Association of Social Isolation With Disability Burden and 1-Year Mortality Among Older Adults With Critical Illness

Key Points

Question  Is social isolation associated with poor functional outcomes and 1-year mortality among older adults with critical illness?

Findings  In this nationally representative cohort study including 997 patients, more social isolation was significantly associated with a higher disability burden and increased likelihood of death within 1 year of hospital admission.

Meaning  The study findings suggest that social isolation is a risk factor for poor outcomes among older adults with critical illness.

Importance  Disability and mortality are common among older adults with critical illness. Older adults who are socially isolated may be more vulnerable to adverse outcomes for various reasons, including fewer supports to access services needed for optimal recovery; however, whether social isolation is associated with post–intensive care unit (ICU) disability and mortality is not known.

Objectives  To evaluate whether social isolation is associated with disability and with 1-year mortality after critical illness.

Design, Setting, and Participants  This observational cohort study included community-dwelling older adults who participated in the National Health and Aging Trends Study (NHATS) from May 2011 through November 2018. Hospitalization data were collected through 2017 and interview data through 2018. Data analysis was conducted from February 2020 through February 2021. The mortality sample included 997 ICU admissions of 1 day or longer, which represented 5 705 675 survey-weighted ICU hospitalizations. Of these, 648 ICU stays, representing 3 821 611 ICU hospitalizations, were eligible for the primary outcome of post-ICU disability.

Exposures  Social isolation from the NHATS survey response in the year most closely preceding ICU admission, which was assessed using a validated measure of social connectedness with partners, families, and friends as well as participation in valued life activities (range 0-6; higher scores indicate more isolation).

Main Outcomes and Measures  The primary outcome was the count of disability assessed during the first interview following hospital discharge. The secondary outcome was time to death within 1 year of hospital admission.

Results  A total of 997 participants were in the mortality cohort (511 women [51%]; 45 Hispanic [5%], 682 non-Hispanic White [69%], and 228 non-Hispanic Black individuals [23%]) and 648 in the disability cohort (331 women [51%]; 29 Hispanic [5%], 457 non-Hispanic White [71%], and 134 non-Hispanic Black individuals [21%]). The median (interquartile range [IQR]) age was 81 (75.5-86.0) years (range, 66-102 years), the median (IQR) preadmission disability count was 0 (0-1), and the median (IQR) social isolation score was 3 (2-4). After adjustment for demographic characteristics and illness severity, each 1-point increase in the social isolation score (from 0-6) was associated with a 7% greater disability count (adjusted rate ratio, 1.07; 95% CI, 1.01-1.15) and a 14% increase in 1-year mortality risk (adjusted hazard ratio, 1.14; 95% CI, 1.03-1.25).

Conclusions and Relevance  In this cohort study, social isolation before an ICU hospitalization was associated with greater disability burden and higher mortality in the year following critical illness. The study findings suggest a need to develop social isolation screening and intervention frameworks for older adults with critical illness.

For many community-dwelling older adults, surviving critical illness is associated with a severe and prolonged course of new disability.¹ Fewer than half of older intensive care unit (ICU) survivors fully recover their prehospitalization functional ability,² despite functional independence being considered one of the highest priority goals among older adults who are recovering from serious illness.³ Additionally, nearly 1 in 3 older ICU survivors die within 3 years of discharge, a rate concentrated in the year after discharge and 3 to 5 times higher than that of the general older adult population.⁴ Identifying older adults most vulnerable to poor outcomes during the ICU hospitalization may facilitate interventions to positively affect the course of recovery.

Social isolation, or an objective lack of social connections, is increasingly recognized as a public health concern for older adults.⁵ In the US, estimates suggest that more than 1 in 5 older adults lack close social ties, such as meaningful relationships with friends and family.^6,7 Among community-dwelling older adults, social isolation is associated with cognitive impairment, disability burden, progression of physical frailty, and elevated mortality rates.^8-10 For older adults recovering from critical illness, social isolation may have a particularly strong association with poor outcomes. For example, socially isolated older adults may have fewer supports to help access postdischarge services, such as ICU follow-up clinics, and they may also be more susceptible to the negative effects of post-ICU syndrome, which describes the disabling sequelae of emotional, cognitive, and physical impairments that follow critical illness.¹¹ However, to our knowledge, the association between social isolation and recovery from critical illness among older adults has not been studied.

To address this knowledge gap, we used data from a nationally representative study of Medicare beneficiaries 65 years and older. Our objectives were to evaluate the association between pre-ICU social isolation and disability burden after discharge from an ICU hospitalization as well as the association of social isolation with 1-year mortality. We hypothesized that increasing social isolation would be associated with a higher disability burden and greater 1-year mortality.

Study data were drawn from the 2011 cohort of the National Health and Aging Trends Study (NHATS), a nationally representative annual survey of late-life disability trends among Medicare beneficiaries.¹² The comprehensive design of this complex survey is described elsewhere.¹³ Briefly, the NHATS 2011 cohort (n = 8245) represents a random sample of adults 65 years and older that was drawn from Medicare enrollment files in 2010, oversampling for the oldest older adults (age >90 years) and non-Hispanic Black individuals. Each survey collected information on demographic characteristics, disability status, health conditions, social connections, and living situation. These survey data were subsequently linked to Medicare claims data. The NHATS is sponsored by the National Institute on Aging (NIA; grant U01AG032947) through a cooperative agreement with the Johns Hopkins Bloomberg School of Public Health. The Johns Hopkins University institutional review board approved the NHATS protocol, and all participants provided written informed consent. Use of NHATS data for this analysis was approved by the Yale University institutional review board.

We first identified inpatient hospitalizations that included an ICU stay between May 2011 and December 2017 using critical care revenue center codes from the linked Medicare inpatient claims file, excluding stays in an intermediate or psychiatric ICU. We restricted the sample to ICU admissions of 1 day or longer.

The primary predictor was a validated 6-item composite measure of social isolation drawn from NHATS survey items during the year most closely preceding ICU admission.⁶ These items correspond to 4 core domains of social connectedness: marriage or live-in partnership, contacts with family and friends, membership in a religious organization, and participation in other community groups.¹⁴ Participants were asked during each annual NHATS survey about the presence of a live-in partner or spouse, whether they talked with family or friends about important matters, whether they visited with friends or family, and whether they recently participated in religious services or social activities (eMethods in the Supplement). Participants scored 1 point for each negative response, with higher scores (range, 0-6) indicating greater social isolation.

The primary outcome of interest was the count of disabilities in 7 activities of daily living (ADLs) and mobility activities at the NHATS interview following discharge from ICU hospitalization. During each annual interview, participants were asked about a need for help from another person in completing each of 7 tasks: getting cleaned up, using the toilet, dressing, eating, getting around inside, going outside, and getting out of bed. Participants were coded as having a disability if they reported needing help from another person or had not completed the task at all within the last month. Total disability counts (range, 0-7) were summed for the interview preceding ICU admission and the round following ICU discharge. For participants who died during the post-ICU follow-up period, disability data were obtained from standardized proxy interviews about participants during the last month of life.

The secondary outcome was time from hospital admission to death, censored at 365 days. Death dates were obtained from the Medicare enrollment files.

We selected descriptive characteristics and model covariates from the NHATS data. The covariates included those that prior work^2,15,16 or clinical experience pointed to as potential confounders of the association between social isolation and outcomes. Participant age, sex, self-reported race (for descriptive purposes), chronic condition count, preadmission disability count, and rural residence were collected from the NHATS interview immediately preceding hospitalization. Preadmission frailty was assessed using modified Fried criteria developed from NHATS interview responses and physical performance measures,¹⁷ with participants classified as having frailty (3 of 5 or more frailty criteria) or not having frailty (2 of 5 or fewer). We identified older adults with probable dementia through an algorithm developed by the NHATS study team.^18,19 From Medicare claims, we also extracted information on the use of mechanical ventilation from procedure codes in inpatient claims and total hospital length of stay from the hospitalization record. From Medicare Master Beneficiary Summary Files, we captured whether the patient was eligible for Medicaid as a proxy for socioeconomic disadvantage.

Selection of the analytic sample is depicted in Figure 1. We identified all patient ICU stays from 2011 through 2017 contributed by the 2011 NHATS cohort. Participants could contribute multiple ICU stays over this period, but for participants who had multiple admissions within the same 12-month period, only the first was retained. We excluded participants with missing data for all 7 ADL and mobility tasks before admission. We then further limited the sample to older adults who were community dwelling at the time of ICU hospitalization. The remaining person-admissions constituted the sample for the secondary outcome of time to death from hospital admission. For the primary outcome of post-ICU disability, we further limited the sample to older adults who survived to hospital discharge and had a disability assessment during an NHATS interview within 365 days, excluding those discharged to hospice. Participants who had a disability in all 7 assessed activities before hospitalization were also excluded, as this population cannot become functionally worse and prior work has shown that patients with maximum disability retain severe disability or die after critical illness.¹ We additionally excluded participants with missing data for all 7 functional activities in the posthospitalization interview. The most common reasons for these missing data were patient and proxy refusals (n = 47) or a transition into a nursing home (n = 17).

Demographic and clinical characteristics of the eligible participants were recorded. Then, to minimize risk of bias associated with missing data (which did not exceed 3% for any variable), and based on an assumption of missing at random, multiple imputation was used to produce 5 imputed data sets. Imputations leveraged all predictor variables in the data set for all eligible NHATS participants across the first 8 years of data collection. The descriptive statistics and imputation were performed using SAS, version 9.4 (SAS Institute).

We initially plotted the mean count of disability with 95% confidence limits by increasing degree of social isolation. We subsequently evaluated the association between social isolation before hospitalization and count of post-ICU disability using a Poisson regression with generalized estimating equations and an exchangeable correlation structure, the latter to account for those contributing multiple ICU admissions. The multivariable model of post-ICU disability count was adjusted for age, sex, probable dementia, frailty, hospital length of stay, use of mechanical ventilation, chronic condition count, prehospitalization disability count, and rural residence. Because NHATS interviews proxies to ascertain the function of decedents during the last month of life, there was no truncation of follow-up data because of death. The adjusted rate ratios (RRs) reported represent the proportional increase in post-ICU disability that was associated with each incremental rise in the social isolation score (from 0-6). We repeated the multivariable modeling in a sensitivity analysis to evaluate potential biases associated with missing data for older ICU survivors who were admitted to nursing homes (n = 17) during the follow-up period. In this sensitivity analysis, outcome data for this subset of survivors were imputed under assumptions of missingness at random.

The unadjusted association between social isolation score and 1-year mortality from date of hospital admission was presented as a Kaplan-Meier test using the log rank statistic. This outcome was subsequently evaluated with a proportional hazards model that adjusted for the same covariates as the primary outcome in which follow-up was censored at 365 days. Compliance with the assumption of proportional hazards was verified with martingale residuals. Results are presented as adjusted hazard ratios (aHRs) that represent the proportional increase in hazard of mortality associated with incremental rise in social isolation.

To adjust for the complex survey design, all multivariable models included analytic weights, strata, and clustering parameters that were specific to the 2011 NHATS cohort. The multivariable models were separately fit to each of the imputations, with successive combining of the estimates using Rubin rules as implemented in SAS-callable SUDAAN (release 11.0.3) for the primary outcome of disability count and in SAS for the 1-year mortality outcome. In all cases, statistical significance was defined as a 2-sided P < .05. Data analysis was conducted from February 2020 through February 2021.

The primary analysis of disability included 648 ICU hospitalizations contributed by 543 older adults, which represented 3 812 611 survey-weighted ICU hospitalizations from 2011 through 2017 (Table). The median age (interquartile range [IQR]) of the sample participants was 81 (75.5-86.0) years (range, 66-102 years), 331 of 648 (51%) were women, and 134 of 648 (21%) identified as non-Hispanic Black. The median (IQR) of preadmission social isolation scores was 3 (2-4), and most participants in the sample had no disabilities before hospitalization (median count, 0; IQR, 0-1). In the primary analytic sample of ICU survivors, 60 (9%) underwent mechanical ventilation; of those discharged alive to nonhospice settings, 254 (39%) were discharged to facilities and 394 (61%) to the community. Among ICU survivors, 102 (16%) were readmitted to the ICU within 1 year; the proportion of high social isolation scores was similar among those who were readmitted and those who were not readmitted.

Greater social isolation was associated with a higher mean count of post-ICU disabilities (Figure 2), ranging from an average of 1 disability for the most socially integrated to more than 3 for the most socially isolated. In adjusted models, higher social isolation scores were associated with significantly higher counts of disability after discharge. The count of post-ICU disability was 7% higher with each incremental rise in the social isolation score from 0 to 6 (adjusted RR, 1.07, 95% CI, 1.01-1.15). Compared with the most socially integrated older adults (score, 0 of 6), moderate social isolation among older ICU survivors (score of 3 of 6) was associated with a 23% higher burden of disability, and the most severe social isolation (score of 6 of 6) was associated with a 50% higher disability burden. The sensitivity analysis that included participants who were admitted to nursing homes during the follow-up period yielded similar results (adjusted RR, 1.08; 95% CI, 1.01-1.15).

For the time to death outcome, 997 ICU hospitalizations contributed by 834 older adults (representing 5 705 675 survey-weighted admissions) were included in the analytic sample. Increasing social isolation was associated with a significantly higher risk of death within 1 year of hospital admission (Figure 3). In adjusted models, each incremental rise in preadmission social isolation scores (from 0 to 6) was associated with a 14% increase in the hazard of death in the year following hospital admission (aHR, 1.14; 95% CI, 1.03-1.25). Compared with the most socially integrated patients (score of 0 of 6), moderate social isolation among older ICU survivors (score of 3 of 6) was associated with a 48% higher hazard of death, while the most severe isolation (score of 6 of 6) was associated with a 119% higher hazard of death.

In this nationally representative study of older ICU survivors, we found that greater social isolation was associated with a significant increase in disability burden and risk of mortality in the year after a critical illness. Importantly, the risks associated with social isolation persisted after accounting for geriatric vulnerabilities, illness severity, and socioeconomic disadvantage. These findings suggest that identifying and ameliorating social isolation may play a role in improving outcomes after critical illness.

Our findings complement an extensive body of work demonstrating that socially isolated community-dwelling older adults are more likely to report lower health-related quality of life and are at greater risk of disability, dementia, and mortality.^5,10,20 While the data used in our work do not permit us to examine the mechanisms that may be driving the association between social isolation and poor outcomes after critical illness, it is plausible that social isolation complicates recovery in 3 distinct ways. First, the biological impairments associated with social isolation (eg, negative changes in neuroendocrine function and an increase in systemic inflammation biomarkers)⁵ may increase vulnerability to disability and mortality. Second, socially isolated older adults may be more likely to experience challenges accessing important health care services (eg, primary care, rehabilitation) after a critical illness, so disabling impairments, such as muscle weakness, may be more likely to be left unaddressed. Our prior work has found that living alone, one component of social isolation, is associated with lower rehabilitation use among older ICU survivors.²¹ Third, those with few social connections may have trouble navigating the complicated process of acquiring important supports, such as durable medical equipment. Unmet needs for assistance with ADLs (eg, need for a shower chair) have been identified as a concern among ICU survivors²²; this concern is likely amplified in the absence of close friends or family members who can assist with advocacy for or installation of needed equipment.

Addressing social isolation has increased in urgency, considering recent research that suggests that between 21% and 24% of community-dwelling older adults in the US are socially isolated.^6,7 Concerns about social isolation among older adults have intensified during the recent COVID-19 pandemic.²³ Infection control and risk-mitigation strategies, such as shelter-in-place orders, were well-meaning but inevitably increased social isolation.²⁴ Yet, there is a paucity of intervention frameworks to guide identification and treatment of socially isolated older adults.

A recent report by the National Academy of Medicine concluded that hospital stays constitute an ideal opportunity to identify social isolation and initiate referrals that merge individual and community-level interventions.⁵ Indeed, hospitalization may represent the only opportunity to identify the most socially isolated older adults. Nevertheless, screening for social isolation is not a common practice during hospitalization; although caregiver supports are often documented, social interconnectedness is rarely assessed. This study suggests that social isolation should be routinely assessed during hospital admissions; prior work has suggested that it is feasible to include social isolation measures in electronic medical records.²⁵

If social isolation is identified during an ICU stay, how best to address this vulnerability is an open question. However, there are promising intervention frameworks that could be adapted. First, there is growing interest in developing peer support models for ICU survivors. These models, designed to offer emotional and social supports for older adults,²⁶ may have an ancillary benefit of increasing social connectedness. Such groups often occur in outpatient settings, and it is not yet clear whether participation is feasible for older ICU survivors who have deficits in mobility and ADL function. However, emerging virtual formats may allow greater participation among those who have difficulty attending clinic-based groups.²⁷ Second, at least one large medical center has begun offering weekly phone calls to isolated older adults by community volunteers, and another has developed a “Geriatric Buddy” program to facilitate social engagement.²⁵ Interventions that encourage meaningful human interactions are unlikely to cause harm and should be considered for socially isolated older adults after an ICU stay. Third, transportation interventions, such as assisting older ICU survivors with completing applications for door-to-door public transit options, may help support attendance at follow-up medical appointments and provide an avenue through which to facilitate social connections. Future research should evaluate whether these interventions improve outcomes among older ICU survivors.

This study is not without limitations. First, social isolation was assessed during a 1-year period before ICU admission and could have changed before the index admission. Second, we excluded a few participants with maximal disability (ie, those who had disability in all 7 functional activities) during the pre-ICU NHATS survey from the primary analysis of post-ICU disability because it was not possible for them to become functionally worse, so our disability results may not be generalizable to older adults who have maximum disability at baseline. However, this group was eligible for and included in the mortality analysis. Third, some older adults died during the postdischarge follow-up, and their functional limitations during the last month of life were reported by proxies, which may introduce recall bias. However, the NHATS methods draw from prior studies that have shown concordance between proxy reports of basic ADL function and patient reports.^28,29

The results of this cohort study suggest that social isolation was common among older survivors of critical illness and was associated with a higher burden of disability and an increased risk of death in the year following ICU admission. These findings suggest an urgent need to develop in-hospital screening tools and intervention frameworks to support socially isolated older adults who are recovering from critical illness.

Accepted for Publication: July 18, 2021.

Published Online: September 7, 2021. doi:10.1001/jamainternmed.2021.5022

Corresponding Author: Lauren E. Ferrante, MD, MHS, Section of Pulmonary, Critical Care, and Sleep Medicine, Yale School of Medicine, 300 Cedar St, TAC-431 South, PO Box 208057, New Haven, CT 06520 (lauren.ferrante@yale.edu).

Author Contributions: Drs Falvey and Ferrante had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Falvey, Ferrante.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Falvey, Murphy, Ferrante.

Critical revision of the manuscript for important intellectual content: Cohen, Leo-Summers, Murphy, Ferrante.

Statistical analysis: O'Leary, Murphy.

Obtained funding: Falvey, Ferrante.

Administrative, technical, or material support: Leo-Summers.

Supervision: Ferrante.

Conflict of Interest Disclosures: Dr Falvey reported grants from the National Institute on Aging (NIA) and the Foundation for Physical Therapy Research during the conduct of the study. Dr Cohen reported grants from NIA during the conduct of the study. Dr Murphy reported grants from the National Institutes of Health during the conduct of the study. Dr Ferrante reported grants from NIA and the Yale Claude D. Pepper Older Americans Independence Center during the conduct of the study. No other disclosures were reported.

Funding/Support: The National Health and Aging Trends Study (NHATS) was supported by grant U01AG032947 from the NIA through a cooperative agreement with the Johns Hopkins Bloomberg School of Public Health.

Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Contributions: We thank Thomas M. Gill, MD, Yale School of Medicine, for his assistance with acquisition of NHATS data. He was not compensated for his work on this project.