The long-standing national multicenter Cardiovascular Risk in Young Finns Study (YFS) was originally designed to provide evidence on the importance and timing of early-life genetic and environmental exposures in the development of cardiovascular diseases [21]. In 1980, 3596 participants (boys and girls) aged 3–18 years were recruited from five cities and their surrounding rural communities. The original cohort G1 was followed up every 3–6 years. The latest follow-up study was conducted in 2018–2020, in which the data collection was expanded to three familiarly related generations: grandparents (G0; aged 59–92 years), parents (the original YFS participants G1; aged 41–56 years), and offspring (G2; aged 3–37 years). In total, N = 12,853 subjects were invited, and N = 7341 (57.1%) provided data in the first multigenerational YFS field study. Individuals with diagnosed cognitive impairment and/or severe restrictions in movement abilities were excluded from the study. A total of 6753 (52.5%) participants attended the study visit, while 588 participants provided questionnaire data only. The generation-specific participation rates for participants who attended the study visit were 64.2% for the original participants (G1; n = 2064; females 54.8%), 54.5% for the parents of the original participants (G0; n = 2146; females 60.6%), and 44.6% for the offspring of the original participants (G2; n = 2543; females 54.9%).

Cognitive performance was assessed during the clinical study visit. Due to the blood samples included in the study protocol, the participants arrived at the study visit after fasting and having avoided smoking for at least 4 h. They were also instructed to avoid heavy physical activity and drinking alcohol beginning the previous evening before the study visit. Before the cognitive testing, the subjects were provided with a light snack containing a whole meal oat-based snack biscuit, a fruit/berry oat drink, or weak fruit/berry juice. In total, N = 6610 (98%) of the participants who came to the clinical study visit also provided data on cognitive performance. Since the cognitive test battery was modified for 3–6-year-old children, they were excluded from the present study. Hence, the generation-specific numbers of participants who successfully went through the cognitive testing protocol reported in this study were N = 2030 (95%) for the original YFS participants (G1), N = 2025 (83%) for their parents (G0) and N = 2431 (99%) for their children (G2).

Cognitive testing was performed using a computer-based cognitive test battery (CANTAB®). The CANTAB® is a computerized, predominantly nonlinguistic and culturally neutral test focusing on a wide range of cognitive domains. The test was performed using a validated touchscreen computer system. The full test battery includes more than 20 individual tests, from which a suitable test battery for each particular study may be selected. The test battery selected for the extended YFS field study included five separate tests that are sensitive enough to capture variation even within the cognitively healthy cohort. The tests selected for the YFS test battery measure four cognitive domains: (1) visual and episodic memory and visuospatial associative learning, (2) short-term and spatial working memory and problem solving, (3) reaction and movement speed and accuracy, and (4) visual processing, recognition, and sustained attention. The test battery was identical for all participants. A study nurse administered the test to all participants and ensured that there was no misunderstanding related to performing the test. Voice-over instructions were provided by the CANTAB® software in Finnish.

First, the participants completed the motor screening (MOT) test, which measures psychomotor speed and accuracy. In the YFS cognitive testing protocol, the MOT test was considered a training procedure that introduced the test equipment to the participants. Simultaneously, the MOT test was used as a screening tool to indicate any difficulties in vision, movement, comprehension, or ability to follow the test instructions. During the MOT test, a series of red crosses were shown in different locations on the screen, and the participants were advised to touch, as quickly as possible, the center of the cross every time it appeared. The MOT test was identical for all participants regardless of age. After the MOT test, four separate tests each measuring a specific cognitive domain were conducted.

Learning and memory was assessed using paired associates learning (PAL) test, which measures visual and episodic memory and visuospatial associative learning and includes aspects of both delayed response procedures and conditional learning. Either 2, 4, 6, 8, or 12 patterns were displayed sequentially in boxes placed on the screen during the PAL test. After that, the patterns were presented in the center of the screen, and the participants were supposed to point to the box in which the particular pattern was previously seen. The test moves on to the next stage if all the patterns are placed in the right boxes. In the case of an incorrect response, all the patterns were redisplayed in their original locations, and another recall phase was performed. The test terminated if the patterns were still incorrectly placed after 4 presentation and recall phases.

Working memory was assessed using spatial working memory (SWM) test, which is used to measure the ability to retain spatial information and manipulate items stored in working memory, problem solving and self-organized search strategies. During the SWM test, the participants were presented with 3, 4, 6, 8, or 12 randomly distributed colored boxes on the screen. After that, the participants were supposed to search for tokens hidden in the boxes. When a token was found, it was supposed to be moved to fill an empty panel on the right-hand side of the screen. Once the token had been removed from the box, the participant had to recall that the computer would never hide a new token in a box that previously contained one; therefore, the participants were not supposed to revisit the same boxes.

Information processing was evaluated using rapid visual information (RVP) test, which is used to assess visual processing, recognition, and sustained attention. In this test, the participant was presented with three number sequences (3–5–7, 2–4–6, and 4–6–8) next to a large box where the number 1–9 appeared in a random order at a rate of 100 numbers per minute. Whenever any of the particular sequences were presented, the participant was supposed to press a touchscreen button. Altogether, nine target sequences were presented at 100-s intervals during the 6-min assessment phase. During the practice phase, the participant was given visual cues (i.e., colored or underlined numbers) to help him or her recognize the particular sequence. At the assessment phase, the cues were no longer presented.

Reaction time was evaluated using reaction time (RTI) test, which assesses the speed of response and movement as well as accuracy on a task where the stimulus was unpredictable (five-choice location task). Five large circles were presented on the screen, and the participant was supposed to press down a touchscreen button at the bottom of the screen and wait until a small yellow spot appeared in any of the five large circles. When the yellow spot appeared, the participant was supposed to touch the yellow spot as soon as possible with the same hand that was pressing the touchscreen button.

Age was defined in full years at the end of the year 2018. Education years were queried from all participants aged 18 years and older. In Finland, the mean age at which secondary education is completed is 28 years. According to our data, 492 (92%) of the 533 participants who reported studying were at most 28 years old. Thus, we considered the own education of the participants aged 28 years and older. For participants aged 18–28 years, their own education was provided if the participant did not report full- or part-time studying. For all participants aged under 18 years and participants aged 18–28 years who reported studying, parental education years were considered (the maximum years of parental education for those participants with data for both parents). Hence, for adults (excluding students), education reflects self-acquired cognitive reserve, while for under 18-year-old participants and students, parental education reflects socioeconomic status.

The CANTAB® test produced several variables, the detailed information of which is presented in Online Resource 1. We used Flury’s common principal component analysis [22] to derive the principal component scores for (i) across all domains/whole CANTAB® test battery and (ii) separately for each measured cognitive domain/each of the four separate subtests. The main idea of this method is to conduct a principal component analysis for a dataset arranged in multiple groups. This approach allows the groups to have different means, variances, and correlations but assumes that the principal components are the same in those groups. Given that cognitive performance may differ between generations and among participants of different ages, we defined six groups as the input for the analysis: G0, G1, G2 (aged 25–37 years; adulthood), G2 (18–24 years; young adulthood), G2 (13–17 years; adolescence) and G2 (7–12 years; childhood). The G2 generation (children of the original YFS participants) was divided into four age windows because we wanted to recognize the possible variation in cognitive performance within different developmental phases. We standardized the variables of cognitive performance before the analysis and subsequently windsorized a few outlying values (> 10 SDs from the mean) to ± 10 to control for any disproportionate influence. Common principal component analysis was implemented with the multigroup package in R (version 4.1.3). Participants’ first, second, and third principal component scores were extracted, and the first principal component scores were chosen to represent overall cognition as well as cognitive domains representing learning and memory, working memory, information processing, and reaction time. This selection was based on the largest amount of explained variance compared to the following principal components as well as reasonable variable loadings of the relevant key variables within each cognitive test. The sign of each principal component was assigned so that higher principal component scores indicate better test performance (e.g., less errors, shorter latency times, and better accuracy). For reaction time, the distribution of participants’ first principal scores was skewed, while for the other domains, the first principal scores were normally distributed according to visual inspection.

We estimated the means of cognitive performance in different domains across the age range of 7–92 years nonparametrically by Loess smoothing, which uses locally weighted polynomial regression in the estimation of the smooth fit. Visualization of the cognitive performance trajectories was based on these estimates. For the subsequent analyses, the first principal component scores for each domain were standardized such that 20–29-year-old participants (20 ≤ age < 30) had a mean value of 0 and standard deviation of 1, as cognitive performance was observed to be at its highest during this age period based on the visualization. Therefore, the values reported for each generation describe how many standard deviations the given principal component scores differ from those of 20–29-year-old participants. In addition, cognitive performance was visualized in similar manner using the second and third principal component scores, and additional analyses were conducted if visual inspection suggested age-related changes in these components.

The associations between age, sex, and education and overall cognitive performance and the specific cognitive domains were studied using generation-specific linear models. Furthermore, for the G2 generation, models were created for each of the four age groups to capture the increasing developmental trajectory of cognitive performance at a young age. The exposure variables age, sex, and education were entered simultaneously into the cognitive domain-specific models. Age was first linearly represented, after which a quadratic term for age was applied (age2) to allow for a nonlinear association between age and cognitive performance (as suggested by Loess fits). Model residuals were homoscedastic and normally distributed by visual inspection except for reaction time, the distribution was skewed. Therefore, we tested the 1/x transformation for reaction time. Due to easier interpretability, untransformed results are shown in Tables 3 and 4. Linear models were generated using SAS 9.4 (SAS Institute, Cary, North Carolina, USA) software, and the level of statistical significance was set at p ≤ 0.05.

We estimated intergenerational correlations among the first principal component scores representing each cognitive domain by calculating partial Pearson’s correlation (partial Spearman’s rank correlation for reaction time) between G0 and G1, G1 and G2, and G0 and G2. The analyses were stratified for sex and adjusted for age. Hence, we obtained age-adjusted correlations between mothers and daughters, mothers and sons, fathers and daughters, and fathers and sons in G0 and G1 as well as in the G1 and G2 generations. For G0 and G2, sex-specific and age-adjusted correlations were calculated between grandparents and grandchildren. The intergenerational correlation analyses were conducted using the ppcor package (version 1.1) in R. Correlations with p ≤ 0.05 were considered to be statistically significant.