Cohort Profile: Dementia Risk Prediction Project (DRPP)

Key Features.

• The Dementia Risk Prediction Project (DRPP) was established to bring together data from 16 longitudinal cohorts of diverse middle-aged and older adults to provide a resource for research of dementia and its vascular and lifestyle risk factors, and develop a dynamic dementia risk prediction model.

• Sixteen cohorts with participants from 49 of the 50 US states, France, Iceland, England and the Netherlands with baseline exams ranging from 1948 to 2006 were included. Participants were recruited across multiple sites with multiple in-person assessments of clinical, genetic and behavioural risk factors, follow-up of >10 years and ongoing Alzheimer’s disease and related dementias ascertainment.

• Collectively, there are 119 061 individuals with data available at baseline. The DRPP provides direct access to harmonized individual-level data for 95 134 individuals from 14 of the 16 cohorts. Median baseline age is 59 years (interquartile range: 49–70) with a maximum age at follow-up of >90 years.

• Frequency of follow-up varies by cohort, with exams occurring every 2–12 years; the number of in-person exams ranges from 2 to 32.

• To apply for permission to access our data on our secure portal, please visit our website at drpp.northwestern.edu.

Why was the cohort set up?

Dementia is a major public health problem; despite declines in the age-specific incidence, its prevalence will continue to increase due to ageing of populations.^1–3 Worldwide, an estimated 55 million people are living with Alzheimer’s or other dementias and, as the population continues to age, this number is expected to rise to 78 million in 2030 and 139 million in 2050.⁴ The number of individuals living with dementia is increasing, leading to greater burden of morbidity, caregiving needs and healthcare utilization. Although we have made strides in reducing mortality and morbidity from other diseases such as cardiovascular disease, stroke and cancer, the proportion of deaths due to dementia has increased by >145% in the USA in the past 20 years⁵ and is presently the seventh leading cause of death among all diseases globally.⁴

Dementia encompasses a set of complex chronic disorders with neuropathology that is often ‘mixed’, with contributions from vascular and neurodegenerative pathologies. Specifically, >70% of Alzheimer’s disease and related dementias (ADRD) cases are estimated to have mixed vascular and Alzheimer’s disease (AD) pathology.⁶ Further, many ADRD risk factor profiles are stronger predictors in midlife than in late life. Therefore, the risk factor profiles are also complex and variable. Although non-modifiable risk factors such as age, sex, race/ethnicity and apolipoprotein E (APOE) genotype impact dementia risk, it has been estimated that 40% of all dementia cases could be prevented or delayed by targeting modifiable risk factors.⁷ Modifiable risk factors that contribute to the largest proportion of dementia cases include midlife hearing status, diabetes, midlife hypertension, midlife obesity, depression, physical inactivity, smoking, low education, excessive alcohol consumption, head injury and air pollution.⁷ For example, an analysis of 7878 Japanese American men in the Honolulu-Asia Aging Study (HAAS) found that untreated midlife hypertension alone contributes to 27% of dementia risk.⁸ In a review of English-language systematic reviews and meta-analyses, these modifiable risk factors combined contributed to 54.1% of the population attributable risk for dementia in the USA and 50.7% worldwide.⁹

Several studies have examined longitudinal trajectories as well as the visit-to-visit variability of risk factors in the long prodrome preceding ADRD diagnosis.^10,11 Among them, trajectories of blood pressure, body mass index (BMI) and cholesterol have been studied extensively.^10,12,13 Interestingly, most of those studies show an age-dependent pattern in the association between these risk factors and cognitive impairment.¹² For example, in a paper using data from Whitehall II participants, different BMI trajectories were identified among people who ultimately developed ADRD as compared with those who did not.¹³ Similarly, collaborators in the HAAS and the Atherosclerosis Risk in Communities Study (ARIC) have both demonstrated that specific blood pressure trajectories in mid- to late life are related to incident dementia.^14,15 Importantly, these mid-life risk factor trajectories are also related to estimated brain amyloid deposition, signalling that these same risk factors impact AD in addition to vascular forms of dementia.¹⁶ These findings and others like them highlight the importance of considering the life-course trajectory of vascular risk factors rather than current or cross-sectional blood pressure or weight to define one’s ADRD risk, especially in late life.

Young and midlife risk exposures, interactions of various risk factors with each other and ageing all contribute to the risk of developing dementia. This level of complexity motivates comprehensive and dynamic risk assessment using well-characterized cohort studies with available data from multiple time points to develop rich, robust risk prediction models. Retrospective harmonization of a multiple of such studies facilitates sufficient statistical power, provides the ability for cross-validation or replication and increases heterogeneity of the pooled sample.¹⁷ None of the prior dementia risk models has incorporated longitudinal risk factor patterns.¹⁸ Longitudinal risk factor patterns, as discussed above, are highly associated with the risk of dementia later in life independently of baseline risk factor levels.^10,12,13 It is likely that using risk factor trajectories will improve our ability to identify individuals at high risk of future ADRD and discriminate between high-risk and low-risk individuals to direct resources and preventive interventions at individuals who would benefit most within our resource-constrained healthcare systems.

The Dementia Risk Prediction Project (DRPP) was thus formed to create a rigorously harmonized data set for developing and validating an accurate and personalized, dynamic dementia risk prediction model that incorporates longitudinal risk factor measurements and easily updates as new measurements are accrued. Additionally, the DRPP data will also be made accessible to approved researchers to collaborate and conduct research that may not be possible in smaller, individual cohorts.

Who is in the cohort?

The DRPP is a pooled consortium initially founded with individual-level data from 16 well-characterized prospective, natural history cohorts of diverse middle-aged and older adults (Table 1). Cohorts were included if they performed multiple in-person assessments of vascular risk factors and had a follow-up of >10 years and ongoing ADRD ascertainment. Table 1 briefly describes each the main research focus, location, size, baseline exam year and length of follow-up of each cohort.

Table 1.

Characteristics of the 14 derivation studies and 2 validation studies

Study name	Main research focus	Location	Number of participants a	Baseline exam year	Person-years of follow-up
Derivation studies
Age, Gene/Environment Susceptibility-Reykjavik Study (AGES-Reykjavik)	Single-centre prospective population study of older men and women in Iceland	Reykjavik, Iceland	5764	2002–06	17 246
Atherosclerosis Risk in Community (ARIC)	Prospective epidemiological study designed to investigate the relationships between atherosclerosis and its clinical outcomes, as well as various cardiovascular risk factors	USA: Forsyth County, North Carolina; Jackson, Mississippi; eight northern suburbs of Minneapolis, Minnesota; and Washington County, Maryland	15 667	1987–99	222 623
Cardiovascular Health Study (CHS)	Observational study of risk factors for cardiovascular disease in persons aged ≥65 years	USA: Sacramento, California; Hagerstown, Maryland; Winston-Salem, North Carolina; and Pittsburgh, Pennsylvania	5888	1989–90	65 917
Framingham Heart Study (FHS)		USA: Framingham, Massachusetts
Original	Longitudinal study designed to identify the common risk factors for cardiovascular disease	USA: Framingham, Massachusetts	5079	1948	146 400
Offspring	The offspring of the Original cohort and their spouses	USA: Framingham, Massachusetts	4856	1971	141 225
New Offspring Spouse (NOS)	If the spouse of an offspring was never enrolled in the FHS and if at least two of his/her biological children participated in Exam 1 of Gen III, then that spouse was invited to participate in the NOS Exam 1; the addition of their data and cell lines may improve the statistical power for studying families in the FHS	USA: Framingham, Massachusetts	101	2003	832
Generation 3	Third generation of participants, which will provide greater resources of phenotypic and genotypic information	USA: Framingham, Massachusetts	4064	2002	47 202
Omni	New group of participants, unrelated to the original cohort, reflecting the increasing diversity of the community	USA: Framingham, Massachusetts and surrounding towns	494	1994	5427
Omni 2	To expand from the Omni Cohort 1 and to represent an ethnically diverse group of ≥10% of the size of the Gen 3 cohort; enrolment of a second cohort of Omni participants started in 2003 and ended in July 2005	USA: Framingham, Massachusetts and surrounding towns	404	2003	3967
Honolulu Heart Program/Honolulu-Asia aging Study (HHP/HAAS)	Longitudinal epidemiological study in aged Japanese American men investigating the prevalence, incidence and risk factors for dementia	USA: Hawaii	3734	1991–93	16 426
Multi-Ethnic Study of Atherosclerosis (MESA)	Prospective study investigating the prevalence, correlates and progression of subclinical cardiovascular disease	USA: New York, New York; Baltimore, Maryland; Chicago, Illinois; Los Angeles, California; Minneapolis-St Paul, Minnesota; Winston-Salem, North Carolina	6814	2000–02	73 639
Reasons for Geographic and Racial Differences in Stroke (REGARDS)	Observational study of risk factors for stroke in persons aged ≥45 years	USA: all 48 contiguous states, 56% from Stroke Belt region	30 183	2003–07	151 124
Sacramento Area Latino Study on Aging (SALSA)	Longitudinal study that investigates the incidence and risk factors of physical and cognitive impairment, dementia and cardiovascular diseases	USA: Sacramento, California	1778	1998–99	9718
Whitehall II	Ongoing longitudinal study of British civil servants that investigates social and occupational influences on health and disease	London, England	10 308	1985–88	218 966
Validation studies
The Three-City Study (3C)b	Relationship between vascular diseases and dementia in persons aged ≥65 years	France: Bordeaux, Dijon, Montpellier	9294	1999–2001	67 710
Rotterdam Study (RS)b	Prospective cohort study on risk and determinants of disease, from cardiovascular disease to neurodegenerative disease and beyond	Rotterdam, The Netherlands	14 633	1989–2009	185 463
RS I	Original cohort of adults aged ≥55 years from the Ommoord district in Rotterdam	Rotterdam, The Netherlands	7978	1989–92	102 897
RS-II	Extension of the cohort with persons from the study district who had become 55 years old since the start of the study or those aged ≥55 years who had moved into the study district	Rotterdam, The Netherlands	2725	2000–03	42 561
RS-III	All persons aged ≥45 years living in the study district who had not been examined already (i.e. mainly comprising those aged 45–60 years)	Rotterdam, The Netherlands	3930	2006–08	40 005

^aIncludes individuals aged ≥18 years.

^bExternal validation study.

The DRPP includes four European cohorts, each based in different countries. The Three-City Study (3C) includes individuals from three French cities: Bordeaux, Dijon and Montpellier; 61% of baseline participants were female.¹⁹ The Age, Gene/Environment Susceptibility-Reykjavik Study (AGES-Reykjavik) is a single-centre study of older men and women (58% female at baseline) in Iceland.²⁰ The Rotterdam Study (RS) is an ongoing population-based study in the city of Rotterdam, the Netherlands, that recruited participants aged ≥45 years into three cohorts (RS-I, RS-II, RS-III).²¹ The Whitehall II study is an ongoing longitudinal study of British civil servants based in London at recruitment to the study; at baseline, the study population was 89% White and 33% female.²²

The remaining 12 cohorts are based in the USA, some are single-centre and others are larger, multicentre studies. The ARIC is based in North Carolina, Mississippi, Minnesota and Maryland; participants aged 45–65 years (55% female and 27% non-White) were recruited at the first visit between 1987 and 1989 and it is ongoing.²³ The Cardiovascular Health Study (CHS) is based in California, Maryland, North Carolina and Pennsylvania, and consisted of 58% women, 16% African American and 31% with cardiovascular disease (CVD) at baseline.²⁴ The HAAS is based in Hawaii and includes older Japanese American men.²⁵ The Multi-Ethnic Study of Atherosclerosis (MESA) is a multicentre study based in New York, Maryland, Illinois, California, Minnesota and North Carolina; at baseline, the study population was 38% White, 28% African American, 23% Hispanic and 11% Asian.²⁶ The Reasons for Geographic and Racial Differences in Stroke (REGARDS) includes participants from all 48 contiguous states; 55% are women, 41% are Black and 56% are from the Stroke Belt region.²⁷ The Sacramento Area Latino Study on Aging (SALSA) recruited a representative sample of individuals of predominantly Mexican heritage residing in Sacramento, California.²⁸

The final six US cohorts include the original Framingham Heart Study (FHS)²⁹ and five of its offshoots [Offspring (FOS), New Offspring Spouse (NOS), Generation 3 (Gen 3), Omni and Omni 2], all based in Framingham, Massachusetts and its surrounding towns. The FHS recruited the original participants for the baseline exam in 1948. The FOS started in 1971 and includes the offspring of the original cohort and their spouses. The NOS included the spouse of members of the FOS by 2003 if they were never included prior and if at least two of their children participated in the Gen 3, a third generation of participants recruited in 2002. In 1994 and 2003, two new cohorts (Omni and Omni 2) were included to reflect the increasing diversity of the Framingham community.³⁰

Variables in 14 of the 16 cohorts were harmonized internally by DRPP analysts and the remaining two (RS and 3C) are external cohorts to be used for validation of risk prediction models. Individual-level data are not available for the external cohorts, but the variables have been harmonized to variable definitions of DRPP by local teams and validation analyses will be run in parallel to the core DRPP data at these two external sites.

Together in DRPP, the consortium cohort represents 119 061 individuals with baseline data from 49 of 50 US states, France, Iceland, England and the Netherlands. Of these, 86 646 individuals have at least one global cognitive assessment score [DRPP used Mini-Mental State Examination, Cognitive Abilities Screening Instrument, Modified Mini-Mental State Examination (3MSE) and Montreal Cognitive Assessment to harmonize a single global cognitive assessment; more details can be found in the Supplementary Methods S1, available as Supplementary data at IJE online]. The median baseline age was 59 years (IQR: 49–70) with a maximum age at follow-up of >90 years. Baseline data were collected over multiple decades, generations and epochs ranging from 1948 to 2006. This cohort includes 20 188 (21%) individuals who identify as Black, 5348 (6%) Asian, 3746 (4%) Hispanic, 182 (0.2%) Other and 65 567 (69%) White. The two validation cohorts did not collect race and ethnicity data and are therefore not included here.

How often have they been followed up?

Each individual study has its own follow-up time and exam schedule, and these vary significantly (Figure 1). The DRPP includes data from 1948 through to 2020, with >1 million person-years of follow-up time. The number of exams ranges from two in the AGES-Reykjavik and REGARDS to 32 in the original FHS cohort. Exams were, in most cases, separated by a year or two but, in others, by as many as 12 years. Some cohorts included phone calls between in-person examinations for additional outcome surveillance and cognitive testing.

Figure 1.

Timing and frequency of risk factor measurement, by cohort. AGES, Age, Gene/Environment Susceptibility-Reykjavik Study; ARIC, Atherosclerosis Risk in Community; APOE, apolipoprotein E; BMI, body mass index; BP, blood pressure; CHS, Cardiovascular Health Study; F/U, follow-up; FHS, Framingham Heart Study; FOS, Framingham Offspring Study; NOS, Framingham New Offspring Spouse; FHS-Gen3, Framingham Heart Study Generation 3; FHS-Omni, Framingham Heart Study Omni; FHS-Omni 2, Framingham Heart Study Omni 2; HAAS, Honolulu Heart Program/Honolulu-Asia aging Study; MESA, Multi-Ethnic Study of Atherosclerosis; REGARDS, Reasons for Geographic and Racial Differences in Stroke; SALSA, Sacramento Area Latino Study on Aging; 3C, The Three-City Study; RS, Rotterdam Study; Yr, year. Note: Study exams use the same language as in their original cohort. Unless stated otherwise, Exam 1 or Phase 1 for each cohort is its baseline visit. RS lists visits as Study-Cohort-visit number (ex. RS-I-2 for Rotterdam Study Cohort 1, Visit 2)

What has been measured?

These cohorts collected data at multiple in-person assessments in various domains including demographics (age, gender, race/ethnicity, educational attainment), clinical risk factors (blood pressure, cholesterol, glucose, HbA1C, use of medications), genetic (APOE) and behavioural risk factors (diet, smoking, alcohol use, physical activity) and ongoing dementia, stroke and cardiovascular event ascertainment. Further information regarding the methods that were used to measure and ascertain these variables in each cohort can be found in Supplementary Tables S1–S9 (available as Supplementary data at IJE online). Of the variables collected by each cohort, 42 common variables have been harmonized by the DRPP.

Thirteen of the 16 cohorts assessed incident dementia according to the Diagnostic and Statistical Manual of Mental Disorders, fourth edition (DSM-IV) criteria [many with AD specifically according to National Institute of Neurological Disorders and Stroke–Alzheimer Disease and Related Disorders (NINCDS-ADRDA) criteria], which include cognitive testing, clinical assessment and interviewing.³¹ Three of the cohorts rely on the International Classification of Diseases, Tenth Revision (ICD-10) codes. Because ICD-10 codes are likely to represent under-ascertainment of the outcome, we will conduct sensitivity analyses with and without these three cohorts to test the robustness of our models to case ascertainment methods. Cohort-specific dementia definitions can be found in Supplementary Table S10 (available as Supplementary data at IJE online).

DRPP data were harmonized over a 2-year period (September 2020 to June 2022) following the Maelstrom guidelines for retrospective data harmonization.¹⁷ Fortier et al. describe a process, similar to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA)³² or Meta-analyses of Observational Studies in Epidemiology (MOOSE)³³ statements, which codify and guide successful data harmonization.¹⁷

During the grant application period, our team defined the research questions, objectives and protocol; assembled information; and selected studies that fit our inclusion criteria (listed above). After engaging all 16 cohorts, we identified the variables needed and requested corresponding data sets with the most detail possible. The level of detail varied from data dictionaries to exact exam questions. Data were inventoried and their harmonization potential was evaluated based on data collection methods and compatibility with data from other cohorts. In some cases, additional data and information were requested to achieve high-quality pooled variables.

What has it found?

The DRPP includes 10 436 incident cases of dementia, 10 357 stroke, 23 991 incident cases of coronary heart disease and 37 654 incident cases of CVD. To date, the DRPP has harmonized 42 variables using methods appropriate to the data available (Supplementary Table S11, available as Supplementary data at IJE online). The consortium demographics are described in Table 2. Throughout the harmonization process, we examined trends by age for continuous variables (Figure 2 shows mean total cholesterol over time, by cohort), patterns of disease over time after defining each chronic disease (hypertension, hypercholesterolemia, and diabetes) (Figure 3 shows hypertension diagnoses by cohort) and event rates by age and cohort (Figure 4 shows the dementia rate per 1000 person-years by age and cohort). These patterns are consistent with prior work.

Table 2.

Baseline demographic data, by cohort

	Derivation studies														Validation studies
Variable	AGES	ARIC	CHS	FHS	FOS	NOS	FHS—Gen3	FHS—Omni	FHS—Omni2	HAAS	MESA	REGARDS	SALSA	Whitehall	3C	RS I	RS-II	RS-III
Total number	5764	15 667	5888	5079	4856	101	4064	494	404	3734	6814	30 183	1778	10 308	9294	7978	2725	3930
Age, years	77 (5.9)	54.2 (5.8)	72.8 (5.6)	44.2 (8.6)	36.9 (9.9)	64.8 (9.4)	40.2 (8.8)	51.4 (9.1)	42.5 (13.8)	77.9 (4.7)	62.2 (10.2)	64.8 (9.4)	70.7 (7.1)	45 (6.1)	74.3 (5.6)	70.6 (9.8)	65.1 (8.4)	57.1 (7.2)
Sex
Female	57.7%	55.1%	57.6%	54.8%	51.7%	53.5%	53.3%	57.7%	56.4%	0	52.8%	55.1%	58.5%	33.1%	60.7%	61.1%	56.1%	57.3%
Male	42.3%	44.9%	42.4%	45.2%	48.3%	46.5%	46.7%	42.3%	43.6%	100%	47.2%	44.9%	41.5%	66.9%	39.3%	38.9%	43.9%	42.7%
Race/ethnicity
White	100%	72.8%	82.8%	100%	99.4%	98.0%	99.3%	1.2%	1.7%	0	38.5%	58.5%	0	89.1%	NA	79.2%	73.7%	81.1%
Asian	0	0.2%	0.1%	0	0	0	0.1%	15.2%	25.7%	100%	11.8%	0	0	5.7%	NA	0.8%	1.2%	1.3%
Black	0	26.9%	15.6%	0	0.2%	0	0.1%	35.0%	24.0%	0	27.8%	41.5%	0	3.6%	NA	0.1%	0.4%	1.3%
Other/unknown	0	0.1%	0.5%	0	0.1%	0	0.3%	7.5%	3.0%	0	0	0	0	1.7%	NA	19.9%	24.8%	16.3%
Hispanic	0	0	1.1%	0	0.3%	2.0%	0.2%	41.1%	45.5%	0	22%	0	100%	0	NA	NA	NA	NA
Education
High school	14.2%	46.4%	40.9%	70.0%	34.5%	21.8%	14.6%	17.6%	10.6%	48.2%	25.1%	34.3%	22.4%	26.6%	55.2%	38.1%	45.0%	35.7%
No school/grade school	68%	9.7%	15.4%	0	0	0	0	0	0	34%	11%	4.2%	60.9%	47.5%	26.1%	25.3%	8.8%	10.8%
Technical/vocational/college/graduate/professional	10.4%	43.7%	43.4%	27.0%	53.1%	49.5%	85.3%	53.8%	67.8%	17.8%	63.5%	61.5%	16.5%	25.8%	18.4%	32.5%	45.0%	52.9%
Missing/unknown	7.5%	0.2%	0.3%	3.0%	12.3%	28.7%	0.2%	28.5%	21.5%	0	0.3%	0.1%	0.1%	0	0.2%	4.2%	1.2%	0.6%

AGES, Age, Gene/Environment Susceptibility-Reykjavik Study; ARIC, Atherosclerosis Risk in Community; CHS, Cardiovascular Health Study; FHS, Framingham Heart Study; FOS, Framingham Offspring Study; NOS, Framingham New Offspring Spouse; FHS-Gen3, Framingham Heart Study Generation 3; FHS-Omni, Framingham Heart Study Omni; FHS-Omni2, Framingham Heart Study Omni 2; HAAS, Honolulu Heart Program/Honolulu-Asia aging Study; MESA, Multi-Ethnic Study of Atherosclerosis; REGARDS, Reasons for Geographic and Racial Differences in Stroke; RS, Rotterdam Study; SALSA, Sacramento Area Latino Study on Aging; 3C, The Three-City Study.

Figure 2.

Mean total cholesterol by age group, by cohort.^{a  a}Data shown for 14 derivation studies only. AGES, Age, Gene/Environment Susceptibility-Reykjavik Study; ARIC, Atherosclerosis Risk in Community; CHS, Cardiovascular Health Study; FHS, Framingham Heart Study; HAAS, Honolulu Heart Program/Honolulu-Asia aging Study; MESA, Multi-Ethnic Study of Atherosclerosis; REGARDS, Reasons for Geographic and Racial Differences in Stroke; SALSA, Sacramento Area Latino Study on Aging

Figure 3.

Hypertension, by cohort. ^aHypertension status is based on 2017 American Heart Association (AHA)/American College of Cardiology (ACC) definitions: Normal BP, <120/80 and NO antihypertension medication use; Elevated BP, 120-129 AND <80 mmHg and NO antihypertension medication use; Hypertension stage 1, 130-139 AND/OR 80-89 mmHg OR antihypertension medication use; Hypertension stage 2, ≥140 AND/OR ≥90 mm Hg OR antihypertension medication use. (Data shown for 14 derivation studies only.) AGES, Age, Gene/Environment Susceptibility-Reykjavik Study; ARIC, Atherosclerosis Risk in Community; BP, blood pressure; CHS, Cardiovascular Health Study; FHS, Framingham Heart Study; HAAS, Honolulu Heart Program/Honolulu-Asia aging Study; MESA, Multi-Ethnic Study of Atherosclerosis; REGARDS, Reasons for Geographic and Racial Differences in Stroke; SALSA, Sacramento Area Latino Study on Aging

Figure 4.

Dementia rate per 1000 person-years, by age category at baseline and cohort.^{a a}Data shown for 13 derivation studies only. REGARDS dementia data are forthcoming. AGES, Age, Gene/Environment Susceptibility-Reykjavik Study; ARIC, Atherosclerosis Risk in Community; CHS, Cardiovascular Health Study; FHS, Framingham Heart Study; HAAS, Honolulu Heart Program/Honolulu-Asia aging Study; MESA, Multi-Ethnic Study of Atherosclerosis; REGARDS, Reasons for Geographic and Racial Differences in Stroke; SALSA, Sacramento Area Latino Study on Aging

After validation and refinement of models based on baseline data, we aim to develop dynamic longitudinal risk prediction models using statistical and machine-learning techniques that will incorporate long-term risk factor trajectories to identify individuals at high risk of ADRD.³⁴

What are the main strengths and weaknesses?

The infrastructure of the DRPP is designed with reproducibility and generalizability in mind. It includes openly available software, described earlier, to facilitate reproducible harmonization, as well as the harmonization of external cohort data. Additionally, the DRPP has pooled and harmonized a large quantity of data on incident dementia and dementia risk factors under multiple domains that have been made available on a cloud-based computing platform for other approved collaborators to use.

The consortium includes multiple multisite studies of a diverse group of individuals representing a wide range of ages, race/ethnicity, socio-economic status and geographic diversity. Together, DRPP includes >1 million person-years of follow-up, with individuals contributing behavioural and clinical data at multiple time points across the life course. As much of the dementia risk prediction work up to this point has been done in quite uniform populations, the diversity and inclusion of traditionally under-represented groups in DRPP are an important contribution.

The cohorts that are included in the DRPP also feature a broad range of measurement methods and calendar times of measurement. These heterogeneous period and cohort effects allow our risk prediction modelling to be broadly generalizable and not limited to a single cohort or sample population. External validation is a necessary step in building risk prediction models in order to ensure reproducibility and generalizability,³⁵ and the availability of two external validation cohorts with harmonized data is unique.

As with all studies, there is a potential for misclassification, particularly for difficult-to-measure behavioural characteristics such as physical activity or lifestyle variables such as diet. We worked with subject-matter experts to harmonize these data at the item level. When scores were used, such as for diet, they were selected based on their association with ADRD and brain health.^36–39 As mentioned above, the heterogeneity in ascertainment of dementia between studies is an important limitation that must be considered in each analysis.

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

The DRPP welcomes collaborations and analyses of these data as well as the addition of new cohorts to our consortium.

Investigators can apply for use of the DRPP data or review the data documentation at drpp.northwestern.edu. Approved collaborators will have access to a data set on the virtual computing platform built specifically for the DRPP.⁴⁰ Our computing platform allows individuals to access data stored on a secure server at Northwestern, to perform statistical analysis using a variety of software options and to save summary-level output on the same server. All individual-level data are securely locked down and cannot be downloaded, e-mailed or shared in any way. After analyses are completed on the virtual DRPP platform, output is then reviewed by the Northwestern DRPP team to confirm that no individual-level data leave our server and only then is it sent to the investigator. For more information or to submit a proposal, please visit our website drpp.northwestern.edu or contact study Principal Investigator Dr Norrina Allen (norrina-allen@northwestern.edu).

Ethics approval

The DRPP has been approved by the Institutional Review Board (IRB) at Northwestern University. In addition, the individual cohorts have been approved by their local IRBs and received informed consent from participants.

Data availability

See ‘Can I get a hold of the data?’ above.

Supplementary data

Supplementary data are available at IJE online.

Author contributions

A.K. managed collection of data from individual cohorts and prepared the text. J.S. inventoried individual-level data, conceived of and wrote the ‘psHarmonize’ R package, harmonized variables and oversaw the harmonization of external cohort data. P.C. inventoried individual-level data and harmonized the lifestyle variables. S.S. conceived of the project, reviewed inclusion criteria, provided subject-matter expertise and completed pre-statistical harmonization of cognitive variables. M.M. harmonized general cognitive variables. A.S. advised on 3C data and facilitated harmonization. A.G. advised on harmonization best practices. A.A. provided access to and insight on the SALSA data. A.S.-M. advised on the Whitehall II data. M.A.I. advised on the RS. C.H. advised on the 3C. C.T. advised on the 3C. C.S. advised on the FHS data and assisted with data access. D.A.L. provided harmonization support and content expertise. D.L.J. provided content expertise. E.B. provided harmonization support and content expertise. F.A.S. provided content expertise. F.J.W. harmonized the RS data and provided insight on data structure. J.H. provided data support for the FHS. L.J.L. advised on content and the HAAS and AGES data. L.Z. provided risk prediction and statistical expertise. M.H. advised on the SALSA data. O.L.L. advised on the CHS data. S.D. advised on the 3C data and harmonization. S.Seshadri advised on the FHS. S.E.J. advised on the REGARDS and assisted with data access. T.M.H. advised on the MESA and ARIC data, and assisted with data access. V.G. advised on the AGES data. D.S. advised on the statistical rigor of the harmonization process and oversaw the harmonization process. N.A. directed the implementation of the study. All authors provided critical review of the manuscript.

Funding

The Dementia Risk Prediction Project (DRPP) is supported by the National Institute for Neurologic Disorders and Stroke (NINDS) via grants 1R61NS120245-01/R33NS120245. Dr. Sedaghat reports funding from the U.S. National Institutes of Health- National Institute of Aging (NIH/NIA) R01AG079108-01. Dr. Levine reports funding support from the U.S. National Institutes of Health - National Institute of Neurological Disorders and Stroke (NIH/NINDS) 1R01 NS 102715 -01 and the U.S. National Institutes of Health - National Institute on Aging (NIH/NIA) 1 RF1 AG068410-01. Dr. Gross was supported by the National Institute on Aging (NIA) R01AG030153 and NIA R01AG070953. Dr. Hughes was supported by relevant grant funding from the National Institute on Aging (NIA): P30AG072947, R01AG054069, R01AG058969 and U01HL096812. Dr. Lopez receives support from NIA: R01AG20098. Dr. Satizabal receives support from NIA (R01 AG059727 and R01 AG082360) and NINDS (UF1/UH1 NS125513). Drs. Satizabal, Himali, and Seshadri are partially supported by the South Texas Alzheimer’s Disease Research Center (P30 AG066546). Drs. Seshadri and Himali receive support from The Bill and Rebecca Reed Endowment for Precision Therapies and Palliative Care. Dr. Himali is supported by an endowment from the William Castella family as William Castella Distinguished University Chair for Alzheimer’s Disease Research, and Dr. Seshadri by an endowment from the Barker Foundation as the Robert R Barker Distinguished University Professor of Neurology, Psychiatry and Cellular and Integrative Physiology. This work was funded by the National Heart Lung and Blood Institute (Framingham Heart Study Contracts No. N01-HC-25195, No. HHSN268201500001I, and No. 75N92019D00031), Boston University School of Medicine, and grants from the National Institute on Aging (NIA; R01 AG054076, R01 AG049607, U01 AG052409, R01 AG059421, RF1 AG063507, RF1 AG066524, U01 AG058589), the National Institute of Neurological Disorders and Stroke (NINDS; R01 NS017950). The Atherosclerosis Risk in Communities study has been funded in whole or in part with Federal funds from the National Heart, Lung, and Blood Institute, National Institutes of Health, Department of Health and Human Services, under Contract nos. (75N92022D00001, 75N92022D00002, 75N92022D00003, 75N92022D00004, 75N92022D00005). The authors thank the staff and participants of the ARIC study for their important contributions. CHS research was supported by contracts HHSN268201200036C, HHSN268200800007C, HHSN268201800001C, N01HC55222, N01HC85079, N01HC85080, N01HC85081, N01HC85082, N01HC85083, N01HC85086, 75N92021D00006, and grants U01HL080295 and U01HL130114 from the National Heart, Lung, and Blood Institute (NHLBI), with additional contribution from the National Institute of Neurological Disorders and Stroke (NINDS). Additional support was provided by R01AG023629 from the National Institute on Aging (NIA). A full list of principal CHS investigators and institutions can be found at CHS-NHLBI.org. The Multi-Ethnic Study of Atherosclerosis (MESA) was supported by contracts 75N92020D00001, HHSN268201500003I, N01-HC-95159, 75N92020D00005, N01-HC-95160, 75N92020D00002, N01-HC-95161, 75N92020D00003, N01-HC-95162, 75N92020D00006, N01-HC-95163, 75N92020D00004, N01-HC-95164, 75N92020D00007, N01-HC-95165, N01-HC-95166, N01-HC-95167, N01-HC-95168 and N01-HC-95169 from the National Heart, Lung, and Blood Institute, and by grants UL1-TR-000040, UL1-TR-001079, and UL1-TR-001420 from the National Center for Advancing Translational Sciences (NCATS). The authors thank the other investigators, the staff, and the participants of the MESA study for their valuable contributions. A full list of participating MESA investigators and institutions can be found at http://www.mesa-nhlbi.org. This paper has been reviewed and approved by the MESA Publications and Presentations Committee. Disclaimer: The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Conflict of interest

None declared.