The bidirectional relationships between social isolation and cognitive function among older adults in China: separating between-person effects from within-person effects

Abstract

This study investigates the bidirectional relationship between social isolation and cognitive function among older adults in China, utilizing data from the Chinese Longitudinal Healthy Longevity Survey (CLHLS). The baseline survey, which commenced in 2008, tracked 1,662 participants over four waves of data collection spanning a decade. We employed Cross-Lagged Panel Models (CLPM) and Random Intercept Cross-Lagged Panel Models (RI-CLPM) to analyze these relationships at both the between-person and within-person levels. CLPM results indicated significant cross-lagged effects between social isolation (β = − 0.119, p < 0.001; β = − 0.162, p < 0.001) and cognitive function (β = − 0.073, p < 0.001; β = − 0.091, p < 0.001) at the between-person level over the last three waves. Specifically, higher prior levels of social isolation were associated with a significant decline in subsequent cognitive function, and vice versa. Furthermore, RI-CLPM results showed that, after controlling for random intercepts and covariates, only social isolation had a significant negative impact on cognitive function across all waves (β = − 0.051, p < 0.05; β = − 0.047, p < 0.05; β = − 0.028, p < 0.05). Overall, this study demonstrates that, when considering both between-person and within-person effects, social isolation exerts a stronger lag effect on cognitive function among older adults in China. This suggests that, over a specific timeframe, reducing social isolation is crucial for promoting healthy aging in this population.

Introduction

Currently, approximately 50 million people worldwide suffered from dementia, incurring an estimated global economic cost of $1 trillion annually¹. With an aging population, the number of dementia patients was projected to increase to 152 million by 2050¹. China alone accounted for about 25% of the global dementia population, presenting a significant challenge for policy makers, healthcare professionals, and family members^2,3. Cognitive decline is one of the earliest and most apparent symptoms of dementia, severely impacting life quality^4,5 and increasing the risk of other diseases and mortality^6,7. Therefore, it is crucial to identify potentially modifiable risk factors for cognitive decline at an early stage.

Social isolation encompasses the structural characteristics of social relationships, including reduced social contact, living alone, and infrequent participation in social activities⁸. In recent years, social distancing policies implemented due to the COVID-19 pandemic have heightened public concern regarding social isolation and its potential impact on the health of older adults^9,10. The 2020 Lancet Commission report identified low social contact as a modifiable risk factor for dementia¹¹, aligning with the social convoy model theory¹² and supported by a substantial body of research^13–15. However, some scholars contended that the gradual onset of dementia may lead to social isolation, rather than social isolation caused dementia^16,17. Additionally, other researchers proposed that social isolation negatively impacts cognitive function, while cognitive impairment may also exacerbate social isolation, creating a “vicious cycle”^18,19. Therefore, it is essential to explore the potential bidirectional relationship between social isolation and cognitive function to clarify the causal sequence and the strength of each impact.

Most empirical studies examining the association between social isolation and cognitive function are derived from cross-sectional studies or investigations of uni-directional longitudinal effects^14,15,20. Only a few studies have addressed the bidirectional relationships between social connections and cognitive function in older adults. Among those that do explore this bidirectional relationship, the focus is predominantly on the link between loneliness and cognitive function, yielding mixed results^21–23. It is important to note that loneliness reflects the functional aspects of social connections, which is determined by the quality of social relationships⁸, and is related to but distinct from social isolation. Moreover, most previous studies have utilized Cross-Lagged Panel Models (CLPM) to investigate the reciprocal relationships between two variables without clearly distinguishing between between-person and within-person effects. However, CLPM is more susceptible to confounding variables, as it assesses differences across individuals rather than dynamic changes within the same individual²⁴. Random Intercept Cross-Lagged Panel Models(RI-CLPM) builds upon the CLPM by incorporating random intercepts, which explicitly differentiates between-person effects and within-person effects²⁵. By controlling for baseline differences among individuals, RI-CLPM allows for an analysis focused on changes within the same individual over time²⁶. This approach allows for a closer examination of the actual causal relationship between social isolation and cognitive function over time^25,27.

To address this research gap, our study utilizes 10 years of four-wave follow-up data from the Chinese Longitudinal Healthy Longevity Survey (CLHLS). CLPM and RI-CLPM were employed to investigate the bidirectional relationships between social isolation and cognitive function at both the between-person and within-person effects. The objective was to provide scientific evidence that can inform the development of more targeted public health strategies.

Methods

Data source and sample selection

The data for this study were sourced from the CLHLS, a nationwide prospective longitudinal survey²⁸. Older adults aged 60 and above were recruited from 23 provinces across China. The combined population of these regions accounts for approximately 85% of China’s total national population. Data on demographic characteristics, socioeconomic status, lifestyle, and health status were collected via face-to-face interviews by study assessors. The survey was initiated in 1998 and conducted every 2–4 years. The CLHLS was approved by the Research Ethics Committee of Peking University (IRB00001052-13074), and all participants or their proxy respondents provided written informed consent. Data in this study were drawn from the last four survey waves, from 2008 to 2018. After excluding samples with missing key variable data and ineligible respondents, the final sample comprised 1662 older adults aged 60 and above. The specific process of sample selection is illustrated in Fig. 1. In this study, we utilized G*Power version 3.1.9.7 to calculate the sample size and statistical power. We assumed an effect size of 0.1, a significance level of 0.05, and a statistical power of 0.95, with two groups and four measurement occasions. Conservatively, we set the intraclass correlation coefficient to 0. Based on these parameters, the minimum sample size required was determined to be 432 participants. To further ensure the robustness of our results, we estimated an attribution rate of 20%, ultimately adjusting the minimum sample size to 540 participants. Our study met the minimum sample size requirement.

Fig. 1.

Flowchart of the selection of study population from participants of the CLHLS.

Measurements

Social isolation

Social isolation was assessed with five dimensions, which was widely used in empirical research^18,29. These dimensions include living arrangements, spousal status, frequency of contact with children and siblings, and participation in social activities such as organized events and playing cards or mahjong. Participants living alone, without a spouse, infrequently contacting children and siblings, and not participating in social activities received one point for each criterion. The total social isolation score ranged from 0 to 5, with higher scores indicating greater levels of social isolation.

Cognitive function

Cognitive function was assessed with the Chinese version of the Mini-Mental State Examination (CMMSE), a well-validated and reliable instrument^30,31. The CMMSE comprises 24 questions across 7 subscales. The orientation subscale includes 4 points for time orientation and 1 point for place orientation. The naming foods subscale involves naming as many foods as possible within 1 min, with a maximum of 7 points. The registering three words subscale allows for a maximum of 3 points. The attention and calculation subscale requires participants to subtract 3 from 20 repeatedly, with a maximum of 5 points. The copying a figure subscale has a maximum of 1 point. The recall subscale involves the delayed recall of the three words mentioned earlier, with a maximum of 3 points. Finally, the language subscale allocates 2 points for naming objects, 1 point for repeating sentences, and 3 points for following commands. The total CMMSE score ranges from 0 to 30, with higher scores indicating better cognitive function.

Covariates

To minimize the potential influence of covariates on relationships between social isolation and cognitive function, and to enhance the simplicity and interpretability of the models³², this study controlled covariates that were simultaneously related to social isolation and cognitive function in the 2008 baseline survey^33,34. Demographic characteristics included gender (male or female), age (60–70 years, 71–80 years, 81–90 years, or above 90 years), educational level (none, 1–6 years, or more than 6 years), and current residence (urban, town, or rural). Health status was measured by Activities of Daily Living (ADL), which is a composite score of independence in activities such as bathing, dressing, toileting, mobility, continence, and feeding²⁹, and depressive symptoms, assessed through three questions: “Do you feel fearful or anxious?” “Do you feel lonely or isolated?”, and “Do you feel increasingly useless as you age?”³⁵. Health behaviors included smoking status (yes or no) and drinking habits (yes or no).

Statistical analysis

This study utilizes CLPM and RI-CLPM to investigate the bidirectional relationship between social isolation and cognitive function at both the between-person and within-person levels^25,36. CLPM assumes that variables mutually influence each other over time, with individuals beginning from similar initial conditions, thus primarily capturing the between-person levels rather than changes unique to individuals³⁶. In contrast, RI-CLPM extends CLPM by incorporating random intercepts to distinctly differentiate between between-person and within-person levels^25,37. This model addresses baseline differences among individuals, enabling the analysis to concentrate on temporal changes within the same older adults, thereby emphasizing the within-person effects^25,27.

First, we conducted descriptive statistical analyses to characterize study participants. Continuous variables were described using means and standard deviations, while categorical variables were described using frequencies and percentages. Next, we performed Pearson correlation analyses of social isolation and cognitive function between and within time points. Subsequently, we employed CLPM and RI-CLPM to explore the bidirectional relationships between social isolation and cognitive function both between individuals and within individuals. To control the potential impact of covariates on effect sizes, we constructed four models.

Model 1 utilized CLPM to observe the effects of social isolation and cognitive function between persons. This model included autoregressive paths, cross-lagged effects and synchronous correlations. Model 2 incorporated baseline covariates based on Model 1. Model 3 employed RI-CLPM to observe the effects of social isolation and cognitive function within persons. Unlike CLPM, RI-CLPM introduced random intercepts, focusing more on within-person changes, thereby reducing the interference of between-person differences on the estimation of lag effects^25,27. Model 4 incorporated baseline covariates based on Model 3.

In this study, models are estimated using robust maximum likelihood estimation (MLR), which is applicable to non-normal and non-independent data. We used the following indices to evaluate the fit of the four models³⁸: Comparative Fit Index (CFI), Tucker-Lewis Index (TLI), Root Mean Square Error of Approximation (RMSEA) and Standardized Root Mean Square Residual (SRMR). A CFI > 0.90, a TLI > 0.90, a RMSEA < 0.08 and a SRMR < 0.08 are considered good. When comparing the four models, lower AIC and BIC values indicate better fit, suggesting a superior model¹⁹. According to guidelines on cross-lagged effects, this study employs benchmark values of 0.03 (small effect), 0.07 (medium effect), and 0.12 (large effect) to represent these effects³⁹. In longitudinal studies, effect sizes are generally much smaller than those in cross-sectional studies, and cross-lagged effects should be significantly lower than correlation coefficients. This is due to the autoregressive effects accounting for a large portion of the variance, attributed to the stability of the variables⁴⁰. All analyses were conducted using SPSS 27.0 and Mplus 8.3, with P < 0.05 indicating statistical significance.

Results

Characteristics of study participants

Table 1 demonstrates significant differences in several baseline characteristics between the analyzed and excluded samples in this study. The analyzed sample comprises a higher proportion of males, predominantly aged between 60 and 80 years, who have attained higher educational levels and are more likely to reside in city. This group exhibits superior performance in activities of daily living and lower scores on depression symptoms, alongside higher rates of smoking and alcohol consumption.

Table 2.

Pearson correlation coefficients between social isolation and cognitive function within and between four time points.

	Mean ± SD	SI1	SI2	SI3	SI4	CF1	CF2	CF3	CF4
SI1	1.62 ± 1.13	1
SI2	1.72 ± 1.16	0.734**	1
SI3	1.92 ± 1.16	0.535**	0.626**	1
SI4	2.16 ± 1.11	0.452**	0.518**	0.556**	1
CF1	27.22 ± 3.53	-0.275**	-0.223**	-0.166**	-0.194**	1
CF2	27.16 ± 3.73	-0.215**	-0.218**	-0.206**	-0.180**	0.257**	1
CF3	26.95 ± 4.03	-0.201**	-0.208**	-0.228**	-0.213**	0.237**	0.434**	1
CF4	24.89 ± 6.63	-0.246**	-0.267**	-0.255**	-0.263**	0.256**	0.358**	0.444**	1

SI, Social isolation; CF, cognitive function; SD, standard deviation.

*p < 0.05, **p < 0.01.

Table 1.

Comparison of baseline characteristics of included and excluded samples (N = 1662).

Variables	Category	Analyzed sample(N = 1662)	Excluded sample(N = 15172)	χ2/F	P
Gender n(%)	Female	879 (52.89)	8799 (57.99)	15.985	0.000***
	Male	783 (47.11)	637 3(42.01)
Age n(%)	60–70	520 (31.29)	1159 (7.64)	2335.04	0.000***
	71–80	722 (43.44)	2161 (14.24)
	80–90	355 (21.36)	3921 (25.84)
	Above 90	65 (3.91)	7931 (52.27)
Educational level n(%)	0 yeara	803 (48.32)	9714 (64.03)	161.318	0.000***
	1–6 years	469 (28.22)	3144 (20.72)
	Above 6 years	390 (23.47)	2314 (15.25)
Current residence n(%)	City	1114 (67.03)	9082 (59.86)	60.087	0.000***
	Town b	336 (20.22)	2957 (19.49)
	Rural	212 (12.76)	3133 (20.65)
ADL c mean ± SD		0.04 ± 0.39	1.10 ± 2.51	295.09	0.000***
Depressive symptoms mean ± SD		6.35 ± 2.40	8.26 ± 3.65	431.992	0.000***
Smoking n(%)	No	1280 (77.02)	12,638 (83.30)	41.283	0.000***
	Yes	382 (22.98)	2534 (16.70)
Drinking n(%)	No	1292 (77.74)	12,661 (83.45)	34.454	0.000***
	Yes	370 (22.26)	2511 (16.55)

*p < 0.05, **p < 0.01, *** p < 0.001; SD, standard deviation

^a0 year, individuals who have not received formal schooling, including those who are illiterate or unable to read.

^btown, areas of population density that fall between city and rural areas, typically including small cities, suburbs, or market towns.

^cADL, Activities of daily living

Correlation between social isolation and cognitive function

Table 2 illustrates the Pearson correlation coefficients between social isolation and cognitive function. At each time point, social isolation and cognitive function exhibit a significant negative correlation (p < 0.01). Additionally, prior social isolation is significantly negatively correlated with subsequent cognitive function (p < 0.01), and prior cognitive function is significantly negatively correlated with subsequent social isolation (p < 0.01).

Model test of the relations between social isolation and cognitive function

Table 3 summarizes the fit indices for the four models. The results indicate that Model 1 has a relatively modest fit (CFI = 0.933, TLI = 0.849, RMSEA = 0.103, SRMR = 0.06). However, Orth et al. suggest that it is acceptable for some fit indices in the CLPM model to fall below the threshold⁴¹. Meanwhile, the fit of Model 2 is relatively poor (CFI = 0.754, TLI = 0.695, RMSEA = 0.092, SRMR = 0.152), which may be attributed to the increased complexity of the model from the introduction of covariates, resulting in an excess of parameters that adversely affects the overall fit³⁶. Additionally, this could also be related to the inherent limitations of the CLPM, which fails to adequately account for stable individual differences, rather than being solely attributable to the increase in covariates⁴². Therefore, Model 1 was selected to observe the effects of social isolation and cognitive function between persons. Model 3 shows a good fit (CFI = 0.979, TLI = 0.934, RMSEA = 0.076, SRMR = 0.046), with all indicators meeting the required thresholds, indicating that the addition of random intercepts significantly improved the model fit. Model 4, which includes baseline covariates based on Model 3 and imposes constraints on the lagged paths of social isolation to cognitive function, demonstrates the best fit among the four models (CFI = 0.965, TLI = 0.946, RMSEA = 0.043, SRMR = 0.048). Compared to Model 3, Model 4 has lower AIC and BIC values and takes baseline covariates into account. Hence, Model 4 was chosen to observe the within-person effects of social isolation and cognitive function.

Table 3.

Model fit indices of the four models.

Model	χ2	df	CFI	TLI	RMSEA	SRMR	AIC	BIC
Model 1	222.347	12	0.933	0.849	0.103	0.060	55123.8	55297.1
Model 2	1109.355	74	0.754	0.695	0.092	0.152	55104.4	55288.6
Model 3	96.448	9	0.979	0.934	0.076	0.046	54952.8	55142.4
Model 4	237.937	59	0.965	0.946	0.043	0.048	54038.7	54304.1

Model 1 = Cross-lagged panel model (CLPM).

Model 2 = Cross-lagged panel model (CLPM) adjusting for gender, age, educational level, current residence, activities of daily living, depressive symptoms, smoking and drinking.

Model 3 = Random intercept cross-lagged panel model (RI-CLPM).

Model 4 = Constraint coefficient incorporated random intercept cross-lagged panel model (RI-CLPM) adjusting for gender, age, educational level, current residence, activities of daily living, depressive symptoms, smoking and drinking.

Bidirectional relationship between social isolation and cognitive function

Table 4 presents the standardized coefficients for the four models examining social isolation and cognitive function among older adults in China, including autoregressive paths, cross-lagged paths, and synchronous correlations. Based on the fit indices of these models, we utilized Model 1 and Model 4 to observe the effects of social isolation and cognitive function between and within persons, as depicted in Figs. 1 and 2, respectively.

Table 4.

Standardization coefficient results of social isolation and cognitive function

Parameter	Model 1		Model 2		Model 3		Model 4
	β	SE	β	SE	β	SE	β	SE
Between-person	/	/	/	/	-0.476***	0.045	-0.111	0.084
Autoregressive effects
SI1→SI2	0.728***	0.013	0.726***	0.013	0.538***	0.026	0.514***	0.026
SI2→SI3	0.610***	0.016	0.606***	0.017	0.346***	0.037	0.320***	0.035
SI3→SI4	0.535***	0.018	0.531***	0.018	0.213***	0.042	0.202***	0.038
CF1→CF2	0.214***	0.032	0.211***	0.031	-0.112*	0.051	-0.037	0.044
CF2→CF3	0.408***	0.034	0.407***	0.033	0.218***	0.034	0.248***	0.032
CF3→CF4	0.408***	0.030	0.406***	0.030	0.289***	0.026	0.276***	0.026
Cross-lagged effects
SI1→CF2	-0.156***	0.024	-0.150***	0.024	-0.090*	0.041	-0.051*	0.025
SI2→CF3	-0.119***	0.023	-0.113***	0.023	-0.050	0.035	-0.047*	0.023
SI3→CF4	-0.162***	0.021	-0.159***	0.021	-0.111***	0.029	-0.028*	0.014
CF1→SI2	-0.022	0.017	-0.02	0.017	0.019	0.028	0.029	0.026
CF2→SI3	-0.073***	0.020	-0.072***	0.020	-0.055	0.030	-0.046	0.028
CF3→SI4	-0.091***	0.020	-0.089***	0.019	-0.051	0.032	-0.050	0.030
Synchronous correlations
SI1 WITH CF1	-0.275***	0.023	-0.270***	0.023	-0.146***	0.037	-0.119***	0.033
SI2 WITH CF2	-0.087***	0.024	-0.086***	0.024	-0.043	0.035	-0.033	0.030
SI3 WITH CF3	-0.100***	0.022	-0.099***	0.022	-0.091**	0.030	-0.078**	0.028
SI4 WITH CF4	-0.119***	0.021	-0.119***	0.021	-0.109***	0.029	-0.058*	0.028

SI, social isolation; CF, cognitive function.

Model 1 = Cross-lagged panel model (CLPM).

Model 2 = Cross-lagged panel model (CLPM) adjusting for gender, age, educational level, current residence, activities of daily living, depressive symptoms, smoking and drinking.

Model 3 = Random intercept cross-lagged panel model (RI-CLPM).

Model 4 = Constraint coefficient incorporated random intercept cross-lagged panel model (RI-CLPM) adjusting for gender, age, educational level, current residence, activities of daily living, depressive symptoms, smoking and drinking.

*p < 0.05, **p < 0.01, ***p < 0.001.

Fig. 2.

Cross-lagged panel model (CLPM) *p < 0.05, **p < 0.01, ***p < 0.001.

As shown in Fig. 2, the cross-lagged effects between persons over the last three waves are significant for both social isolation (β = -0.119, p < 0.001; β = -0.162, p < 0.001) and cognitive function (β = -0.073, p < 0.001; β = -0.091, p < 0.001). An increase in previous social isolation is associated with a significant decline in subsequent cognitive function. Similarly, a decline in previous cognitive function is associated with a significant increase in subsequent social isolation.

However, the results of Model 4 differ from those of Model 1. To improve the model’s fit and enhance the interpretability of our findings, we standardized the lagged effects of social isolation on cognitive function. Specifically, we imposed constraints on these paths to ensure their consistency over time^43,44. As depicted in Fig. 3, with the inclusion of random intercepts and covariates, only social isolation (β = -0.051, p < 0.05; β = -0.047, p < 0.05; β = -0.028, p < 0.05) has a significant within-person lag effect on cognitive function. Specifically, previous social isolation has a significantly negative effect on subsequent cognitive function, whereas the effect of previous cognitive function on subsequent social isolation, although negative, is not significant.

Fig. 3.

Constraint coefficient incorporated random intercept cross-lagged panel model (RI-CLPM) adjusting for gender, age, educational level, current residence, activities of daily living, depressive symptoms, smoking and drinking *p < 0.05, **p < 0.01, ***p < 0.001.

Discussion

This study utilized a decade of four-wave follow-up data from the CLHLS to systematically explore the bidirectional relationship between social isolation and cognitive function among older adults in China, employing both the CLPM and the RI-CLPM. The findings indicate a significant lag effect between social isolation and cognitive function, although the outcomes differ between the CLPM and RI-CLPM analyses.

Specifically, the CLPM analysis reveals that social isolation has a substantial negative impact on subsequent cognitive function, and conversely, a decline in cognitive function exacerbates social isolation. The lag effect of social isolation on cognitive function is more pronounced than that of cognitive function on social isolation. In contrast, the RI-CLPM analysis uncovers more complex within-person dynamic changes, highlighting the importance of individual baseline characteristics. According to the RI-CLPM results, social isolation significantly affects cognitive function over time, but the impact of cognitive function on social isolation is not significant. By controlling for confounding factors such as gender, age, education level, current residence, activities of daily living, depressive symptoms, smoking, and drinking, the robustness of the RI-CLPM results is ensured. This suggests that social isolation is a critical antecedent of cognitive decline, whereas the influence of changes in cognitive function on social isolation is relatively weak.

The CLPM analysis indicates that social isolation among older adults in China significantly negatively impacts subsequent cognitive function, and conversely, a decline in cognitive function exacerbates social isolation. Our findings align largely with existing literature^18,19,45,46. This bidirectional relationship can be explained from multiple perspectives. First, regarding the impact of social isolation on cognitive function, the cognitive reserve theory posits that social isolation reduces interaction with the outside world, limits brain stimulation and exercise, and thereby diminishes cognitive reserve, ultimately affecting cognitive function^18,19. Additionally, the stress-buffering theory suggests that social isolation weakens the social support system, leaving individuals without effective buffers against stress, resulting in subsequent cognitive decline^47,48. Moreover, social isolation may lead to unhealthy lifestyles, such as lack of exercise and sleep disorders, which also negatively impact cognitive function^8,45. Conversely, regarding the impact of cognitive function on social isolation, the socioemotional selectivity theory indicates that as people age, those with declining cognitive function may prioritize interactions with a few close individuals, reducing extensive social interactions and thereby exacerbating social isolation⁴⁹. Additionally, cognitive decline may make it difficult for individuals to maintain and manage complex social networks, further reducing social interactions^50,51. Therefore, this bidirectional relationship indicates that the mutual influence between social isolation and cognitive function is a complex and dynamic process that warrants further research and attention.

The CLPM results indicate that the lag effect of social isolation on cognitive function among older adults in China is significantly greater than the lag effect of cognitive function on social isolation. This discrepancy may arise because social isolation has a direct and immediate impact on cognitive function, while changes in cognitive function take longer to significantly affect the degree of social isolation. Additionally, the three-year interval between assessments may be insufficient to capture the short-term impact of cognitive function on social isolation. Furthermore, we must acknowledge the limitations of existing assessment tools. Although the Chinese version of Mini-Mental State Examination (CMMSE) is widely used for evaluating cognitive function, it may not capture subtle changes, which could have limited short-term impacts on social isolation. Therefore, future research should consider using more sensitive assessment tools, such as the Montreal Cognitive Assessment (MoCA), to detect these subtle changes.

According to the RI-CLPM analysis, social isolation among the older adults in China has a significant lag effect on cognitive function, whereas the lag effect of cognitive function on social isolation is negative but not significant. Compared to CLPM, RI-CLPM introduces random intercepts to control for between-person differences and emphasize within-person changes, thereby reducing the interference of between-person differences in estimating lag effects^25,27. This regulatory mechanism likely explains the differing results between the two models. Additionally, different within-person associations exist among participants^21,50. Specifically, a decline in cognitive function may lead to a deterioration in social relationship quality for some participants, while others may receive more emotional and practical support. Therefore, these between-person and within-person differences together contribute to the non-significance of the lag effect of cognitive function on social isolation.

Our study reveals the complex interplay between social isolation and cognitive function among older adults in China, which has important practical implications for formulating social policies and interventions. Specifically, the bidirectional relationship at the within-person level may help identify actionable intervention areas, while the bidirectional relationship at the between-person level may help recognize those who might benefit from such interventions^19,52. In our within-person analysis, we found that elderly individuals facing social isolation are at significant risk of cognitive decline. This underscores the importance of closely monitoring individuals’ social isolation status for timely interventions. In the between-person analysis, we observed a negative cycle between social isolation and cognitive function, with social isolation having a more pronounced impact on cognitive function compared to their peers. This indicates that older adults experiencing social isolation requires special attention and support. Based on our research findings, we recommend regularly assessing the social interactions and support networks of older adults. Social policies should prioritize those experiencing social isolation to effectively mitigate cognitive decline and improve their quality of life.

Strengths and limitations

This study has several significant advantages. First, we utilized 10 years of national data, providing a wealth of longitudinal information that allows for a comprehensive understanding of the dynamic relationship between social isolation and cognitive function. Second, our analysis not only focused on the impact of social relationships on cognitive function but also examined the reciprocal effect of cognitive function on social relationships. This bidirectional analysis offers a more holistic understanding of the adaptive mechanisms of cognitive aging. Lastly, we employed both the CLPM and the RI-CLPM to analyze bidirectional relationships at both the between-person and within-person levels²⁵, thereby more accurately capturing associations at different levels.

However, this study also has several limitations. First, the information on social isolation relies on self-reports from participants, which may introduce retrospective recall and social desirability biases, thereby affecting data accuracy. Second, as with any longitudinal cohort study, loss to follow-up is inevitable, and missing data may lead to bias, resulting in the final sample being a relatively healthy group compared to the original data, which reduced the generalizability of the study results. Additionally, there were only four measurement points over the 10-year period, and this relatively long interval may not provide in-depth insights into processes occurring on shorter time scales. Lastly, RI-CLPM primarily investigates within-person associations that dominate the population, but there may be different within-person associations among various individuals, which are not fully captured by the model. Moreover, we opted for a constrained version of RI-CLPM to maximize statistical power; although the resulting model fit is satisfactory, such constraints may limit the model’s flexibility. In the future, we will adopt digital tracking methods to provide more objective and continuous data and consider combining RI-CLPM with methods that capture betwen-person heterogeneity in within-person associations⁵³, to offer more precise and in-depth insights.

Conclusions and implications

In summary, using 10 years of four-wave follow-up data from the CLHLS, we found that social isolation exerts a stronger lag effect on cognitive function, whether exploring between-person or within-person effects. Both social isolation and dementia are major global public health challenges. This suggests that, in a specific timeframe, reducing social isolation is more important for healthy aging in older adults. Future research should further explore the underlying mechanisms between social isolation and cognitive function to better understand and address this complex relationship.

Author contributions

X.X. Contributed to conceptualization, data curation, formal analysis, methodology, project administration, resources, software, supervision, validation, visualization, writing the original draft, and review and editing of the draft. F.W. Contributed to data curation, formal analysis, methodology, resources, software, validation, visualization, and review and editing of the draft. Z.Z. Contributed to data curation, resources, and review and editing of the draft. A.X. Contributed to conceptualization, data curation, funding acquisition, project administration, resources, validation, supervision, visualization, and review and editing of the draft. All authors approved the final version.

Funding

This research was supported by the National Social Science Fund of China (2018VJX065).

Data availability

The datasets that support this article are publicly available from the project of the CLHLS. The questionnaires and datasets are free download at website: https://opendata.pku.edu.cn/dataset.xhtml?persistentId=doi:10.18170/DVN/WBO7LK.

Declarations

Competing interests

The authors declare no competing interests.

Institutional review board

This study was conducted according to the guidelines laid down in the Declaration of Helsinki, and all procedures involving research study participants were approved by the biomedical ethics committee of Peking University (IRB00001052–13074).

Informed consent

In the baseline and follow-up questionnaires, all participants—or their legal representatives—signed written permission forms.