Introduction

The growing prevalence of cognitive decline among older adults presents a significant public health challenge, with profound implications for individual quality of life, family relationships, and the allocation of healthcare resources. While traditional perspectives have considered cognitive decline as an inevitable consequence of aging1,2, recent research challenges this notion, suggesting vast inter- and intraindividual heterogeneity of cognitive trajectories in later life3. The diversity in cognitive outcomes, reflected in some individuals experiencing rapid decline and others sustaining robust cognitive function, illustrates the complex and varied cognitive trajectories observable among older adults. However, as4,5,6 highlight, the complexity of patterns of these cognitive trajectories remains insufficiently understood, as even when controlling for genetic, lifestyle and environmental influences, differences still arise, leaving a gap in our knowledge of the cognitive aging process.

The investigation into cognitive trajectories during aging is critical for several reasons. Firstly, it enables researchers to evaluate the impact of significant life events on cognitive performance7. For example,7 highlights how discrete events, such as retirement, may contribute to changes in cognitive functions by meaningfully altering daily routines and social interactions. Supporting this,8 found that in older individuals, reductions in social interactions are associated with lower cognitive performance. Secondly, it allows for early detection of cognitive decline to implement timely interventions that could halt or reverse cognitive deterioration, as emphasized by7,9,10. Early identification of cognitive decline can allow for cognitive training or lifestyle modifications, which can significantly improve cognitive outcomes11. Lastly, it has the potential to reveal non-linear cognitive patterns over time, enhancing our understanding of the cognitive aging process12. 12 suggest that cognitive changes may involve periods of stability, improvement, or rapid decline. As such, longitudinal studies provide valuable insights into these patterns, helping researchers identify critical periods for intervention and to understand the underlying mechanisms of cognitive aging.

The reasons described above highlight the value in examining the inter- and intraindividual differences that contribute to the observed heterogeneity in cognitive performance among the aging population. Interindividual variability reflecting the possible differences in cognitive performance between two or more individuals, and intraindividual changes, referring to the variability in cognitive performance over time, highlight the complexity of the cognitive aging process13.

Despite the recognized importance of capturing these types of variability, effectively studying them remains a challenge. The methodologies traditionally employed in the study of cognitive trajectories have often been sub-optimal in capturing heterogeneity, as14 critique. Researchers have focused on average cognitive trajectories or subgroup analyses, which fail to account for the substantial variability in cognitive trajectories15,16,17. A viable alternative may be the identification of clusters of participants with similar cognitive trajectories, which could more accurately reflect the differences in cognitive performance over time within the aging population12. This alternative methodological approach could capture key drivers of cognitive trajectories with reduced inherent bias, providing a more nuanced and accurate representation of cognitive aging18,19,20.

Different methods have been applied for profiling cognitive trajectories. For example, 21,22,23 employed linear mixed models, growth mixture models, and latent class analysis, respectively, to identify distinct profiles in aging populations, to consider both inter- and intra-individual variability. However, despite investigating different outcomes, memory function21, episodic memory and verbal fluency23, and Mini-Mental State Examination (MMSE) scores22, these studies identified profiles with linear trajectories. In contrast, researchers such as24 used growth curve models to uncover three nonlinear trajectories of functional independence, based on the degree of assistance required. While these models allow for the investigation of inter- and intra-individual variability, they often encounter challenges with convergence when modeling nonlinear trajectories. Furthermore, even when advanced methods are used, studies frequently report simplified linear trajectories or average within-group results due to the inherent complexity of interpreting nonlinear trends. Therefore, it is critical not only to select methodologies that can address complex population heterogeneity but also to ensure these models can also lead to interpretable results.

Due to these methodological limitations, the literature on cognitive trajectories in aging is characterized by conflicting findings. Some, such as25,26,27,28, argue for predominant stability in cognitive function, viewing decline as a sign of pathology. In contrast, a few recent findings such as those from29,30 documented heterogeneity even after accounting for pathological factors. Authors such as 31,32 add to this complexity by suggesting that in addition to the heterogeneity in cognitive aging present at between- and within-individual levels, it is also present in the interactions between individuals and their environments. They propose that individuals may respond to similar environmental or lifestyle factors in markedly different ways, highlighting the need to consider both individual environmental contexts and the differences in these interactions.

Given the individualized responses to environmental factors, it is essential to understand how these factors shape cognitive trajectories in aging. The concept of individual environmental opportunities and constraints, as discussed by33, offers a framework for understanding the multifaceted nature of these influences. This framework posits that cognitive trajectories result from inherent individual differences and are significantly shaped by the availability and quality of environmental resources and challenges. Research by19,34,35,36, highlight the potential of lifestyle modifications to support cognitive performance, and discuss how social determinants of health, such as education level, socioeconomic status, and access to healthcare may mediate this relationship. In particular, 37 emphasize that low education is directly associated with low socioeconomic status, which, in turn, is linked to poorer healthcare access and, consequently, lower cognitive performance. Further,38,39 review how risk and protective factors contribute to cognitive changes, identifying fourteen factors that may be responsible for over 40% of worldwide dementias. Using the Health and Retirement Study (HRS), 40,41 highlight that social deprivation was negatively associated with longitudinal cognitive function. Using this same database, 42 identified that physical activity initiated during midlife may reduce cognitive decline in older adulthood. These findings suggest that the influence of such factors may be observed across cognitive trajectories, with individuals in steeper cognitive decline trajectories being more strongly associated with detrimental factors such as smoking43, while those maintaining higher cognitive performance being more closely linked to protective factors such as education and physical activity22,44. These insights underscore the importance of considering both individual and environmental factors in understanding cognitive aging trajectories.

Despite the promise of these findings, the literature on lifestyle factors and cognitive performance in healthy older adults is fraught with limitations, including methodological issues, small effect sizes and inconsistent results45,46,47,48,49. These discrepancies may stem from methodological shortcomings that also fail to account for the aging population’s heterogeneity, often relying on small, non-representative samples and designs that do not capture the long-term nature of cognitive change50,51. Further, while a range of lifestyle and environmental factors have been linked to improved cognitive aging52,53, most studies focus on their relationship with baseline cognitive performance, assuming that cognitive aging is a uniform process (e.g., 54,55). This overlooks potential heterogeneity in how individuals’ cognitive trajectories respond to such factors over time.

To address these challenges, our study seeks to assess the heterogeneity in cognitive performance over time among older adults by identifying representative patterns of cognitive trajectories. Further, we explore the influence of lifestyle factors on these trajectories. By analyzing longitudinal data from 1746 individuals aged 60 and over at three time points across eight years, we aim to identify characteristic cognitive trajectories and examine the interplay of lifestyle factors that shape them. Building on prior research, we hypothesize that 1) Distinct cognitive performance trajectories will emerge within the aging population, with some individuals showing faster rates of cognitive decline, while others maintain stable or higher levels of cognitive function over time; and 2) The sensitivity of cognitive trajectories to environmental factors will differ. The group experiencing the fastest cognitive decline will be more strongly affected by detrimental factors, the group with the highest cognitive performance will be more positively influenced by beneficial factors, and the middle group will show a moderate response to both beneficial and detrimental factors over time.

Public significance statement

Challenging the one-size-fits-all approach, this study reveals significant heterogeneity in cognitive aging, identifying distinct trajectories uniquely influenced by lifestyle factors. These findings highlight that depending on the individual trajectories followed, different needs will be required to mitigate cognitive decline. This study undersores the need for a paradigm shift that considers personalized interventions tailored to individual needs. By embracing this heterogeneity, we can meaningfully improve public health strategies that effectively promote cognitive health across the aging population.

Methods

Software and dataset

All analyses were conducted using R and RStudio for group-based trajectory modeling (R packages used included traj and lcmm), Python for multinomial logistic regression analysis (Python libraries used included scikit-learn and statsmodels), and Tableau for data visualization.

We analyzed data from the Health and Retirement Study (HRS, available at https://hrs.isr.umich.edu/about), a longitudinal database that collected comprehensive information on employment history, disability, retirement plans, and net worth, income, and health of the U.S. population over the age of 50 56. Our objective was to leverage the rich data on lifestyle factors provided by the HRS to investigate their associations with distinct cognitive trajectories in aging. To achieve this, we extracted information on cognitive performance and lifestyle factors from the Preload, Physical Health, Leave-Behind Questionnaires, and Cognition modules of 2012, 2016, and 2020 waves of the HRS. We selected individuals who were present in all three waves to enable a longitudinal analysis of cognitive performance and lifestyle factors over time.

Participants and selection criteria

In total, our representative dataset consisted of 57,189 observations from 25,180 participants, encompassing 36 independent variables (see Table 1). The HRS sample design oversamples African American and Hispanic populations to support research on racial and ethnic disparities, successfully recruiting and retaining these groups57. To further expand minority participation, additional screening was implemented in the 2016 and 2022 cohorts, resulting in a sample of approximately 22,500, including 4,700 African American and 2,600 Hispanic participants. Given our research question and hypothesis, a substantive portion of the independent variables were focused on participants’ evaluations of their life circumstances, subjective well-being and lifestyles. We obtained this information from a specific portion of the database called the “leave-behind questionnaire”, which was completed by rotating 50% of core participants during each biennial wave58. As a result, this data was not available for every participant at every wave, and its lower response rates compared to the core survey limited the availability of complete datasets. To ensure sufficient participants and data, we focused on the 2012, 2016 and 2020 waves, which provided the most robust and recent data for our analysis. Additionally, we adhered to methodological recommendations by12, which highlight that a minimum of three time points is necessary to accurately model cognitive trajectories. The choice of 2012, 2016 and 2020 meets this criterion and allows for sufficient temporal spacing that captures meaningful changes in cognitive performance, while minimizng the effects of participant attrition over time. These intervals align with the approximate 4-years gap in usable data from the “leave-behind questionnaire”, allowing for the investigation of cognitive aging trajectories and their relationship to distinct lifestyle and contextual factors.

Table 1 List of binary covariates and proportion of  “yes”.
Full size table

We excluded respondents and variables that had over 20% of missing data, resulting in a final dataset of 5,238 observations corresponding to 1,746 respondents (1,746 × 3 = 5,238). As part of our selection criteria, we included individuals above the age of 60 (\(\text{mean age}=67.0; \text{s.d.}= 11.58; \text{mean education}=12.7; \text{s.d.}= 3.24\)). The sample included in this study had 1,056 female participants (60.4%). Participants aged 80 or older comprised 19.6% of the sample (see Supplementary Fig. 1 for the distribution of the Age variable in our sample). Educational attainment was used as a proxy for SES in this sample because, unlike alternative income variables, it did not require creating a composite score59. It shows that 13% of respondents did not complete high school, which typically correlates with lower SES, while 53% had some college or higher education, often associated with higher SES due to greater earning potential and career opportunities. Lastly, all included participants were considered healthy, presenting no neuropathological diagnosis. Further, health satisfaction in the sample indicates that 53.3% of respondents report being satisfied with their health, 32.8% rate their health neutrally, and 13.9% report dissatisfaction with their health.

Cognitive performance

We computed a metric of an episodic memory test to assess cognitive performance. This metric consisted of adding the total number of words correctly recalled by the respondents during the immediate and delayed recall tasks across all waves (eg.,60,61). Participants were randomly assigned one of four standardized word lists, each containig 10 words. Participants were asked to recall the words immediately and after a short delay within the same session. The composite score represents the sum of correctly recalled words during the immediate and delayed recall tasks, with scores at each time point ranging from 0 to 20. This composite measure provided a more nuanced evaluation of memory function compared to the original variables, which ranged from 0 to 10. In addition, the obtained composite scores do not suffer from ceiling or flooring effects, i.e., a skew in the distribution due to the level of difficulty in the test leading to most of the participants scoring either extremely good or extremely poor.

Episodic memory was selected as the cognitive performance measure for this study as it is a highly sensitive indicator of cognitive aging. Research has consistently shown that episodic memory is particularly vulnerable to age-related decline and has therefore been used as a proxy for cognitive changes in aging populations (e.g.,32,62,63,64). Accordingly, the term “cognitive trajectories” in this manuscript refers to patterns of cognitive change over time, with episodic memory serving as the representative domain for this investigation. It does not imply a comprehensive assessment of all cognitive domains but rather reflects a well-validated approach for studying variability in cognitive aging.

Covariates

We conducted a literature review to identify lifestyle factors that are commonly associated with cognitive performance (see Supplementary Table 1 for details). We selected 36 lifestyle factors, which have been previously used in an analysis on the same database, which were found as relevant to cognitive performance32. These factors were drawn from the HRS database, which includes a variety of question formats tailored to different types of queries. Specifically, the covariates were collected in two main formats: binary and Likert scales with varying response options. For example, questions such as “How often do you do physical activity?”, “How often do you walk for 20 min or more?”, and “Do you have ongoing financial strain?” illustrate the range of question structures. Responses to these questions include Likert scales (e.g., “Never,” “Hardly Ever,” “Sometimes,” “Often,” “Every Day”, and “Daily,” “Several times a week,” “Once a week,” “Several times a month,” “At least once a month,” “Not in the last month,” “Never/Not Relevant”), and binary responses (e.g., “No, didn’t happen,” “Yes, but not upsetting,” “Yes, somewhat upsetting,” and “Yes, very upsetting.”).

To simplify the analysis, all variables, except age and education, were transformed into binary values. Responses such as “Often” or “Every Day” were re-labeled as 1, while “Sometimes,” “Hardly Ever,” or “Never” were re-labeled as 0. Similarly, “Daily” or “Several times a week” were re-labeled as 1, while less frequent responses were re-labeled as 0. Blank responses were retained as missing values. For binary questions, all “Yes” responses, were re-labeled as 1, and “No” responses were re-labeled as 0. These transformations ensured consistency across variables, facilitating analysis while maintaining the integrity of the data. A summary of the binarized covariates is presented in Table 1.

Our primary analysis focused on assessing the relationship between baseline covariates (2012) and cognitive trajectories. To verify the robustness of our findings, we replicated the analysis using covariates assessed at 2016 and 2020, providing a longitudinal perspective on lifestyle-cognitive trajectory associations. These verification results can be found in the supplementary material document.

Group-based trajectory modelling

We employed a previously developed statistical approach to categorize all participants into distinct clusters based on similar patterns of age-related cognitive performance changes65. Specifically, we applied a group-based trajectory modeling (GBTM) to identify cognitive ageing trajectories, using cognitive performance measures collected at three time points. GBTM identifies latent subgroups within a population, each characterized by a unique trajectory of change over time, rather than assuming a single homogeneuous population or focusing solely on individual variability. For each participant, we calculated the probability of belonging to each trajectory group based on their cognitive scores over time and fitted a quadratic model for each group to estimate parameters such as intercepts, linear and quadratic terms, group size, and probabilities of group membership. The number of latent groups was not determined a priori, instead, we tested models with increasing numbers of groups and selected the optimal solution using Bayesian Information Criteria (BIC) and Akaike Information Criteria (AIC) scores66. Each participant was then assigned to the cognitive trajectory group for which they had the highest probability of membership.

This approach has been previously employed in cognitive aging research. For example, 67 identifed MMSE trajectories and examined associated risk factors such as age, gender, and comorbidities in a healthy older adult population.68 explored how levels of engagement in social and intellectual activities influence cognitive trajectories. More recently,19 extended GBTM to model dual trajectories, investigating the relationship between physical activity and cognitive function during aging. GBTM is particularly valuable in this field as it captures heterogeneity in cognitive trajectories by identifying subgroups within populations that are highly variable, and where trajectories of cognitive decline are not uniform. This is not possible with traditional methods such as Linear Mixed Models which assume a single population-wide trajectory with individual variability around it.

Additionally, GBTM provides a flexible, data-driven approach to subgroup identification, allowing the number and characteristics of the trajectory groups to emerge from the data, rather than requiring a priori assumptions. By defining distinct group-level trajectories, GBTM provides insights into population heterogeneity, assuming minimal variability within each subgroup, with residual differences representing deviations from the shared group trajectory. This makes GBTM especially suited for understanding the complexity of cognitive aging trajectories in heterogeneous populations.

Multinomial logistic regression analysis

After categorizing all participants into distinct cognitive performance groups, we applied a multinomial logistic regression model to examine the associations between baseline lifestyle factors (scores from 36 independent variables in 2012) and the identified cognitive trajectories. The model compared the likelihood of belonging to each cognitive trajectory by calculating the log-odds ratios of being in one group versus another. Specifically, we compared the odds of being in the unmodulated cognitive change or high cognitive performance with late decline trajectories relative to the low cognitive performance with early decline trajectory. This can be expressed as:

$$\log\left(\frac{\text{Probability (Unmodulated Cognitive Change)}}{\text{Probability (Low Cognitive Performance with Early Decline )}}\right)$$

and

$$\log\left(\frac{\text{Probability (High Cognitive Performance with Late Decline )}}{\text{Probability (Low Cognitive Performance with Early Decline)}}\right)$$

Marginal effects were then computed for each independent variable to determine the direction and strength of influence on the probability of remaining in or transitioning out of a specific cognitive trajectory. These effects reflect the likelihood of moving between cognitive trajectories as each independent variable changes, while holding all other variables constant.

False discovery rate (FDR) correction

Given the inclusion of multiple covariates in our analysis, we applied a FDR correction to control for the increased likelihood of false positives arising from multiple hypothesis tests. To adjust the p-values, we used the Benjamini-Hochberg (BH) procedure69, which controls the expected proportion of false discoveries (incorrectly rejected null hypotheses) among all significant results. Details on how the BH was implemented can be found in the supplementary material of this manuscript.

Overall, the BH adjustment helps ensure that p-values are corrected for multiple comparisons while keeping the proportion of false discoveries under control. A p-value was considered significant if the adjusted p-value was below the pre-specified threshold of \(\alpha =0.05\). By applying this procedure, we ensured that the overall rate of false positives remained controlled, even with a large number of covariates in the analysis.

Ethics statement

The current study, which complies with the provision of TCPS2 (2018), was exempt from the requirements of Research Ethics Review.

Results

Significantly distinct representative cognitive trajectories

To investigate the association between lifestyle factors and cognitive performance, we employed a cluster analysis approach to identify distinct cognitive trajectories in aging. Specifically, we used a Group-Based Trajectory Model (GBTM) to cluster participants based on their cognitive scores as a function of age. To determine the optimal number of clusters, we evaluated several models using both Byesian Information Criteria (BIC) and Akaike Information Criteria (AIC). We found that three clusters provided the best fit (for more info, please refer to the Supplementary Material – Supplementary Figs. 1 and 2). Each participant was attributed three probabilities, each indicating how strongly they belonged to one of the three groups. We then assigned each individual to the cluster with the highest probability. The resulting clusters, which we refer to as trajectories of low cognitive performance with early decline (in red), unmodulated cognitive change (in yellow) and high cognitive performance with late decline (in blue), are illustrated in Fig. 1. This figure shows the estimated cognitive performance scores as a function of age, where the x-axis represents age and the y-axis represents cognitive performance.

Fig. 1
figure 1

Trajectories of cognitive performance clustered in three significantly distinct groups representing Low Cognitive Performance with Early Decline (Red), Unmodulated Cognitive Change (Yellow) and High Cognitive Performance with Late Decline (Blue).

Full size image

Although the trajectories in Fig. 1 have similar curvatures, they differed in the dynamics of overall increases and decreases in cognitive performance with age, as illustrated by Table 2. Specifically, the trajectory of low cognitive performance with early decline experienced a decline in cognition of about 27% from the ages of 60 to 80 years old, which increased by approximately 49% between the ages of 80 to 90. The trajectory of unmodulated cognitive change experienced a decline of about 10% from the ages of 60 to 80 and a decline of 23% from the ages of 80 to 90. Lastly, the trajectory of high cognitive performance with late decline demonstrated a decline in cognition of approximately 7% and 17% for the age periods of 60 to 80 and 80 to 90 years, respectively (please refer to Eqs. 5, 6, and 7 of supplemenatry material for details on the obtained equations for each trajectory of cognitive performance). The proportion of individuals in each group was similar for the trajectories of low cognitive performance with early decline and high cognitive performance with late decline , both comprising approximately 20% of the population (17% (N = 297) and 20% (N = 264), respectively), while the trajectory of unmodulated cognitive change contained 62% (N = 915) of the individuals in our sample.

Table 2 Percentage of change in cognitive performance between the ages of 60—80, and 80—90.
Full size table

Association of lifestyle factors specific to each cognitive trajector

We further investigated each cognitive trajectory by examining the impact of baseline lifestyle factors (wave 2012). To explore the association of selected lifestyle factors (see Table 1) on the distinct cognitive performance trajectories, we applied a multinomial logistic regression analysis (Fig. 2). Further, we performed an analysis of average marginal effects to determine how changes in lifestyle factors affect the likelihood of remaining or not in a given cognitive trajectory (Fig. 3).

Fig. 2
figure 2

Log-Odds of Significant Lifestyle Factors for Membership in Unmodulated Cognitive Change or High Cognitive Performance with Late Decline Trajectories, Compared to Low Cognitive Performance with Early Decline.

Full size image
Fig. 3
figure 3

Average marginal effects obtained from the multinomial logistic regression model for each cognitive trajectory group.

Full size image

The regression analysis revealed that from the initial 36 covariates, six were significant (see Supplementary Table 2 for full table of results). After applying the p-value correction, the number of significant factors was reduced to four. Given the comparative nature of the multinomial logistic regression, the coeficient associated to the significant variables reflects the likelihood of being in a given trajectory compared to the reference trajectory (trajectory of low cognitive performance with early decline ). As observed in Fig. 2, the factors that positively impact the trajectory of unmodulated cognitive change and display stronger statistical significance include Often Use Computer and Years of Education (\(p\le 0.001\)). The factors that positively impact the trajectory of high cognitive performance with late decline and display stronger statistical significance include Often do Word Games (\(p\le 0.05\)), and Often Use Computer and Years of Education (\(p\le 0.001\)). Age (\(p\le 0.001\)) had a negative coefficient associated to it, reflecting that increased age is associated with a greater likelihood of being in trajectory of low cognitive performance with early decline rather than the trajectory of high cognitive performance with late decline . For the factors that did not display any statistically significant impact on the cognitive trajectories, please see the Supplementary material. Additional corroborating references are included for each covariate in the supplementary material.

The marginal effect analysis (Fig. 3 and Supplementary Table 3) showed that, in the trajectory of low cognitive performance with early decline , two covariates were significant (Often Use Computer and Years of Education). In the trajectory of unmodulated cognitive change, no factors were significant. The trajectory of high cognitive performance with late decline exhibited three significant covariates (Years of Education, Often do Word Games, and Age). Significant covariates contributed either positively (when the magnitude of impact is greater than zero) or negatively (when the magnitude of impact is less than zero) to an individual’s cognitive performance trajectory. For example, Often Use Computer, represented in light green in Fig. 3, was negatively associated with the low cognitive performance with early decline trajectory. This indicates that using a computer contributes negatively to an individual being in this trajectory (and as such, the more an individual uses a computer, the less likely they are to be in the trajectory of low cognitive performance with early decline). Opposingly, the positive association observed in the trajectory of high cognitive performance with late decline with respect to using a computer, indicates that doing so contributes positively to an individual remaining in this cognitive performance trajectory. Similar interpretations can be made for the remaining factors.

Our results replicated the main literature findings, suggesting that significant independent variables may express specific associations with the different cognitive trajectory groups. For example, increases in Age is negatively associated with the trajectory of high cognitive performance with late decline. Additionally, having more Years of Education is positively associated with the trajectory of high cognitive performance with late decline and negatively associated with the trajectory of low cognitive performance with early decline. Often Do Word Games is positively associated with the trajectory of high cognitive performance with late decline.

Finally, results in Fig. 3 demonstrate that the trajectories of low cognitive performance with early decline and high cognitive performance with late decline exhibited the most pronounced impact of external influences, resulting in cognitive performance trajectories that varied in amplitude and decline rates. Conversely, our results identified a third group resistant to external influences over time, resulting in a trajectory of unmodulated cognitive change. This group of individuals appears to be the least sensitive to the effects of lifestyle factors, both in terms of number of significant factors and the magnitude of impact. We have provided these additional results in the supplementary material (see Supplementary Table 3) accompanying this work. To assess the robustness of these results, we conducted several verification analyses, including an evaluation of variables across cognitive trajectories and examination of marginal effects for baseline lifestyle factors at waves 2016 and 2020. The results indicate that the central trajectory had the fewest significant associations with covariates in both years compared to the other two trajectories, confirming the patterns observed in the main analysis. Detailed results can be found in Supplementary Tables 4, 5, 6, and 7.

Discussion

Challenging the conventional view, our study contributes to the growing evidence that cognitive aging is not uniform, but instead shaped by individual life experiences. The primary contribution of this study is the identification of distinct cognitive performance trajectories – low cognitive performance with early decline, unmodulated cognitive change and high cognitive performance with late decline – and their differential responsiveness to environmental influences (environmental susceptibility) over time, highlighting heterogeneity in cognitive aging. These findings confirm our first hypothesis which posits that distinct cognitive performance trajectories will emerge within the aging population characterized by different rates of cognitive decline.

Further, two of the identified cognitive trajectories, low cognitive performance with early decline and high cognitive performance with late decline, were more strongly associated with external influences (identified by the increased magnitude of effects observed in these trajectries in Fig. 3), and presented differences in amplitude and decline rates (as observed in Fig. 1). Conversely, the trajectory of unmodulated cognitive change (see Fig. 1), exhibited weaker associations with external influences over time (see reduced magnitude of effects in Fig. 3). This cognitive heterogeneity adds complexity to the traditional view of a normative cognitive decline with age (as suggested by authors such as 7,70,71, suggesting that trajectories may vary significantly based on individual responses to environmental influences.

In specific, our research has revealed that individual-specific responses to environmental exposures can shape cognitive trajectories, confirming and expanding on our second hypothesis. This hypothesis suggests that the sensitivity of cognitive trajectories to environmental factors will vary: the group with the fastest cognitive decline is more strongly associated with detrimental factors, the group with the highest cognitive performance benefits more from positive factors, and the middle group shows a moderate response to both positive and negative associations over time.

  • Low Cognitive Performance with Early Decline: This trajectory (marked by an amplitude of 7.62 and a maximum percentage of change of 49%) was characterized by a rapid decrease in cognitive performance, reaching a threshold of probable pathology at age 82. As illustrated in Fig. 3, factors such as computer use and years of education negatively impact the likelihood of remaining in this trajectory, suggesting that a lack of engagement in these activities accelerates cognitive decline for those following this cognitive trajectory. This suggests that individuals on the low cognitive performance with early decline trajectory are more susceptible to the benefits of concentration activities and formal schooling. It is possible that it may not be “computer use” itself that contributes negatively to the trajectory of low cognitive performance with early decline. Instead, “computer use” might reflect underlying factors such as social interaction through the use of social media. In fact,72 discuss that older adults that experience a reduced social network and become less engaged with the community may be more likely to present increased rates of cognitive decline.

  • High Cognitive Performance with Late Decline: This trajectory (marked by an amplitude of 13.85 and a maximum percentage of change of 17%) shows maintenance of high cognitive performance in the observed period, remaining clear of pathology throughout, as seen in Fig. 1. Figure 3 suggests that engaging in activities such as word games and increased years of education, is positively associated with the permanence in this trajectory. However, increased age was negatively associated with the likelihood of maintaining high cognitive performance. This indicates that cognitive engagement through specific cognitively stimulating activities such as word games, can help sustain high cognitive abilities, despite aging (Fig. 3). Authors such as 29,73 confirm our findings that higher education is associated with this trajectory, and discuss that persons in this group may engage in healthier lifestyles than their less educated counterparts. Further,74 also discuss how productive engagement in activities that require sustained mental effort (such as word games), contribute to this trajectory. In fact, the authors highlight that interventions that rely on sustained cognitive challenges (and novelty) will be more facilitative of improved cognitive performance when compared to non-cognitively challenging activities74.

  • Unmodulated cognitive change: In Fig. 1 this trajectory is represented by a moderate decline in cognitive performance, with a maximum change of 23% (and amplitude of 10.63). The cognitive performance of individuals in this trajectory gradually decreases with age, approaching the pathology threshold at age 100. The response to lifestyle factors of individuals following this trajectory is absent, underscoring its weaker associations with environmental influences (see Fig. 3). This suggests that for those in the trajectory of unmodulated cognitive change, cognitive performance may be associated with factors less susceptible to lifestyle interventions, potentially including genetics. 32,75 identified a subgroup of the older adult population whose cognitive performance appeared to be unaffected by alterations in the environment in distinct cross-sectional studies. 32 discuss the environmental influence on cognitive performance at the behavior level, whereas 75 discuss this interaction at the genetic level. The current study builds on these findings by identifying a subgroup of individuals whose longitudinal cognitive trajectory appears to be resistant to environmental factors.

The results presented in this work suggest that the type of factor and the magnitude of association vary across trajectories, as opposed to what was previously suggested by authors such as 76 that consider a similar influence of lifestyle factors on the whole population. Given the absence of a uniform influence, it is possible that the heterogeneity observed in cognitive trajectories across the population, may reflect differences in how lifestyle choices are associated with cognitive outcomes over time . The extent of this association varies across different cognitive performance trajectories, and is characterized by their unique responses to environmental influences over time16. For instance, while social engagement is generally associated with better cognitive performance (eg.: 47,77), its benefits, represented through computer use, are most pronounced in individuals with a trajectory of low cognitive performance with early decline. This group shows greater potential for cognitive improvement as opposed to individuals in other trajectories. Also, concentration activities are linked with better cognitive performance (e.g.: 79,80). However, this association is most pronounced in individuals with a trajectory of high cognitive performance with late decline. Individuals in other trajectories appear to have negligible responses to this activity. Recognizing the trajectory-specific association of lifestyle factors on cognitive performance, could help clarify inconsistent findings in the literature35,38.

The dichotomy in responsiveness to environmental factors in the context of cognitive trajectories in aging is a novel finding that expands our understanding of environmental susceptibility in cognitive aging and underscores the need to tailor future research and interventions to individual differences32. Drawing from developmental science theories81,82, the trajectories of low cognitive performance with early decline and high cognitive performance with late decline trajectory appear to be more strongly associated with their environment, experiencing greater losses or gains, based on their lifestyle choices. Conversely, the trajectory of unmodulated cognitive change, is more weakly associated with both negative and positive lifestyles factors included in this study. 16,82,83 use the terms “orchids” and “dandelions” as analogies to describe individual differences in environmental sensitivity. “Orchids” represent individuals who are highly sensitive and vulnerable to their surroundings, while “dandelions” represent those who are more resilient and less affected by environmental factors. The trajectory-specific differences observed in our study align with this analogy, with individuals in the low cognitive performance with early decline and high cognitive performance with late decline groups demonstrating 'orchid-like’ associations with their environment. These individuals experience pronounced cognitive gains or losses based on external factors, as shown in Figs. 1 and 3. Conversely, the 'dandelion-like’ unmodulated cognitive change trajectory remains relatively stable, regardless of environmental influences.The identification of clusters of cognitive performance trajectories in aging can lead to the capture key differences in how (and why) people change in cognition over time13. These findings highlight the importance of considering heterogeneity in cognitive aging trajectories as the individual-specific environmental sensitivity can inform the development of targeted cognitive health strategies 84,85. Authors such as 47,86,87 suggest that interventions aimed at improving and maintaining cognitive engagement are valuable for the cognitive performance of older adults. We build on these recommendations by emphasizing the need for interventions that are tailored to the interaction between environmental influences and individual cognitive trajectories 88,89.

Further, the observed differences may not be only a result of environmental sensitivity, but could also be influenced by genetic factors90. Genetic variability may contribute to how individuals respond to their environments, as environmental sensitivity itself may have a genetic basis (for a more comprehensive review, see 91). This genetic variability could help explain differences in cognitive trajectories by influencing reponses to both positive (e.g.: cognitively engaging tasks) and negative (e.g.: financial strain) environmental conditions75,81,90,92. In the current study, individuals in low cognitive performance with early decline and high cognitive performance with late decline trajectories may exhibit genetic traits that allow for greater cognitive variability in response to environmental factors. These individuals might be better equipped to respond to positive changes such as social and intellectual engagement, but also more susceptible to negative changes, such as sedentarism and stress, due to this genetic expression. Conversely, individuals following a trajectory of unmodulated cognitive change may exhibit less sensitivity to negative conditions, potentially offering resistance to detrimental factors. However, this resistance might also prevent them from benefiting from positive environmental changes92. While our study does not directly assess genetic influences, acknowledging genetic variability as a potential driver, aligns with broader research on individual differences in cognitive outcomes (e.g.: 75,93,94). Future research may consider investigating the role of individual genetic backgrounds in shaping cognitive trajectories in aging.

In conclusion, by identifying distinct trajectories, including one largely unaffected by environmental factors, our study provides a novel perspective on cognitive aging by emphasizing its complexity as a heterogeneous process shaped by individual responsiveness. This challenges the traditional view of a uniform cognitive decline with age and highlights the importance of considering individual-specific responses to environmental exposures. The identification of trajectories of low cognitive performance with early decline, high cognitive performance with late decline and unmodulated cognitive change underscores the meaningful heterogeneity in cognitive aging and the significant role of lifestyle and genetic factors. Our findings suggest that cognitive aging trajectories are associated with the relationship between individual environmental sensitivity and lifestyle factors, potentially contributing to the need for considering trajectory-specific approaches in cognitive health research, rather than assuming a one-size-fits-all solution. For example, individuals identified as following the low cognitive performance with early decline trajectory might benefit from research exploring whether interventions that promote social engagement or participation in cognitively stimulating activities could be effective. Similarly, maintaining beneficial behaviors and minimizing detrimental ones could be particularly relevant for individuals in the high cognitive performance with late decline group preserve their cognitive performance. While our study did not include an intervention component, these examples highlight the need for future studies that investigate whether tailoring strategies to address specific needs could be beneficial. However, it is important to acknowledge that engagement in everyday activities does not necessarily translate to the same effects when structured interventions are introduced 95.

Further research should explore the biological and genetic markers associated with these trajectories to better understand the underlying mechanisms of cognitive aging. Such an understanding may help determine whether interventions tailored to cognitive trajectories could be effective. This distinction may be critical for policymakers and clinicians aiming to enhance the well-being of older adults, highlighting the need for personalized strategies in cognitive health maintenance and improvement.

Limitations

The selected methodological approach can only establish correlational relations rather than causational ones. While some covariates were observed to correlate with specific trajectories of cognitive aging, these findings should be interpreted as associations rather than direct effects. To better understand the causal relationships underlying cognitive trajectories, future research should incorporate clinical or intervention trial methodologies. Further, we recognize that individuals who participate in the HRS longitudinal large-scale study are relatively healthy, educated, and active 96,97 which may limit the generalizability of our findings. Furthermore, the collection frequency of lifestyle factors (every four years for the same group of individuals) contributes to increased attrition rates, which could bias the sample towards more motivated and higher-functioning individuals. Future work may benefit from the inclusion of an additional number of time points as a means to better identify representative patterns of cognitive trajectories. Expanding the scope to include individuals with greater health and socio-economic challenges, could improve representativeness of the findings.

In addition, this study used episodic memory as a proxy for broader cognitive trajectories, focusing on heterogeneity in cognitive aging patterns rather than a domain-specific analysis. While episodic memory is a sensitive and well-validated marker of age-related cognitive decline, it does not encompass all cognitive domains. Future research could include multiple measures to provide a more comprehensive view of cognitive aging98. Also, a limitation of the methodological approach employed is that baseline cognitive scores were inherently part of the trajectory identification process. Including baseline scores as a separate covariate in subsequent analyses would risk over-adjustment. Future research might consider alternative modeling approaches to separately examine baseline cognitive levels and isolate within-person changes. Further, it would also be valuable to explore the underlying mechanisms and better understand how concentration-related activities support cognitive performance.

Lastly, a limitation of this study is that the final data collection wave occurred in 2020, during the COVID-19 pandemic, which may have influenced cognitive performance through reduced social interaction, increased stress, and changes in daily routines (e.g., 99,100,101). While our analysis captures long-term cognitive aging patterns, the 2020 data may reflect pandemic-related effects. Future studies should account for such external crises when interpreting aging trajectories.