Residential trajectories across the life course and their association with cognitive functioning in later life

Previous work has found that later life urban–rural differences in cognitive health can be largely explained by indicators of cognitive reserve such as education or occupation. However, previous research concentrated on residence in limited, specific, periods. This study offers a detailed investigation on the association between urban (vs. rural) residence from birth, and cognitive functioning in older age. Using data from the Survey of Health Ageing and Retirement in Europe we created residential trajectories from birth to survey enrolment with a combination of sequence and cluster analysis. Using mixed-effects models, we investigated the association between residential trajectories in early, mid, and later life and three cognitive functioning outcomes: immediate recall, delayed recall, and verbal fluency. In a sample of 38,165 participants, we found that, even after accounting for differences related to education and occupation, rural (vs. urban) residence in early life remained associated with poorer cognitive performance later in life. This suggests that growing up in rural regions leads to a long-term disadvantage in cognitive functioning. Thus, public health policies should consider that urban–rural inequalities in early life may have long-lasting associations with inequalities in cognitive health in old and very old age.


Assessment of Cognitive Functioning in SHARE
Immediate and delayed recall were assessed using a list of 10 words [1] read out by the interviewer. For the immediate recall score, respondents were asked to repeat as many words as they could remember right after they were read out to them. For delayed recall, respondents were asked to repeat the same list of 10 words after having responded to other questions assessing cognitive functioning. In the verbal fluency test [2], respondents were tasked with naming as many animals as possible in one minute. The score for this test was the total number of animals named that were validated by the interviewer, without any penalty for repetitions or wrong names. The Pearson correlations between the different outcomes are show in Table S1.

Sequence Analysis & Residential Trajectories
The residential trajectories were coded as sequences using the retrospective information supplied by respondents of the life-calendar, which is part of the SHARELIFE modules of waves three and seven. Respondents were asked about each place residence since their birth. They further provided information on when they began or stopped living at a certain place, and importantly, on the type of area a specific place of residence was in (big city, suburbs or peri-urban area, large town, small town, or rural area).
This information was imported in TraMineR [4] to analyse and classify the trajectories of respondents' area of residence. To cluster the trajectories, we calculated the distance between sequences using a spell-length optimized variant of optimal matching [5] with constant substitution costs. Because there were more than 40,000 sequences complete sequences for all respondents who completed the life history calendar, we used the WeightedCluster R package to select the unique sequences and calculate weights in relation to the number of times each unique sequence appeared in the original data. The clustering itself was done using Ward's method with the fastcluster R package using the calculated weights [6].
The choice of the number of groups was based on cluster quality measures computed with the WeightedCluster R package [7]. According to the cluster quality measures (Tables S2 and S3), the five clusters provided the best solution for midlife trajectories and is the best for early-life trajectories. Therefore, we partitioned the sequences into five clusters for both life phases. After a visual inspection of the sequences contained in each cluster (see Figures S1 and S2), we named the clusters as "big city" (cluster 1 for early-life trajectories and cluster 1 for midlife trajectories), "suburbs" (cluster 4 for early-life trajectories and cluster 3 for midlife trajectories), "large town" (cluster 3 for early-life trajectories and cluster 2 for midlife trajectories), "small town" (cluster 5 for early-life trajectories and cluster 4 for midlife trajectories), and "rural" residence (cluster 2 for early-life trajectories and cluster 5 for midlife trajectories) for both early-life and midlife trajectories. We then re-used the cluster assignments in the final analysis as an explanatory variable. The code for the analyses is available at: https://doi.org/10.17605/OSF.IO/TC8YE

Figure S1 Sequence density plots (chronograms) of early-life trajectories.
For each cluster, the bars represent the proportion of respondents living in one of the five residential areas at a specific age in early life (birth to age 20). Based on this figure we see that cluster 1 mainly contains respondents who lived in a big city in early life, cluster 2 respondents who lived in rural areas, cluster 3 respondents who lived in large towns, cluster 4 respondents who lived in the suburbs, and cluster 5 respondents who lived in small towns.

Figure S2 Sequence density plots (chronograms) of midlife trajectories.
For each cluster, the bars represent the proportion of respondents living in one of the five residential areas at a specific age after early life (age 21 to study entry). Based on this figure we see that cluster 1 mainly contains respondents who lived in a big city, cluster 2 respondents who lived in large towns, cluster 3 respondents who lived in the suburbs, cluster 4 respondents who lived in small towns, and cluster 5 respondents who lived in rural areas.