The core challenge is deceptively simple to state and fiendishly hard to solve. An ideal dataset for lifecourse research would prospectively measure all exposures, outcomes, and covariates across every critical life period, with enough temporal resolution to distinguish the timing of an effect from its duration, minimal attrition over decades, and strong generalizability to a target population. No such dataset exists. Cohort studies that enroll people in middle or late life, which supply most of the outcome data in aging research, routinely lack information on earlier decades. As a result, researchers face a menu of imperfect options, each suited to different questions, and the review organizes those options into four major approaches: birth cohort studies, data linkages, late-life retrospective measures, and the combination of age-overlapping cohorts.

Birth cohort studies are widely regarded as the gold standard. They enroll participants at birth or even before conception and follow the same individuals across their lives, collecting prospectively designed data on surveys, health, and biology with clear temporal ordering. The Dunedin Multidisciplinary Health and Development Study, which has tracked 1,037 New Zealanders born in 1972 and 1973 since birth, has yielded findings connecting persistent cannabis use to midlife cognitive reserve and smaller hippocampal volume, and its midlife structural MRI measures position the cohort to illuminate how dementia risk accumulates. In the United States, no comparable long-running national birth cohort exists, but the NIH Environmental Influences on Child Health Outcomes Program and the Adolescent Brain Cognitive Development Study are collecting deep early-life data in diverse populations that will mature into powerful resources for aging research decades from now.

The gold standard’s price tag is time. Birth cohorts initiated in the 1970s are only now reaching ages at which dementia incidence begins to rise, and the review notes they would need another two decades or more before incidence is high enough to statistically power analyses of health effects. Sustained funding across half a century is rare, and other problems compound the delay. Participants born in a narrow range of calendar years are shaped by secular trends in diet, healthcare access, and air pollution that can compromise generalizability to other eras, a problem the review suggests addressing through parallel harmonized analyses across multiple cohorts, such as recent work on early-life excess body weight and midlife cognition spanning three British birth cohorts born in 1946, 1958, and 1970. Recruitment from selected populations, like the Kaiser Permanente members enrolled in the Child Health and Development Study between 1959 and 1967, further limits representativeness, and decades of attrition can induce selection bias precisely because the factors driving long-term participation, such as health status, income, and education, are themselves determinants of the outcomes under study.

Data linkage offers a faster route to prospectively generated exposure information. By connecting late-life study participants to records originally created for other purposes, researchers can recover early-life data spanning tax and financial records, schooling and employment histories, vital records, and even consumer credit data on disruptive life events like bankruptcies and liens, which would be nearly impossible to collect retrospectively. Nordic countries represent the most comprehensive version of this approach, with government registries capturing health, education, and economic data on entire populations from birth, offering minimal attrition and exceptional representativeness. Geographic linkage is another powerful variant: members of the Lothian Birth Cohort 1936 provided residential histories that were connected to historical air pollution data to examine lifecourse pollution exposure and dementia risk, while large US studies including the Health and Retirement Study have been linked to historical census records to study childhood lead exposure, early-life socioeconomic conditions, and New Deal employment in relation to late-life cognition.

Yet linkage carries its own distinctive biases. Merging datasets typically requires identifiers such as names, birth dates, or addresses, which may be non-unique, incomplete, or measured with error, and when unique identifiers are unavailable, probabilistic linkage becomes possible but demands meticulous validation. Administrative data exist only for people who interacted with the collecting institution, so earnings records exclude those without covered employment and credit histories exclude the unbanked, introducing selection bias that can disproportionately drop very low-income and underrepresented individuals. Privacy and re-identification concerns constrain what can be linked and how, sometimes forcing researchers into secure enclaves or privacy-preserving methods that permit greater linkage error. And because these records were never designed for research, the variables collected are limited to what administration required, poorly suited to capturing psychosocial constructs like loneliness or perceived discrimination, and vulnerable to quality problems such as diagnostic codes with high specificity but low sensitivity.

When no prospective early-life data exist at all, researchers can turn to retrospective measures collected in late life. Recalled measures, gathered through life history questionnaires such as those deployed in the Health and Retirement Study and its international sister studies, are logistically attractive: inexpensive, fast, and flexible enough to capture subjective experiences across childhood, adolescence, and early adulthood in a single instrument. But the measurement error is real and sometimes differential. A striking comparison found stronger effects of childhood adversity on cognitive decline in studies using objective court records than in studies using subjective recall, and even concrete questions, such as whether a parent ever remarried, show only moderate reliability. In dementia research the problem is magnified because cognitive impairment itself degrades the accuracy of recall, creating error that is differential by the outcome. Survival bias compounds the threat, since recalled data can only come from people who live long enough to report it, a mechanism that may explain attenuated associations between smoking and dementia at older ages and potential underestimation of education’s influence on cognitive decline. Strategies like life calendars, specific prompting, and neutral language can help, and quantitative bias analysis offers a way to assess robustness when recall is the only option.

Biological measures collected in late life provide a rarer but intriguing window into earlier exposures. Bone lead, which accumulates over the lifespan, can be measured non-invasively to estimate cumulative lead exposure, and emerging work on teeth and bone is extending this logic to other social and environmental exposures. Mendelian randomization offers an indirect route, using germline genomic data collected in late life as an instrument to estimate effects of otherwise unavailable measures such as childhood obesity on late-life outcomes. But the review is candid about the limits: late-life biomarkers capture cumulative exposure at the moment the sample was taken without revealing timing or intensity across life periods, restricting which lifecourse models can be tested, and many constructs of interest, such as cumulative blood pressure elevation, have no known biological measure at all. Other late-life measures like DNA methylation and allostatic load reflect downstream processes of earlier life factors rather than the exposures themselves.

The most methodologically innovative approach approximates complete lifecourse data by combining multiple age-overlapping longitudinal cohorts. For trajectory questions, techniques like cohort-sequential integrative data analysis align observed measures across cohorts by age to model population-level trajectories of BMI, cardiovascular risk, or cognition across age ranges no single study covers. For effect estimation, an emerging synthetic cohort strategy links participants with exposure data in one cohort to participants with outcome data in another, using matching-based imputation such as hot deck methods to borrow early-life exposure information from similar individuals. The National Longitudinal Survey of Youth 1979 and the Health and Retirement Study, which share birth years and comparable midlife measures, illustrate how this can work in practice. The catch is the strength of the assumptions: data fusion theory requires that the linking variables capture all connections between exposure and outcome, both mediating mechanisms and confounding paths, and that these variables be known, available in both cohorts, measured without error, and harmonizable, while birth cohort effects must not distort the associations being estimated.

The review’s ultimate message is that there is no one-size-fits-all answer, only a disciplined match between research question and data. The authors provide a decision flowchart that begins with the research question and walks through whether the exposure was prospectively measured, whether historical sources permit accurate linkage, whether retrospective measures are feasible given survival and recall biases, and whether cohorts can be validly combined. Crucially, the authors emphasize that combinations of approaches often outperform any single one: retrospective recall can be validated through linkage, birth cohorts can gain new exposures through administrative connections, and triangulating evidence across studies with different strengths and biases strengthens causal inference, as exemplified by the UK Longitudinal Linkage Collaboration’s connections across more than twenty longitudinal studies. As exposome research, which seeks to characterize the totality of interacting exposures across domains and life periods, pushes lifecourse science further, these data tradeoffs will only grow in importance, and the methodological scaffolding laid out in this review may well determine how much of the early-life story of late-life disease scientists can finally see.

Subject of Research: Data approaches for lifecourse research on early-life determinants of late-life health outcomes: opportunities, limitations, and tradeoffs

Article Title: Data approaches for lifecourse research on early-life determinants of late-life health outcomes: opportunities, limitations, and tradeoffs

Article References: Hayes-Larson, E., Amin, S., Nichols, E., Adar, S. D., Chiang, Y.-Y., Flood, S. M., Gross, A. L., Claus Henn, B., Keller, K. P., Knapp, D., Lee, J., Shih, R., Weiss, J., Weuve, J., & Kezios, K. L. (2026). Data approaches for lifecourse research on early-life determinants of late-life health outcomes: opportunities, limitations, and tradeoffs. SSM – Population Health, Article 101985. https://doi.org/10.1016/j.ssmph.2026.101985

Image Credits: AI Generated

DOI: 10.1016/j.ssmph.2026.101985

Keywords: Data, approaches, lifecourse, research, early-life, determinants, late-life, health, outcomes, opportunities, limitations, tradeoffs