Multimorbidity,Social Engagement,and Age-Related Cognitive Decline in Older Adults from the Rancho Bernardo Study of Healthy Aging

Abstract

Background:

Multimorbidity is associated with increased rate of cognitive decline with age. It is unknown whether social engagement, which is associated with reduced risk of dementia, modifies associations between multimorbidity and cognitive decline.

Objective:

To examine the associations of multimorbidity with longitudinal cognitive test performance among community-dwelling older adults, and to determine whether associations differed by levels of social engagement.

Methods:

We used data from the Rancho Bernardo Study of Healthy Aging, a community-based prospective cohort study. Starting in 1992–1996, participants completed a battery of cognitive function tests at up to 6 study visits over 23.7 (mean = 7.2) years. Multimorbidity was defined as≥2 of 14 chronic diseases. Social engagement was assessed using items based on the Berkman-Syme Social Network Index. Multivariable linear mixed-effects models were used to test associations of multimorbidity and cognitive performance trajectories. Effect measure modification by social engagement was evaluated.

Results:

Among 1,381 participants (mean age = 74.5 years; 60.8% women; 98.8% non-Hispanic White), 37.1% had multimorbidity and 35.1% had low social engagement. Multimorbidity was associated with faster declines in Mini-Mental State Examination (MMSE; β= –0.20; 95% CI –0.35, –0.04), Trail-Making Test Part B (β= 10.02; 95% CI 5.77, 14.27), and Category Fluency (β= –0.42; 95% CI –0.72, –0.13) after adjustment for socio-demographic and health-related characteristics. Multimorbidity was associated with faster declines in MMSE among those with low compared to medium and high social engagement (p-interaction < 0.01).

Conclusions:

Multimorbidity was associated with faster declines in cognition among community-dwelling older adults. Higher social engagement may mitigate multimorbidity-associated cognitive decline.

Keywords

Alzheimer’s disease cognitive aging epidemiology morbidity multimorbidities

INTRODUCTION

Approximately 62% of adults aged 65 to 74 years in the United States (US) have multimorbidity, which is defined as the presence of at least two chronic conditions, such as diabetes and arthritis [1, 2]. In a study of multimorbidity among Medicare beneficiaries, the most prevalent combination of 3 conditions was hypertension, hyperlipidemia, and ischemic heart disease [1]. Multimorbidity is associated with 38% higher risk of mild cognitive impairment (MCI) or dementia [3]. It is plausible that multimorbidity represents the accumulation of conditions that, for example, affect dysregulation in cardiovascular- and stress-related mechanisms that increase susceptibility to aging-related outcomes such as cognitive decline [4]. Older adults with less frequent social engagement, defined as the maintenance of social connections and participation in social activities [5], may have a higher number of morbidities [6]. Social engagement includes elements of one’s social network, such as marital status, contact with friends and family as well as church involvement and group activities, that affect health and well-being [7, 8]. Involvement in these types of social activities has been associated with lower risk of dementia [9 –11], which can operate through mechanisms such as stress reduction [10].

Multimorbidity has been associated with higher risk of dementia in multiple longitudinal cohort studies of women and men, ranging from 2,176 to 10,095 participants [3 , 13]. For example, a multimorbidity index comprising 17 chronic conditions was associated with 38% higher risk of mild cognitive impairment or dementia among 2,176 older community-dwelling adults in the Mayo Clinic Study of Aging (mean age = 78.5±5.2 years) [3]. One 12-year study in the Swedish National Study on Aging and Care in Kungsholmen (SNAC-K) found that cardiometabolic-specific multimorbidity was associated with 73% higher risk of cognitive impairment and 86% higher risk of progression to dementia [12]. Several longitudinal studies have found that multimorbidity (derived from 9 to 16 conditions, such as cancer, hypertension, and stroke) was associated with lower cognitive performance across multiple domains as well as indices of global cognitive ability [4 , 15]. However, it is unknown if these associations are modified by social engagement. Understanding whether social engagement plays a role in multimorbidity-associated cognitive decline would be important in the context of the US Surgeon General’s Advisory on the poor health outcomes of declining social connection [16]. This would inform the development of dementia prevention strategies that are focused on increasing social engagement.

The objective of this study was to examine the associations of multimorbidity with longitudinal cognitive test performance among community-dwelling older adults, and to determine whether associations differed by levels of social engagement. We hypothesized that multimorbidity would be associated with faster declines in the cognitive measures that were administered to these study participants, and that declines would be faster among those with lower levels of social engagement than among those who report more frequent social engagement. The available cognitive tests are presumed measures of global cognition, executive function/speed, and category fluency.

METHODS

Study population

This prospective cohort study used data from the Rancho Bernardo Study of Healthy Aging (RBS) [17]. The RBS is a community-based prospective cohort study that began in 1972–1974 when residents of Rancho Bernardo, a suburb of San Diego, California were enrolled in a nationwide lipid research study. Beginning in 1984, surviving, community-dwelling participants were followed at approximately 4-year intervals, completing detailed demographic and health assessments. Cognitive assessment was first included in Visit 5 (1988–1992) and included in all subsequent visits. For the present study, we used data from Visit 7 (1992–1996; n = 1,781) as our baseline, with follow-up to the most recent visit, Visit 12 (2014–2016; n = 221; for a maximum follow-up time of 24 years). Visit 7 was chosen as the baseline because it contained relevant data on comorbidities. We excluded 400 participants who did not have complete cognitive data at baseline, which resulted in a final analytic sample of 1,381. Excluded participants, relative to included participants, were less likely to have multimorbidity (26.3% versus 37.1%; p < 0.01), had a lower mean age of 61.2 (SD = 12) years (p < 0.01), but did not differ by sex (p = 0.95) or social engagement (p = 0.25).

This study was approved by the University of California, San Diego Institutional Review Board. All RBS participants provided written informed consent prior to participating at each study visit.

Measures

Cognitive outcomes

Cognitive function was assessed at each visit using a short battery of standardized neuropsychological tests, including the 1) Mini-Mental State Examination (MMSE; global cognitive function); 2) Trail-Making Test Part B of the Halsted Reitan Battery (Trails B; executive function/speed); and 3) Category Fluency Test (CFT). The MMSE assesses orientation, registration, attention, calculation, language, and recall, and is scored from 0 to 30, with higher scores indicating higher global cognitive function [18, 19]. MMSE scores < 24 suggest cognitive impairment or dementia [20]. For Trails B, participants are instructed to draw a line to connect, in ascending order, an alternating series of letters and numbers on a page [21]. Trails B performance was measured as time (seconds) needed to complete this task, with shorter time suggesting better executive function (set shifting) and faster processing speed. The CFT assesses semantic memory [22]. CFT performance was defined as the number of unique animals named within 1 minute; a greater number of animals named suggests higher category fluency.

Multimorbidity

It has been recommended that multimorbidity indices should contain at least 12 chronic conditions [2]. Therefore, we defined multimorbidity as the presence of≥2 of the following 14 common chronic conditions at baseline: arthritis, asthma, cancer (excluding skin cancer), chronic kidney disease, chronic obstructive pulmonary disease (COPD), depression, diabetes, dyslipidemia, heart attack, heart failure, hypertension, liver disease, osteoporosis, and stroke. Similar diseases have been included in other multimorbidity indices in prior literature [3 , 23]. We defined evidence of significant dyslipidemia symptoms as the presence of all the following conditions at baseline: abnormal total cholesterol (>200 mg/dL), high-density lipoprotein (HDL; < 40 mg/dL for men, < 50 mg/dL for women), low-density lipoprotein (>100 mg/dL), and non-HDL (>130 mg/dL) [24]. In sensitivity analyses, we considered a cardiovascular disease multimorbidity variable defined as the presence of at least 2 related conditions at baseline (i.e., heart attack, heart failure, hypertension, stroke) versus 1 or none of these conditions. All outcomes were self-reported, excluding dyslipidemia which was based on clinical laboratory measures.

Social engagement

As described in prior RBS work [25], social engagement was assessed at Visit 7 using questions adapted from the Berkman-Syme Social Network Index [7]. We examined four different sources of self-reported social engagement based on a prior analysis in the Framingham Heart Study [26]: 1) social group involvement (0 = not active or inactive; 1 = somewhat active or very active); 2) participation in religious meetings or services (0 = not active or inactive; 1 = somewhat active or very active); 3) number of close family and friends (0 =≤2 friends and relatives; 1 =≥3 friends and relatives); and 4) marital status (0 = separated/divorced/widowed/single; 1 = married). All questions and responses are provided in Supplementary Table 1. We summed the four sources of social engagement (minimum and maximum values = 0, 4) and then categorized to three levels (low =≤2; medium = 3; or high = 4), such that higher scores suggest greater social engagement.

Covariates

Covariate data were collected at baseline. We selected covariates for the present study based on prior literature that examined the association between multimorbidity and cognitive decline or dementia [3 , 23]. Covariates of interest included age, sex, body mass index (BMI), education, smoking status, alcohol use, and physical activity (regular physical activity≥3 times per week). We also included a retest variable, which indicated whether a participant completed a cognitive test in a prior visit or not, to control for practice-related differences in performance [27].

Statistical analysis

Participant characteristics were described using mean (standard deviation [SD]), median (minimum and maximum values) or counts (percentages) by multimorbidity status. Comparisons by multimorbidity status were performed using Kruskal-Wallis tests for continuous variables and Pearson’s χ²-test for categorical variables.

We used linear mixed-effects models to estimate associations of multimorbidity with each longitudinal cognitive test performance outcome. Linear mixed-effect models accommodate missing data as well as inconsistent measurement intervals within and across participants, which allowed us to use all available data under the assumption that data were missing at random [28]. Three successively adjusted models were fit with random effects for each participant and time to account for differences in cognitive performance at baseline (intercept) as well as rate of change (slope). Model 1 included the fixed effects of multimorbidity, time (years from baseline), time-squared (to assess non-linear trends), and a multimorbidity by time interaction term, adjusted for the retest variable, age, sex, and education. We did not include a multimorbidity by time-squared interaction given no significant main effect in fully adjusted models for MMSE (p = 0.82) or CFT (p = 0.22), and a small difference in Bayesian information criterion (BIC) values in comparing Trails B models with and without the term (BIC = 40359.17 versus BIC = 40362.82, respectively). Time was standardized by grand mean centering (grand mean = 5.3 years) and dividing by the standard deviation (SD = 5.2 years) across all observations for all participants. Model 2 included Model 1 and additionally adjusted for BMI, smoking status, alcohol use, and physical activity. Model 3 included Model 2 and additionally adjusted for social engagement. We did not adjust for race and ethnicity due to the homogeneity of the study sample. For visualization, we plotted predicted values of Model 3 over standardized time for each outcome.

To assess effect measure modification by social engagement, we stratified Model 2 by subgroups of social engagement and plotted predicted values among subgroups of low, medium, and high social engagement. To determine whether effect modification by social engagement was significant, p values for interaction (p_Interaction) were calculated using likelihood ratio tests to compare models with and without a 3-way interaction of multimorbidity by time by continuous social engagement in the full (i.e., unstratified) sample. For the MMSE model among those with high social engagement only, we only included random intercepts but not random slopes due to smaller variability in the MMSE in this subgroup compared to other tests and singular fit for the model with both random slopes and intercepts. For this subgroup model, there were slightly lower BIC values for the model with random slopes only compared to the model with random slopes and intercepts (BIC = 4233.820 versus BIC = 4243.39, respectively). In exploratory analyses, we assessed effect measure modification by different sources of social engagement. Specifically, we stratified analyses for each of these variables: 1) social group involvement; 2) participation in religious meetings or services; 3) number and frequency of contact with family and friends; and 4) marital status. For these analyses, models included covariates in Model 2 as well as the three sources of social engagement that were not examined as the stratifying factor.

We performed several sensitivity analyses. Given evidence of associations between cardiometabolic multimorbidity and higher dementia risk [12], we repeated our main analysis using the presence of at least 2 cardiovascular disease-related conditions at baseline (i.e., heart attack, heart failure, hypertension, stroke) versus 1 or none of these conditions. In exploratory analyses, we also tested effect measure modification by sex and marital status, given that the association of marital status with cognition may differ by sex [9, 29]. To evaluate potential bias due to reverse causation, we repeated analyses of Trails B and CFT trajectories among participants scoring≥24 on the MMSE at Visit 7 (n = 1,351) [20]. To assess selection bias due to loss to follow-up, we repeated Model 3 weighted by stabilized inverse probability of selection weights (IPSWs) for each participant at each visit [30]. Stabilized IPSW create a pseudo-population of similar size at each visit after censoring [31]. To construct the IPSW, we calculated the inverse of propensity scores based on logistic regression to predict being selected (i.e., uncensored) at each visit, conditional on baseline multimorbidity status, selection in the prior visit, sex, age, education, BMI, smoking status, alcohol use, physical activity, and social engagement. Weights were stabilized by the probability of selection at each visit, conditional on baseline multimorbidity status and selection at the prior visit, and then truncated at the 5th and 95th percentiles. To assess bias due to the competing risk of mortality, we used joint models where we simultaneously fit a Cox proportional hazards model for mortality during the follow-up period and linear mixed-effects Model 1 for each outcome [32]. To examine results among older adults only, analyses were repeated after excluding participants who were < 50 years old at baseline (n = 13). Finally, because those who did not report participation in religious activities could not have been categorized into the high level of social engagement based on our definition, we assessed effect measure modification using a social engagement variable that did not consider participation in religious activities.

All statistical analyses were conducted in R (version 4.2.2; R Core Team, 2022). The lme4 package was used for linear mixed-effects analyses [33], the sjPlot package was used for plotting predicted values [34], and the JM, survival and nlme packages were used for joint modeling [35 –37]. Statistical tests were two-tailed, and results with p values < 0.05 were considered significant.

RESULTS

Participant characteristics

Of the 1,381 participants, 37.1% had multimorbidity at baseline (Table 1). Most participants were non-Hispanic White (98.8%), women (60.8%), and 74.5 years (SD = 9.4) of age, on average, at baseline. There were no differences between women and men by age at baseline (mean [SD] age 74.6 [9.7] versus 74.2 [9.0] years; p = 0.25). Although the sample had a wide age range, most participants were aged 50 and older (n = 1368).

Table 1

Baseline characteristics by multimorbidity status of participants in the Rancho Bernardo Study of Health Aging

	Multimorbidity status at Baseline
Characteristic		No (N = 869)	Yes (N = 512)	Overall (N = 1381)
Age (y)	Mean (SD)	73.1 (9.75)	76.8 (8.25)	74.5 (9.39)
	Median [Min, Max]	73.9 [30.9, 93.5]	77.8 [41.4, 96.1]	75.0 [30.9, 96.1]
Sex	Female	512 (58.9%)	328 (64.1%)	840 (60.8%)
Education	<12 years	44 (5.1%)	14 (2.8%)	58 (4.2%)
	12 years	212 (24.7%)	115 (22.6%)	327 (23.9%)
	13–15 years	271 (31.6%)	162 (31.9%)	433 (31.7%)
	16 years	197 (23.0%)	138 (27.2%)	335 (24.5%)
	≥17 years	134 (15.6%)	79 (15.6%)	213 (15.6%)
Body mass index (kg/m²)	Median [Min, Max]	24.6 [15.5, 45.9]	24.6 [16.2, 44.6]	24.6 [15.5, 45.9]
Smoking status	Never	367 (42.6%)	210 (41.1%)	577 (42.1%)
	Former	423 (49.1%)	276 (54.0%)	699 (50.9%)
	Current	71 (8.2%)	25 (4.9%)	96 (7.0%)
Alcohol consumption	Non-drinker	108 (12.5%)	100 (19.5%)	208 (15.1%)
	Daily	318 (36.9%)	165 (32.2%)	483 (35.2%)
	2–4 per week	95 (11.0%)	47 (9.2%)	142 (10.3%)
	1-2 per week	125 (14.5%)	60 (11.7%)	185 (13.5%)
	1-2 per month	108 (12.5%)	74 (14.5%)	182 (13.3%)
	<1 per month	107 (12.4%)	66 (12.9%)	173 (12.6%)
Physical activity	≥3 times/week	651 (75.4%)	347 (67.8%)	998 (72.6%)
Social engagement index	Low (score≤2)	295 (34.3%)	190 (37.3%)	485 (35.5%)
	Medium (score = 3)	357 (41.6%)	202 (39.6%)	559 (40.8%)
	High (score = 4)	207 (24.1%)	118 (23.1%)	325 (23.7%)
Social group activity	Somewhat or very active	664 (77.0%)	410 (80.1%)	1074 (78.2%)
Participation in religious meetings or services	Somewhat or very active	359 (41.6%)	239 (46.7%)	598 (43.5%)
Close friends and relatives	≥3	747 (86.7%)	447 (87.5%)	1194 (87.0%)
Marital Status	Married	648 (74.7%)	318 (62.2%)	966 (70.1%)

n missing: education = 15 (1.1%); body mass index = 3 (0.2%); smoking status = 9 (0.7%); alcohol consumption = 8 (0.6%); physical activity = 6 (0.4%); social engagement index = 12 (0.9%); group activity = 7 (0.5%); participation in religious meetings or services = 7 (0.5%); close friends and relatives = 8 (0.6%); marital status = 3 (0.2%).

Compared to participants without multimorbidity, those with multimorbidity were older (mean [SD] age = 76.8 [8.3] versus 73.1 [9.8] years; p < 0.01), more likely to be former smokers (54.0% versus 49.1%; p = 0.03), less likely to report daily alcohol consumption (32.2% versus 36.6%; p < 0.01), and less likely to be married (37.8% versus 25.3%; p < 0.01) at baseline. The social engagement index did not differ by multimorbidity status (p = 0.30). Among those with multimorbidity, the three most common chronic conditions were hypertension (58.8%), arthritis (43.8%), and history of cancer (31.3%; Supplementary Table 2). Participants with multimorbidity had shorter mean follow up time of 5.5 (SD = 5.7) years compared to those without multimorbidity (mean = 8.2 [SD = 6.5] years). Follow up time also decreased as social engagement decreased (high engagement mean = 8.3 [SD = 6.4] years; medium engagement = 7.5 [SD = 6.2] years; low engagement = 6.3 [SD = 6.5] years). There were differences for baseline cognitive measures of Trails B (p < 0.01) and CFT (p = 0.04), but not MMSE (p = 0.12), by the social engagement index after adjustment for age, sex, and education (Supplementary Table 16).

Baseline multimorbidity and cognitive function by time

Over the average 7.2 (SD = 6.4) years of follow up for the sample, multimorbidity at baseline was associated with greater decline in cognitive performance (Table 2, Fig. 1). In Model 1, multimorbidity was associated with faster declines in MMSE (β= –0.19; 95% CI –0.35, –0.04), Trails B (β= 9.75; 95% CI 5.51, 13.99), and CFT (β= –0.40; 95% CI –0.70, –0.11) after adjustment for the retest variable, age, sex, and education. Results were similar after adjustment for BMI, smoking status, alcohol use and physical activity in Model 2, and after further adjustment for social engagement in Model 3.

Fig. 1

Trajectories of cognition function test performance over years since baseline (standardized) by multimorbidity status at baseline in the Rancho Bernardo Study of Healthy Aging. Plots are based on model coefficients adjusted for the retest variable, age, sex, education, body mass index (standardized), smoking, alcohol use, physical activity, and social engagement. The y-axis for Trails B is inverted so that for all graphs, steeper downward sloping lines indicate poorer performance.

Table 2

Associations of multimorbidity with longitudinal cognitive test performance among participants of the Rancho Bernardo Study of Healthy Aging

Outcome	Model	Term	β (95% CI)
MMSE	1	Multimorbidity	0.02 (–0.18; 0.22)
		Time	–0.50 (–0.59; –0.41)
		Time²	0.18 (0.13; 0.22)
		Multimorbidity×time	–0.19 (–0.35; –0.04)
	2	Multimorbidity	<0.01 (–0.20; 0.20)
		Time	–0.5 (–0.59; –0.42)
		Time²	0.17 (0.13; 0.22)
		Multimorbidity×time	–0.20 (–0.35; –0.04)
	3	Multimorbidity	0.01 (–0.19; 0.21)
		Time	–0.50 (–0.59; –0.41)
		Time²	0.18 (0.13; 0.22)
		Multimorbidity×time	–0.20 (–0.36; –0.04)
Trails B	1	Multimorbidity	11.21 (4.83; 17.59)
		Time	11.66 (9.35; 13.98)
		Time²	3.71 (2.59; 4.83)
		Multimorbidity×time	9.75 (5.51; 13.99)
	2	Multimorbidity	11.29 (4.87; 17.71)
		Time	11.76 (9.43; 14.09)
		Time²	3.68 (2.56; 4.80)
		Multimorbidity×time	10.02 (5.77; 14.27)
	3	Multimorbidity	10.78 (4.39; 17.17)
		Time	11.80 (9.46; 14.13)
		Time²	3.68 (2.55; 4.80)
		Multimorbidity×time	9.91 (5.66; 14.17)
CFT	1	Multimorbidity	–0.44 (–0.89; 0.02)
		Time	–1.36 (–1.53; –1.19)
		Time²	–0.05 (–0.14; 0.04)
		Multimorbidity×time	–0.40 (–0.70; –0.11)
	2	Multimorbidity	–0.39 (–0.85; 0.06)
		Time	–1.36 (–1.53; –1.19)
		Time²	–0.05 (–0.14; 0.04)
		Multimorbidity×time	–0.42 (–0.72; –0.13)
	3	Multimorbidity	–0.37 (–0.83; 0.09)
		Time	–1.36 (–1.54; –1.19)
		Time²	–0.05 (–0.14; 0.04)
		Multimorbidity×time	–0.42 (–0.72; –0.13)

Time is defined as time from baseline in years, standardized by centering using the grand mean and dividing by the standard deviation. Model 1 adjusted for retest (yes; no), age (years), sex (female; male), and education years (<12; 12; 13–15; 16;≥17). Model 2 included Model 1 and adjusted for BMI (kg/m2; standardized), smoking status (never; former; current), alcohol use (non-drinker; daily; 2–4 per week; 1-2 per month; < 1 per month), and physical activity (<3 times/week;≥3 times/week). Model 3 included Model 2 and adjusted for social engagement (≤2; 3; 4).

Effect measure modification by social engagement

In models stratified by social engagement, multimorbidity, relative to no multimorbidity was associated with greater decline on MMSE among those with low social engagement only (β= –0.38; 95% CI –0.70, –0.06; p_Interaction < 0.01, Table 3, Fig. 2). There were no significant differences in rate of decline by multimorbidity among those with medium (β= –0.14; 95% CI –0.39, 0.12) or high (β= –0.12; 95% CI –0.35, 0.12) social engagement, adjusted for the retest variable, age, sex, education, BMI, smoking status, alcohol use, and physical activity. The multimorbidity by time by social engagement interaction was not statistically significant for Trails B (p_Interaction = 0.07) or CFT (p_interaction = 0.35).

Fig. 2

Trajectories of cognition function test performance over years since baseline (standardized) by multimorbidity status at baseline (red = no; blue = yes) stratified by social engagement level in the Rancho Bernardo Study of Healthy Aging. Plots are based on model coefficients adjusted for the retest variable, age, sex, education, body mass index (standardized), smoking, alcohol use, and physical activity. The y-axis for Trails B is inverted so that for all graphs, steeper downward sloping lines indicate poorer performance.

Table 3

Associations of multimorbidity with longitudinal cognitive test performance by social engagement among participants of the Rancho Bernardo Study of Healthy Aging

	Social engagement
		Low (n = 485)	Medium (n = 559)	High (n = 325)
Outcome	Term	β (95% CI)	β (95% CI)	β (95% CI)	p _interaction
MMSE	Multimorbidity	0.27 (–0.07; 0.62)	–0.18 (–0.49; 0.12)	–0.12 (–0.51; 0.28)	0.0064
	Time	–0.55 (–0.73; –0.36)	–0.51 (–0.65; –0.37)	–0.43 (–0.58; –0.29)
	Time²	0.15 (0.07; 0.23)	0.15 (0.07; 0.22)	0.19 (0.11; 0.27)
	Multimorbidity×time	–0.38 (–0.70; –0.06)	–0.14 (–0.39; 0.12)	–0.12 (–0.35; 0.12)
Trails B	Multimorbidity	6.55 (–5.06; 18.16)	11.57 (1.83; 21.30)	17.50 (4.50; 30.50)	0.0691
	Time	13.96 (9.34; 18.59)	11.59 (7.76; 15.42)	10.10 (6.54; 13.67)
	Time²	3.48 (1.30; 5.67)	5.66 (3.83; 7.48)	1.43 (–0.31; 3.18)
	Multimorbidity×time	12.13 (3.89; 20.37)	8.20 (1.05; 15.35)	10.13 (3.82; 16.45)
CFT	Multimorbidity	–0.01 (–0.79; 0.77)	–0.24 (–0.96; 0.49)	–1.04 (–1.98; –0.10)	0.3513
	Time	–1.27 (–1.58; –0.96)	–1.55 (–1.83; –1.28)	–1.19 (–1.51; –0.87)
	Time²	–0.14 (–0.31; 0.02)	–0.14 (–0.29; 0.01)	0.14 (–0.03; 0.31)
	Multimorbidity×time	–0.45 (–0.97; 0.08)	–0.21 (–0.70; 0.27)	–0.68 (–1.22; –0.14)

Time is defined as time from baseline in years, standardized by centering using the grand mean and dividing by the standard deviation. Models are adjusted for retest (yes; no), age (years), sex (female; male), education years (<12; 12; 13–15; 16;≥17), BMI (kg/m²; standardized), smoking status (never; former; current), alcohol use (non-drinker; daily; 2–4 per week; 1-2 per month; < 1 per month), and physical activity (<3 times/week;≥3 times/week). p_interaction is derived from likelihood ratio tests comparing models with and without a 3-way interaction of multimorbidity by time by continuous social engagement.

Exploratory analysis: Effect modification by individual sources of social engagement

There was no evidence to suggest effect measure modification by social group involvement (ps_interaction > 0.07, Supplementary Table 3). Participants with≤2 close friends and relatives had faster multimorbidity-associated declines in MMSE compared to participants with≥3 close friends and relatives, adjusted for the retest variable, age, sex, education, BMI, smoking status, alcohol use, physical activity, social group involvement, participation in religious meetings or services, and marital status (p_interaction < 0.01, Supplementary Table 4). Among those who were married, multimorbidity was associated with faster declines in Trails B (p_interaction < 0.01) adjusted for the retest variable, age, sex, education, BMI, smoking status, alcohol use, social group involvement, participation in religious meetings or services, and number of close family and friends (Supplementary Table 5). The multimorbidity by time by participation in religious meetings or services interactions were not significant for any outcome (ps_interaction > 0.15, Supplementary Table 6).

Sensitivity analysis

Cardiovascular disease multimorbidity was associated with faster declines in MMSE and Trails B, but not CFT in fully adjusted models (Supplementary Table 7). In sex-stratified analyses, women had faster multimorbidity-associated declines in MMSE, Trails B, and CFT than men (Supplementary Table 8). However, interactions by sex were not significant (Supplementary Table 8). In exploratory analyses further stratified by sex and marital status, multimorbidity was associated with faster Trails B declines among women who were married but not among those who were not married, whereas multimorbidity was associated with faster declines among non-married men vs. married men (p_interaction < 0.01, Supplementary Table 9). After restricting analyses to participants who scored≥24 on the MMSE at baseline, multimorbidity was associated with faster declines for Trails B and CFT in fully adjusted models, consistent with our main analysis (Supplementary Table 10). Results were also similar after applying stabilized IPSWs to our analyses, suggesting minimal impact of selection bias due to loss to follow-up (Supplementary Table 11). Associations were similar in joint models that accounted for the increased risk of mortality among those with multimorbidity (versus no multimorbidity) after controlling for age, sex, and education (HR = 1.61; 95% CI 1.41–1.85; Supplementary Table 12). Excluding individuals < 50 years of age (n = 13) had minimal effects on the results (Supplementary Table 13), with the exception that the interaction of social engagement with multimorbidity status was significant for both MMSE (p_interaction < 0.01) and Trails B (p_interaction = 0.01, Supplementary Table 14). Multimorbidity-related declines in MMSE were only significant for those in the low social engagement group (Supplementary Table 14). Rates of multimorbidity-related Trails B decline were greater among the low social engagement group, followed by high and medium groups (Supplementary Table 14). In the full cohort, results were similar when examining effect modification by the social engagement index that excluded consideration of religious social activities (Supplementary Table 15). With this social engagement index, significant effect modification was observed for MMSE (p_interaction < 0.01) and Trails B (p_interaction = 0.02); multimorbidity-related declines in MMSE and Trails B were faster among those with low social engagement, followed by the medium then high social engagement groups (Supplementary Table 15).

DISCUSSION

In this prospective cohort study of 1,381 older adults, multimorbidity was associated with faster declines in global cognitive function, executive function/speed, and category fluency over an average 7.2 years of follow-up. These findings varied by level of social engagement. Participants with lower social engagement had a larger magnitude of multimorbidity-associated declines in global cognitive function and executive function/speed (in those over 50 years of age) compared to those with medium and high social engagement. Findings for category fluency did not significantly vary by social engagement.

Our results are generally consistent with prior literature that multimorbidity relative to no multimorbidity is associated with greater cognitive decline [4 , 15]. However, results vary by specific tests and study population. For example, in one analysis in the Health and Retirement Study of 14,265 participants with approximately 14 years of follow up, a multimorbidity-weighted index was associated with faster declines in global cognitive function, immediate recall, and working memory, but not delayed recall [15]. In contrast, in a study of 756 participants from the Baltimore Longitudinal Study of Aging with an average of 3 years of follow up, multimorbidity was associated with faster declines in verbal fluency, but not global function, executive function, spatial ability, or memory [4]. It is important to note that cross-study comparisons are challenging given differences in morbidities and tests of cognitive domains that were assessed. However, these studies, in conjunction with our own, support Fabbri et al.’s [38] hypothesis that multimorbidity coincides with multiple aging phenotypes, highlighting the importance of reducing multimorbidity in our aging populations.

The present study extends previous work by examining whether the associations of multimorbidity with cognitive function differ by social engagement, which has been associated with lower risk of dementia [11]. Among participants with low social engagement, multimorbidity was associated with faster declines in our measure of global cognitive function compared to those with higher social engagement. Multimorbidity was also associated with faster decline in our measure of executive function in analyses restricted to older adults (<50 years). This is consistent with meta-analyses that suggest that higher social engagement is associated with lower risk of dementia [9, 11], and a cross-sectional study of 838 older adults in the Rush Memory and Aging Project that found that higher frequency of social activity and perceived social support were associated with better global cognitive function [39]. In exploratory analyses that stratified by sources of social engagement, we found that the modifying effects of social engagement differed by type of social contact and by cognitive outcome. Participants with≤2 close friends and relatives had faster multimorbidity-associated declines than those with more close friends and relatives. Future studies should consider examining which features of social engagement activities have the most impact.

Although the interaction was not significant, likely due in part to lower power, we found that women generally had faster multimorbidity-associated cognitive decline compared to men. Conversely, one study of 2,176 participants in the Mayo Clinic Study of Aging found that multimorbidity was associated with higher MCI risk among men compared to women over a median 4 years of follow up [3]. Older women may be more impacted by changes to marital status if they outlive their husbands and become widowed, potentially reducing their social contact or engagement [9]. Unexpectedly, after stratifying by marital status, we found that multimorbidity was associated with faster declines in executive function/speed among women who were married compared to not married. This is at odds with others who found that being married is associated with lower risk of dementia compared to being widowed or single [29]. After we stratified by sex, we observed faster declines among married women and unmarried men. However, it is important to note that the sample sizes were low in our stratified analyses. These interactions need to be investigated in larger study populations.

Multimorbidity likely affects cognitive function through dysregulation in multiple pathways including cardiovascular and stress-related systems [4] as well as metabolic mechanisms (e.g., via diabetes) [40]. Social engagement may modify the relationship between multimorbidity and cognitive health given its association with positive health behaviors such as physical activity [41], which play a role in cardiovascular health [42] and multimorbidity risk [43]. Participating in enriching social activities that are intellectually stimulating or meaningful may promote slower cognitive decline, especially among older adults [44]. Fratiglioni et al. [10] posit that the social, mental, and physical lifestyle components in late life operate via common pathways through the cognitive reserve, vascular, and stress hypotheses to affect dementia risk. Perry et al. [45] hypothesize that social connectedness, via social bridging and bonding, can affect neurodegenerative, neuroendocrine, and hypothalamic-pituitary-adrenal stress response pathways related to cognitive aging. Under the stress hypothesis, for example, social engagement can lead to positive emotion states that reduce stress [10]. Reducing chronic stress can lower Alzheimer’s disease risk [46]. According to the glucocorticoid cascade hypothesis, stress can induce corticosterone hypersecretion, potentially resulting in loss of hippocampal neurons [47]. More work is needed to understand the mechanisms by which different sources of social engagement may mitigate multimorbidity-associated cognitive decline.

There are several limitations to our study. First, bias due to reverse causation is possible if cognitive impairment resulted in lower levels of social engagement. However, results were similar after excluding participants with evidence of cognitive impairment at baseline, as determined by a score < 24 on the MMSE [20]. Second, there was differential follow-up time by baseline multimorbidity status and potential for selection bias due to loss to follow-up. However, results were similar after applying stabilized IPSW and joint modelling. Third, there is no standard method of operationalizing multimorbidity [1, 38]. Results may differ for multimorbidity indices including different diseases. Although we were underpowered to examine specific combinations of morbidities, we found similar results using a multimorbidity index that considered cardiovascular disease conditions only. Fourth, it is likely that multimorbidity and social engagement change over time. Additionally, the relationships between multimorbidity, social engagement and cognitive decline may follow a different temporal path than we have hypothesized and tested here. For example, multimorbidity or lower baseline cognitive function may affect one’s ability to partake in certain types of social contact or activities, and thus affect cognitive function downstream. The wide range of potential mechanisms make it difficult to determine which aspects of social engagement play the most prominent role in these associations. Lastly, the RBS population is primarily non-Hispanic, White, and well-educated, and results may not generalize to other populations.

There were several strengths to our study, including up to 23.7 years of longitudinal follow-up. This allowed us to leverage repeated assessments of cognitive function and assess longitudinal associations between multimorbidity and cognitive decline. We also considered morbidities that were included in multimorbidity indices in other study populations.

In this study of community-dwelling older adults, participants with multimorbidity had faster declines in global cognitive ability, executive function/speed, and category fluency, relative to those without multimorbidity. Multimorbidity-associated declines in global cognitive function and executive function/speed (among participants > 50 years of age) were more pronounced among those with lower social engagement compared to those with higher levels of social engagement. These findings contribute to the US Surgeon General’s call for more research on social connection indicators that influence health outcomes [16]. With the projected rise in older adults with multimorbidity and dementia [48, 49], further study is needed to confirm these findings in more diverse populations, and to determine whether social engagement interventions may mitigate multimorbidity-associated cognitive decline.

Footnotes

ACKNOWLEDGMENTS

The authors have no acknowledgments to report.

FUNDING

This work was supported by a National Institutes of Health, National Institute on Aging, T32 Predoctoral Training Fellowship (T32 AG058529 to A.I.B.P.). A.H.S. and L.K.M. were supported by National Institute on Aging grants RF1AG079149 and RF1AG074345. H.P. was supported by the National Cancer Institute (K01 CA234317), the SDSU/UCSD Comprehensive Cancer Center Partnership (U54 CA132384 & U54 CA132379), and the Alzheimer’s Disease Resource Center for Advancing Minority Aging Research at the University of California San Diego (P30 AG059299). Data collection for the Rancho Bernardo Study of Healthy Aging was provided primarily by the National Institutes of Health (including grant numbers: HV012160, AA021187, AG028507, AG007181, DK31801, HL034591, HS06726 and HL089622). Archiving and sharing of RBS data was supported by AG054067. RBS data is available through the RBS website: .

CONFLICT OF INTEREST

Alexander Ivan B. Posis is an Editorial Board Member of this journal but was not involved in the peer-review process nor had access to any information regarding its peer-review.

All other authors have no conflict of interest to report.

DATA AVAILABILITY

The data supporting the findings of this study are openly available from The Rancho Bernardo Study (RBS) of Healthy Aging at .

The supplementary material is available in the electronic version of this article: .

References

Salive

(2013) Multimorbidity in older adults. Epidemiol Rev 35, 75–83.

Fortin

, Stewart

, Poitras

M-E

, Almirall

, Maddocks

(2012) A systematic review of prevalence studies on multimorbidity: Toward a more uniform methodology. Ann Fam Med 10, 142–151.

Vassilaki

, Aakre

, Cha

, Kremers

, St. Sauver

, Mielke

, Geda

, Machulda

, Knopman

, Petersen

, Roberts

(2015) Multimorbidity and risk of mild cognitive impairment. J Am Geriatr Soc 63, 1783–1790.

Fabbri

, An

, Zoli

, Tanaka

, Simonsick

, Kitner-Triolo

, Studenski

, Resnick

, Ferrucci

(2016) Association of accelerated multi-morbidity and age-related cognitive decline in older, non-demented participants from the Baltimore Longitudinal Study of Aging. J Am Geriatr Soc 64, 965–972.

Bassuk

, Glass

, Berkman

(1999) Social disengagement and incident cognitive decline in community-dwelling elderly persons. Ann Intern Med 131, 165–173.

Cherry

, Walker

, Brown

, Volaufova

, LaMotte

, Welsh

, Su

, Jazwinski

, Ellis

, Wood

, Frisard

(2013) Social engagement and health in younger, older, and oldest-old adults in the Louisiana Healthy Aging Study. J Appl Gerontol 32, 51–75.

Berkman

, Syme

(1979) Social networks, host resistance, and mortality: A nine-year follow-up study of Alameda County residents. Am J Epidemiol 109, 186–204.

Bath

, Deeg

(2005) Social engagement and health outcomes among older people: Introduction to a special section. Eur J Ageing 2, 24–30.

Livingston

, Huntley

, Sommerlad

, Ames

, Ballard

, Banerjee

, Brayne

, Burns

, Cohen-Mansfield

, Cooper

, Costafreda

, Dias

, Fox

, Gitlin

, Howard

, Kales

, Kivimäki

, Larson

, Ogunniyi

, Orgeta

, Ritchie

, Rockwood

, Sampson

, Samus

, Schneider

, Selbæk

, Teri

, Mukadam

(2020) Dementia prevention, intervention, and care: 2020 report of the Lancet Commission. Lancet 396, 413–446.

10.

Fratiglioni

, Paillard-Borg

, Winblad

(2004) An active and socially integrated lifestyle in late life might protect against dementia. Lancet Neurol 3, 343–353.

11.

Penninkilampi

, Casey

A-N

, Singh

, Brodaty

(2018) The association between social engagement, loneliness, and risk of dementia: A systematic review and meta-analysis. J Alzheimers Dis 66, 1619–1633.

12.

Dove

, Marseglia

, Shang

, Grande

, Vetrano

, Laukka

, Fratiglioni

, Xu

(2023) Cardiometabolic multimorbidity accelerates cognitive decline and dementia progression. Alzheimers Dement 19, 821–830.

13.

Hassen

, Fayosse

, Landré

, Raggi

, Bloomberg

, Sabia

, Singh-Manoux

(2022) Association between age at onset of multimorbidity and incidence of dementia: 30 year follow-up in Whitehall II prospective cohort study. BMJ 376, e068005.

14.

Espeland

, Justice

, Bahnson

, Evans

, Munshi

, Hayden

, Simpson

, Johnson

, Johnston

, Kritchevsky

the Look AHEAD Research Group (2021) Eight-year changes in multimorbidity and frailty in adults with type 2 diabetes mellitus: Associations with cognitive and physical function and mortality. J Gerontol A Biol Sci Med Sci 77, 1691–1698.

15.

Wei

, Levine

, Zahodne

, Kabeto

, Langa

(2020) Multimorbidity and cognitive decline over 14 years in older Americans. J Gerontol A Biol Sci Med Sci 75, 1206–1213.

16.

US Department of Health and Human Services (2023) Our epidemic of loneliness and isolation: The US Surgeon General’s advisory on the healing effects of social connection and community. https://www.hhs.gov/sites/default/files/surgeon-general-social-connection-advisory.pdf

17.

Barrett-Connor

(2013) The Rancho Bernardo Study: 40 years studying whywomen have less heart disease than men and how diabetes modifies women’s usual cardiac protection. Glob Heart 8, 10.1016/j.gheart.2012.12.002.

18.

Folstein

, Folstein

, McHugh

(1975) “Mini-mental state”: A practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res 12, 189–198.

19.

Teng

, Chui

(1987) The Modified Mini-Mental State (3MS) examination. J Clin Psychiatry 48, 314–8.

20.

Tombaugh

, McIntyre

(1992) The mini-mental state examination: A comprehensive review. J Am Geriatr Soc 40, 922–935.

21.

Reitan

(1958) Validity of the Trail Making Test as an indicator of organic brain damage. Percept Mot Skills 8, 271–276.

22.

Butters

, Granholm

, Salmon

, Grant

, Wolfe

(1987) Episodic and semantic memory: A comparison of amnesic and demented patients. J Clin Exp Neuropsychol 9, 479–497.

23.

Bunn

, Burn

A-M

, Goodman

, Rait

, Norton

, Robinson

, Schoeman

, Brayne

(2014) Comorbidity and dementia: A scoping review of the literature. BMC Med 12, 192.

24.

Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (2001) Executive Summary of The Third Report of The National Cholesterol Education Program (NCEP) Expert Panel on Detection, Evaluation, And Treatment of High Blood Cholesterol In Adults (Adult Treatment Panel III). JAMA 285, 2486–2497.

25.

Alattar

, Bergstrom

, Laughlin

, Kritz-Silverstein

, Richard

, Reas

, Harris

, Barrett-Connor

, McEvoy

(2020) Hearing impairment and cognitive decline in older, community-dwelling adults. J Gerontol A Biol Sci Med Sci 75, 567–573.

26.

Loucks

, Sullivan

, D’Agostino

Sr , Larson

, Berkman

, Benjamin

(2006) Social networks and inflammatory markers in the Framingham Heart Study. J Biosoc Sci 38, 835–842.

27.

Reas

, Laughlin

, Bergstrom

, Kritz-Silverstein

, Barrett-Connor

, Mcevoy

(2017) Effects of sex and education on cognitive change over a 27-year period in older adults: The Rancho Bernardo Study HHS Public Access. Am J Geriatr Psychiatry 25, 889–899.

28.

Little

RJA

, Rubin

(2019) Statistical analysis with missing data, John Wiley & Sons, Hoboken, NJ, USA.

29.

Sommerlad

, Ruegger

, Singh-Manoux

, Lewis

, Livingston

(2018) Marriage and risk of dementia: Systematic review and meta-analysis of observational studies. J Neurol Neurosurg Psychiatry 89, 231–238.

30.

Cole

, Hernán

(2008) Constructing inverse probability weights for marginal structural models. Am J Epidemiol 168, 656–664.

31.

Hernán

, Robins

(2020) Causal inference: What if, Taylor & Francis.

32.

Ibrahim

, Chu

, Chen

(2010) Basic concepts and methods for joint models of longitudinal and survival data. J Clin Oncol 28, 2796–2801.

33.

Bates

, Mächler

, Bolker

, Walker

(2014) Fitting Linear Mixed-Effects Models using lme4. J Stat Softw 67, 1–48.

34.

Lüdecke

(2022) sjPlot: DataVisualization for Statistics in Social Science. R package version 2.8.15, https://CRAN.Rproject.org/package=sjPlot.

35.

Rizopoulos

(2010) JM: An R package for the joint modelling of longitudinal and time-to-event data. J Stat Softw 35, 1–33.

36.

Pinheiro

, Bates

, DebRoy

, Sarkar

R Core Team (2021) nlme: Linear and Nonlinear Mixed Effects Models; R package version 3.1-152.

37.

Therneau

(2023) A package for survival analysis in R; R package version 3.5-5.

38.

Fabbri

, Zoli

, Gonzalez-Freire

, Salive

, Studenski

, Ferrucci

(2015) Aging and multimorbidity: New tasks, priorities, and frontiers for integrated gerontological and clinical research. J Am Med Dir Assoc 16, 640–647.

39.

Krueger

, Wilson

, Kamenetsky

, Barnes

, Bienias

, Bennett

(2009) Social engagement and cognitive function in old age. Exp Aging Res 35, 45–60.

40.

Zilliox

, Chadrasekaran

, Kwan

, Russell

(2016) Diabetes and cognitive impairment. Curr Diab Rep 16, 87.

41.

Samuel

, Dennison Himmelfarb

, Szklo

, Seeman

, Echeverria

, Diez Roux

(2015) Social engagement and chronic disease risk behaviors: The Multi-Ethnic Study of Atherosclerosis. Prev Med 71, 61–66.

42.

Lloyd-Jones

, Allen

, Anderson

CAM

, Black

, Brewer

, Foraker

, Grandner

, Lavretsky

, Perak

, Sharma

, Rosamond

American Heart Association (2022) Life’s Essential 8: Updating and Enhancing the American Heart Association’s Construct of Cardiovascular Health: A Presidential Advisory From the American Heart Association. Circulation 146, e18–e43.

43.

Delpino

, de Lima

APM

, da Silva

BGC

, Nunes

, Caputo

, Bielemann

(2022) Physical activity and multimorbidity among community-dwelling older adults: A systematic review with meta-analysis. Am J Health Promot 36, 1371–1385.

44.

Hertzog

, Kramer

, Wilson

, Lindenberger

(2008) Enrichment effects on adult cognitive development: Can the functional capacity of older adults be preserved and enhanced? Psychol Sci Public Interest 9, 1–65.

45.

Perry

, McConnell

, Coleman

, Roth

, Peng

, Apostolova

(2022) Why the cognitive “fountain of youth” may be upstream: Pathways to dementia risk and resilience through social connectedness. Alzheimers Dement 18, 934–941.

46.

Wilson

, Evans

, Bienias

, Mendes de Leon

, Schneider

, Bennett

(2003) Proneness to psychological distress is associated with risk of Alzheimer’s disease. Neurology 61, 1479–1485.

47.

Sapolsky

, Krey

, McEwen

(1986) The neuroendocrinology of stress and aging: The glucocorticoid cascade hypothesis. Endocr Rev 7, 284–301.

48.

Kingston

, Robinson

, Booth

, Knapp

, Jagger

for the MODEM project (2018) Projections of multi-morbidity in the older population in England to 2035: Estimates from the Population Ageing and Care Simulation (PACSim) model. Age Ageing 47, 374–380.

49.

Gauthier

, Rosa-Neto

, Morais

, Webster

(2021) World Alzheimer Report 2021. Journey through the diagnosis of dementia, London, England.