Generation differences in determinants of cognitive aging: Comparisons of five population-based representative Swedish cohorts born 44-years apart and followed from age 70 on multiple occasions
Studies show large birth cohort differences in cognitive aging. In terms of level, onset of decline, and rate of change. Studies also reveal multiple interacting factors accounting for observed individual differences. We know little about how the relative importance of these factors differ across generations—where some become more/less important as determinant of cognitive aging. Reasons for current absence of knowledge is due to lack of relevant data, which is difficult to obtain since it requires multiple follow-ups of cohort born decades apart. The H70 is one of few studies that allows such comparisons. The ongoing project, initiated in 1971, includes multiple follow-ups, on cognitive, health, lifestyle, and demographic factors from age 70 across five population-based cohorts born in 1901, 1906, 1923, 1930, and 1944. In the proposed project, we take advantage of collected data and model time trends across cohorts in effects of health-related, lifestyle, and demographic factors on individual differences in cognitive aging. We use growth curve modelling techniques and aim to replicate findings within NEAR. Project outcomes contribute to: 1) a better understanding of factors leading to cognitive decline; 2) improved public health recommendations; 3) predictions concerning relative importance of determinants for cognitive aging and dementia risk in future populations. This information is essential for informed decision making concerning health and welfare in greying societies.
Final report
Purpose and development of the project
The purpose of this project was to use existing longitudinal population-based data to examine patterns of cognitive aging and to identify determinants of individual differences across birth cohorts. The project was based on secondary data analyses and involved three researchers working part-time within the project (Dr. Valgeir Thorvaldsson, Dr. Johan Skoog, and Dr. Linn Zulka). Overall, the project developed according to plan and resulted in several analyses and scientific publications. The primary focus was to advance understanding of both individual level mechanisms and broader contextual factors that shape cognitive aging trajectories across generations.
Implementation
A substantial portion of the work in this project was devoted to data identification, cleaning, and harmonization. Parts of the data are historical in nature, with initial data collections dating back to the early 1970s. Many variables were not digitally documented, requiring manual retrieval of information from archival sources. In addition, merging datasets collected more than 30 years apart posed methodological challenges, including inconsistencies in measurement and documentation. Considerable effort was therefore invested in ensuring comparability across cohorts through careful harmonization procedures. Beyond data preparation, time was primarily allocated to statistical analyses and manuscript preparation.
Key results and conclusions
The project demonstrates that patterns of cognitive aging differ substantially between contemporary older adults and earlier-born cohorts (i.e., individuals who were old in the 1970s–1990s). These differences are evident across multiple dimensions: overall cognitive performance levels, rates of cognitive change prior to pathological decline, timing of accelerated decline associated with brain pathology, and the rate of decline following such acceleration.
Specifically, contemporary cohorts of older adults show higher levels of cognitive functioning and remain cognitively intact for a longer period of their lifespan. The onset of accelerated decline tends to occur closer to death, consistent with the concept of morbidity compression. However, once decline begins, it progresses more rapidly than in earlier cohorts. These findings have important implications at multiple levels, including for individuals, families, caregiving systems, healthcare services, and broader societal resource planning.
Cognitive functioning was also found to be a strong predictor of survival in old age, but the nature of this association differed across cohorts. In earlier-born cohorts, the level of cognitive performance was the strongest predictor of survival, whereas the rate of change played a secondary role. In more recent cohorts, the rate of cognitive decline emerged as the more important predictor, while baseline level remained relevant but less dominant. This shift suggests that dynamic processes of decline may have become increasingly informative for survival in aging populations.
These findings are important because they highlight that predictors of late-life outcomes are not stable across generations. Models of aging and survival therefore need to account for cohort specific mechanisms rather than assuming uniform processes across time.
Stroke and cognitive aging
The project also examined the impact of stroke on cognitive aging trajectories across birth cohorts. While the incidence of stroke has declined in recent cohorts, prevalence has increased due to longer life expectancy and improved survival following stroke. This provided a unique opportunity to evaluate cohort differences in post-stroke cognitive outcomes.
The results showed that stroke has a substantial negative effect on cognitive functioning in both cohorts, corresponding to approximately a 0.45 standard deviation decrease in general cognitive ability, an effect size likely to have meaningful consequences for everyday functioning. However, a cohort by stroke by age interaction was observed. In earlier cohorts, the negative impact of stroke increased with age, whereas in more recent cohorts the impact decreased with age. This pattern likely reflects improvements in acute stroke care and post-stroke rehabilitation in contemporary populations.
This finding is important because it suggests that medical advances not only influence survival but also modify the cognitive consequences of major health events across the lifespan.
Methodological development
The project has also contributed to methodological development in the analysis of longitudinal cognitive aging and survival. Several advanced statistical approaches were applied and extended to address limitations in conventional modeling frameworks.
First, we implemented random change point models within a Bayesian framework to capture individual differences in the timing of accelerated cognitive decline. These models allow the estimation of person specific transition points from relatively stable functioning to terminal decline, a feature that is difficult, and often infeasible, to estimate using traditional frequentist maximum likelihood approaches. The Bayesian formulation provides greater flexibility, improved convergence, and a natural way to quantify uncertainty in the timing of these change points, which is central to studying morbidity compression.
Second, the project developed and applied cohort moderated joint models of longitudinal change and survival. While joint modeling is increasingly used in aging research, the explicit incorporation of birth cohort as a moderator of the association between cognitive trajectories and survival outcomes is, to our knowledge, novel. This approach allows for direct testing of whether the relationship between cognition and mortality differs across generations. In this context, it provides a powerful framework for identifying cohort specific mechanisms underlying aging and survival, thereby improving both theoretical understanding and predictive accuracy.
Third, the growth process was parameterized not only in terms of level and rate of change, but also using the area under the individual-specific growth curve (AUC). This representation captures the cumulative cognitive burden over time in a parsimonious way and has not previously been implemented in this context. The AUC approach is particularly advantageous for prediction modeling, as it reduces complex longitudinal information into a single interpretable metric that reflects both level and change, and may better capture processes related to terminal decline and overall functional reserve.
Finally, we applied second-order growth curve models within a structural equation modeling (SEM) framework. This approach enables the modeling of latent cognitive constructs over time while accounting for measurement error and changes in the measurement structure across waves. Such models are rarely used in cohort comparative aging research, largely due to the extensive data requirements. In the present project, the availability of harmonized longitudinal data across cohorts made it possible to implement this approach, providing more reliable estimates of true cognitive change and allowing for more valid comparisons across generations.
Together, these methodological developments enhance the precision, flexibility, and interpretability of analyses of cognitive aging and survival, and offer a framework that can be extended to other longitudinal population-based studies.
Future research questions
The project raises several new research questions. One important direction concerns the use of machine learning approaches to predict cognitive decline, morbidity compression, and survival, and how such models can be implemented in clinical and policy contexts. This project represents one of the few systematic evaluations of cohort differences in morbidity compression, highlighting the need to further investigate its individual, healthcare, and societal consequences.
In addition, several potential moderating factors, such as lifestyle, demographics, and biological risk markers, remain to be examined in a cohort sensitive framework. These questions will be pursued in future work building on the current project.
Dissemination and collaboration
Findings from the project have been disseminated primarily through peer-reviewed scientific journals and conference presentations. The project has also facilitated several collaborations. Locally, collaborations were established within the University of Gothenburg, particularly between the Department of Psychology and the Department of Psychiatry and Neurochemistry (Dr. Skoog). Nationally, the project contributed to collaborations within the NEAR consortium (NEAR – The National E-infrastructure for Aging Research). Internationally, collaborations were developed with researchers at Humboldt University in Germany (Dr. Gerstorf), Ohio University in the USA (Dr. Muniz-Terrera), and Northwestern University (Dr. Graham). Ongoing projects continue to build on these collaborations and extend the research agenda established in this work.
The purpose of this project was to use existing longitudinal population-based data to examine patterns of cognitive aging and to identify determinants of individual differences across birth cohorts. The project was based on secondary data analyses and involved three researchers working part-time within the project (Dr. Valgeir Thorvaldsson, Dr. Johan Skoog, and Dr. Linn Zulka). Overall, the project developed according to plan and resulted in several analyses and scientific publications. The primary focus was to advance understanding of both individual level mechanisms and broader contextual factors that shape cognitive aging trajectories across generations.
Implementation
A substantial portion of the work in this project was devoted to data identification, cleaning, and harmonization. Parts of the data are historical in nature, with initial data collections dating back to the early 1970s. Many variables were not digitally documented, requiring manual retrieval of information from archival sources. In addition, merging datasets collected more than 30 years apart posed methodological challenges, including inconsistencies in measurement and documentation. Considerable effort was therefore invested in ensuring comparability across cohorts through careful harmonization procedures. Beyond data preparation, time was primarily allocated to statistical analyses and manuscript preparation.
Key results and conclusions
The project demonstrates that patterns of cognitive aging differ substantially between contemporary older adults and earlier-born cohorts (i.e., individuals who were old in the 1970s–1990s). These differences are evident across multiple dimensions: overall cognitive performance levels, rates of cognitive change prior to pathological decline, timing of accelerated decline associated with brain pathology, and the rate of decline following such acceleration.
Specifically, contemporary cohorts of older adults show higher levels of cognitive functioning and remain cognitively intact for a longer period of their lifespan. The onset of accelerated decline tends to occur closer to death, consistent with the concept of morbidity compression. However, once decline begins, it progresses more rapidly than in earlier cohorts. These findings have important implications at multiple levels, including for individuals, families, caregiving systems, healthcare services, and broader societal resource planning.
Cognitive functioning was also found to be a strong predictor of survival in old age, but the nature of this association differed across cohorts. In earlier-born cohorts, the level of cognitive performance was the strongest predictor of survival, whereas the rate of change played a secondary role. In more recent cohorts, the rate of cognitive decline emerged as the more important predictor, while baseline level remained relevant but less dominant. This shift suggests that dynamic processes of decline may have become increasingly informative for survival in aging populations.
These findings are important because they highlight that predictors of late-life outcomes are not stable across generations. Models of aging and survival therefore need to account for cohort specific mechanisms rather than assuming uniform processes across time.
Stroke and cognitive aging
The project also examined the impact of stroke on cognitive aging trajectories across birth cohorts. While the incidence of stroke has declined in recent cohorts, prevalence has increased due to longer life expectancy and improved survival following stroke. This provided a unique opportunity to evaluate cohort differences in post-stroke cognitive outcomes.
The results showed that stroke has a substantial negative effect on cognitive functioning in both cohorts, corresponding to approximately a 0.45 standard deviation decrease in general cognitive ability, an effect size likely to have meaningful consequences for everyday functioning. However, a cohort by stroke by age interaction was observed. In earlier cohorts, the negative impact of stroke increased with age, whereas in more recent cohorts the impact decreased with age. This pattern likely reflects improvements in acute stroke care and post-stroke rehabilitation in contemporary populations.
This finding is important because it suggests that medical advances not only influence survival but also modify the cognitive consequences of major health events across the lifespan.
Methodological development
The project has also contributed to methodological development in the analysis of longitudinal cognitive aging and survival. Several advanced statistical approaches were applied and extended to address limitations in conventional modeling frameworks.
First, we implemented random change point models within a Bayesian framework to capture individual differences in the timing of accelerated cognitive decline. These models allow the estimation of person specific transition points from relatively stable functioning to terminal decline, a feature that is difficult, and often infeasible, to estimate using traditional frequentist maximum likelihood approaches. The Bayesian formulation provides greater flexibility, improved convergence, and a natural way to quantify uncertainty in the timing of these change points, which is central to studying morbidity compression.
Second, the project developed and applied cohort moderated joint models of longitudinal change and survival. While joint modeling is increasingly used in aging research, the explicit incorporation of birth cohort as a moderator of the association between cognitive trajectories and survival outcomes is, to our knowledge, novel. This approach allows for direct testing of whether the relationship between cognition and mortality differs across generations. In this context, it provides a powerful framework for identifying cohort specific mechanisms underlying aging and survival, thereby improving both theoretical understanding and predictive accuracy.
Third, the growth process was parameterized not only in terms of level and rate of change, but also using the area under the individual-specific growth curve (AUC). This representation captures the cumulative cognitive burden over time in a parsimonious way and has not previously been implemented in this context. The AUC approach is particularly advantageous for prediction modeling, as it reduces complex longitudinal information into a single interpretable metric that reflects both level and change, and may better capture processes related to terminal decline and overall functional reserve.
Finally, we applied second-order growth curve models within a structural equation modeling (SEM) framework. This approach enables the modeling of latent cognitive constructs over time while accounting for measurement error and changes in the measurement structure across waves. Such models are rarely used in cohort comparative aging research, largely due to the extensive data requirements. In the present project, the availability of harmonized longitudinal data across cohorts made it possible to implement this approach, providing more reliable estimates of true cognitive change and allowing for more valid comparisons across generations.
Together, these methodological developments enhance the precision, flexibility, and interpretability of analyses of cognitive aging and survival, and offer a framework that can be extended to other longitudinal population-based studies.
Future research questions
The project raises several new research questions. One important direction concerns the use of machine learning approaches to predict cognitive decline, morbidity compression, and survival, and how such models can be implemented in clinical and policy contexts. This project represents one of the few systematic evaluations of cohort differences in morbidity compression, highlighting the need to further investigate its individual, healthcare, and societal consequences.
In addition, several potential moderating factors, such as lifestyle, demographics, and biological risk markers, remain to be examined in a cohort sensitive framework. These questions will be pursued in future work building on the current project.
Dissemination and collaboration
Findings from the project have been disseminated primarily through peer-reviewed scientific journals and conference presentations. The project has also facilitated several collaborations. Locally, collaborations were established within the University of Gothenburg, particularly between the Department of Psychology and the Department of Psychiatry and Neurochemistry (Dr. Skoog). Nationally, the project contributed to collaborations within the NEAR consortium (NEAR – The National E-infrastructure for Aging Research). Internationally, collaborations were developed with researchers at Humboldt University in Germany (Dr. Gerstorf), Ohio University in the USA (Dr. Muniz-Terrera), and Northwestern University (Dr. Graham). Ongoing projects continue to build on these collaborations and extend the research agenda established in this work.