Innovation in Aging, 2024, 8, igae019

https://doi.org/10.1093/geroni/igae019
Advance access publication 23 February 2024
Original Report

Bidirectional Relationships and Mediating Effects Between

Social Isolation, Loneliness, and Frailty in Chinese Older
Adults
Chaoping Pan, PhD*,
College of Medical Humanities and Management, Wenzhou Medical University, Wenzhou City, Zhejiang Province, China.
*Address correspondence to: Chaoping Pan, PhD. E-mail: [email protected]
Decision Editor: J. Tina Savla, PhD, FGSA

Abstract
Background and Objectives: Social isolation (SI) and loneliness are key factors that contribute to frailty among older adults. Current estimates
regarding how frailty affects SI and loneliness and how SI and loneliness affect frailty may be flawed due to reverse causality. This study aimed
to investigate the bidirectional relationships and mediating effects among SI, loneliness, and frailty among older adults in China.
Research Design and Methods: The study analyzed data from 6 waves of the Chinese Longitudinal Healthy Longevity Survey conducted
between 2002 and 2018. The sample included individuals aged 65 and older. The General Cross-Lagged Panel Model was used to account for
confounding factors and reveal mediating effects.
Results: Our findings specifically indicate a direct effect of SI on frailty, although suggesting that loneliness may indirectly affect frailty through
its influence on SI. Additionally, frailty can lead to increased SI and loneliness.
Discussion and Implications: SI and loneliness are strongly intertwined with frailty among older adults in China. To prevent the development of
frailty, public health initiatives should prioritize reducing SI among older adults. Furthermore, efforts to decrease frailty levels can yield positive
outcomes by mitigating both SI and loneliness among this population.

Translational Significance: This study investigates the bidirectional relationships and mediating effects among social isolation (SI),
loneliness, and frailty among older adults in China. The results suggest that reducing SI may be an effective strategy for preventing
and managing frailty among older adults in this population. Additionally, frailty can lead to increased SI and loneliness, highlighting the
importance of addressing frailty in interventions aimed at reducing SI and loneliness among older adults.

Keywords: Frailty, General Cross-Lagged Panel Model, Loneliness, Social isolation

Frailty is defined by a deterioration in function across various develop interventions that can effectively address this issue
physiological systems, coupled with an increased vulnerabil- and improve the health and well-being of this population.
ity to stressors (1–4). Two main clinical models are commonly According to the social convoy theory, social relation-
used to assess frailty. Fried’s phenotype model defines frailty ships can be divided into objective and subjective dimensions
by considering the presence of at least 3 of the following 5 cri- (8). Social isolation (SI) and loneliness are 2 dimensions
teria: slow gait speed, unintentional weight loss, self-reported of impoverished social relationships, representing objec-
exhaustion, low physical activity, and weak grip strength (5). tive and subjective aspects, respectively (9). SI pertains to
On the other hand, Rockwood’s cumulative deficit model the objective absence of social interaction with others and
measures frailty using a Frailty Index (FI) that reflects the encompasses elements such as living alone, withdrawal from
accumulation of deficits. This index is obtained by calcu- social connections, and limited engagement in social activ-
lating the proportion of deficits present divided by the total ities (10). Conversely, loneliness is a subjective experience
number of deficits considered (6). It is associated with var- that emerges when there is a disparity between one’s desired
ious adverse health outcomes, such as hospitalization, falls, and actual level of social connection and relationship quality
and increased mortality risk (3–5). With the aging population (2). Examining both SI and loneliness is important because
increasing worldwide, frailty has emerged as a pressing public loneliness can be entirely unrelated to SI (11). As the global
health challenge (7). Therefore, it is crucial to determine the population continues to age rapidly, SI has become a prom-
key factors that contribute to frailty among older adults to inent issue among older adults. Research has revealed that
between 10% and 43% of older adults experience SI in the

Received: July 10 2023; Editorial Decision Date: February 7 2024.

© The Author(s) 2024. Published by Oxford University Press on behalf of The Gerontological Society of America.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/),
which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited.
2 Innovation in Aging, 2024, Vol. 8, No. 3

later stages of life (12), with China reporting a notably high loneliness, although loneliness mediates the impact of frailty
prevalence of 42.4% among older adults (13). Similarly, in on SI.
countries like the United States and Europe, approximately
5%–40% of older adults experience feelings of loneliness
(14), although China reports a loneliness rate of 31.3% Method
among older adults (13). Both SI and loneliness have been Data and Participants
found to negatively affect health and well-being, including This study utilized data from CLHLS, a nationally represen-
outcomes such as frailty, disability, quality of life, and cogni- tative longitudinal survey covering 23 of the 31 provinces in
tive impairment (11,15–18). China and approximately 85% of the entire population. The
Indeed, several studies have indicated that higher levels participants included in this data set were individuals who
of SI and loneliness increase the risk of frailty among older were 65 years of age or older. The baseline survey was con-
adults (2,19,20). Although the exact mechanisms behind ducted in 1998, and subsequent data was collected in 7 waves
this association have not been extensively studied, possi- in 2002, 2005, 2008, 2011, 2014, and 2018. For this study,
ble explanations could include health behaviors, stress, and we utilized data from 6 waves of CLHLS conducted from
the maintenance and repair of physiological systems (21). 2000 to 2018. Further details about the CLHLS can be found
Conversely, a few studies have found that health conditions elsewhere (28,29). Exclusion criteria for this study included
may lead to increased SI and loneliness among older adults individuals under the age of 65 at any time point, as they
(22–24). The contribution of frailty to increased SI can be accounted for less than 1% of the total sample. Additionally,
attributed to the fact that frailty often restricts the ability of as this is a longitudinal study, individuals who were tracked
older adults to engage in social activities, visit friends and less than twice were also excluded. Therefore, the number of
family, and participate in physical activity (25). Moreover, the respondents for each wave was: 2002, n = 8136; 2005, n = 11
prediction of loneliness by frailty may be due to the fact that 427; 2008, n = 11 571; 2011, n = 9194; 2014, n = 6551;
frailty could reduce the capacity of older adults to fulfill their 2018, n = 3469.
socioemotional needs (26). As a result, it is possible that the Compared with the remaining participants, those who
relationships between SI, loneliness, and frailty are actually dropped out of the study exhibited certain characteristics.
bidirectional in nature. They were generally older, more likely to be female, had lower
To our knowledge, there are no studies that have utilized education levels, higher levels of loneliness, slightly lower SI
appropriate statistical methods to explore the bidirectional scores, and higher frailty levels. More detailed information
relationship between SI, loneliness and frailty, and examine is presented in Supplementary Table 1 in the Supplemental
their potential mediating roles. Current estimates regarding materials. In this study, it is worth noting that there were
how frailty affects SI and loneliness (frailty → SI, frailty → intermittent missing values of 22.7%, 1.6%, and 11.8% for
loneliness) and how SI and loneliness affect frailty (SI → frailty, SI, and loneliness, respectively. The multiple imputa-
frailty, loneliness → frailty) may be flawed due to reverse cau- tion method was implemented to address potential bias due
sality (2,19,20). Thus, methods accounting for reverse causal- to intermittent missing data by filling in missing data. Further
ity are necessary to establish the causal ordering among these details on the sample characteristics and correlations of the
3 processes. It is also unclear which effect is more influential, analyzed variables can be found in Supplementary Tables 1–4.
which can affect the targeting of preventative measures. If
SI, loneliness, and frailty all play important roles in bidirec- Measures
tional relationships, interventions targeting all 3 factors are
needed to improve the health and well-being among older Frailty Index
adults, rather than focusing on only 1 factor (20,22). Finally, FI is a well-established tool for assessing frailty and has been
failure to consider the bidirectional relationship between SI, validated in previous research (30–32). Typically, a validated
loneliness, and frailty can result in flawed estimates of the FI measures at least 30 diseases and uses an equal-weighted
mediating roles of these variables, which can, in turn, lead to method to calculate scores (19,30,33,34). In our study, we
suboptimal interventions. As a result, gaining a better under- utilized a comprehensive set of 45 health conditions to mea-
standing of the mediating role of these variables can optimize sure FI, which exceeded the minimum threshold and under-
interventions targeted at intervening variables (27). scored the robustness of our analysis. The measurement of
The General Cross-Lagged Panel Model (GCLM) is a pow- FI involved evaluating 8 instrumental activities of daily liv-
erful statistical tool that enables the examination of bidirec- ing, 6 activities of daily living, 6 functional limitations, 11
tional relationships and mediating roles in longitudinal data. self-reported chronic diseases, 7 cognitive functions, 3 men-
In this study, we utilized this method to analyze data from the tal health conditions, 2 functional impairments, 1 self-rated
Chinese Longitudinal Healthy Longevity Survey (CLHLS), health measure, and one investigator-rated health assessment.
a nationally representative survey of older adults in China Each health condition was assigned a score between 0 and 1.
conducted over a 16-year period. By utilizing this method, The total FI score was obtained by summing the individual
our aim was to investigate the directionality of the longitu- scores of all the health conditions and dividing it by 45. The
dinal associations between SI, loneliness, and frailty, as well final FI score ranged from 0 to 1, with higher scores suggest-
as establish their mediating roles in these relationships. The ing greater prevalence of frailty (35).
hypotheses are as follows:
H1: There are bidirectional associations between SI and Social isolation
frailty, as well as between loneliness and frailty. The present study utilized the 5 dimensions recommended by
H2: It is hypothesized that SI mediates the impact of lone- previous literature (11,13,16,36,37) to measure SI in older
liness on frailty, and loneliness mediates the impact of SI adults. These dimensions included: (1) living alone, (2) hav-
on frailty. Additionally, SI mediates the impact of frailty on ing a spouse, (3) frequent contact with children, (4) frequent
Innovation in Aging, 2024, Vol. 8, No. 3 3

contact with siblings, and (5) participating in social activities previous study has demonstrated that incorporating cor-
(eg, playing cards or mahjong, joining organized social events, relation terms is a more effective approach for controlling
or working). Individuals who lived alone were without a confounding factors than including specific confounding
spouse, had infrequent contact with children/siblings, or had variables as covariates (41). The efficacy of this approach
limited social participation were coded as 1. Conversely, a has been increasingly recognized and employed in other
value of 0 was assigned to individuals who did not live alone, studies as well (43,44).
had a spouse, received frequent visits from children/siblings, The cross-lagged coefficients β1, β3, γ1, γ3, µ1, and µ2 are
or engaged in social activities. The total score ranged from 0 of particular interest as they reveal how differences in FI, SI,
to 5, as reported in prior research (11,16,37). and Ln at a given time point predict differences in FI, SI, and
Ln at the next time point. The autoregressive paths β2 , γ2,
Loneliness and µ3 reflect the degree to which individual differences in
Based on previous studies (13,16,36,38,39), this study used expected scores are predicted by differences from past time
a single item to measure loneliness, which asks “Do you feel points. Furthermore, the model allows for the computation of
lonely?” Responses include “never” (0 points), “hardly ever” mediating effects between SI, loneliness, and frailty by calcu-
(1 point), “sometimes” (2 points), “often” (3 points), and lating the cross-lagged coefficients. Before running the analy-
“always” (4 points), with a total score range of 0–4 points. sis, the variables were standardized, resulting in the regression
The single-dimension loneliness scale is widely used in the coefficients being expressed as standard deviation (SD) from
older population, and previous research has shown that it has the mean. This standardization process helps to compare dif-
a high correlation with multidimensional scales (13,40). ferent variables used in the analysis. The study employs 10
000 bootstrapping methods to compute confidence intervals.
Analysis Multiple model fit indices, including the Confirmatory Fit
Index (CFI), the Tucker Lewis Index (TLI), the root mean
In this study, we used GCLM to analyze the data, which has
square error of approximation (RMSEA), and the standard-
been recommended by previous studies (22,41). It has been
ized root mean squared residual (SRMR), were used to con-
increasingly employed in social and health research, especially
firm the goodness of model fit of our analysis (41). A CFI
for investigating the associations between SI and physical func-
value greater than 0.95 was deemed indicative of a good
tion (22). One reason why GCLM is the suitable method for
model fit. Similarly, RMSEA and SRMR values equal to or
this analysis is its ability to model lagged relationships, which
below 0.06 were considered to indicate a good fit, although
enables the exploration of bidirectional and mediating effects
values up to 0.08 were deemed acceptable (27).
(22,41). This is particularly important in understanding the
complex dynamics of SI, loneliness, and frailty. Additionally,
GCLM offers advantages in reducing confounding effects and Results
strengthening causal inferences by effectively controlling for
both stable and time-varying factors (22,41,42). At baseline, the average age of the participants was 81.83
To fit our model, we used MPlus 8 software and followed years. Females comprised 54.8% of the sample. Around
the structural equation modeling framework proposed by 42.5% of the participants had received at least 1 year of
Zyphur et al. (41). In formal terms, the model specification education, although the majority (56.1%) resided in rural
used in this study is expressed as: areas. The participants reported a mean SI score of 2.87 and
a mean loneliness score of 0.98. Frailty was assessed to have
FIit = β1 SIit−1 + β2 FIit−1 + β3 Lnit−1 + θt + µi + it a mean score of 0.14. For more detailed information, refer to
Supplementary Table 1.
SIit = γ1 FIit−1 + γ2 SIit−1 + γ3 Lnit−1 + σt + αi + eit Table 1 shows the goodness-of-fit statistics of GCLM on
the CLHLS data set, covering waves from 2002 to 2018. The
Lnit = µ1 FIit−1 + µ2 SIit−1 + µ3 Lnit−1 + ρt + ωi + τit
indices evaluated include CFI, TLI, RMSEA, and SRMR, with
In the model, the subscripts i and t represent individu- their corresponding values of 0.993, 0.988, 0.020, and 0.027,
als and time, respectively. SI represents social isolation, FI respectively. These indices indicate that the model fits the data
represents Frailty Index, and Ln represents loneliness. The well and provides a reliable representation of the relation-
regression coefficients to be estimated are β1, β2 , β3, γ1, γ2, γ3 ships among the analyzed variables.
, µ1, µ2 and µ3. θ , σ and ρ represent time effects, although µ,
α , and ω capture time-invariant effects. , e , and τ represent
individual-specific error terms. It is important to acknowl-
edge that the model does not incorporate specific time- Table 1. The Goodness-of-Fit Statistics of GCLM (CLHLS, Waves
varying or time-invariant variables as distinct entities. 2002–2018)
Instead, the model controls for confounding factors that can
vary over time and factors that remain constant across time Index Values
by treating them as collective entities. This is accomplished
by including correlation terms among , e , and τ , as well as CFI 0.993
µ, α , and ω to address potential confounding. There were 2 TLI 0.988
main reasons for choosing this method to control potential RMSEA 0.020
confounding. First, due to limitations in the available data SRMR 0.027
set, it was not possible to include all confounding variables
in the analysis, which could introduce bias into the results. Notes: CFI = Confirmatory Fit Index; CLHLS = Chinese Longitudinal
Healthy Longevity Survey; GCLM = General Cross-Lagged Panel
Furthermore, including a large number of confounding vari- Model; RMDEA = root mean square error of approximation; SRMR =
ables can lead to convergence issues with the model. Second, standardized root mean square residual; TLI = Tucker–Lewis Index.
4 Innovation in Aging, 2024, Vol. 8, No. 3

Table 2. The Key Model Parameters of GCLM (CLHLS, Waves Table 3. Key Mediating Effects From GCLM (CLHLS, Waves 2002–2018)
2002–2018)
Path Standardized coefficient (95% CI)
Path Standardized coefficient (95% CI)
Lnt−2 → SIt−1 → FIt 0.006(0.003, 0.010)
SIt−1 → SIt 0.378(0.354, 0.405) SIt−2 → Lnt−1 → FIt 0.003(–0.002, 0.008)
SIt−1 → FIt 0.121(0.086, 0.153) FIt−2 → SIt−1 → Lnt 0.025(0.017, 0.033)
SIt−1 → Lnt 0.235(0.212, 0.254) FIt−2 → Lnt−1 → SIt 0.005(0.003, 0.009)
Lnt−1 → Lnt 0.065(0.049, 0.080) SIt−2 → FIt−1 → Lnt 0.012(0.007,0.018)
Lnt−1 → SIt 0.051(0.038, 0.066) Lnt−2 → FIt−1 → SIt 0.001(–0.001,0.004)
Lnt−1 → FIt 0.011(−0.011, 0.032)
Notes: CLHLS = Chinese Longitudinal Healthy Longevity Survey; GCLM
FIt−1 → FIt 0.349(0.292, 0.402) = General Cross-Lagged Panel Model.
FIt−1 → SIt 0.107(0.080, 0.129)
FIt−1 → Lnt 0.100(0.082, 0.117) of loneliness in the future through SI of the following time
Notes: CLHLS = Chinese Longitudinal Healthy Longevity Survey; GCLM point.
= General Cross-Lagged Panel Model. Overall, these findings provide evidence for the complex
interplay among frailty, SI, and loneliness over time. For more
detailed information on these mediating effects, please refer
to Table 3.
Table 2 displays the standardized GCLM regression
coefficients and their 95% confidence intervals (CIs) for
key parameters concerning frailty, SI, and loneliness over
Discussions
time. The analysis indicates that greater levels of SI at a This study aimed to investigate the bidirectional associ-
given time point were linked with higher frailty in the ations and their mediating effects between SI, loneliness,
next time point ( β1). Specifically, each increment of 1 and frailty among older adults in China using GCLM.
SD in SI increased future frailty by 0.151 SDs (95% CI: Our findings revealed a bidirectional relationship between
0.121, 0.180). Conversely, no significant association was SI and frailty, supporting our first hypothesis. However,
found between loneliness at a given time point and future contrary to our initial hypothesis, we did not observe a
frailty ( β3). direct effect from loneliness to frailty. Furthermore, our
Moreover, higher levels of frailty at a specific time point results indicated that SI mediates the impact of loneliness
were associated with increased SI and loneliness in the next on frailty, while also mediating the impact of frailty on
time point (γ1 and µ1). For every increase of 1 SD in frailty, loneliness. Conversely, loneliness only mediates the impact
SI and loneliness increased by 0.107 SDs (95% CI: 0.080, of frailty on SI. These findings partially confirm our second
0.129) and 0.100 SDs (95% CI: 0.082, 0.117), respectively. hypothesis. These findings have important implications
The analysis also revealed that higher levels of SI at a spe- for understanding the complex and dynamic relationship
cific time point were associated with increased loneliness between these 3 factors and for improving the health and
in the next time point (µ2 = 0.235, 95% CI: 0.212, 0.254). well-being of older adults in China.
Conversely, higher levels of loneliness at a given time point According to the social convoy model, SI and loneliness
were associated with increased SI in the next time point represent objective and subjective aspects of social rela-
(γ3 = 0.051, 95% CI: 0.038, 0.066). For further details, please tionships that may have different effects on health (8). One
see Table 2. hypothesis explaining these different effects is that loneliness
Table 3 shows the key mediating effects and their 95% and SI may have separate pathways to influencing health
CIs from GCLM analysis using data from CLHLS waves in (13,39). For instance, loneliness may be more likely to be
2002–2018. The results indicate that the path Lnt−2 → SIt−1 → associated with other risk factors, such as sleep quality (45),
FIt had a standardized coefficient of 0.006 with a 95% CI of making it less likely to emerge as an independent risk factor.
(0.003, 0.010), suggesting that SI may play a mediating role Conversely, SI may have a stronger likelihood of emerging as
in the relationship between frailty and loneliness over time. an independent risk factor (13). Consistent with this theory,
Specifically, higher levels of loneliness at 1 point in time were our results indicated that higher levels of SI, but not loneli-
associated with higher levels of SI at the following time point, ness, were independently associated with an increase in frailty
which in turn was associated with higher levels of frailty among older adults. This finding is inconsistent with previous
in the future. However, the path SIt−2 → Lnt−1 → FIt had a literature (19,20). Davies et al. indicated that both loneliness
standardized coefficient of 0.003 with a 95% CI of (−0.002, and SI could affect frailty among older adults (20), although
0.008), indicating that SI did not significantly affect frailty Gale and colleagues only found an association between the
through loneliness. frailty phenotype and loneliness (19). The inconsistent results
Moreover, the path FIt−2 → SIt−1 → Lnt had a standardized may be attributed to 2 reasons. Firstly, previous studies did
coefficient of 0.025 with a 95% CI of (0.017, 0.033), sug- not adequately control for the reverse causations that exist in
gesting that higher levels of frailty at 1 point in time were the relationships between SI, loneliness, and frailty, and did
associated with higher levels of SI in the future through lone- not adequately control for confounding factors. Secondly, dif-
liness of the following time point. Additionally, the path FIt−2 ferences in data sets used across studies may also contribute
→ Lnt−1 → SIt had a standardized coefficient of 0.005 with to the inconsistency. For instance, the mean age of our sample
a 95% CI of (0.003, 0.009), indicating that higher levels of at baseline was 81.83, which was notably older than the sam-
frailty at 1 point in time were associated with higher levels ples in previous studies with mean ages of 69.3 (19) and 66.3
Innovation in Aging, 2024, Vol. 8, No. 3 5

(20), respectively. Therefore, our study suggests that reducing Conclusion

frailty among older adults requires more attention to be paid
This study investigated the bidirectional relationships and
to SI rather than loneliness, particularly among the oldest-old
mediating effects among SI, loneliness, and frailty among
population.
older adults in China using a novel statistical method called
Although many studies have explored the mediating
GCLM. Our findings suggest that SI plays a more important
effects between SI, loneliness, and mental health, few
role than loneliness in predicting frailty among older adults.
have examined their relationships with frailty (18,27). To
Interventions targeting the reduction of SI may be necessary
address this gap, we conducted a mediation analysis and
to improve their overall health and well-being. Moreover, we
found that loneliness may indirectly affect frailty through
found that frailty predicts both SI and loneliness, highlighting
SI. This suggests that loneliness has a long-term effect on
the importance of addressing frailty in interventions aimed at
frailty through SI among older adults. Our findings sug-
reducing SI and loneliness among older adults. These findings
gest that interventions such as community-based services,
have significant implications for improving the health and
aimed at improving social connectedness, are required to
well-being among older adults and provide valuable insights
reduce SI among lonely older adults and decrease their
into the complex relationships between these factors.
frailty levels.
Our results also revealed that health conditions can
increase SI and loneliness among older adults, which is Supplementary Material
consistent with other studies (22,23). This finding further
Supplementary data are available at Innovation in Aging
underscores the importance of addressing frailty in inter-
online.
ventions aimed at reducing SI and loneliness among older
adults. Interventions such as community healthcare services
and telehealth services that aim to reduce frailty may be Funding
beneficial in reducing SI and loneliness among older adults
(8,46). None.
Our study had several strengths. First, this study used data
from a large, nationally representative sample of individuals Conflict of Interest
aged 65 and older, spanning 16 years, enabling us to explore
the relationships between frailty, loneliness, and SI in greater None.
depth. Second, our study utilized a novel statistical method
called GCLM, which addressed confounding by reverse cau- Data Availability
sality and controlled for both observable and unobservable
time-invariant and time-varying confounds. This approach The data set used in this paper is publicly available in
strengthened causal inference and provided more accurate https://opendata.pku.edu.cn/dataset.xhtml?persistentId=-
insights into the mediating effects. Finally, we utilized a com- doi:10.18170/DVN/WBO7LK. The study was not preregis-
prehensive FI tool to measure frailty, which has been vali- tered.
dated in previous research (30–32). This instrument takes
into account various aspects of health, including physical,
Author Contributions
cognitive, and mental health, making it a more robust mea-
sure of frailty than other tools that only focus on a single C.P. conceived, designed, and analyzed the data and was
dimension of health. responsible for the interpretation of findings, primary draft-
Our study has several limitations that should be ing of the manuscript, and revisions. The author read and
addressed. First, the article used a single item to measure approved the final manuscript.
loneliness. Despite previous research showing that the single-
dimension loneliness scale has a high correlation with mul-
tidimensional scales (13,40), using a composite measure
References
could have provided a more comprehensive understanding 1. Clegg A, Young J, Iliffe S, Rikkert MO, Rockwood K. Frailty in
of the relationship between loneliness and frailty. Second, elderly people. Lancet. 2013;381(9868):752–762. https://doi.
due to limitations in the available data, the paper did not org/10.1016/S0140-6736(12)62167-9
2. Maltby J, Hunt SA, Ohinata A, Palmer E, Conroy S. Frailty and
consider participants’ contacts with nonfamily members,
social isolation: comparing the relationship between frailty
such as friends or neighbors, when measuring SI. This may and unidimensional and multifactorial models of social iso-
have limited our assessment of SI. Finally, it is important to lation. J Aging Health. 2020;32(10):1297–1308. https://doi.
acknowledge that attrition may introduce bias to the results, org/10.1177/0898264320923245
as individuals who dropped out of the study exhibited dif- 3. Collard RM, Boter H, Schoevers RA, Voshaar RCO. Prevalence of
ferent characteristics compared with those who remained. frailty in community-dwelling older persons: a systematic review. J
For example, the observed null effect of loneliness on frailty Am Geriatr Soc. 2012;60(8):1487–1492. https://doi.org/10.1111/
could be attributed to this attrition. Future research could j.1532-5415.2012.04054.x
address the limitations identified in our study by employing 4. Hoogendijk EO, Afilalo J, Ensrud KE, Kowal P, Onder G, Fried LP.
more comprehensive measures of loneliness, such as mul- Frailty: implications for clinical practice and public health. Lan-
cet. 2019;394(10206):1365–1375. https://doi.org/10.1016/S0140-
tidimensional scales that capture different dimensions of
6736(19)31786-6
loneliness. Additionally, future studies should include par- 5. Fried LP, Tangen CM, Walston J, et al. Cardiovascular Health Study
ticipants’ interactions with friends, neighbors, or other non- Collaborative Research Group Frailty in older adults: evidence for a
family members in the measurement of SI to validate our phenotype. J Gerontol A Biol Sci Med Sci. 2001;56(3):M146–M156.
findings. https://doi.org/10.1093/gerona/56.3.m146
6 Innovation in Aging, 2024, Vol. 8, No. 3

6. Rockwood K, Song XW, MacKnight C, et al. A global clinical of US national survey data. J Gerontol Series a-Biol Sci Med Sci.
measure of fitness and frailty in elderly people. CMAJ: Canadian 2021;76(11):1977–1980. https://doi.org/10.1093/gerona/glab110
Medical Association Journal = journal de l'Association medi- 23. Luo M. Social isolation, loneliness, and depressive symptoms: a
cale canadienne. 2005;173(5):489–495. https://doi.org/10.1503/ twelve-year population study of temporal dynamics. J Geron-
cmaj.050051 tol Ser B, Psychol Sci Soc Sci. 2023;78(2):280–290. https://doi.
7. He B, Ma Y, Wang C, et al. Prevalence and risk factors for frailty org/10.1093/geronb/gbac174
among community-dwelling older people in China: a systematic 24. Merchant RA, Liu SG, Lim JY, Fu XX, Chan YH. Factors asso-
review and meta-analysis. J Nutr, Health Aging 2019;23(5):442– ciated with social isolation in community-dwelling older adults:
450. https://doi.org/10.1007/s12603-019-1179-9 a cross-sectional study. Qual Life Res: Int J Qual Life Aspects
8. Antonucci TC, Ajrouch KJ, Birditt KS. The convoy model: explain- Treatment, Care Rehabil. 2020;29(9):2375–2381. https://doi.
ing social relations from a multidisciplinary perspective. Gerontol- org/10.1007/s11136-020-02493-7
ogist. 2014;54(1):82–92. https://doi.org/10.1093/geront/gnt118 25. Hajek A, Brettschneider C, Eisele M, et al. Social support and func-
9. Valtorta NK, Kanaan M, Gilbody S, Hanratty B. Loneliness, tional decline in the oldest old. Gerontology. 2022;68(2):200–208.
social isolation and social relationships: what are we measur- https://doi.org/10.1159/000516077
ing? A novel framework for classifying and comparing tools. 26. Hoogendijk EO, Suanet B, Dent E, Deeg DJH, Aartsen MJ. Adverse
BMJ Open. 2016;6(4):e010799. https://doi.org/10.1136/ effects of frailty on social functioning in older adults: results from
bmjopen-2015-010799 the Longitudinal Aging Study Amsterdam. Maturitas. 2016;83:45–
10. Holt-Lunstad J, Smith TB, Baker M, Harris T, Stephenson D. 50. https://doi.org/10.1016/j.maturitas.2015.09.002
Loneliness and social isolation as risk factors for mortality: a 27. Santini ZI, Jose PE, Cornwell EY, et al. Social disconnectedness, per-
meta-analytic review. Perspect Psychol Sci: J Assoc Psychol Sci. ceived isolation, and symptoms of depression and anxiety among
2015;10(2):227–237. https://doi.org/10.1177/1745691614568352 older Americans (NSHAP): a longitudinal mediation analysis. Lan-
11. Jarach CM, Tettamanti M, Nobili A, D’Avanzo B. Social isolation cet Public Health. 2020;5(1):E62–E70. https://doi.org/10.1016/
and loneliness as related to progression and reversion of frailty in s2468-2667(19)30230-0
the Survey of Health Aging Retirement in Europe (SHARE). Age 28. Zimmer Z, Martin LG, Nagin DS, Jones BL. Modeling disability
Ageing. 2021;50(1):258–262. https://doi.org/10.1093/ageing/ trajectories and mortality of the oldest-old in China. Demography.
afaa168 2012;49(1):291–314. https://doi.org/10.1007/s13524-011-0075-7
12. Nicholson NR. A review of social isolation: an important but 29. Gao M, Sa Z, Li Y, et al. Does social participation reduce the risk of
under assessed condition in older adults. J Prim Prev. 2012;33(2- functional disability among older adults in China? A survival anal-
3):137–152. https://doi.org/10.1007/s10935-012-0271-2 ysis using the 2005–2011 waves of the CLHLS data. BMC Geriatr.
13. Yu B, Steptoe A, Chen YJ. Social isolation, loneliness, and all- 2018;18(1):224. https://doi.org/10.1186/s12877-018-0903-3
cause mortality: a cohort study of 35,254 Chinese older adults. J 30. Searle SD, Mitnitski A, Gahbauer EA, Gill TM, Rockwood K.
Am Geriatr Soc. 2022;70(6):1717–1725. https://doi.org/10.1111/ A standard procedure for creating a frailty index. BMC Geriatr.
jgs.17708 2008;8:24. https://doi.org/10.1186/1471-2318-8-24
14. Leigh-Hunt N, Bagguley D, Bash K, et al. An overview of system- 31. Hu K, Keenan K, Hale JM, Borger T. The association between
atic reviews on the public health consequences of social isolation city-level air pollution and frailty among the elderly population in
and loneliness. Public Health. 2017;152:157–171. https://doi. China. Health Place. 2020;64:102362. https://doi.org/10.1016/j.
org/10.1016/j.puhe.2017.07.035 healthplace.2020.102362
15. Freak-Poli R, Ryan J, Tran T, et al. Social isolation, social support 32. Bennett S, Song XW, Mitnitski A, Rockwood K. A limit to frailty
and loneliness as independent concepts, and their relationship with in very old, community-dwelling people: a secondary analysis of
health-related quality of life among older women. Aging Mental the Chinese longitudinal health and longevity study. Age Ageing.
Health. 2022;26(7):1335–1344. https://doi.org/10.1080/1360786 2013;42(3):372–377. https://doi.org/10.1093/ageing/afs180
3.2021.1940097 33. Rockwood K, Mitnitski A. Frailty in relation to the accumulation
16. Guo LZ, An L, Luo FP, Yu B. Social isolation, loneliness and of deficits. J Gerontol A Biol Sci Med Sci. 2007;62(7):722–727.
functional disability in Chinese older women and men: a longi- https://doi.org/10.1093/gerona/62.7.722
tudinal study. Age Ageing. 2021;50(4):1222–1228. https://doi. 34. Pizzonia M, Giannotti C, Carmisciano L, et al. Frailty assessment,
org/10.1093/ageing/afaa271 hip fracture and long-term clinical outcomes in older adults. Eur J
17. Berkman LF, Glass T, Brissette I, Seeman TE. From social inte- Clin Invest. 2021;51(4):e13445. https://doi.org/10.1111/eci.13445
gration to health: Durkheim in the new millennium. Soc Sci Med 35. Romero-Ortuno R. An alternative method for Frailty Index cut-off
(1982). 2000;51(6):843–857. https://doi.org/10.1016/s0277- points to define frailty categories. Eur Geriatr Med. 2013;4(5):299–
9536(00)00065-4 303. https://doi.org/10.1016/j.eurger.2013.06.005
18. Yang RM, Wang HC, Edelman LS, et al. Loneliness as a media- 36. Huang Y, Zhu X, Liu X, Li J. The effects of loneliness, social iso-
tor of the impact of social isolation on cognitive functioning of lation, and associated gender differences on the risk of developing
Chinese older adults. Age Ageing. 2020;49(4):599–604. https://doi. cognitive impairment for Chinese oldest old. Aging Ment Health.
org/10.1093/ageing/afaa020 2022;27:1360–1367. https://doi.org/10.1080/13607863.2022.211
19. Gale CR, Westbury L, Cooper C. Social isolation and loneliness as 6396
risk factors for the progression of frailty: the English Longitudi- 37. Steptoe A, Shankar A, Demakakos P, Wardle J. Social isolation,
nal Study of Ageing. Age Ageing. 2018;47(3):392–397. https://doi. loneliness, and all-cause mortality in older men and women.
org/10.1093/ageing/afx188 Proc Natl Acad Sci USA. 2013;110(15):5797–5801. https://doi.
20. Davies K, Maharani A, Chandola T, Todd C, Pendleton N. The org/10.1073/pnas.1219686110
longitudinal relationship between loneliness, social isolation, and 38. Courtin E, Knapp M. Social isolation, loneliness and health
frailty in older adults in England: a prospective analysis. Lancet in old age: a scoping review. Health Social Care Commun.
Healthy Longev. 2021;2(2):e70–e77. https://doi.org/10.1016/ 2017;25(3):799–812. https://doi.org/10.1111/hsc.12311
S2666-7568(20)30038-6 39. Tanskanen J, Anttila T. A prospective study of social iso-
21. Cacioppo JT, Hawkley LC. Social isolation and health, with lation, loneliness, and mortality in Finland. Am J Public
an emphasis on underlying mechanisms. Perspect Biol Med. Health. 2016;106(11):2042–2048. https://doi.org/10.2105/
2003;46(3):S39–S52. https://doi.org/10.1353/pbm.2003.0049 AJPH.2016.303431
22. Cruz BP, Perales F, Alfonso-Rosa RM, Del Pozo-Cruz J. Bidirec- 40. Luo Y, Waite LJ. Loneliness and mortality among older adults in
tional and dynamic relationships between social isolation and phys- China. J Gerontol Series B, Psychol Sci Soc Sci. 2014;69(4):633–
ical functioning among older adults: a cross-lagged panel model 645. https://doi.org/10.1093/geronb/gbu007
Innovation in Aging, 2024, Vol. 8, No. 3 7

41. Zyphur MJ, Allison PD, Tay L, et al. From data to causes I: building 44. Pan C, Cao N. Bidirectional and dynamic relationships between
a general cross-lagged panel model (GCLM). Organ Res Methods. social isolation and frailty among older adults in China. Arch
2020;23(4):651–687. https://doi.org/10.1177/1094428119847278 Gerontol Geriatr. 2024;116:105229. https://doi.org/10.1016/j.
42. Hamaker EL, Kuiper RM, Grasman RPPP. A critique of the cross- archger.2023.105229
lagged panel model. Psychol Methods. 2015;20(1):102–116. 45. Cacioppo JT, Cacioppo S. Loneliness in the modern age: an evo-
https://doi.org/10.1037/a0038889 lutionary theory of loneliness (ETL). Adv Exp Soc Psychol.
43. Cruz BD, Perales F, Alfonso-Rosa RM, Del Pozo-Cruz J. Bidirec- 2018;58:127–197.
tional and dynamic relationships between social isolation and phys- 46. Martin PR, Reece J, Lauder S, McClelland A. A randomised con-
ical functioning among older adults: a cross-lagged panel model trolled trial of a social support intervention. Appl Psychol: Health
of US national survey data. J Gerontol Series A, Biol Sci Med Sci. Well-Being. 2011;3(1):44–65. https://doi.org/10.1111/j.1758-
2021;76(11):1977–1980. https://doi.org/10.1093/gerona/glab110 0854.2010.01044.x