https://orcid.org/0000-0001-6296-0486
Abstract
Superagers are 80 to 89-year-olds with average or better cognition and memory equivalent to individuals 20 to 30 years younger. As sex and modifiable lifestyle/health factors influence cognitive aging and dementia risk, we examined their impact on superager status. Data from participants (n = 469; 67% female) aged 80–89 years old were analyzed from an online database that included demographic and dementia risk factors, and performance on tasks assessing working memory, cognitive inhibition, associative memory, and set shifting. Cross-sectional comparisons were made between superagers and those with typical-for-age cognitive abilities (typical-agers) to examine relationships between sex, superager status, and dementia risk factors. Females performed better than age-matched males on the associative memory task in the 50–69 years old group used for normative comparisons, and in the 80–89 years old group (ps < .001). More females than males were classified as superagers using non-sex-stratified normative comparisons (p = .009), and in sex-stratified normative comparisons (p = .022). Total weighted dementia risk reduced odds of superager status (OR = 0.199, 95% CI [0.046, 0.829]). Other lifestyle dementia risk factors were unrelated to superager status or could not be tested due to low endorsement. The findings support observations that superaging is more common in females, even when controlling for sex differences in memory performance. Future studies of superagers should account for sex differences. Results support being ambitious about dementia prevention, as having fewer modifiable dementia risk factors may be positively associated with superager status.
INTRODUCTION
Cognitive aging research has historically focused on identifying processes that undergo decline with age (Anderson & Craik, 2017), but a more recent focus examines the determinants of successful aging. There is extensive heterogeneity in the approach to the study of successful aging (Depp & Jeste, 2006). In 2012, Harrison and colleagues began the systematic study of individuals in their 80s with episodic memory comparable to individuals 20 to 30 years their junior. With non-memory cognitive abilities also intact, such individuals were classified as “superagers”. The superager phenotype has since been associated with greater regional cerebral volumes (Harrison et al., 2018), slower rate of cortical atrophy (Cook et al., 2017), less pathological burden associated with Alzheimer’s disease (Gefen et al., 2015; Gefen et al., 2021), and genetic mutations in memory signaling pathways (Huentelman et al., 2018), relative to typical aging peers (typical-agers).
As the study of superagers has proliferated, there has been a commensurate increase in methodological differences between studies and the corollary effects on results. For example, a recent study found that application of different operational definitions of “super-aging” resulted in greatly varied prevalence rates of superagers within a given sample (Powell et al., 2023). A significant methodological limitation not broadly considered to date is how sex influences superager group membership. Notably, the majority of studies on superagers have not considered sex differences in verbal episodic memory ability when applying superager classification, with some exceptions (e.g., Maccora et al., 2021; Sun et al., 2016; Zhang et al., 2020). Although many of the published studies consider sex as a covariate in statistical models, there is a paucity that consider sex as a moderator variable or conduct analyses within sex-specific groups. In contrast, the consideration of sex in Alzheimer’s research has garnered increasing support (e.g., Mazure & Swendsen, 2016). This has occurred in the context of accumulating evidence of potential neurobiological differences in males and females, including morphological and functional differences, even if small in nature (Cosgrove et al., 2007; Ritchie et al., 2018). For example, males may show steeper age-related decline in hippocampal volume and global total brain volume relative to females (Armstrong et al., 2019). Thus, studies of the neurobiological correlates of superager status that do not consider sex may confound sex-related differences in normal aging trajectories with differences as a factor of superageing.
Further to this, females, on average, outperform males on tests of verbal episodic memory (the referent test for superaging; Hirnstein et al., 2022) and studies have demonstrated that males experience steeper declines in cognitive aging relative to females across different domains (LaPlume et al., 2022b; McCarrey et al., 2016). The Rey Auditory Verbal Learning Test (RAVLT) is the most common tool used to identify superagers (Andrade et al., 2023). A study of healthy older adults reported that females outperformed age-matched males on the RAVLT with large to very large effects across the ages of 60–89 years old (Gale et al., 2007). More specifically, Gale and colleagues noted the average performance of females in the 80 to 89-year-old group to be 9.5 words, out of a possible of 15, relative to similar-aged males at 7.4 words. Against the common benchmark of 9 words for “superior memory” in 80 to 89-year-olds, more females relative to males would meet this criterion. However, an examination of the effect of normative comparisons on superager classification status has yet to be reported.
Methodological concerns aside, a central interest in studying the superager cognitive phenotype is to identify the characteristics that differentiate superagers from their typical-aging peers. Recent endeavors have attempted to determine which modifiable dementia risk factors are associated with superager status, under the assumption that the absence of risk factors for cognitive decline might differentiate those with exceptional versus average-for-age cognitive abilities. In the largest study to date (n = 1679), Maccora et al. (2021) found that, in females, superagers had higher education, greater units of weekly alcohol intake, and participated in more cognitively engaging activities relative to typical-agers. In males, superagers also had higher education, but reported more social activities per week and fewer depressive symptoms. Cook Maher et al. (2017) found that superagers, relative to typical agers, engaged in more social interactions. Factors such as body mass index, diabetes, hypertension, dyslipidemia, and smoking status have not been found to be significant factors that differentiate superager and typical-ager groups (Calandri et al., 2020; Harrison et al., 2018; Kim et al., 2020; Maccora et al., 2021). With the exception of Maccora and colleagues, most studies have used small samples to explore this question, and again have not used sex-specific normative data.
We sought to extend the extant research by examining the effects of sex-specific normative data on superager status, as well as associations of sex-corrected superager status with dementia risk factors in a sample of 80 to 89-year-olds. The primary aim of the study was to examine the role of sex in superager classification using an existing database from an online cognitive assessment platform. It was anticipated that a greater proportion of females would be identified as superagers relative to males when group membership was decided using non-sex-stratified normative comparisons. In this normative method, group membership is decided against the average performance of all participants aged 50–69, irrespective of sex. This is in contrast to sex-corrected normative comparisons, where participants aged 80–89 are compared against participants 50–69 with the same self-reported sex. Using this method, it was expected that there would be an increase in the proportion of males meeting criteria for superagers, and a relative narrowing of proportions of male and female superagers. These predictions are primarily informed by the aforementioned female advantage in verbal memory tests (Hirnstein et al., 2022) and steeper cross-sectional memory decline in males with age (LaPlume, et al., 2022b).
The secondary aim of our study was to examine the extent to which modifiable dementia risk factors predict sex-corrected superager status. The selection of dementia risk factors was informed by contemporary models of lifetime dementia risk (i.e., Livingston et al., 2020). Livingston and colleagues described the cumulative and unique risk conferred by twelve modifiable dementia risk factors across the lifespan. Eight of the 12 factors identified in Livingston et al., 2020 lifespan dementia risk model were available in the existing dataset and tested as predictors of superager status. It was expected that the typical-ager group would endorse more risk factors relative to the superager group. Primarily informed by Maccora et al., (2021) work, it was expected that low education and current depression would independently decrease odds of superager status.
MATERIALS AND METHODS
Participants
Data from Cogniciti’s Brain Health Assessment, an online cognitive self-assessment (Troyer et al., 2014), were used for the present analyses. The Brain Health Assessment is a freely-available online assessment platform. Individuals can access the website and complete the assessment without indicating their reason for doing so, be it worry about cognition or general curiosity about cognitive status. At the time of data collection, the assessment could only be accessed on desktop or laptop devices. Prior to starting the cognitive assessment, all participants read an information page and answered a series of demographic and health questions. Participants were advised to complete the assessment within one sitting, in a quiet area, free from intrusions or distractions. They were further informed that pausing the assessment would result in the termination of the test-taking attempt. Participants were then asked to take the test when feeling well, to put forth their best effort, to not use external aids, and ensure proper functioning of their computer (including setting their screen zoom to “100%” to ensure accurate presentation of stimuli). The assessment is designed to last approximately 20 to 30 min. The sample of participants reported here represents a secondary analysis of a subset of participants included in a recent study (original n = 93,363; LaPlume et al., 2022c). Data from this prior study included all logged attempts on the Brain Health Assessment platform from 2014 to 2019. No active recruitment methods were used for this research. Data were collected in compliance with institution regulations and ethics review, and the research was completed in accordance with the Helsinki Declaration.
Extensive data cleaning was used to ensure only valid data were analyzed. This involved the exclusion of data where there were disruptions to collection (e.g., the web browser page refreshed during data collection or there was incomplete data collection), iterative trimming of data to remove extreme outliers, and the exclusion of data where the reported age was presumed inaccurate or outside specified ranges (see previous reports for a detailed description; LaPlume et al., 2022c; LaPlume et al., 2022a; Anderson, et al., 2022). Because participants can attempt the Brain Health Assessment more than once, only data from first attempts were included in order to eliminate potential practice effects. Trimming procedures were applied between-subjects for each age decade and cognitive task condition, and was accomplished in an iterative manner using a moving cut-off standard deviation value which was adjusted based on the sample size (LaPlume et al., 2022c). With this recursive moving criterion approach, influential outliers were removed from the dataset, leaving “true” cases within the upper and lower distribution of the sample. Approximately 1%–4% of data per age decade per cognitive task were removed as a result of trimming, representing approximately 20% (n = 15,702) of the original dataset; different participants were removed for different tasks, and only complete cases were retained. After applying all the above cleaning criteria, the sample size from which the present sample was drawn was n = 22,117 (LaPlume et al., 2022c).
For the present study, participants who reported any of the following conditions that may affect cognition were excluded from analyses: neurodegenerative conditions (i.e., Alzheimer’s disease, Huntington’s disease, Parkinson’s disease, multiple sclerosis, dementia with Lewy bodies, or vascular dementia), mild cognitive impairment, bipolar disorder, schizophrenia, cancer treated with chemotherapy, brain tumor, brain surgery, seizures, stroke/TIA, or a neurodevelopmental disorder (i.e., attention deficit disorder or learning disability). Participants in the 80–89 years old group also had to meet a criterion such that they were “cognitively intact,” defined as at least average performance (i.e., age-corrected z greater or equal to −1) across the four Brain Health Assessment cognitive tasks.
The final sample size used for the present analyses included data from 9233 (n = 6500 female) participants aged 50–69 (normative comparison group for memory performance only) and 469 (n = 314 female) participants aged 80–89 (superager analysis group). Within the 80–89-year-old group, participants were classified into one of two groups: (a) superagers, whose associative memory performance was at or above the average level of 50 to 69 years old; (b) typical-agers, whose associative memory performance was below this level. Operationalization of superager and typical-ager group membership status was consistent with Harrison et al.’ (2012) approach.
Self-report measures
Participants completed a brief online self-report health and demographic questionnaire. This questionnaire included eight items that correspond to the Lancet commission life course model of dementia prevention (Livingston et al., 2020): fewer than 12 years of formal education, traumatic brain injury (TBI), hearing loss, alcohol or substance misuse, hypertension, past 5-year history of smoking, diabetes, and current depression. All health-related conditions were assessed as a self-reported lifetime incidence. With the exception of age and smoking history, the modifiable lifestyle risk factors were collected as binary (yes/present; no/absent) outcomes. Smoking history was assessed as recency of smoking (e.g., “I have never smoked,” “I smoke currently,” “I stopped smoking 3 to 5 years ago”) and binarized as those who never smoked or stopped smoking more than 5 years prior to the assessment, and those who currently smoke or have smoked in the past 5 years. A verbatim copy of the questionnaire items and response options is included in Supplemental Materials – Table S1.
Two composite risk scores were created using previously described procedures (LaPlume et al, 2022). The first composite score was a count of the total number of endorsed risk factors (“total factors”; range of 0 to 8). The second was a weighted variable of total endorsed risk factors (“weighted total factors”; range of 0 to 1; LaPlume et al., 2022). The weighed total risk index was computed using the weighted population attributable fraction (weighted PAF) for each risk factor (Livingston et al., 2020) relative to the total risk conferred by the eight risk factors included. With the eight included risk factors, the total PAF was 31%. Individual weighted PAFs from Livingston and colleagues’ report ranged from 1 to 8%. The weighted factors used here were expressed as the fraction of Livingston’s weighted PAF for a risk factor (e.g., 8% for hearing loss) relative to the total PAF (i.e., 31%). This weighted index accounted for the shared variance or communality among the risk factors.
Online cognitive assessment
The Brain Health Assessment included four tasks in a set order: a Spatial Working Memory (Shape Matching) Task; A Stroop Interference task measuring cognitive inhibition; A Face-Name Association (FNA) task measuring associative memory; A Letter-Number Alternation task (a version of Trail Making Test Part B) measuring set shifting. Acceptable reliability and validity of the cognitive tests included in the Brain Health Assessment have been previously described (Paterson et al., 2022; Troyer et al., 2014). Task scores were transformed to z-scores for the purpose of identifying “intact” cognition, described above; z-scores on the Spatial Working Memory, Stroop Interference, and Letter-Number Alternation tasks were transformed by a constant of −1, so that for all tests, higher z-scores reflect better performance. Troyer et al. (2014) provide a comprehensive overview of each task; brief descriptions are presented below.
Spatial working memory
Participants were presented with a 4×3 grid of squares. Six pairs of visual stimuli were “hidden” in this grid. Participants clicked on a square to reveal the hidden stimulus with a goal to identify the six pairs in the fewest clicks possible. Participants completed three trials; two at the beginning of the cognitive assessment and one trial following the last cognitive task (Letter Number Alternation). Only data from the first two trials are used for this analysis. The primary outcome is the number of clicks required to successfully match all paired stimuli over the two trials. Fewer clicks represented stronger performance.
Stroop interference
One, two, or three words were presented on the screen. Participants indicated the number of words by pressing a corresponding button. Words were neutral (e.g., “and”) or numbers (e.g., “one”). Trials included congruent (matched word and number of words), incongruent (mismatched word and number of words), and neutral conditions, presented randomly. Each trial type occurred 30 times, totaling 90 trials. The median response time for incongruent trials is used as the primary measure of cognitive inhibition, with shorter response times corresponding to better performance.
Face-name association
Participants were presented with a photo and a corresponding name simultaneously for three seconds. Twenty-four face-name pairs were shown twice, across two learning trials. A recognition test followed the second learning trial, with 12 “intact” and 12 “recombined” face-name pairings. Participants distinguished between “intact” and “recombined” pairs. Accuracy of the FNA task was selected as the measure of associative memory, and was operationalized as the associative recognition rate across trials (hit rate minus false alarm rate). Higher associative recognition rates correspond to stronger performance. Notably, the accuracy on the FNA has been shown to correlate with other similar memory tasks, such as word-list recognition and incidental paired recall, with moderate to large effects (Paterson et al., 2022).
Letter-number alternation
Based on the Trail Making Test, part B, participants were required to click letter or number stimuli, in alternating ascending order. They were instructed to complete the test as quickly as possible without making mistakes. Numbers ranged from 1 to 8, and letters range from A to H. Incorrect responses were immediately identified and participants were required to self-correct to continue. There was no time limit. Set shifting is reflected in the time to complete the alternating sequence within the letter-number alternation task, with lower times reflecting better performance.
Analyses
This study was pre-registered on the Open Science Framework (osf.io/udq7k), and all analyses were completed using the R language and environment (R Core Team, 2022). The present analyses focus on the effects of sex correction in normative comparisons and select risk factors, and thus deviate from the pre-registered analyses.
To address the first aim, 2×2 chi-square tests were used to compare the proportion of males and females in the superager group relative to the typical-ager group. For the first chi-square test, superager group membership used non-sex-corrected normative comparisons. For the second chi-square test, superager group membership was re-defined using sex-corrected normative comparisons (i.e., males aged 80–89 compared with males aged 50–69 and females aged 80–89 compared with females aged 50–69). Normative comparisons for the FNA task were drawn from the 50–69-year-old participants from the Brain Health Assessment sample. To address the second aim, logistic regression models were used to test the relationship between risk factor endorsement and relative odds of superager status. Sex-corrected superager status was used in the logistic regression models. Model 1 included the total risk factor endorsement, Model 2 included the weighted total risk factor endorsement, and Model 3 included individual Livingston-identified factors (education, hearing loss, hypertension, diabetes, and depression). There was low endorsement of three risk factors (drug or alcohol misuse, recent history of smoking, and history of TBI) across participants and subgroups, hence these were excluded from analysis as individual predictors due to a lack of reliable statistical analysis, but they were retained in the composite risk factor indices.
RESULTS
Sample characteristics
Table 1 contains basic demographic characteristics of the sample (n = 469) split by sex-corrected superager status. Groups did not differ on average age or family history of memory problems. Typical agers were more likely than their superager counterparts to report recent concern with their memory. Superagers, on average, performed better than their typical-aging peers on the Stroop Interference and Letter-Number Alternation tasks, with small to medium effects. Superagers and typical agers did not differ on the Spatial Working Memory task. Figure 1 contains a graphical depiction of FNA performance by group and subgroups. Overall, participants between the ages of 80 to 89 scored significantly lower on the FNA task relative to their 50 to 69-year old counterparts, t(10062) = 25.22, p < .001, d = 0.91. This held true for both men in their 80s relative to men between the ages of 50 and 69, t (3009) = 14.79, p < .001, d = 0.93, and women in their 80s relative to women between the ages of 50 and 69, t(7051) = 20.29, p < .001, d = 0.90. Females performed better than males on the FNA in both the 80 to 89-year-old group, t(829) = −2.66, p < .001, d = 0.20, and the 50 to 69-year-old group, t(9231) = −7.64, p < .001, d = −0.17. Endorsement for each individual Livingston lifestyle factor is presented in Table 1.
Demographic characteristics and risk factor endorsement by sex-corrected group.
| Superager (M ± SD or n (%)) | Typical-ager (M ± SD or n (%)) | p | Effect size (d or V) | |
|---|---|---|---|---|
| Age | 83.1 ± 2.71 | 83.0 ± 2.65 | .928 | 0.008 |
| Family History | 59 (36%) | 95 (31%) | .230 | 0.03 |
| Memory Concerns | 34 (21%) | 99 (32%) | .010 | 0.11 |
| English as Second Language | 8 (5%) | 22 (7%) | .348 | 0.00 |
| Spatial Working Memory | 89.6 ± 24.39 | 91.8 ± 26.51 | .370 | 0.09 |
| Stroop Interference | 1280.1 ± 198.58 | 1334.8 ± 215.59 | .007 | 0.26 |
| Face-Name Association | 76.6 ± 11.01 | 35.1 ± 14.3 | < .000 | 3.13 |
| Letter-Number Alternation | 36.3 ± 12.04 | 39.7 ± 13.06 | .005 | 0.27 |
| Non-sex-corrected norms | ||||
| Males | 40 (26%) | 115 (74%) | .009 | 0.121 |
| Females | 121 (39%) | 193 (61%) | ||
| Sex-corrected norms | ||||
| Males | 42 (27%) | 113 (73%) | .022 | 0.100 |
| Females | 120 (38%) | 194 (62%) | ||
| Lifestyle Risk Factor | ||||
| Education | 9 (6%) | 28 (9%) | ||
| Hearing Loss | 76 (47%) | 172 (56%) | ||
| Alcohol/Drug Misuse* | 2 (1%) | 6 (2%) | ||
| Hypertension | 81 (50%) | 143 (47%) | ||
| TBI* | 0 (0%) | 0 (0%) | ||
| Diabetes | 18 (11%) | 27 (9%) | ||
| Depression (current) | 9 (6%) | 18 (6%) | ||
| Smoking (recent)* | 2 (1%) | 8 (3%) | ||
| Total Endorsed Risk Factors | ||||
| 0 | 37 (23%) | 55 (18%) | ||
| 1 | 65 (40%) | 136 (44%) | ||
| 2 | 51 (31%) | 85 (28%) | ||
| 3 | 6 (4%) | 28 (9%) | ||
| 4 | 3 (2%) | 3 (1%) | ||
| Any Endorsed Risk Factor | 125 (77%) | 252 (82%) | ||
| Weighted Risk | 0.18 (0.15) | 0.21 (0.15) |
| Superager (M ± SD or n (%)) | Typical-ager (M ± SD or n (%)) | p | Effect size (d or V) | |
|---|---|---|---|---|
| Age | 83.1 ± 2.71 | 83.0 ± 2.65 | .928 | 0.008 |
| Family History | 59 (36%) | 95 (31%) | .230 | 0.03 |
| Memory Concerns | 34 (21%) | 99 (32%) | .010 | 0.11 |
| English as Second Language | 8 (5%) | 22 (7%) | .348 | 0.00 |
| Spatial Working Memory | 89.6 ± 24.39 | 91.8 ± 26.51 | .370 | 0.09 |
| Stroop Interference | 1280.1 ± 198.58 | 1334.8 ± 215.59 | .007 | 0.26 |
| Face-Name Association | 76.6 ± 11.01 | 35.1 ± 14.3 | < .000 | 3.13 |
| Letter-Number Alternation | 36.3 ± 12.04 | 39.7 ± 13.06 | .005 | 0.27 |
| Non-sex-corrected norms | ||||
| Males | 40 (26%) | 115 (74%) | .009 | 0.121 |
| Females | 121 (39%) | 193 (61%) | ||
| Sex-corrected norms | ||||
| Males | 42 (27%) | 113 (73%) | .022 | 0.100 |
| Females | 120 (38%) | 194 (62%) | ||
| Lifestyle Risk Factor | ||||
| Education | 9 (6%) | 28 (9%) | ||
| Hearing Loss | 76 (47%) | 172 (56%) | ||
| Alcohol/Drug Misuse* | 2 (1%) | 6 (2%) | ||
| Hypertension | 81 (50%) | 143 (47%) | ||
| TBI* | 0 (0%) | 0 (0%) | ||
| Diabetes | 18 (11%) | 27 (9%) | ||
| Depression (current) | 9 (6%) | 18 (6%) | ||
| Smoking (recent)* | 2 (1%) | 8 (3%) | ||
| Total Endorsed Risk Factors | ||||
| 0 | 37 (23%) | 55 (18%) | ||
| 1 | 65 (40%) | 136 (44%) | ||
| 2 | 51 (31%) | 85 (28%) | ||
| 3 | 6 (4%) | 28 (9%) | ||
| 4 | 3 (2%) | 3 (1%) | ||
| Any Endorsed Risk Factor | 125 (77%) | 252 (82%) | ||
| Weighted Risk | 0.18 (0.15) | 0.21 (0.15) |
Note. *Factors dropped from analyses due to low expected cell values (i.e., exceptionally low base rate). Weighted risk score presented as group mean (standard deviation), corrected for sex. Family history: percent of sample with a reported familial history of memory problems. English as second language (percent yes). For sex-corrected and non-sex-corrected superager and typical ager statuses, percentages reflect percent within sex (i.e., percent of males or percent of females). Effect sizes presented as absolute value. Cognitive outcomes presented: Spatial Working Memory (number of responses required); Stroop Interference (incongruent median reaction time (ms); FNA (associative recognition rate); Letter Number Alternation (time to completion (s)). Overall n = 469.
Demographic characteristics and risk factor endorsement by sex-corrected group.
| Superager (M ± SD or n (%)) | Typical-ager (M ± SD or n (%)) | p | Effect size (d or V) | |
|---|---|---|---|---|
| Age | 83.1 ± 2.71 | 83.0 ± 2.65 | .928 | 0.008 |
| Family History | 59 (36%) | 95 (31%) | .230 | 0.03 |
| Memory Concerns | 34 (21%) | 99 (32%) | .010 | 0.11 |
| English as Second Language | 8 (5%) | 22 (7%) | .348 | 0.00 |
| Spatial Working Memory | 89.6 ± 24.39 | 91.8 ± 26.51 | .370 | 0.09 |
| Stroop Interference | 1280.1 ± 198.58 | 1334.8 ± 215.59 | .007 | 0.26 |
| Face-Name Association | 76.6 ± 11.01 | 35.1 ± 14.3 | < .000 | 3.13 |
| Letter-Number Alternation | 36.3 ± 12.04 | 39.7 ± 13.06 | .005 | 0.27 |
| Non-sex-corrected norms | ||||
| Males | 40 (26%) | 115 (74%) | .009 | 0.121 |
| Females | 121 (39%) | 193 (61%) | ||
| Sex-corrected norms | ||||
| Males | 42 (27%) | 113 (73%) | .022 | 0.100 |
| Females | 120 (38%) | 194 (62%) | ||
| Lifestyle Risk Factor | ||||
| Education | 9 (6%) | 28 (9%) | ||
| Hearing Loss | 76 (47%) | 172 (56%) | ||
| Alcohol/Drug Misuse* | 2 (1%) | 6 (2%) | ||
| Hypertension | 81 (50%) | 143 (47%) | ||
| TBI* | 0 (0%) | 0 (0%) | ||
| Diabetes | 18 (11%) | 27 (9%) | ||
| Depression (current) | 9 (6%) | 18 (6%) | ||
| Smoking (recent)* | 2 (1%) | 8 (3%) | ||
| Total Endorsed Risk Factors | ||||
| 0 | 37 (23%) | 55 (18%) | ||
| 1 | 65 (40%) | 136 (44%) | ||
| 2 | 51 (31%) | 85 (28%) | ||
| 3 | 6 (4%) | 28 (9%) | ||
| 4 | 3 (2%) | 3 (1%) | ||
| Any Endorsed Risk Factor | 125 (77%) | 252 (82%) | ||
| Weighted Risk | 0.18 (0.15) | 0.21 (0.15) |
| Superager (M ± SD or n (%)) | Typical-ager (M ± SD or n (%)) | p | Effect size (d or V) | |
|---|---|---|---|---|
| Age | 83.1 ± 2.71 | 83.0 ± 2.65 | .928 | 0.008 |
| Family History | 59 (36%) | 95 (31%) | .230 | 0.03 |
| Memory Concerns | 34 (21%) | 99 (32%) | .010 | 0.11 |
| English as Second Language | 8 (5%) | 22 (7%) | .348 | 0.00 |
| Spatial Working Memory | 89.6 ± 24.39 | 91.8 ± 26.51 | .370 | 0.09 |
| Stroop Interference | 1280.1 ± 198.58 | 1334.8 ± 215.59 | .007 | 0.26 |
| Face-Name Association | 76.6 ± 11.01 | 35.1 ± 14.3 | < .000 | 3.13 |
| Letter-Number Alternation | 36.3 ± 12.04 | 39.7 ± 13.06 | .005 | 0.27 |
| Non-sex-corrected norms | ||||
| Males | 40 (26%) | 115 (74%) | .009 | 0.121 |
| Females | 121 (39%) | 193 (61%) | ||
| Sex-corrected norms | ||||
| Males | 42 (27%) | 113 (73%) | .022 | 0.100 |
| Females | 120 (38%) | 194 (62%) | ||
| Lifestyle Risk Factor | ||||
| Education | 9 (6%) | 28 (9%) | ||
| Hearing Loss | 76 (47%) | 172 (56%) | ||
| Alcohol/Drug Misuse* | 2 (1%) | 6 (2%) | ||
| Hypertension | 81 (50%) | 143 (47%) | ||
| TBI* | 0 (0%) | 0 (0%) | ||
| Diabetes | 18 (11%) | 27 (9%) | ||
| Depression (current) | 9 (6%) | 18 (6%) | ||
| Smoking (recent)* | 2 (1%) | 8 (3%) | ||
| Total Endorsed Risk Factors | ||||
| 0 | 37 (23%) | 55 (18%) | ||
| 1 | 65 (40%) | 136 (44%) | ||
| 2 | 51 (31%) | 85 (28%) | ||
| 3 | 6 (4%) | 28 (9%) | ||
| 4 | 3 (2%) | 3 (1%) | ||
| Any Endorsed Risk Factor | 125 (77%) | 252 (82%) | ||
| Weighted Risk | 0.18 (0.15) | 0.21 (0.15) |
Note. *Factors dropped from analyses due to low expected cell values (i.e., exceptionally low base rate). Weighted risk score presented as group mean (standard deviation), corrected for sex. Family history: percent of sample with a reported familial history of memory problems. English as second language (percent yes). For sex-corrected and non-sex-corrected superager and typical ager statuses, percentages reflect percent within sex (i.e., percent of males or percent of females). Effect sizes presented as absolute value. Cognitive outcomes presented: Spatial Working Memory (number of responses required); Stroop Interference (incongruent median reaction time (ms); FNA (associative recognition rate); Letter Number Alternation (time to completion (s)). Overall n = 469.
Performance on the face-name association task across age groups, sex, and superager status. Note. Bars represent mean group-level performance; error bars are standard deviation. Sample sizes are as follows: Overall 50–69 year-olds n = 9233; male 50–69 year-olds n = 2733; female 50–69 year-olds n = 6500; overall 80–89 year-olds n = 469; male 80–89 year-olds n = 155; female 80–89 year-olds n = 314; superager n = 162; typical-ager n = 307.
Aim 1: Is there sex bias in superager classification?
Using the mean performance (i.e., non-sex-corrected data) on the FNA task in the 50 to 69-year-old participants as the benchmark, proportionally more females (121; 39% of females) than males (40; 26% of males) were categorized as superagers, |$\chi$|2 (1) = 6.904, p = .009, with a small effect size, V = 0.121. When superager status was determined using sex-corrected normative data (i.e., females aged 80 to 89 compared with females aged 50 to 69, and males aged 80 to 89 compared to males aged 50 to 69), there were still proportionally more females (120; 38% of females) relative to males (42; 27% of males) meeting criteria for superager status, |$\chi$|2(1) = 5.194, p = .022, V = 0.100. Relative to non-sex-corrected comparisons, using sex-corrected normative data resulted in one female moved from the superager group to the typical-ager group, and two males moved from the typical-ager group to the superager group. Table 1 contains a breakdown of superager status versus typical ager status stratified by sex and normative comparison.
Aim 2: Association of sex-corrected superager status with lifestyle factors
Results of the logistic regression analyses are included in Table 2. The total number of risk factors endorsed was unrelated to superager status (p = .287). By contrast, the weighted endorsement of risk factors negatively predicted superager status (p = .034). None of the individually tested risk factors were significantly associated with superager status.
Logistic regression model outcomes
| b | SE | z | p | OR | OR 95% CI [LL,UL] | ||
|---|---|---|---|---|---|---|---|
| Whole Sample | |||||||
| Model 1 | Total Risk | −0.116 | 0.109 | −1.060 | .287 | 0.891 | [0.718, 1.101] |
| Model 2 | Total-Weighted Risk | −1.373 | 0.649 | −2.116 | .034 | 0.253 | [0.070, 0.895] |
| Model 3 | Education | −0.626 | 0.403 | −1.553 | .121 | 0.535 | [0.230, 1.136] |
| Hearing loss | −0.370 | 0.197 | −1.880 | .060 | 0.691 | [0.469, 1.015] | |
| Hypertension | 0.128 | 0.198 | 0.647 | .518 | 1.137 | [0.771, 1.677] | |
| Diabetes | 0.279 | 0.332 | 0.842 | .400 | 1.322 | [0.681, 2.520] | |
| Depression | 0.007 | 0.426 | 0.016 | .987 | 1.007 | [0.418, 2.271] | |
| b | SE | z | p | OR | OR 95% CI [LL,UL] | ||
|---|---|---|---|---|---|---|---|
| Whole Sample | |||||||
| Model 1 | Total Risk | −0.116 | 0.109 | −1.060 | .287 | 0.891 | [0.718, 1.101] |
| Model 2 | Total-Weighted Risk | −1.373 | 0.649 | −2.116 | .034 | 0.253 | [0.070, 0.895] |
| Model 3 | Education | −0.626 | 0.403 | −1.553 | .121 | 0.535 | [0.230, 1.136] |
| Hearing loss | −0.370 | 0.197 | −1.880 | .060 | 0.691 | [0.469, 1.015] | |
| Hypertension | 0.128 | 0.198 | 0.647 | .518 | 1.137 | [0.771, 1.677] | |
| Diabetes | 0.279 | 0.332 | 0.842 | .400 | 1.322 | [0.681, 2.520] | |
| Depression | 0.007 | 0.426 | 0.016 | .987 | 1.007 | [0.418, 2.271] | |
Note. Model 1 included total risk factors as the sole predictor (Total). Model 2 included the weighted total risk factor score. Model 3 included the individual Livingston model factors as individual predictors. Overall n = 469. Significant p-values are bolded.
Logistic regression model outcomes
| b | SE | z | p | OR | OR 95% CI [LL,UL] | ||
|---|---|---|---|---|---|---|---|
| Whole Sample | |||||||
| Model 1 | Total Risk | −0.116 | 0.109 | −1.060 | .287 | 0.891 | [0.718, 1.101] |
| Model 2 | Total-Weighted Risk | −1.373 | 0.649 | −2.116 | .034 | 0.253 | [0.070, 0.895] |
| Model 3 | Education | −0.626 | 0.403 | −1.553 | .121 | 0.535 | [0.230, 1.136] |
| Hearing loss | −0.370 | 0.197 | −1.880 | .060 | 0.691 | [0.469, 1.015] | |
| Hypertension | 0.128 | 0.198 | 0.647 | .518 | 1.137 | [0.771, 1.677] | |
| Diabetes | 0.279 | 0.332 | 0.842 | .400 | 1.322 | [0.681, 2.520] | |
| Depression | 0.007 | 0.426 | 0.016 | .987 | 1.007 | [0.418, 2.271] | |
| b | SE | z | p | OR | OR 95% CI [LL,UL] | ||
|---|---|---|---|---|---|---|---|
| Whole Sample | |||||||
| Model 1 | Total Risk | −0.116 | 0.109 | −1.060 | .287 | 0.891 | [0.718, 1.101] |
| Model 2 | Total-Weighted Risk | −1.373 | 0.649 | −2.116 | .034 | 0.253 | [0.070, 0.895] |
| Model 3 | Education | −0.626 | 0.403 | −1.553 | .121 | 0.535 | [0.230, 1.136] |
| Hearing loss | −0.370 | 0.197 | −1.880 | .060 | 0.691 | [0.469, 1.015] | |
| Hypertension | 0.128 | 0.198 | 0.647 | .518 | 1.137 | [0.771, 1.677] | |
| Diabetes | 0.279 | 0.332 | 0.842 | .400 | 1.322 | [0.681, 2.520] | |
| Depression | 0.007 | 0.426 | 0.016 | .987 | 1.007 | [0.418, 2.271] | |
Note. Model 1 included total risk factors as the sole predictor (Total). Model 2 included the weighted total risk factor score. Model 3 included the individual Livingston model factors as individual predictors. Overall n = 469. Significant p-values are bolded.
DISCUSSION
The overarching goal of this study was to examine the influence of sex and dementia risk factors on superager classification. To accomplish this, data from an online cognitive assessment were used to probe sex-based bias in superager allocation methods and examine the association of lifestyle risk factors with sex-corrected superager status. We found that females were more likely than males to be classified as superagers when either sex-corrected or non-sex-corrected norms were used. Thus, our findings support other reports of a female advantage for superaging (e.g., Maccora et al., 2021), including when comparisons are made based on sex-specific normative data. The alignment of our findings with prior studies showing a female superager advantage is notable, given key methodological differences, such as cognitive tools used to identify superagers and differing participant ages. This finding is perhaps also consistent with past research that has documented sex-specific differences in cognitive aging trajectories (LaPlume et al., 2022b; McCarrey et al., 2016).
We did not find evidence of sex-based bias in superager classification methods, based on the type of normative correction applied, although a small number of participants were re-classified depending on the use of a sex-based normative correction. The effect of this misclassification is likely small in large samples. However, many studies of the superager construct use small samples and different tools for superager classification, and misclassification in these groups of participants may conceivably impact outcomes reported in these studies. Indeed, sex-based differences in performance on the FNA task observed here were small. In contrast, the RAVLT remains the most used cognitive instrument to identify superagers (Andrade et al., 2023) and large sex-based differences in delayed recall performance have been reported (Gale et al., 2007). Thus, while misclassification was small in the present sample, this may be amplified with other memory measures where large sex-based differences in performance are observed. Dissociating true sex differences from methodological differences or other confounds remains an important area in general for future research with superagers.
There is growing emphasis on disaggregating sex effects in cognitive aging and related research, which have direct consequences for clinical neuropsychological practice (e.g., diagnostic error in amnestic mild cognitive impairment; Sundermann et al., 2019). To date, few studies on the superager construct explicitly report correcting for sex in classification methods (e.g., Maccora et al., 2021; Zhang et al., 2020). Considering sex in cognitive aging research remains critical, particularly when group membership definition is contingent upon cognitive ability. While some standard neuropsychological tests used in clinical practice include sex-corrected normative data, the continued explicit consideration of sex in cognitive research and test development remains an important and ongoing pursuit. Where sex-adjusted norms are not available, researchers may consider employing their own sex-corrected normative methods or using tests with such data available.
We also examined the relationship between sex-corrected superager status and lifestyle dementia risk factors. A higher weighted total risk index predicted lower probability of superager status. This is consistent with our hypothesis that the presence of factors that confer risk for cognitive decline and dementia would be more prevalent in typical agers relative to superagers. However, no individual risk factors from the Livingston model that were tested were related to superager status. This is perhaps consistent with accumulating evidence highlighting the synergistic detrimental effect of cumulative dementia risk on cognitive aging. Individual dementia risk factors tend to co-occur (e.g., Morris et al., 2016), and greater aggregate risk is associated with poorer cognitive outcomes (e.g., LaPlume et al., 2022c; Peters et al., 2019). Sex may also influence aggregation of risk factors, with males reporting greater co-occurrence of risk factors relative to females (e.g., Morris et al., 2016). Importantly, this dose–response relationship may also hold for the inverse association, where the presence of more protective factors is associated with better cognitive outcomes (Peters et al., 2019), which is consistent with the results reported here.
Select risk factors could not be tested as individual predictors due to lack of available data or exceptionally low endorsement. The low endorsement of these and other risk factors could be due to a high-performing sample, or that individuals with such histories were excluded due to covariation of these risk factors with clinically significant cognitive decline (e.g., diagnosis of dementia or mild cognitive impairments). Future work with higher reported individual risk factor frequencies with larger, balanced, samples could examine sex differences for individual risk factors (e.g., hearing loss and cognitive impairment in females relative to males; Al-Yawer et al., 2022; Juneja et al., 2021; Özgedik et al., 2022).
Limitations to the current study warrant consideration. First, the individuals captured in this sample may not be representative of the population and hence the results may not generalize broadly. For example, those with concerns about their cognition may be more likely to partake in the Brain Health Assessment. Further, people aged 80 to 89 years who can independently access computer-based assessments may differ from those who are unable to due to economic or cognitive constraints. This, in turn, may reflect a higher-functioning sample less likely to have histories of significant risk factors that impact cognition. Second, reliance on a self-report retrospective database may impact outcomes. For example, we were unable to characterize or examine the role of race or ethnicity, as these data were not collected. In addition, there is potential for self-report bias with regards to personal health history, as there was no method to independently validate reported conditions. Relatedly, there was no available method to ensure that participants strictly adhered to pre-test instructions, such as applying their best effort, ensuring that testing was completed in a distraction-free environment, or that individuals’ computers were optimized to complete the cognitive assessment. These concerns are somewhat mitigated by stringent inclusion and exclusion criteria, including careful scrutiny of the data for quality. Third, our methods diverge from past work on the superager construct by virtue of differing modes of assessment (online versus paper-and-pencil tests) and type of memory assessment (associative memory versus delayed free recall). One primary advantage of online assessment is the potential to reach diverse populations, in large numbers; completing in-person comprehensive assessment on the same scale would be limited by way of resources and time. The Brain Health Assessment also has been validated against standardized in-person test versions (Paterson et al., 2022; Troyer et al., 2014), and performance-based online data has been shown to produce comparable results in unsupervised at-home and in-lab contexts in older adults (e.g., Cyr et al., 2021). Nonetheless, comprehensive neuropsychological examination in the context of superager research would provide a more nuanced approach to examining cognition, risk factors, and sex differences in this group. Efforts to recruit large samples using comprehensive assessment methods are needed.
CONCLUSION
In clinical practice, neuropsychologists and other clinicians can counsel their clients on evidence-based recommendations for managing concretely-identifiable dementia risk factors (e.g., Livingston et al., 2020). While many clients in clinical practice may wish to avoid or delay cognitive decline, many may also seek ways to optimize or improve their cognitive health. The study of superagers provides a unique avenue to generate knowledge on the characteristics or traits that are associated with exceptional aging. The current study contributes to the growing literature on the superager construct by investigating potential sex-based sources of bias in superager allocation methods and tests of risk factors. In line with contemporary evidence, our results also highlight the potential cumulative effects of multiple risk factors on superager status. While some tests of risk factors were well-powered, others could not be investigated. As a concretely identified successful point in memory aging, the superager construct warrants continued examination as a means to ultimately identify ways to promote healthy aging.
FUNDING
This work was supported by a Discovery Grant from the Natural Sciences and Engineering Research Council of Canada awarded to NDA (grant number RGPIN-2023-05241).
CONFLICT OF INTEREST
The authors declare no conflicts of interest.
AUTHOR CONTRIBUTION
Matthew McPhee (Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Writing – original draft, Writing – review & editing), Larissa McKetton (Data curation, Project administration, Resources, Writing – review & editing), Annalise LaPlume (Data curation, Methodology, Writing – review & editing), Angela Troyer (Conceptualization, Methodology, Resources, Writing – review & editing), Nicole Anderson (Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Supervision, Writing – review & editing).
