Introduction
With the aging of much of the world’s population, the prevalence of dementia is projected to sharply increase by mid-century (Hebert et al., Reference Hebert, Weuve, Scherr and Evans2013). The need to confront this growing public health challenge has stimulated research on potentially modifiable risk factors for dementia. Much of this research has focused on social engagement (Hertzog et al., Reference Hertzog, Kramer, Wilson and Lindenberger2009). In old age, there is substantial evidence that smaller social network size (Fratiglioni et al., Reference Fratiglioni, Wang, Ericsson, Maytan and Winblad2000; Crooks et al., Reference Crooks, Lubben, Petitti, Little and Chiu2008; Rafnsson et al., Reference Rafnsson, Orrell, d’Orsi, Hogervorst and Steptoe2020) and higher level of emotional isolation (Rafnsson et al. Reference Rafnsson, Orrell, d’Orsi, Hogervorst and Steptoe2020; Tilvis et al., Reference Tilvis2004; Wilson et al., Reference Wilson2007; Shankar et al., Reference Shankar, Hamer, McMunn and Steptoe2013; Holwerda et al., Reference Holwerda2014; Sutin et al., Reference Sutin, Stephan, Luchetti and Terraccciano2020; Sundstrom et al., Reference Sundstrom, Adolfsson, Nordin and Adolfsson2020) are associated with increased risk of cognitive decline and dementia. However, with few exceptions (Sutin et al., Reference Sutin, Stephan, Luchetti and Terraccciano2020) this research has been based on predominantly white persons of non-Latin descent. It is uncertain, therefore, whether social engagement is related to dementia in racial and ethnic minorities.
In this paper, we examine the correlation between levels of social and emotional isolation and likelihood of dementia. As part of the Pathology, Alzheimer’s and Related Dementias Study (PARDoS), knowledgeable informants of 1,493 older Brazilian decedents underwent a uniform structured interview to assess social and emotional isolation, history of major depression, and dementia and mild cognitive impairment (MCI). In analyses, we tested whether social and emotional isolation were related to the likelihood of dementia and MCI.
Methods
Decedents and informants
PARDoS enrolls deceased older persons who died from natural non-violent causes (e.g. excluding accidents, suicide, homicide) in the state of Sao Paulo, Brazil, and includes cases originally enrolled in the Study of Ancestry and Neurodegenerative Diseases. After consent by a legal representative of the decedent, informants were invited to participate in a structured interview while waiting for completion of brain and other organ removal. The study was approved by the Brazilian national ethics committee Comissao Nacional de Etica em Pesquisa. Because participants are deceased, the study is exempt from human subjects review in the USA.
These analyses are based on 1,493 individuals aged 65 years or older at death who had a consent signed by a legal representative for brain removal and examination of the decedent plus an informant who consented to do the interview. They died at a mean age of 79.5 (SD = 8.9; range: 65–110) mostly of cardiovascular disease (62.9%), infectious diseases (21.7%), and cancer (2.5%). They had completed a mean of 4.9 years of formal education (SD = 3.9; range: 0–25); 52.4% were women. Proxy reported race was black or mixed in 34.1% and white in 65.9%.
The relationships of informants to decedents were as follows: child (71.6%), grandchild (8.5%), sibling (5.4%), spouse (4.5%), other relative or in-law (1.7%), and other (8.1%). They reported knowing the decedent for a median of 46 years (interquartile range: 39–83). In the last year of the decedent’s life, 70.9% of informants reported daily contact with the decedent, 28.0% reported weekly contact, and 1.2% reported monthly contact.
Clinical interview
An informant of the decedent had a 60–90 minute structured interview with a study nurse. If more than one representative of the decedent was available, we primarily considered the responses of the person with the most frequent contact with the decedent. The interview assessed basic demographic information including age at death, sex, education, and race; social and emotional isolation; history of major depression; and clinical diagnoses of dementia and MCI.
Assessment of social and emotional isolation
Social isolation was assessed with standard social network questions about frequency of contact with children, other family members, and friends (Cornoni-Huntley et al., Reference Cornoni-Huntley, Brock, Ostfeld, Taylor and Wallace1986). Social network size was defined as the number of individuals from these three categories seen at least once per month (Barnes et al., Reference Barnes, Mendes de Leon, Wilson, Bienias and Evans2004; Bennett et al., Reference Bennett, Schneider, Tang, Arnold and Wilson2006).
Emotional isolation was assessed with the 3-item short form (Hughes et al., Reference Hughes, Waite, Hawkley and Cacioppo2004) of the Revised UCLA Loneliness Scale (Russell et al., Reference Russell, Peplau and Cutrona1980). Informants were asked to rate on a 3-point scale how often the decedent felt: lack of companionship, left out, and isolated from others. Item scores (0–2) were summed to yield the total score (0–6). We are not aware of previous use of informant report to assess loneliness, but informants are commonly used to assess other subjective states such as depression (Gilley et al., Reference Gilley1995) and personality traits (Costa and McCrae, Reference Costa and McCrae1992). The self-report version of this short form has been shown to have sound psychometric properties (Russell et al., Reference Russell, Peplau and Cutrona1980). The informant version of the scale used in the present study had adequate internal consistency (Cronbach’s coefficient alpha = 0.86 in full group, 0.84 in black participants, 0.87 in white participants) which is comparable to estimates for the self-report version of the scale (Tilvis et al., Reference Tilvis2004).
Assessment of major depression history
We adapted the depression section of the Structured Clinical Interview for DSM (Robins et al., Reference Robins, Helzer and Croughan1981) for informant report (Saldanha et al., Reference Saldanha2020). The informant was asked about past depressive symptoms. The diagnosis of major depression required persistent depressed mood or loss of interest accompanied by at least four additional depressive symptoms.
Assessment of dementia and MCI
We diagnosed dementia and MCI in the decedent with the informant portion of the Clinical Dementia Rating Scale (Morris et al., Reference Morris1992)]. The scale rates level of function in six domains: memory, orientation, judgment and problem solving, community affairs, home and hobbies, and personal care. An algorithm converts domain scores to an overall rating of no cognitive impairment (score = 0), MCI (score = 0.5), or dementia (score > 0.5). Dementia diagnoses based on the informant portion of the Clinical Dementia Rating Scale have been shown to have good agreement with diagnoses based on in-person examination (Waite et al., Reference Waite1999; Ferretti et al., Reference Ferretti2010).
Statistical analysis
To test the hypotheses that smaller social network size and higher level of loneliness are associated with higher likelihood of dementia and MCI, we constructed a series of logistic regression models. The outcome of the first set of models was no cognitive impairment versus dementia. All models had terms for age at death, sex, education, and race. The core model also had terms for social network size and loneliness score. We subsequently repeated the core model with a term for depression and then in separate analyses added terms for the interaction of race with social network size and the interaction of race with loneliness. We conducted additional analyses with terms for friend network size plus demographic variables in one model and family network size plus demographic variables in a second model. We then repeated each of these models with terms for loneliness and depression.
The outcome for the second set of logistic regression models was no cognitive impairment versus MCI. All models included terms for the four demographic variables. The core model had terms for social network size and loneliness. We added a term for depression to the core model and then repeated the model, first with a term for race × social network size and then with a term for race × loneliness. We constructed two additional models, one with a term for friend network size and the other with a term for family network size.
Results
Social and emotional isolation
Decedents were reported to have at least monthly contact with a median of eight individuals (interquartile range = 9.0). Neither age (Spearman r = −0.05, P = 0.063) nor education (Spearman r = −0.03, P = 0.208) was related to social network size. Black race (χ2 [1] = 10.8, P = 0.001) was associated with larger social network size, but sex (χ2 [1] = 0.2, P = 0.637) was not. Loneliness scores ranged from 0 to 6 (median = 0.0, interquartile range = 1.0) with higher scores indicating more loneliness. Loneliness was not related to age (Spearman r = 0.01, P = 0.587) but was to education (Spearman r = −0.05, P = 0.047). Level of loneliness in black and white people did not differ (χ2 [1] = 0.2, P = 0.663). Women were more lonely than men (χ2 [1] = 13.3, P < 0.001). Smaller social network size was associated with higher loneliness score (Spearman r = −0.18, P < 0.001).
Clinical classification
On the Clinical Dementia Rating Scale, 947 individuals met criteria for no cognitive impairment, 122 had MCI, and 424 had dementia. Consistent with prior research, the diagnostic subgroups differed in age (F[2, 1490] = 74.8, P < 0.001) and education (F[2, 1490] = 4.2, P = 0.015).
Relation of isolation to dementia
To estimate the associations of social and emotional isolation with dementia, we constructed a logistic regression model. The outcome was persons with no cognitive impairment versus persons with dementia. The model included terms for social network size and loneliness score plus terms for the potentially confounding effects of age at death, sex, education, and race. As shown for model A in Table 1, both smaller network size and higher level of loneliness were associated with higher likelihood of dementia. The size of each effect was modest. Thus, a typical decedent with a small social network (size = 3, 10th percentile) was about 11.3% more likely to have dementia than a decedent with a large social network (size = 23, 90th percentile) which was equivalent to being 2.4 years older. A lonely decedent (score = 3, 90th percentile) was about 9.4% more likely to have dementia than a decedent who was not lonely (score = 0, 10th percentile) which was equivalent to being 2.0 years older.
Relation of social and emotional isolation to likelihood of dementia*
OR, odds ratio; CI, confidence interval.
* From separate logistic regression models adjusted for age at death, sex, education, and race.
Because depression is related to social isolation (Domenech-Abella et al., Reference Domenech-Abella2017), emotional isolation (Cacioppo et al., Reference Cacioppo, Hughes, Waite, Hawkley and Thisted2006), and dementia (Saczynski et al., Reference Saczynski2010), we repeated the initial analysis with a term added for a history of major depression (present in 10.4%). As shown for model B in Table 1, a history of depression was associated with a more than twofold increase in the likelihood of dementia, but after adjusting for depression the associations of social and emotional isolation with dementia persisted.
With few exceptions (Sutin et al., Reference Sutin, Stephan, Luchetti and Terraccciano2020), knowledge about the relation of social and emotional isolation to dementia is based on studies of predominantly non-Latin white persons. Therefore, we repeated model B from Table 1, first with a term for the interaction of race (i.e. black+mixed versus white) with social network size and then again with a term for the interaction of race with loneliness. There were no interactions (each P > 0.70).
Prior research has suggested that engagement with friends is more strongly related to late life cognitive health than engagement with family (Zahodne et al., Reference Zahodne, Ajrouch, Sharifian and Antonucci2019; Sharifian et al., Reference Sharifian, Kraal, Zaheed, Sol and Zahodne2020). In separate analyses, larger networks of both friends (odds ratio [OR] = 0.947, 95% confidence interval [CI]: 0.923, 0.973) and family (OR = 0.976, 95% CI: 0.956, 0.996) were related to lower likelihood of dementia. When we added terms for loneliness and depression, the association of friend network size with dementia persisted (OR = 0.959, 95% CI: 0.928, 0.977), but the association of family network size with dementia was no longer significant (OR = 0.982, 95% CI: 0.962, 1.002).
Relation of isolation to MCI
To test the associations of social and emotional isolation with MCI, we constructed a logistic regression model contrasting the MCI subgroup with the no cognitive impairment subgroup. Model A in Table 2 shows that higher level of loneliness was related to higher likelihood of MCI, but social network size was not. When a term for depression was added to the model, neither social network size nor loneliness was related to MCI (model B in Table 2).
Relation of social and emotional isolation to likelihood of mild cognitive impairment*
OR, odds ratio; CI, confidence interval.
* From separate logistic regression models adjusted for age at death, sex, education, and race.
Discussion
We interviewed knowledgeable informants of nearly 1,500 older Brazilian decedents to test the hypotheses that social and emotional isolation are associated with higher likelihood of dementia. More than one-fourth of decedents met dementia criteria. Both smaller social network size and higher level of loneliness were associated with higher likelihood of dementia even after controlling for depression. The results support the idea that social and emotional isolation are associated with late life dementia.
Most prior research reporting an association of loneliness with dementia is based on groups of predominantly non-Latin white participants (Rafnsson et al., Reference Rafnsson, Orrell, d’Orsi, Hogervorst and Steptoe2020; Tilvis et al., Reference Tilvis2004; Wilson et al., Reference Wilson2007; Shankar et al., Reference Shankar, Hamer, McMunn and Steptoe2013; Holwerda et al., Reference Holwerda2014; Sundstrom et al., Reference Sundstrom, Adolfsson, Nordin and Adolfsson2020). In an exception to this pattern, analyses of data from the Health and Retirement Study did not suggest either ethnic or racial differences in the association of loneliness with dementia (Sutin et al., Reference Sutin, Stephan, Luchetti and Terraccciano2020). That the association of loneliness with dementia was observed in older Brazilians and did not vary between black and white persons provides further evidence that the association of loneliness with dementia generalizes across racial and ethnic lines.
Smaller social network size has been associated with dementia in some studies (Fratiglioni et al., Reference Fratiglioni, Wang, Ericsson, Maytan and Winblad2000; Crooks et al., Reference Crooks, Lubben, Petitti, Little and Chiu2008; Rafnsson et al., Reference Rafnsson, Orrell, d’Orsi, Hogervorst and Steptoe2020). However, other studies have not observed the association (Wilson et al, Reference Wilson2007; Holwerda et al., Reference Holwerda2014), and meta-analyses have reflected these inconsistent results (Kuiper et al., Reference Kuiper2015; Penninkilampi et al., Reference Penninkilampi, Casey, Singh and Brodaty2018). In the present study, smaller social network size was robustly associated with higher likelihood of dementia even after accounting for loneliness and depression. That this association was observed in individuals of Latin descent and did not differ between black and white persons suggests that the association of social network size with dementia generalizes across racial and ethnic lines.
Recent research has suggested that the size of friend networks is more strongly related to late life cognitive health than the size of family networks (Zahodne et al., Reference Zahodne, Ajrouch, Sharifian and Antonucci2019; Sharifian et al., Reference Sharifian, Kraal, Zaheed, Sol and Zahodne2020). In the present analyses, both friend and family network sizes were related to dementia in initial models, but only friend network size remained significant in the fully adjusted analyses.
Knowledge about social and emotional isolation in MCI is limited. In cross-sectional studies, higher level of loneliness has been associated with lower level of cognitive function (Boss et al., Reference Boss, Kang and Branson2015) and higher likelihood of MCI (Yu et al., Reference Yu, Lam and Lee2016). In the present analyses, we found that loneliness, but not social isolation, was associated with higher likelihood of MCI, consistent with prior research. However, this association was no longer significant after adjusting for depression, possibly because the smaller number of MCI cases (n = 121 compared to 424 with dementia) limited statistical power. Further research on social and emotional isolation in MCI is needed.
The bases of the correlation between social and emotional isolation on the one hand and dementia on the other hand are uncertain. One possibility is that social and emotional isolation are not risk factors for dementia but consequences of the neurodegenerative and cerebrovascular conditions that underlie dementia. In support of this reverse causality hypothesis, there is evidence that social network size tends to decrease in old age (Wrzus et al., Reference Wrzus, Hanel, Wagner and Neyer2013; Sander et al., Reference Sander, Schupp and Richter2017; Casey et al., Reference Casey, Liu, Kochan, Sachdev and Brodaty2020) and that declining network size is associated with cognitive decline (Casey et al., Reference Casey, Liu, Kochan, Sachdev and Brodaty2020; Rohr et al., Reference Rohr2020). However, loneliness appears to be relatively stable in adulthood and old age (Mund et al., Reference Mund, Freuding, Mobius, Horn and Neyer2020) and evidence that lower cognitive function predicts subsequent increase in loneliness has been inconsistent (Wilson et al., Reference Wilson2007; Donovan et al., Reference Donovan, Wu, Rentz, Sperling, Marshall and Glymour2017; Yin et al., Reference Yin, Lassale, Steptoe and Cadar2019). In addition, neither social network size (Bennett et al., Reference Bennett, Schneider, Tang, Arnold and Wilson2006) nor loneliness (Wilson et al., Reference Wilson2007) appears to be related to postmortem neuropathological markers of dementia. At present, therefore, support for the reverse causality hypothesis is mixed. Other (not mutually exclusive) possibilities are that social engagement somehow enhances cognitive reserve by modifying the deleterious impact of neuropathologies on cognition (Bennett et al., Reference Bennett, Schneider, Tang, Arnold and Wilson2006) or by a positive association with cognition that is independent of dementia-related neuropathologies (Wilson et al., Reference Wilson2007). Understanding the link between social engagement and late life cognitive function may suggest novel strategies for delaying dementia onset in old age.
This study has strengths and limitations. Results are based on a large racially diverse group of participants. The diagnoses of dementia and MCI were based on uniform implementation of previously validated clinical procedures. Social and emotional isolation were assessed with standard psychometrically sound measures. Caregiver burden which was not assessed may have influenced informant report. In addition, reliance on informant report may have increased measurement error and assessing isolation and dementia/MCI at the same point in time may also have biased our estimate of the association between them. An important limitation is that analyses are based on a selected group. In the future, we will be able to examine the associations of social and emotional isolation with dementia-related neuropathologies in PARDoS. Data collection is ongoing.
Conflict of interest
None.
Source of funding
The study was supported by National Institute on Aging grant R01AG54058. The sponsors had no role in the study design; in the collection, analysis, and interpretation of the data; in the writing of the report; and the decision to submit the article for publication.
Description of author(s)’ roles
RSW: study concept and design, analysis and interpretation of data drafting manuscript for intellectual content. AWC: analysis and interpretation of data, critical revision of manuscript for intellectual content. CS: data collection, critical revision of manuscript for intellectual content. SEL: analysis and interpretation of data, critical revision of manuscript for intellectual content. LLB: analysis and interpretation of data, critical revision of manuscript for intellectual content. JMF: study concept and design, data collection, analysis and interpretation of data, critical revision of manuscript for intellectual content. DAB: study concept and design, obtaining funding, critical revision of manuscript for intellectual content.
Acknowledgments
The authors thank the informants and study staff in Brazil and the USA.
