measure of health status. With this measure, the project intends to identify patterns of healthy ageing trajectories and their determinants, the critical points in time when changes in trajectories are produced, and to propose timely clinical and public health interventions to optimize and promote healthy ageing. To achieve this, a new cohort has been composed from harmonized datasets of existing international longitudinal cohorts related to health and ageing.

The ATHLOS project follows the World Health Organization’s definition of healthy ageing by studying healthy ageing as an ongoing process of developing and maintaining the functional ability that enables wellbeing in older age. This ongoing process interacts with the environment in which people live and can either favour health or be harmful to it. Environments are highly influential on individual behaviour, exposure to health risks, access to quality health and social care and the opportunities that ageing brings. Healthy ageing is thus not a unitary phenomenon but must be deconstructed into its components: mental (e.g. cognitive decline), physical (e.g. activities of daily living) and social functioning (e.g. participation in community activities).

National and international research funding agencies and governments have supported several follow-up studies of population cohorts since the early 1990s [e.g. the ‘Health and Retirement Study’ (HRS)]. HRS has been used as a model for many other longitudinal studies in a number of countries, such as the ‘English Longitudinal Study of Ageing’ (ELSA),, the ‘Japanese Study of Aging and Retirement’ (JSTAR), the ‘Mexican Health and Aging Study’ (MHAS), the ‘China Health and Retirement Longitudinal Study’ (CHARLS), the ‘Longitudinal Aging Study in India’ (LASI) or the ‘Korean Longitudinal Study of Ageing’ (KLOSA), also called the ‘HRS-family’ studies. More recently, multi-country projects have also been initiated, such as the Study on Global Ageing and adult health (SAGE) funded by the World Health Organization, the Survey of Health, Ageing and Retirement in Europe (SHARE) funded by the European Commission and the 10/66 dementia research study.

Although these studies have been powered to provide relevant national estimates, sample sizes might be limited for assessing the joint effect of several predisposing and protective factors. Additionally, although cross-country comparisons provide evidence of how contextual and health care factors impact population health, the few existing multi-country studies are limited to a selected group of countries and require a significant amount of time, co-ordination and financial resources.

Recently, strategies to harmonize data a posteriori from different longitudinal studies have been proposed to overcome some of the challenges stated above. For example, the Gateway to Global Ageing (G2AGING) is a platform funded by the National Institute on Aging, National Institutes of Health that aims to achieve data harmonization of longitudinal studies on ageing and to facilitate cross-national comparisons in population survey data. To date, G2AGING has harmonized the HRS datasets with the datasets of the other nine ‘HRS-family’ studies. In a broader context, an international research programme, called Maelstrom Research, provides systematic harmonization methodology and tools with the aim of leveraging the creation of research collaborations. In the context of ageing, Maelstrom Research has facilitated research consortia including the Integrative Analysis of Longitudinal Studies of Aging and Dementia (IALSA), which harmonized 9 studies, and the Promoting Mental Well-being and Healthy Ageing in Cities (MINDMAP), which incorporates 10 studies. These consortia have a specific focus on ageing and health and cover populations mostly from North America and Europe.

The ATHLOS consortium constitutes a new collaborative research project that, among other things, uses the Maelstrom Research resources. Unlike G2AGING, Maelstrom Research offers open-source software and guidelines to harmonize data according to concrete research aims. Thus, a harmonized dataset comprising at least 17 longitudinal population studies, from Europe and international countries, was created. These studies include information on common health conditions, as well as a detailed assessment of participants’ functioning. Integrating data from existing cohort studies leads to greater sample size and statistical power to more precisely estimate the determinants and risk factors of healthy ageing. Furthermore, ageing trajectories can be compared between different countries and populations to evaluate if different cultures have diverse risk factors impacting the population’s healthy ageing.

Who is in the ATHLOS cohort?

The cohort comprises more than 411 000 individuals who participated in 17 general population longitudinal studies in 38 countries. The studies are the 10/66 Dementia Research Group Population-Based Cohort Study, the Australian Longitudinal Study of Aging (ALSA), the ATTICA Study, CHARLS, Collaborative Research on Ageing in Europe (COURAGE), ELSA, Study on Cardiovascular Health, Nutrition and Frailty in Older Adults in Spain (ENRICA), the Health, Alcohol and Psychosocial factors in Eastern Europe Study (HAPIEE), the Health 2000/2011 Survey, HRS, JSTAR, KLOSA, MHAS, SAGE, SHARE, the Irish Longitudinal Study of Ageing (TILDA) and the Uppsala Birth Cohort Multigenerational Study (UBCoS).,

Each study includes one or more populations and provides data on health determinants and age-related events. An overview of the included studies and their target populations is provided in Table 1. Table 2 presents sample sizes and response rates at baseline for each study and population. The median percentage of response rate at each study’s baseline was 75%, and the range was from 53% (SAGE-Mexico) to 96% (10/66-Rural China). It should be noted that the sample sizes of the CHARLS, ELSA, Health 2000/2011, HRS, JSTAR, MHAS and SHARE studies were increased in posterior waves of data collection. Supplementary Table S1, available as Supplementary data at IJE online, presents sample sizes, number of new participants, deceased participants and drop-outs for each study, population and wave.

Table 1.

List of studies included in the ATHLOS project

Studies		Countries/populationsa	Recruitmentd	Refreshment
Acronym	Name
10/66	The 10/66 Dementia Research Group Population-Based Cohort Study	Cuba, India, China, Dominican Republic, Venezuela, Peru, Mexico and Puerto Rico	All 65+ respondents in a household	No
ALSA	The Australian Longitudinal Study of Aging	Australia: Participants drawn from the South Australian Electoral Roll	All 65+ respondents in a household	No
ATTICA	The ATTICA Study	Greece: Metropolitan Athens area	18+ participants	No
CHARLS	The China Health and Retirement Longitudinal Study	China: All counties except Tibet	45+ participants and spouses	Wave 2
COURAGE	Collaborative Research on Ageing in Europe	Spain and Poland	18+ participants	No
ELSA	The English Longitudinal Study of Ageing	UK and Northern Ireland	50+ participants and spouses	Wave 3, 4, 6
ENRICA	Study on Cardiovascular Health, Nutrition and Frailty in Older Adults in Spain	Spain	60+ participants	No
HAPIEE	The Health, Alcohol and Psychosocial factors in Eastern Europe Study	Poland, Czech Republic and Lithuania	45–69 participants	No
HEALTH 2000-11	The Health 2000–2011 Survey	Finland	30+ participants	Wave 2
HRS	The Health and Retirement Survey	United States: 6 birth sub-cohorts	50+ participants and spouses	All waves
JSTAR	The Japanese Study of Aging and Retirement	Japan: 5 cities sub-cohort, 2 cities sub-cohort and 3 cities sub-cohortb	50–75 participants	No
KLOSA	The Korean Longitudinal Study of Ageing	South Korea	45+ participants and spouses	No
MHAS	The Mexican Health and Aging Study	Mexico	50+ participants and spouses	Wave 3
SAGE	WHO Study on Global Ageing and Adult Health	South Africa, Ghana, China, India, Russia and Mexico	All 50+ respondents in a household (small sample 18+)	No
SHARE	The Survey of Health, Ageing and Retirement in Europe	20 countriesc	50+ participants and spouses	All waves
TILDA	The Irish Longitudinal Study of Ageing	Ireland	50+ participants and spouses	No
UBCOS	The Uppsala Birth Cohort Multigenerational Study	Sweden: Births at the Uppsala Academic Hospital between 1915 and 1929	Hospital records, census records, and register data. Spouses, descendants and spouses of descendants	Descendants cohort

Although several studies were conducted in the same countries, the probability that the same individual participated in more than one study is likely very small because all study designs included a probability sample from the general population.

5 cities: Adachi-Kanazawa-Shirakawa-Sendai-Takikawa; 2 cities: Tosu-Naha; 3 cities: Chofu-Tondabayashi-Hiroshima.

Countries included in the SHARE study from waves 1 to 5: Denmark, Sweden, Austria, France, Germany, Switzerland, Belgium, the Netherlands, Spain, Italy, Greece, Israel, Czech Republic, Poland, Ireland, Estonia, Hungary, Slovenia, Portugal and Luxembourg.

Values are ages in years.

Table 2.

Coverage time of interview, sample sizes and response rates at baseline of each study and population included in the ATHLOS cohort

Study /	Population	Year of interview																											Sample sizea at baseline	Response rate at baseline
		1915-29	1930-90	1991	1992	1993	1994	1995	1996	1997	1998	1999	2000	2001	2002	2003	2004	2005	2006	2007	2008	2009	2010	2011	2012	2013	2014	2015
10/66	Cuba															W1						W2							2813	94
	India																W1				W2								2004	72
	Urban China																W1					W2							1160	74
	Rural China																W1					W2							1002	96
	Dominican Rep.																W1					W2							2011	95
	Venezuela																	W1				W2							1965	80
	Urban Peru																	W1				W2							1381	80
	Rural Peru																	W1				W2							552	88
	Urban Mexico																	W1				W2							1003	84
	Rural Mexico																		W1			W2							1000	86
	Puerto Rico																			W1					W2				2009	93
ALSA						W1	W2	W3	W4		W5			W6			W7		W8	W9		W10	W11			W12	W13		2087	55
ATTICA															W1				W2					W3					3037	75
CHARLS																								W1		W2	c W3	c W4	18245	81
COURAGE	Spain																							W1			W2		4753	70
	Poland																							W1				W2	4071	67
ELSA															W1		W2		W3		W4		W5		W6		W7		12099	66
ENRICA																					W1				W2			W3	2519	60
HAPIEE	Poland														W1				W2										10728	61
																		Mortality and Cardiovascular followup b
	Czech Republic														W1					W2									8857	55
																		Mortality and Cardiovascular followup b
	Lithuania																	W1											7111	65
																		Mortality and Cardiovascular followup b
HEALTH 2000/2011													W1											W2					8028	93
HRS	HRS sub-sample				W1		W2		W3		W4		W5		W6		W7		W8		W9		W10		W11		c W12		12787	82
	AHEAD					W2		W3			W4		W5		W6		W7		W8		W9		W10		W11		c W12		8297	80
	CODA										W4		W5		W6		W7		W8		W9		W10		W11		c W12		2364	73
	WBB										W4		W5		W6		W7		W8		W9		W10		W11		c W12		2622	70
	EBB																W7		W8		W9		W10		W11		c W12		3400	75
	MBB																						W10		W11		c W12		5102
JSTAR	5 citiesd																			W1		W2		W3					3862	60
	2 citiese																					W1		W2					1440
	3 citiesf																							W1					1966
KLOSA																			W1		W2		W3		W4		c W5		10254	64
MHAS														W1		W2									W3		c W4		15146	89
SAGE	South Africa																			W1							c W2		4227	75
	Ghana																			W1							c W2		5573	81
	China																			W1							c W2		15050	93
	India																			W1							c W2		12198	68
	Russia																			W1							c W2		4947	83
	Mexico																					W1					c W2		5448	53
SHARE																	W1		W2		W3		W4			W5		c W6	30816	62
TILDA																							W1		W2		c W3		8504	62
UBCOS	Birth generation	W1 - W4		W5									W6																20732	-
	Descendants		W1 - W4	W5									W6																33052

Sample sizes derived from datasets provided by the study owners. Spouses of participants can be included.

The HAPIEE study has a continuous mortality and cardiovascular follow-up from 2005 to 2015.

Dataset will eventually be included.

5 cities: Adachi-Kanazawa-Shirakawa-Sendai-Takikawa.

2 cities: Tosu-Naha.

3 cities: Chofu-Tondabayashi-Hiroshima.

All studies are cohorts based on questionnaires except for the UBCoS study, which collects routine health and social data for all babies born in the Uppsala Academic Hospital between the years 1915 and 1929, and their descendants. The UBCoS data were converted into periods of data collection to resemble the design of the other studies.

Finally, the study on the Identification of health and disability determinants on ageing in Italy (IDAGIT) will be subsequently included in the cohort.

How often have participants been followed up?

Most of the longitudinal studies included in the ATHLOS harmonized dataset started between 2000 and 2010 and have at least 2 waves of data collection (see Table 2). ALSA and HRS started much earlier, in the 1990s, and have more than 10 waves of data collection. SAGE has only 1 wave of data harmonized to date. However, new waves of data are expected to be harmonized in the future.

Regarding UBCoS, as register data have been collected approximately every 10 years from 1960 to 2008, we distributed the data in 6 waves.

What has been harmonized?

The data harmonization requires an a priori definition of the variables of interest and their possible values. Thus, the ATHLOS consortium defined a wide range of variables, called DataSchema variables, which included all health conditions, sociodemographic variables, personal functioning and contextual factors. These are usually assessed in population studies. Variables that have international standards or have been created by well-known scales and measured tests were employed in the harmonization process. For example, the International Classification of Functioning, Disability and Health (ICF) biopsychosocial model and the conceptualization of health suggested by the World Health Organization were used for characterizing the functioning-related variables.

The DataSchema variables were classified as follows: (i) sociodemographic and economic characteristics; (ii) lifestyle and health behaviours; (iii) health status and functional limitations; (iv) diseases; (v) death; (vi) physical measures; (vii) psychological measures; (viii) laboratory measures; (ix) social environment and life events; and (x) other administrative information. In Table 3, a list of core variables within the aforementioned domains, together with the individual studies, is provided.

Table 3.

List of the core variables for the harmonized ATHLOS datasets and the studies including potential information to be harmonized in at least one population or wave

Domain	Sub-domains	10/66	ALSA	ATTICA	CHARLS	COURAGE	ELSA	ENRICA	HAPIEE	H2000/11	HRS	JSTAR	KLOSA	MHAS	SAGE	SHARE	TILDA	UBCoS
Sociodemographic and economic characteristics	Birth	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓
	Sex	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓
	Marital status	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓
	Education	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓
	Living alone	✓	✓	×	✓	✓	✓	✓	×	✓	✓	×	✓	×	✓	✓	✓	✓
	Employment/retirement	✓	✓	×	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓
	Wealth	✓	✓	×	✓	✓	✓	×	×	✓	✓	✓	✓	✓	✓	✓	✓	✓
Lifestyle and health behaviours	Tobacco	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓
	Alcohol	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓
	Physical activity	✓	✓	✓	✓	✓	✓	✓	×	✓	✓	✓	✓	✓	✓	✓	✓	✓
Health status and functional limitations	Memory	✓	×	×	✓	✓	✓	×	×	✓	✓	×	×	✓	✓	✓	✓	×
	Dizziness	×	✓	×	×	×	✓	×	×	×	✓	×	×	×	×	✓	×	×
	Orientation	✓	✓	×	✓	✓	✓	✓	×	×	✓	✓	×	×	×	✓	✓	×
	Walking speed	×	×	×	✓	✓	✓	✓	×	✓	✓	×	×	✓	✓	✓	✓	×
	Energy	✓	✓	×	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	×
	Sleep	✓	✓	×	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	×
	Pain	✓	✓	×	✓	✓	✓	✓	✓	✓	✓	×	✓	✓	✓	✓	✓	×
	Incontinence	✓	✓	×	✓	×	✓	✓	×	✓	✓	×	✓	✓	×	✓	✓	×
	Hearing/sight	✓	✓	×	✓	✓	✓	✓	×	✓	✓	✓	✓	✓	✓	✓	✓	×
	Mobility	✓	✓	×	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	×
	Activities of Daily Living (ADL)	✓	✓	×	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	×
	Instrumental ADL	✓	✓	×	✓	✓	✓	✓	×	✓	✓	✓	✓	✓	✓	✓	✓	×
	Cognitive impairment	✓	×	×	✓	✓	✓	✓	✓	×	✓	✓	×	✓	✓	✓	✓	×
	Self-reported health	✓	✓	×	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	×
	Falls	✓	✓	×	✓	×	✓	✓	×	✓	✓	×	✓	✓	×	✓	✓	×
Diseases	Diabetes	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓
	Respiratory	✓	✓	×	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓
	Hypertension	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓
	Joint disorders	✓	✓	×	✓	✓	✓	✓	×	✓	✓	✓	✓	✓	✓	✓	✓	✓
	Cardiovascular disease	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓
	Cancer	×	✓	×	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	×	✓	✓	✓
Death	Living status	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	×	✓	×	✓
Physical measures	Body measures	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	×
	Grip strength	×	✓	×	✓	✓	✓	✓	×	✓	✓	✓	✓	✓	✓	✓	✓	✓
	Blood pressure	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	×
Psychological measures	Screening measure of cognition	×	✓	×	×	✓	×	✓	×	✓	×	×	✓	×	×	×	✓	×
	Depression	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	×
	Anxiety	✓	×	✓	×	✓	✓	✓	×	✓	×	×	×	×	✓	×	✓	×
Laboratory measures	Glucose, cholesterol, …	×	×	✓	×	×	✓	✓	✓	✓	×	×	×	✓	×	×	✓	×
Social environment and life events	Social network	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	×
	Social support	✓	✓	×	✓	✓	✓	✓	✓	✓	×	✓	✓	✓	✓	✓	✓	×
	Social participation	✓	✓	×	✓	✓	✓	×	✓	✓	×	✓	✓	✓	✓	✓	✓	×
	Social trust/cohesion	×	×	×	×	✓	✓	×	✓	✓	×	✓	×	×	✓	✓	×	×
	Life events	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	×
	Loneliness	✓	✓	×	✓	✓	✓	×	✓	✓	✓	✓	✓	✓	✓	✓	✓	×
Administrative variables	ID participant/household, date of interview, etc.	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓	✓

What has ATHLOS found? Key findings and publications

ATHLOS includes data from all populated continents, with Europe being the most represented. Sociodemographic information by continent and country is shown in Table 4. The median year of birth was around the 1940s, with people from America being older (born in the 1930s) and those in Australia much older (born in 1914). Overall, the median age at baseline was about 60 years. Sweden exhibits a younger average age at baseline, as UBCoS cohorts were based on register data starting in 1960. The percentage of female participants was slightly above 50%, other than in Australia and Ghana, which had lower percentages. The average percentage of primary education or less stood at about 37%, but in general there was heterogeneity even in countries from the same study as in SHARE. In Europe, for example, the lowest percentage was observed in Germany (2%) and the highest percentage in Spain (58%); in South America, the percentage was very high in Venezuela (81%) and Dominican Republic (90%).

Table 4.

Descriptive statistics of some sociodemographic variables by continent and country

Continent	Country	n	Year of birth (median)	Age at participant's baseline (median)	Female (%)	Primary education or less (%)	Studies involved
Europe	Austria	6411	1945	63	58	14	SHARE
	Belgium	8720	1948	60	55	21	SHARE
	Czech Republic	18092	1946	60	56	14	HAPIEE, SHARE
	Denmark	5553	1948	60	54	13	SHARE
	Estonia	7075	1945	65	59	6	SHARE
	Finland	9673	1948	47	54	47	Health2000
	France	8105	1946	61	57	40	SHARE
	Germany	8690	1946	62	54	2	SHARE
	Greece	6969	1949	55	54	38	ATTICA, SHARE
	Hungary	3076	1948	63	57	2	SHARE
	Ireland	9638	1948	62	46	29	SHARE, TILDA
	Italy	7158	1945	63	55	48	SHARE
	Lithuania	7111	1945	61	55	12	HAPIEE, SHARE
	Luxembourg	1610	1950	62	53	37	SHARE
	Netherlands	6547	1946	61	54	14	SHARE
	Poland	17532	1947	58	54	20	COURAGE, HAPIEE, SHARE
	Portugal	2080	1947	64	57	56	SHARE
	Slovenia	3755	1948	63	56	10	SHARE
	Spain	15952	1944	65	54	58	COURAGE, ENRICA, SHARE
	Sweden	66243	1945	16	50	35	SHARE, UBCoS
	Switzerland	4571	1946	62	55	11	SHARE
	United Kingdom	18489	1944	59	54	38	ELSA
Eurasia	Russia	4947	1946	62	64	9	SAGE
Asia	China	38990	1951	59	53	60	10/66, CHARLS, SAGE
	India	14202	1947	55	61	58	10/66, SAGE
	Israel	3857	1946	61	55	21	SHARE
	Japan	7268	1945	63	52	25	JSTAR
	South Korea	10254	1945	61	56	45	KLOSA
North America	United States of America	37317	1938	56	56	27	HRS
	Cuba	2813	1930	74	65	58	10/66
	Dominican Republic	2011	1931	74	66	90	10/66
	Mexico	28817	1944	59	58	72	10/66, MHAS, SAGE
	Puerto Rico	2009	1932	76	67	44	10/66
South America	Peru	1933	1932	74	61	56	10/66
	Venezuela	1965	1935	71	64	81	10/66
Africa	Ghana	5573	1950	60	49	47	SAGE
	South Africa	4227	1947	60	57	62	SAGE
Oceania	Australia	2087	1914	78	49	36	ALSA
Total		411320	1945	58	54	37	The 17 studies

Advanced analytical approaches have already been applied to some studies of the ATHLOS dataset to test the methodology for developing a single measure of health status and to identify different patterns of health trajectories over time. This measure will allow for the comparison of health status across populations and longitudinal studies included in ATHLOS. Specifically, these analyses have already been conducted on harmonized datasets comprising ELSA and HRS studies. Evidence suggests that the average health scores and trajectories are sensitive to age and that the health status measure is a good predictor of mortality.,

Additionally, a large systematic review (with more than 90 000 articles screened) was conducted to summarize and synthesize the current evidence on social, biological, behavioural, psychological and sociodemographic determinants of healthy ageing. This systematic review indicated limited research about healthy ageing in low- and middle-income countries and confirmed the heterogeneity in the conceptualization and definition of healthy ageing.

What are the main strengths and weaknesses of ATHLOS?

The harmonized dataset in the ATHLOS project constitutes a new cohort that has been created by collecting data from 17 longitudinal studies from five continents. The harmonization approach and tools used in this project were adapted from the methodology developed by Maelstrom Research. This approach is systematic and rigorous to ensure that harmonized variables are comparable.

It should be noted that the harmonization is a retrospective process, as studies were not initially designed to be harmonized. The heterogeneity in study design, instruments and data collection limits the amount and quality of information that can be pooled. Thus, we are conducting thorough documentation of the whole process, not only for the sake of reproducibility and transparency, but also to estimate the quality of harmonization for every variable.

What are the main problems inherent to the harmonization?

In the course of the harmonization process, we encountered several challenges. First, the harmonization potential is a trade-off between the number of studies (quantity) that can be included and the content equivalence (precision) within the study-specific variables. For example, education can be harmonized using standard criteria, such as the ISCED2011, creating a categorical variable based on the highest qualification or generating a continuous variable for years of education. Greater precision in the definition of education would entail a lower number of studies that could be included. Second, some variables were at times conceptually different across studies, even though they described the same underlying construct. For example, employment may be addressed directly (e.g. are you employed?) or indirectly (e.g. are you retired?). The same applies to energy level, which can be addressed in terms of presence of energy (e.g. do you have energy for daily life?) or inversely (e.g. did you feel tired out or low in energy?). In this case, our intention was to address the variable in aggregate and not the way in which the question was asked. Further, ethical and legal issues may restrict the sharing and pooling of individual data. For example, studies may not publicly provide biomarker or mortality information of participants who have been lost to follow-up. Therefore, managing and pooling large datasets from different studies poses significant challenges, but the advantages seem worthwhile if we consider the global coverage and the gain in statistical power.

Can I get hold of the data? Where can I find out more?

A platform of free software applications, developed by Maelstrom Research, is used to store the original datasets, guide the harmonization process and create a web portal for the studies from the ATHLOS Consortium, as well as the final harmonized databases. These software applications have General Public Licences and can therefore be used and freely modified according to the ATHLOS project needs. The web catalogue can be found at: https://athlos.pssjd.org. Full access to statistical summaries and reports of the harmonization process for each variable in each study requires registration. Documentation of the whole harmonization process for each variable in each study is publicly shared at: https://github.com/athlosproject/athlos-project.github.io/

No individual dataset can be downloaded from these websites. Harmonized datasets with individual data are stored on a secure server. At this stage of the project, only researchers and collaborators of the ATHLOS Consortium can download harmonized datasets, unless study owners provide their consent. Thus, external users should contact the Scientific Committee (athlos@pssjd.org), comprised of members of the ATHLOS Consortium, to access the harmonized datasets. Alternatively, users could access original datasets directly from the study owners and follow the documentation and codes published in the abovementioned `github` webpage.

Profile in a nutshell

The Ageing Trajectories of Health – Longitudinal Opportunities and Synergies (ATHLOS) cohort harmonizes existing longitudinal data from 17 international cohort studies.

It aims to achieve a better understanding of the impact of ageing on health and to propose timely clinical and public health interventions to optimize and promote healthy ageing.

The cohort comprises more than 411 000 individuals from 38 countries. Most of the studies started between 2000 and 2010 and have between 2 and 13 waves of data collection. New waves of data collected during the ATHLOS project and other studies will be incorporated in updated versions of the harmonized dataset.

Harmonized datasets include variables classified in the following areas: (i) sociodemographic and economic characteristics; (ii) lifestyle and health behaviours; (iii) health status and functional limitations; (iv) diseases; (v) death; (vi) physical measures; (vii) psychological measures; (viii) laboratory measures; (ix) social environment and life events; and (x) other administrative information.

The catalogues of the studies and final harmonized databases, together with documentation of the whole harmonization process, can be found in the web portal: (https://athlos.pssjd.org). External users interested in using the harmonized datasets should contact the ATHLOS Scientific Committee: (athlos@pssjd.org).

Funding

This work was supported by the five-year Ageing Trajectories of Health: Longitudinal Opportunities and Synergies (ATHLOS) project. The ATHLOS project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 635316. See Appendix for more information about funding in each study.

Click here for additional data file.