Social Networks and Longitudinal Neuropsychiatric Symptom Trajectories across the Cognitive Continuum

Abstract

Introduction

Neuropsychiatric symptoms (NPSs) significantly contribute to caregiver burden and the risk of institutionalization in patients with cognitive dysfunction. Although social isolation is linked to cognitive decline, its associations with longitudinal NPS trajectories are poorly understood. This study explores the association between social network characteristics and longitudinal NPS trajectories across the cognitive continuum.

Methods

We conducted a retrospective observational study of 618 adults (aged ≥ 50) evaluated at a memory clinic between 2017 and 2024. Social networks were assessed, and NPS were measured using the Neuropsychiatric Inventory (NPI). Linear mixed-effects models examined associations between social networks and NPI trajectories, adjusting for demographics, comorbidities, medications, and cognitive status.

Results

At baseline, 485 (78.5%) participants had an adequate social network. A cross-sectional multivariable analysis showed that an adequate social network was independently associated with lower NPI scores (β = −1.716, p = 0.029). Longitudinally, adequate social networks were associated with lower NPI scores (β = −2.156, p = 0.008), though the time × social network interaction was not significant overall (p > 0.05). However, a subgroup analysis revealed a significant interaction (β = −0.124, p < 0.001) in participants without baseline NPS. No significant interactions were observed by cognitive status (p > 0.05).

Conclusion

Social networks are associated with the cross-sectional NPS burden and predict longitudinal symptom trajectories. These findings suggest that social network characteristics could be a modifiable target for attenuating the risk of NPS emergence.

Plain Language Summary

Being socially connected is associated with fewer behavioral issues in people with cognitive decline. People with cognitive decline often experience changes in mood or behavior, such as anxiety, irritability, or depression. Healthcare professionals call these changes neuropsychiatric symptoms. They can be upsetting for both patients and their families and may contribute to an earlier need for institutional care. In this study, we examined whether staying socially connected with family members and friends was associated with the development of these symptoms. We found that people who had regular contact with others tended to have fewer emotional and behavioral problems. Among those who did not have neuropsychiatric symptoms at the beginning of the study, symptoms appeared more slowly in individuals who remained socially engaged than in those who had limited social contact. Because the total number of new symptoms recorded during the observation period was relatively low, extended research is needed to fully verify these patterns over time. Overall, the findings indicate that maintaining active social relationships is linked to a reduced occurrence of behavioral challenges in individuals experiencing cognitive decline.

Introduction

Dementia is a growing global health crisis, with an estimated 57 million people affected worldwide in 2019 and projected to reach 153 million by 2050 due to population aging [1–3]. Neuropsychiatric symptoms (NPS) are important non-cognitive symptoms but are often underrecognized. They affect up to 90% of patients during the disease course and are associated with aggravated functional impairment, diminished quality of life, increased caregiver burden, early institutionalization, and increased healthcare costs and burden [2–5].

Social networks encompass the structural and functional aspects of interpersonal relationships [6]. Previous studies have consistently reported that poor social networks are associated with cognitive dysfunction, pathology progression, and an elevated incidence of dementia [6–9].

However, social networks in relation to longitudinal trajectories of NPS have not received much research attention. Previous studies have used cross-sectional designs or focused on the prevalence of dementia, rather than on neuropsychiatric outcomes [8–10]. Additionally, whether the potential effects of social networks vary by a patient’s baseline NPS profile or underlying cognitive status remains unclear. Clarifying these longitudinal effects is clinically important for psychosocial interventions aimed at preventing or alleviating NPS [7]. Therefore, focusing specifically on older adults evaluated at a memory clinic for cognitive concerns, this study aimed to: (1) evaluate the cross-sectional association between social network status and NPS; (2) determine whether social networks are associated with longitudinal NPS trajectories; and (3) investigate whether baseline NPS status or underlying cognitive status modifies those associations.

Methods

Study Design

We conducted a retrospective observational study using comprehensive electronic medical record data from the Memory Clinic of Korea University Ansan Hospital. As part of our standard clinical pathway, patients presenting with cognitive complaints undergo a standardized diagnostic protocol. During routine outpatient visits, all structured interviews and neuropsychological tests were administered by trained clinical psychologists. Relying on dedicated psychologists, rather than the treating physicians, ensured uniform assessment protocols across all patients and effectively minimized potential examiner bias.

The study protocol received approval from the Institutional Review Board at the participating hospital (IRB No. “removed for review”). Because of the retrospective design of the data collection, the requirement for written informed consent was waived. All procedures adhered to the ethical principles of the Declaration of Helsinki.

Participants

Participants who visited the clinic between March 2017 and July 2024 were screened. The inclusion criteria were as follows: (1) aged 50 years or older at the initial visit; (2) had information on the baseline social networks; and (3) completed at least two NPS evaluations during the follow-up period. The exclusion criteria were individuals with incomplete clinical documentation and those lost to follow-up after the baseline assessment. Based on these criteria, 618 participants were included in this study.

All participants were evaluated on the basis of their medical history, physical and neurological examinations, laboratory tests, brain imaging, and neuropsychological assessments conducted by trained psychologists using the Consortium to Establish a Registry for Alzheimer’s Disease Assessment Packet - Korean version (CERAD-K) neuropsychological battery [11]. Consensus diagnoses by expert neurologists and a neuropsychologist, based on independent review of the neuropsychological test results and comprehensive clinical evaluations, were used to classify cognitive status as cognitively normal (CN), with no significant objective cognitive impairment and completely preserved activities of daily living; mild cognitive impairment, diagnosed using Petersen’s criteria [12]; or dementia, diagnosed according to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition criteria for major neurocognitive disorder [13].

Social Networks

Social networks were assessed using a 4-item index adapted from the Berkman-Syme Social Network Index [14, 15]. The original index assessed marital status, contacts with friends and relatives, church membership, and other group affiliations. We modified this to better capture social integration: (1) cohabitation status (living with family versus living alone); (2) frequency of contact with family members or friends; (3) frequency of interactions with neighbors; and (4) regular involvement in community or religious activities. Each item was scored as 0 (absent/infrequent) or 1 (present/frequent), yielding a total score ranging from 0 to 4. For the comparative analysis, we dichotomized participants into two distinct categories: those with poor social networks (scores of 0–1; designated as SNS(−)) and those with adequate social networks (scores of 2–4; designated as SNS(+)). This threshold selection was determined based on established epidemiological and community-based cohort studies [16, 17].

Neuropsychiatric Symptoms

NPS were evaluated using the Neuropsychiatric Inventory (NPI), a validated caregiver-informant-based interview assessing 12 neuropsychiatric domains: delusions, hallucinations, agitation/aggression, depression/dysphoria, anxiety, elation/euphoria, apathy/indifference, disinhibition, irritability/lability, aberrant motor behavior, nighttime behavioral disturbances, and appetite/eating abnormalities [18]. For each domain, frequency (scored 1–4) and severity (scored 1–3) are assessed, with domain scores calculated by multiplying frequency and severity. For subgroup analyses, we dichotomized the participants into two distinct categories: an asymptomatic group (total NPI score = 0; designated as NPS(−)) and a symptomatic group (total NPI score ≥ 1; designated as NPS(+)).

Demographic and Clinical Variables

Age, sex, and years of formal education were obtained as demographic data. The presence of medical comorbidities, including hypertension, diabetes mellitus, and dyslipidemia, was established by corroborating physician-assigned International Classification of Diseases, 10th Revision codes with current electronic prescription logs (online suppl. Table S1; for all online suppl. material, see https://doi.org/10.1159/000553583).

Concomitant medications were systematically categorized according to the World Health Organization Anatomical Therapeutic Chemical classification system (online suppl. Table S2). Because of their well-documented influence on behavioral symptoms, our analysis placed particular emphasis on psychotropic drugs (antipsychotic, antidepressant, and antianxiety agents) and cognitive enhancers (memantine and the cholinesterase inhibitors donepezil, rivastigmine, and galantamine) [3, 4].

Cognitive function was evaluated using the Montreal Cognitive Assessment (MoCA), a 30-point screening tool on which higher scores indicate superior cognitive performance [19]. Additionally, dementia severity was staged from 1 to 7 using the Global Deterioration Scale (GDS), with higher scores indicating greater clinical severity [20].

Statistical Analysis

Baseline characteristics were summarized using descriptive statistics and evaluated across the two social network status (SNS(−) versus SNS(+)). Continuous variables are presented as means with standard deviations or medians with interquartile ranges (IQRs), depending on normality assessed via the Shapiro-Wilk test. To evaluate group-level differences, we applied Pearson’s χ² tests for categorical variables and independent t-tests or Kruskal-Wallis tests for continuous measures, as dictated by the data distribution.

To identify factors associated with baseline NPI scores, we conducted univariable and multivariable linear regression analyses. For the multivariable models, we used stepwise selection to evaluate independent associations between NPI scores and predictor variables of demographics, medical comorbidities, concomitant medications, cognitive status, neuropsychological test scores, and social networks.

For longitudinal analyses, linear mixed-effects models (LMMs) were used to appropriately evaluate repeated-measures data with unbalanced time points and missing follow-up data while accounting for within-subject correlation [21]. Under missing-at-random assumptions, LMMs provide valid parameter estimates without requiring imputation because they use all available data from each participant. The dependent variable was the total NPI score at each visit. After adjusting for covariates significantly associated with baseline NPS in the cross-sectional analyses, the fixed effects included time since baseline, social network group, and their interaction to estimate group differences in symptom change rates. Random intercepts were used to account for participants’ baseline variability. Additionally, subgroup analyses were performed to stratify the participants by: (1) baseline NPS status: NPI(−) versus NPI(+); and (2) cognitive status: CN, mild cognitive impairment, and dementia. The parameters were estimated using restricted maximum likelihood.

Statistical significance was defined as a two-sided p < 0.05. All analyses were performed using R version 4.3.0 (R Foundation for Statistical Computing, Vienna, Austria) with the lme4 package [22].

Results

Table 1 presents the baseline characteristics stratified according to social network status. Of the 618 participants, 133 (21.5%) and 485 (78.5%) had poor and adequate social networks. Those groups did not differ significantly in terms of median age, sex distribution, years of education, or prevalence of hypertension, diabetes, or dyslipidemia.

Table 1.

Baseline demographic and clinical characteristics of the study participants stratified by social network status

Variable	Total (N = 618)	SNS(−) (N = 133)	SNS(+) (N = 485)	p value
Demographics
Age, year	71.00 [63.00–77.00]	70.00 [62.00–77.00]	71.00 [63.00–77.00]	0.354
Sex, female, n (%)	335 (54.2)	64 (48.1)	271 (55.9)	0.136
Education, year	7.00 [5.00–12.00]	8.00 [5.00–12.00]	7.00 [5.00–12.00]	0.755
Medical comorbidities, n (%)
Hypertension	279 (45.1)	55 (41.4)	224 (46.2)	0.371
Diabetes	113 (18.3)	25 (18.8)	88 (18.1)	0.963
Dyslipidemia	312 (50.5)	63 (47.4)	249 (51.3)	0.475
Concomitant medications, n (%)
Antithrombotic agents	280 (45.3)	55 (41.4)	225 (46.4)	0.349
Cholinesterase inhibitors	187 (30.3)	51 (38.3)	136 (28.0)	0.029
Memantine	35 (5.7)	9 (6.8)	26 (5.4)	0.682
Antipsychotic agents	53 (8.6)	14 (10.5)	39 (8.0)	0.464
Antidepressant agents	213 (34.5)	53 (39.8)	160 (33.0)	0.170
Antianxiety agents	205 (33.2)	51 (38.3)	154 (31.8)	0.185
Antiparkinson agents	224 (36.2)	48 (36.1)	176 (36.3)	1.000
Cognitive status, n (%)	​	​	​	0.067
CN	138 (22.3)	25 (18.8)	113 (23.3)	​
MCI	306 (49.5)	60 (45.1)	246 (50.7)	​
Dementia	174 (28.2)	48 (36.1)	126 (26.0)	​
Neuropsychological tests
MoCA	18.00 [12.00–22.00]	18.00 [11.00–22.00]	18.00 [12.00–22.00]	0.713
GDS	3.00 [3.00–4.00]	3.00 [3.00–4.00]	3.00 [3.00–4.00]	0.009
NPI	2.00 [0.00–8.00]	3.00 [0.00–12.00]	2.00 [0.00–7.00]	0.007

Data are presented as the median (interquartile range) for continuous variables, and as number (percentage) for categorical variables.

CN, cognitively normal; MCI, mild cognitive impairment; MoCA, Montreal Cognitive Assessment; GDS, Global Deterioration Scale; NPI, neuropsychiatric inventory; SNS(−), poor social network; SNS(+), adequate social network.

Notably, a slight but statistically significant baseline imbalance was observed in clinical severity; participants in the SNS(−) group exhibited higher GDS scores compared to those in the SNS(+) group (p = 0.009). The SNS(−) group had a higher proportion of people receiving cholinesterase inhibitors (p = 0.029), but no significant differences were observed in other medications, including memantine, antipsychotics, antidepressants, antianxiety, and antiparkinson agents. Although the distribution of cognitive status did not differ between groups (p = 0.067), participants in the SNS(−) group had significantly higher GDS scores (median 3.00, IQR: 3.00–4.00) compared to those in the SNS(+) group (median 3.00, IQR: 3.00–4.00), with a numerically higher proportion having dementia (36.1% vs. 26.0%).

Associations with Baseline NPS

In the univariable analysis (Table 2), several variables were significantly associated with baseline NPI scores: cholinesterase inhibitor use, memantine use, antipsychotic use, antidepressant use, antianxiety use, dementia, MoCA scores, GDS scores, and social networks. In the multivariable analysis (Table 2), memantine use, antidepressant use, antianxiety use, dementia, and social networks remained independently associated with NPI scores. The negative association between NPI scores and SNS(+) indicates that better social networks were associated with a lower NPS burden at baseline, independent of other factors.

Table 2.

Univariable and multivariable linear regression analyses for factors associated with baseline Neuropsychiatric Inventory scores

Variable	Univariable β (95% CI)	p value	Multivariable β (95% CI)	p value
Demographics
Age	0.004 (−0.071 to 0.079)	0.921	​	​
Sex, female	−1.100 (−2.483 to 0.283)	0.119	​	​
Education, year	−0.025 (−0.172 to 0.122)	0.742	​	​
Medical comorbidities
Hypertension	−1.029 (−2.414 to 0.356)	0.145	​	​
Diabetes mellitus	−0.336 (−2.109 to 1.436)	0.710	​	​
Dyslipidemia	−1.137 (−2.516 to 0.241)	0.106	​	​
Concomitant medications
Antiparkinson drug	0.661 (−0.774 to 2.096)	0.366	​	​
Antithrombotics agents	−1.349 (−2.732 to 0.034)	0.056	​	​
Cholinesterase inhibitors	2.959 (1.469 to 4.449)	<0.001	​	​
Memantine	8.437 (5.513 to 11.360)	<0.001	4.347 (1.439 to 7.254)	0.003
Antipsychotic drug	4.236 (1.794 to 6.678)	0.001	​	​
Antidepressant drug	4.126 (2.707 to 5.544)	<0.001	3.009 (1.647 to 4.371)	<0.001
Antianxiety drug	3.007 (1.563 to 4.452)	<0.001	2.237 (0.872 to 3.603)	0.001
Cognitive status
CN	1.000	​	​	​
MCI	−0.239 (−1.904 to 1.426)	0.778	​	​
Dementia	6.222 (4.366 to 8.079)	<0.001	5.216 (3.714 to 6.717)	<0.001
Neuropsychological tests
MoCA	−0.225 (−0.326 to 0.124)	<0.001	​	​
GDS	2.702 (2.063 to 3.342)	<0.001	​	​
Social networks
SNS(+)	−2.572 (−4.247 to −0.898)	0.003	−1.716 (−3.257 to 0.175)	0.029

CN, cognitively normal; MCI, mild cognitive impairment; MoCA, Montreal Cognitive Assessment; GDS, Global Deterioration Scale; SNS, social network status.

Longitudinal NPS Trajectories

All participants completed at least one follow-up visit. The median follow-up period for the entire cohort was 22.0 months (IQR: 11.0–37.0 months), with a total of 1,505 follow-up observations (median 2.0 visits per participant, IQR: 2.0–2.0). Specifically, the median follow-up duration was 20.0 months (IQR: 11.0–36.0 months) for the SNS(−) group and 22.0 months (IQR: 11.0–37.0 months) for the SNS(+) group (p = 0.208). SNS(+) was significantly associated with a lower NPS burden (β = −2.156, p = 0.008) (Table 3). However, the interaction between social networks and time was not statistically significant (p > 0.05), indicating that social network status did not significantly modify the rate of NPS change over time in the overall study population (Table 3).

Table 3.

Longitudinal associations between social network status and NPI scores, as shown by linear mixed-effects models

Fixed effects	Estimate (95% CI)	p value
Memantine use	2.607 (−0.061 to 5.276)	0.055
Antidepressant use	−1.507 (−2.827 to 0.188)	0.025
Antianxiety use	−0.574 (−1.913 to 0.766)	0.401
Dementia	2.441 (1.199 to 3.682)	<0.001
Time (months)	0.003 (−0.046 to 0.052)	0.904
SNS(+)	−2.156 (−3.747 to 0.565)	0.008
Time × SNS(+)	−0.016 (−0.072 to 0.041)	0.584

CI, confidence interval; LMM, linear mixed-effects model; NPI, neuropsychiatric inventory; SNS, social network status.

Subgroup Analysis by Baseline NPS Status

To assess whether social network effects varied by baseline NPS, we stratified the participants by baseline NPS status (Table 4). Among the 618 participants, 287 (46.4%) were NPS(−), and 331 (53.6%) were NPS(+).

Table 4.

Longitudinal effects of social network status on total NPI trajectories, stratified by baseline NPS status

Stratum/Fixed effects	Estimate (95% CI)	p value
Baseline NPS(−)
Time (months)	0.153 (0.095 to 0.211)	<0.001
SNS(+)	−0.842 (−2.322 to 0.639)	0.265
Time × SNS(+)	−0.124 (−0.186 to 0.062)	<0.001
Baseline NPS(+)
Time (months)	−0.041 (−0.167 to 0.084)	0.514
SNS(+)	−4.670 (−9.546 to 0.207)	0.060
Time × SNS(+)	0.064 (−0.102 to 0.230)	0.446

LMM, linear mixed-effects model; NPI, neuropsychiatric inventory; NPS, neuropsychiatric symptoms; SNS, social network status.

As illustrated in Figure 1 and detailed in Table 4, in the NPS(−) subgroup, a significant time × social network interaction was identified (β = −0.124, p < 0.001). Participants with SNS(−) demonstrated a significant increase in NPI scores over time (β = 0.153 points per month, p < 0.001), whereas those with SNS(+) showed a substantially attenuated rate of symptom emergence (0.029 points per month). This represents an 81.0% [≒ (0.153–0.029)÷0.153] decrease in the rate of NPS development in the SNS(+) group. Notably, there was no significant difference in baseline NPI scores between the social network groups in this subgroup (β = −0.842, p = 0.265), suggesting that the observed longitudinal difference reflects a true protective association, rather than baseline confounding.

Fig. 1.

a, b Longitudinal trajectories of total Neuropsychiatric Inventory scores stratified by baseline neuropsychiatric symptoms status. Being socially connected is associated with fewer behavioral issues in people with cognitive decline. People with cognitive decline often experience changes in mood or behavior, such as anxiety, irritability, or depression. Healthcare professionals call these changes neuropsychiatric symptoms. They can be upsetting for both patients and their families and may contribute to an earlier need for institutional care. In this study, we examined whether staying socially connected with family members and friends was associated with the development of these symptoms. We found that people who had regular contact with others tended to have fewer emotional and behavioral problems. Among those who did not have neuropsychiatric symptoms at the beginning of the study, symptoms appeared more slowly in individuals who remained socially engaged than in those who had limited social contact. Because the total number of new symptoms recorded during the observation period was relatively low, extended research is needed to fully verify these patterns over time. Overall, the findings indicate that maintaining active social relationships is linked to a reduced occurrence of behavioral challenges in individuals experiencing cognitive decline.

In contrast, within the NPS(+) subgroup, the interaction between time and social networks was not statistically significant (p > 0.05), although a significant main effect of social networks was observed (β = −4.670, p = 0.060). These findings suggest that although the SNS(+) is associated with a low overall symptom burden, it does not significantly alter the longitudinal trajectory of NPS once symptoms have manifested.

Subgroup Analysis by Cognitive Status

Subgroup analyses across cognitive status groups revealed no statistically significant time × social network interactions in any cognitive group (all, p > 0.05) (online suppl. Table S3).

Discussion

This study examined the relationship between social networks and NPS. Adequate social networks were cross-sectionally associated with a lower NPS burden. The longitudinal effect of adequate social networks was evident only in individuals without baseline NPS. These findings suggest that adequate social networks might attenuate symptom emergence, but they do not substantially alter the trajectory of established symptoms once they have manifested.

In this study, the cross-sectional observation that adequate social networks are associated with a lower NPS burden aligns with previous studies that link social factors to NPS [10, 23, 24]. Previous research indicates that greater social connectedness, including marriage, correlates with reduced NPS severity [23]. Furthermore, loneliness, the subjective counterpart to objective social isolation, has been identified as a robust predictor of specific neuropsychiatric manifestations, particularly delusions, hallucinations, and depression [24]. Notably, participants with poorer social networks exhibited greater baseline clinical severity, including higher GDS scores and a higher prevalence of dementia. Nevertheless, the associations between social network characteristics and NPS remained significant after adjustment for these factors, suggesting that the observed relationship is not merely a reflection of underlying disease severity.

This study showed that the differential effect of social networks depends on baseline NPS. Among baseline NPS(−), an adequate social network was significantly associated with a slower rate of symptom emergence over time. Adequate social networks may be a protective factor against the development of NPS in vulnerable populations. Conversely, for baseline NPS(+), social network status did not significantly affect longitudinal symptom trajectories.

Several hypotheses may explain these divergent outcomes. The protective effect observed in the NPS(−) group may reflect the benefits of active social engagement, which provides cognitive stimulation, emotional support, and behavioral regulation, thereby buffering against the development of NPS [7, 25]. Conversely, poor social networks intensify neuropsychiatric vulnerability to psychological stress by removing structured daily routines and reducing opportunities for meaningful activities [26, 27]. This concept is strongly corroborated by real-world data from the COVID-19 pandemic, when enforced social isolation precipitated both the emergence and rapid worsening of NPS [26–28].

However, once NPS have clinically manifested, as seen in the NPS(+) in this study, they are likely driven by progressive underlying neurodegenerative processes that are not particularly amenable to psychosocial modification [4, 29]. Established NPS arise from complex interactions involving advanced structural changes in the brain and profound neurobiological disruptions [4, 30], and these powerful pathological forces may overwhelm an adequate social network’s protective capacity. Consequently, this study indirectly suggests that adequate social networks alone are insufficient for patients with pre-existing NPS; they require a multimodal management approach that seamlessly integrates both pharmacological and non-pharmacological strategies.

For cognitive status, this study shows an association between a dementia diagnosis and steeper NPI trajectories, which is consistent with the established paradigm that NPS severity escalates alongside cognitive decline [4, 30]. Interestingly, however, when the participants were stratified by cognitive status, we found no significant time-by-social-network interactions. Thus, the protective association of an adequate social network with NPS emergence is primarily contingent on baseline neuropsychiatric status, rather than the specific stage of cognitive dysfunction.

Although some previous reports suggested that the influence of social networks might fluctuate depending on dementia severity [7, 9, 10], our heterogeneous cohort and relatively short follow-up duration might have constrained the statistical power required to detect stage-specific nuances. Furthermore, because the prevalence of specific NPS domains varies significantly across different dementia etiologies [4], future studies must investigate whether the protective benefits of social networks are domain- or etiology-specific.

The influence of concurrent pharmacological treatments warrants careful interpretation. Cross-sectionally, the association of memantine, antidepressants, and antianxiety agents with higher baseline NPI scores likely reflects confounding by indication, as these medications are typically prescribed to patients already presenting with NPS [3, 4, 31]. Longitudinally, the negative association between antidepressant use and NPI trajectories suggests a potential therapeutic benefit, aligning with meta-analyses supporting their efficacy in managing behavioral symptoms [4]. Nevertheless, these findings must be interpreted with caution given the observational design and potential for residual confounding.

This study has several limitations. First, the clinical magnitude of the interaction effect (β = −0.124) requires careful contextualization. Given that the minimal clinically important difference for total NPI scores is estimated to be 8 points [32]. Given the observed monthly difference in symptom progression rates between the poor and adequate social network groups (approximately 0.124 points per month), this divergence would take approximately 5 years to reach the minimal clinically important difference. While larger cohorts with extended follow-up are required to confirm long-term accumulation, from a preventive neurology perspective, attenuating the emergence of highly distressing behavioral symptoms by even a few years represents a crucial clinical window to preserve functional independence and postpone caregiver burden. Second, the single-center, memory clinic-based design introduces a clear selection bias. The participants represent individuals with pre-existing cognitive concerns, which limits the extrapolation of our findings to the general population. Baseline imbalances in GDS scores and cholinesterase inhibitor prescription rates suggest a risk of residual confounding or reverse causality; we cannot entirely rule out that poorer social networks are a secondary consequence of advanced cognitive decline and early behavioral withdrawal rather than a pure baseline predictor. Additionally, our reliance on a stepwise selection method carries an inherent risk of overfitting. Third, our social network index captured only structural components, neglecting qualitative dimensions (e.g., relationship satisfaction, emotional support) known to mediate behavioral trajectories. Furthermore, evaluating only the total NPI score may obscure domain-specific variations across dementia etiologies, which should be investigated in future prospective work. Fourth, the missing-at-random assumption could not be formally verified due to electronic medical record-based data limitations regarding immediate dropouts, leaving a possibility of non-random attrition bias. Lastly, our cognitive subgroup analyses may have been underpowered to detect subtle stage-specific nuances, particularly within the small CN subgroup (n = 138).

In conclusion, structural social networks are independently associated with baseline NPS burden and longitudinal trajectories, particularly through attenuating symptom emergence in initially asymptomatic individuals. In practice, addressing social networks requires integrating brief social screenings into routine dementia care. Clinicians can proactively identify social vulnerabilities and refer isolated patients to local senior activities or structured memory day-care programs. Combining these early, targeted psychosocial interventions with multimodal therapies for symptomatic patients provides a realistic non-pharmacological strategy to alleviate long-term neuropsychiatric burdens.

Acknowledgments

We thank all the participants in this study.

Statement of Ethics

The study protocol received approval from the Institutional Review Board at Korea University Ansan Hospital (IRB No. 2025AS0172). Because of the retrospective design of the data collection, the requirement for written informed consent was waived. All procedures adhered to the ethical principles of the Declaration of Helsinki.

Conflict of Interest Statement

The authors have no conflicts of interest to declare.

Funding Sources

This study was not supported by any sponsor or funder.

Author Contributions

Guarantor of integrity of the entire study, study concepts and design, manuscript editing, and data analysis: Hyo-Jeong Lee and Moon-Ho Park.

Funding Statement

This study was not supported by any sponsor or funder.

Data Availability Statement

The data that support the findings of this study are not publicly available due to privacy reasons but are available from the corresponding author upon request.