Lifestyle, Cognitive, and Psychological Factors Associated With a Resilience Phenotype in Aging

Study Design

This was a cross-sectional study on a population-based sample of older adults without dementia and aged between 83 and 87 years who participated in the fifth wave of the InveCe.Ab study, a population-based cohort enrolled in 2009 (Invecchiamento Cerebrale in Abbiategrasso, i.e., Brain Aging in Abbiategrasso; ClinicalTrials.gov, NCT01345110). Here, we mainly used data from the last assessment performed in 2022, which included all relevant variables for the resilience construct. However, in light of our aim, we reconstructed indices of cumulative exposure to stressors using data from the entire study period. Figure 1 shows the study design and measurements used for the present study, within the framework of the InveCe.Ab project.

Open multimedia modal

Participants and Procedures

Participants were recruited from the InveCe.Ab study, a population-based cohort of people born between 1935 and 1939 who lived in the city of Abbiategrasso when the study began (November 1, 2009). The aim was to estimate the incidence of dementia and to explore sociodemographic, clinical, and lifestyle factors associated with aging and dementia (Guaita et al., 2013). Participants enrolled underwent multidimensional assessments at baseline (n = 1,321), and after 2, 4, and 8 years; they engaged in blood sampling, geriatric visits, neuropsychological assessments, and social and lifestyle interviews.

For our purposes, all InveCe.Ab participants (age range: 83–87 years) who were alive and underwent at least one follow-up assessment after baseline were contacted again in 2022 to participate in the fifth wave of the study. When the study began (mid-February 2022), there were 782 eligible participants.

All the eligible individuals received a personalized letter explaining the study’s aims and procedures for the fifth wave. Then, trained personnel called each participant to provide further explanations and, in the case of acceptance, to schedule the appointments for a multidimensional assessment comprising blood sampling at home by a nurse, a social and lifestyle survey by phone lasting approximately 30 min, and in-person visits to the Golgi Cenci Foundation with the neuropsychologist and the physician (geriatrician or neurologist) on the same day, lasting a total of 3 hr. Visits could also be provided at home or in a nursing home to meet older adults’ needs.

The study procedures were carried out in accordance with the principles outlined in the Declaration of Helsinki of 1964 and the relevant amendments. The study was approved by the Ethics Committee of Milano Area 3 on January 19, 2022 (number: 26-19012022). Informed consent was signed before the study visits.

Multidimensional Assessment

The geriatrician collected a guided and detailed medical history supported by clinical documentation, recorded current medications, reviewed the results of the blood tests, and performed a physical examination with special attention to neurologic signs and symptoms. Comorbidity was assessed with the Cumulative Illness Rating Scale (CIRS; Parmelee et al., 1995). Functional status was assessed with the Basic Activity of Daily Living (BADL) and Instrumental Activity of Daily Living (IADL) scales (Katz et al., 1970^; Lawton and Brody, 1969).

Neuropsychologists administered the 15-item form of the Geriatric Depression Scale (GDS) and a comprehensive neuropsychological test battery to evaluate global cognition (Mini-Mental State Examination, MMSE) and the main cognitive domains: language (category fluency), memory (Rey-Auditory Verbal Learning Test), attention (attentional matrices), executive functions (Trail Making Test, Raven Coloured Progressive Matrices), visuospatial abilities, and constructional praxis (Clock Drawing Test; Rey-Osterrieth Complex Figure Test). Age- and education-adjusted scores were used according to Italian norms to determine the presence of impairment for each test. See the study protocol for more details on the instruments used (Guaita et al., 2013).

After the visits, the neuropsychologist and the physician made independent working diagnoses within the following categories: dementia, based on DSM-V criteria (Sachdev et al., 2014); mild cognitive impairment, based on the Petersen criteria (Petersen, 2004); cognitive impairment—not dementia (Chertkow et al., 2008); clinically relevant depression or subthreshold depression (Vaccaro et al., 2017); and psychosis (ICD10: F20–F29). Participants without cognitive impairment and without major mental disorders (clinically relevant depression or psychosis) were classified as cognitively normal.

After multidimensional assessment completion, an expert geriatrician (A.G.) together with a clinical neuropsychologist (L.P.) reviewed all individual records, and a final diagnosis within the aforementioned categories was reached.

Resilience Phenotype

For the present study, we included all participants without dementia based on the multidimensional assessment performed. Then, we classified resilient individuals as those who maintained normal cognition, mental health, and functional independence in very old age (83–87 years). This represents a positive adaptation to the aging process, considering that all of the participants are oldest-old in a specific age quintile according to the study design. Normal cognition was ascertained by the agreement of the geriatrician and the neuropsychologist who reviewed the individual charts, as specified in the subparagraph on multidimensional assessment. Functional independence was determined if all the BADLs were preserved. Mental health was ascertained by the absence of clinically relevant depression or psychosis at the final diagnosis.

Resilience-Enhancing Factors

During the fifth wave, to enrich the characterization of resilience in aging, we added specific scales to the standard InveCe.Ab instruments are detailed in the study protocol (Guaita et al., 2013).

Table 1 displays the measures used as predictors of resilience in the present study. Unless otherwise specified, we used the data collected during the fifth wave (2022). We grouped the measures into the three chief dimensions of cognitive, physical, and psychological health, based on the a priori hypothesis about their potential actions in relation to the different dimensions. In the cognitive domain, we included variables classically used as indirect proxies of cognitive reserve, representing the amount and complexity of cognitive-stimulating activity performed throughout one’s life, such as years of education, occupational level, leisure time cognitive activities, and social contact later in life (Pinto et al., 2024). As a measurement of early-life cognitive enrichment, we used the childhood socioeconomic condition indicator (Aartsen et al., 2019). Moreover, because the intelligence quotient is frequently employed as a cognitive reserve proxy, we used the raw baseline score of the Raven Coloured Progressive Matrices (Basso et al., 1987), which assesses logical reasoning and is considered a measure of fluid intelligence (one of the two components of general intelligence). In the physical health domain, we included self-reported measures of physical activity and Mediterranean diet adherence (Schröder et al., 2011) due to their established protective effects against several aging-related diseases. We also included waist circumference, handgrip test (Lemmink et al., 2001), and walking speed test (Camicioli et al., 1997), in line with recent contributions that suggest measuring physical functionality directly through standardized performance and strength tests, to better capture the real biology of the function (Patrizio et al., 2021).

Open multimedia modal

Finally, as predictors of psychological health, we chose measurements known to increase individuals’ ability to adapt to adverse and stressful situations and protect against mental illness. The 14-item Resilience Scale measures self-reported adaptive coping styles to counteract adversity, with specific questions on five dimensions (Callegari et al., 2016): purpose, perseverance, self-reliance, equanimity, and existential aloneness (authenticity). These are all psychosocial characteristics associated with high resilience among older adults (Laird et al., 2019^; MacLeod et al., 2016). Personality traits and attachment styles are stable individual traits shaped by gene[FIGURE DASH]environment interactions in early life and are associated with the risk of mental illness during one’s life (Laird et al., 2019). We measured three personality traits using the Eysenck Personality Questionnaire–Revised (EPQ-SF-R; Dazzi et al., 2004): extraversion (a tendency to be sociable, carefree, convivial, easy-going, and impulsive), neuroticism (low emotional stability and a predisposition toward experiencing negative affect), and psychoticism (a tendency to be hostile, untrusting, unemotional, lacking in empathy, and unfriendly). Adult attachment styles refer to an individual’s emotional regulation abilities, which arise from the intimate bonds built with caregivers very early during infancy, modulating the internal state and subsequent behaviors in close relationships (Bartholomew and Horowitz, 1991). An early supportive and responsive social environment, leading to a secure attachment profile, encourages both mental and cognitive health (Walsh et al., 2019). Conversely, insecure attachment has been linked to dysregulated physiological responses to stress, risky health-related behaviors, and susceptibility to serious physical illness (Pietromonaco and Beck, 2019).

Stressors

We used indices based on instruments administered in each wave to gauge the cumulative exposure to physical and psychosocial stressors during the aging process. We also include COVID-19 infection as an important source of stress, especially for the oldest individuals, who are at increased risk of hospitalization and death due to COVID-19 (Minnai et al., 2022).

Physical stressors were based on two comorbidity measures: the CIRS severity index and the number of drugs used. The CIRS (Parmelee et al., 1995) is a standardized tool compiled by a physician to rate the level of pathology and impairment of 13 major organ groups and in the psychiatric/behavioral domain on a scale ranging from 1 (no impairment) to 5 (extremely severe impairment). The severity of overall physical illness was computed as the mean of the 13 items (excluding the psychiatric/behavioral domain). For our purposes, we used the average CIRS severity score of all available assessments for each participant.

Moreover, during the geriatric visit, the number of pharmacologically active ingredients consumed in the past 2 weeks was recorded, and a list was compiled of 53 prespecified principles chosen among those drug categories most likely to interfere with cerebral activity. We used the average number of pharmacologically active ingredients recorded in all available assessments.

Psychosocial stressors were represented by the score of the Geriatric Adverse Life Events Scale (GALES) which investigates the stressful life events experienced over the previous 12 months with a 26-item checklist (Devanand et al., 2002). For the present study, we measured cumulative exposure to psychosocial stressors as the sum of the adverse events reported in all available assessments.

Finally, as a potential stressor, we included contracting COVID-19, which was assessed as a yes/no question during the geriatric visit in 2022.

Statistical Analysis

All statistical analyses were performed using SPSS Statistics 20.0 (IBM SPSS Statistics for Windows, version 20.0, BM Corp. Released 2011, Armonk, NY, US). Preliminarily, descriptive univariate analysis was performed to compare the resilient and non-resilient subgroups using Student’s t-test for continuous variables and the chi-squared test for categorical ones.

To address our research questions, we performed several steps of analysis. First, we were interested in exploring the underlying structure of the measures considered to be predictors of resilience (see Table 1 for details). To make the statistical analyses more robust and to include the entire sample evaluated, we handled missing data by performing a data imputation through Predictive Mean Matching with Full Conditional Specification (MCMC), which specifies a predictive model for each missing variable considering all other relevant variables to be predictors (i.e., the six factors, sociodemographics, stressors, and the main clinical scales of MMSE, GDS, BADL). This ensures a plausible posterior predictive distribution. Then, we performed an exploratory factor analysis (EFA) because, to the best of our knowledge, no previous study has concurrently assessed all these physical, cognitive, and psychological factors in the same sample of oldest-old individuals. We examined the correlation matrix for randomness using Bartlett’s test of sphericity, and the KMO statistic had to be greater than or equal to 0.50. The correlation matrix was factored using the principal axis factoring method after we determined that it could be factored. We assumed that the factors considered would be correlated, so an oblimin rotation was used. We applied Kaiser’s criterion (an eigenvalue greater than 1) to establish how many factors should be retained. We defined a threshold of 0.3 for variable loadings to justify the inclusion of the factors obtained and eventually rerun the analysis accordingly (Osborne, 2014). To analyze the factors as new variables, we extracted the scores using Bartlett’s method.

This step of analysis provided answers to RQ1 and reduced the dimensions of the dataset by obtaining a limited number of factors with higher internal consistency and low intercorrelation, thus addressing the issue of multicollinearity among the predictors.

Second, to investigate the relative importance of each factor in explaining resilience (RQ2), we constructed a logistic regression model with the resilience phenotype (1 = yes; 0 = no) as the dependent variable and factor scores as the independent variables, with sociodemographics (age, sex) and stressor indices as covariates. To test the performance of the models, we studied sensitivity and specificity using ROC curves.

Study Sample

Recruitment took place between mid-February 2022 and January 2023. Among the 782 eligible individuals, 27 died (3.5%), 52 were unreachable (6.6%), and 202 refused to take part (25.8%). Thus, 501 participants (64.1% of the eligible population) participated in the fifth wave, and 97 received a final diagnosis of dementia (19.4%), resulting in a sample of 404 oldest-old individuals without dementia who were eligible for the present investigation. Among them, 153 were categorized as resilient (38%) based on prespecified criteria. Before imputation, 11% of the values in the dataset were missing. Table 2 shows the comparisons of the measures of interest according to resilience status. Resilient individuals were more likely to be male and slightly younger. Overall, they showed a more favorable profile for most of the measures used as predictors of resilience. They were relatively healthier based on the CIRS index and on the number of drugs assumed, whereas no differences were found in terms of exposure to stressful life events or COVID-19 infection.

Open multimedia modal

The Structure of the Resilience-Related Factors

The first EFA on the 20 selected variables resulted in six factors, with two variables showing loadings of less than 0.3 (neuroticism and psychoticism). We reran the EFA excluding these variables, confirming the number and composition of factors resulting from the first analysis.

The correlation matrix was not random according to the results of Bartlett’s test of sphericity: χ²(190) = 1,322.818, p < .001; in addition, the KMO statistic was 0.672, above the minimal threshold for factor analysis. As a result, we decided that factor analysis would be appropriate for the correlation matrix.

Kaiser’s criterion suggests retaining six factors that, after rotation, accounted for 59% of the total variance. Table 3 shows the factor loadings from the pattern matrix. Factor 1 (17% of the total variance) included all the variables related to cognitive enrichment and stimulation in early and middle life, namely, education, occupation, childhood socioeconomic conditions, and fluid intelligence; thus, it was named “cognitive reserve.” Factor 2 (12% of the total variance) grouped psychological resilience, confidence, and extraversion; thus, we named it “affective reserve.” Factor 3 (9% of the total variance) included the need for approval, relationship as secondary, and preoccupation with relationships subscales of the Attachment Style Questionnaire (ASQ); this factor was named “insecure attachment.” Factor 4 (8% of the total variance) was named “current lifestyle” because it included physical activity, leisure time activities, social contact, adherence to the Mediterranean diet, and walking speed. Considering the direction of the loadings, we changed the sign of this factor to improve the interpretative reading of its relationship with the outcome (i.e., higher scores represent a more active lifestyle). Factor 5 (7% of the total variance), including handgrip strength and waist circumference, can be defined as “physical reserve.” Finally, Factor 6 (6% of the total variance) included only the ASQ subscale of discomfort with closeness; thus, it can be considered a measure of “Avoidant Attachment.” The correlation among the factors after rotation ranged between −0.216 and 0.183 (see Supplementary Table S1).

Open multimedia modal

Predictors of Resilience Phenotype

Table 4 displays the results of the logistic regression model. Cognitive reserve, affective reserve, and current lifestyle are the factors that significantly and independently predict the resilience phenotype (p < .05), controlling for sociodemographics and cumulative exposure to physical stressors, stressful life events during the aging process, and COVID-19 infection, with an AUC of 0.70 (95% CI: 0.64–0.76).

Open multimedia modal

Features of Resilient Participants

The “resilience phenotype” was directly ascertained through a comprehensive multidimensional assessment in 38% of the sample under study, which was composed of individuals aged between 82 and 87 years who were without dementia. Resilient individuals were more likely to be male and slightly younger, despite the narrow age range considered. As expected, and in line with our hypothesis, they showed a more favorable profile in most of the measures considered to be predictors of resilience. Moreover, they were slightly healthier based on comorbidity measures. At the same time, they did not differ from the non-resilient subgroup for exposure to stressful life events during the observation period (12 years) or for COVID-19 infection.

The Structure of the Resilience-Related Factors

The first finding of the study pertains to the structure of the main cognitive, psychological, and physical resilience-enhancing factors in old age (RQ1). For the exploratory factor analysis, we grouped the selected variables into six separate factors in a predictable and plausible way, favoring a straightforward interpretation of the underlying constructs. The cognitive reserve factor is composed of variables related to early-life cognitive enrichment, with education showing the highest loading, in line with previous research (Pinto et al., 2024). The affective reserve factor contains both trait-like factors shaped by gene—environment interactions in early life (attachment and personality traits) and state-like factors (i.e., belief and coping measured by the RS-14) known to protect against depression later in life (Laird et al., 2019). All self-report questionnaires on current habits and activities load primarily on the same factor, together with the walking speed test, a key indicator of physical health in older adults strictly related to neurosensory, muscle-skeletal, and cardiometabolic functions (Mount et al., 2019). Hence, this could be viewed as a comprehensive measure of adherence to a healthy lifestyle later in life, an established protective factor against dementia and other age-related diseases (Kivipelto et al., 2018). Finally, the factor “physical reserve” contains two measures exclusively related to physical health and functionality, namely, waist circumference and handgrip strength.

The loading values indicate direct relationships of both waist circumference and strength with the underlying factor, implying that a common proxy for overweight could be protective in our sample. This finding is in line with prior studies reporting an age-dependent relationship between body mass index and dementia risk over the course of life (Fratiglioni et al., 2020^; Livingston et al., 2020^; Rolandi et al., 2020).

Taken together, these findings contribute to advancing theoretical knowledge on the structure of physical, lifestyle, cognitive, and psychosocial protective factors in old age. Indeed, to the best of our knowledge, no previous study has concurrently measured all these independent but interrelated resilience-enhancing factors in the same sample of older adults. From a methodological standpoint, this first step of the analysis was crucial to reduce the dimensionality of our dataset to obtain a limited number of relevant predictors for the resilience phenotype, with high internal consistency and low intercorrelation, thus addressing the issue of multicollinearity among the measures of interest (Fratiglioni et al., 2020).

Predicting the Resilience Phenotype

Cognitive reserve, affective reserve, and current lifestyle significantly and independently predicted the resilience phenotype in our sample of oldest-old people, with similar impacts on the probability of being resilient and controlling for cumulative exposure to physical and psychosocial stressors during the study period, from 70–75 years to 83–87 years of age (see Figure 1). Notably, we derived these findings from multivariable models of primary data, allowing us to directly control for the main confounding factors and to estimate the unique contribution of each predictor.

Cognitive reserve and modifiable lifestyles were established as protective factors against dementia; several theoretical and empirical models have highlighted their interrelationship in determining better cognitive outcomes during the aging process (Arenaza-Urquijo et al., 2015^; Song et al., 2022).

The original finding of the present study is that the affective reserve factor—composed of variables typically related to mental health and psychological wellbeing—exerts a comparable and independent protective effect in determining resilient aging. A recent review focused on the protective effect of secure attachment on dementia introduced the concept of affective reserve to explain how favorable affective resources might protect against cognitive decline (Walsh et al., 2019). Our results could provide the first experimental insight into the existence of such protective factor in the context of aging.

Our results also have crucial translational implications. First, they further support the importance of being ambitious about disease prevention and promotion of successful aging, in line with the life-course approach for reducing dementia risk (Fratiglioni et al., 2020). Indeed, both the cognitive and affective reserve factors identified here are mostly composed of measures related to early-life exposure, which may be targeted by specific public health interventions to encourage an enriched, safe environment for younger generations. In this view, a new preventive model termed “young adult brain capital” focused on the importance of optimizing young adults’ brains and mental health to empower the next generation to become active agents in reducing their own risk of developing age-related diseases (Farina et al., 2023).

At the same time, we further confirm that it is never too late to start prevention by showing that the maintenance of an active, healthy lifestyle in very old age (80+) is an independent contributor to the resilience phenotype. This age group is a fast-growing portion of the population, posing an enormous challenge for health and social care systems but receiving limited attention in research studies. In line with our findings, a recent systematic review—which included 27 observational cohort studies—showed that eating a healthy diet with plenty of fruits and vegetables, engaging in physical and leisure/social activities, or having a combination of these lifestyle habits is associated with improved cognitive health in the oldest-old individuals (Ye et al., 2023). Targeting lifestyle habits to improve health and wellbeing (rather than to reduce disease incidence) throughout the life course may be the most effective and parsimonious approach, with positive outcomes for a broad range of age-related diseases (Hachinski and Avan, 2022).

Strengths and Limitations

The present study has several methodological strengths. To the best of our knowledge, this is the first attempt to concurrently assess the structure and impact of the chief resilience-enhancing factors in the same cohort, thus allowing us to measure their direct and independent effects on the outcome of interest. Another original aspect is the operationalization of the resilient outcome as a multidimensional construct encompassing physical, cognitive, and psychological health. This is due to the opportunity to rely on a population-based cohort prospectively followed with comprehensive geriatric and neuropsychological assessments. At the same time, we bridge theoretical knowledge from different but interrelated research fields, offering practical suggestions to prevent age-related diseases and disabilities. Finally, we were able to recruit and evaluate hundreds of individuals aged between 83 and 87 years through a personalized approach and reciprocal trust consolidated over 12 years (Guaita et al., 2013). As such, we have provided important information on a selected portion of the aged population, which has displayed the greatest growth in recent years due to increased life expectancy but is rarely or marginally included in studies on aging (Ye et al., 2023).

Some limitations should also be discussed. First, although relying on a longitudinal cohort, for our aim, we were able to perform only cross-sectional analysis because we included some measures of interest only in the most recent waves of assessment (2018: personality traits, adherence to the Mediterranean diet; 2022: childhood socioeconomic conditions, handgrip strength, psychological resilience, and attachment style). Moreover, we focused on a narrow age range in a specific geographic area, thus reducing the generalizability of our findings. Finally, the current study adopted a new definition of resilience, thus it is important to replicate the results in other national contexts/samples whereas using a similar operationalization.

Acknowledgments

The authors would like to thank all the collaborators and professionals who participated in the fifth wave of the InveCe.Ab study for their valuable work. Recruitment and telephone survey: Valeria Marzagalli, Valeria Terzoli, Valentina Biagioli, and Francesca Conti. Geriatric visits: Arcangelo Ceretti, Mauro Ceroni, Maria Cottino, Riccardo Perelli, Emanuele Poloni, and Silvia Vitali. Blood sampling: Annalisa Davin, Valentina Fantini, Riccardo Rocco Ferrari, and Xhulja Profka; Administrative staff: Federica Zagari and Luciana Mordocco. We further thank Riccardo Rocco Ferrari for Figure 1 graphical editing.