Cluster-derived profiles of suicidal ideation and quality of life among older adults living alone in South Korea

Abstract

Older adults living alone are often regarded as a vulnerable group exposed to psychosocial risks, yet they are not a homogeneous population. This study aimed to identify distinct clusters of psychosocial well-being within this group and examine key determinants using the World Health Organization Quality of Life (WHOQOL) framework and data from the Intensive Managed Care for Elderly Living Alone (IMCELA) program. A non-hierarchical cluster analysis was conducted based on scores from the Suicidal Ideation Scale (SIS) and the Older People’s Quality of Life questionnaire (OPQOL), resulting in three clusters: a baseline cluster with low SIS and low OPQOL, a High Suicidal Ideation cluster, and a High OPQOL cluster. Multinomial logistic regression revealed that higher meal frequency and lower depression scores significantly reduced the likelihood of belonging to the High Suicidal Ideation cluster. Meanwhile, high OPQOL was shaped by a broader set of factors, with perceived social support as the strongest predictor, followed by cognitive functioning, loneliness, and depressive symptoms. These findings suggest that while clinical risk factors such as depression and poor nutrition are key in addressing suicidal ideation, improving quality of life requires comprehensive strategies that enhance emotional and social resources. The results offer practical guidance for developing targeted interventions and policies to support the psychosocial well-being of older adults living alone.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-34036-w.

Introduction

Older adults living alone are generally regarded as a vulnerable population in terms of physical, psychological, and social aspects¹. Existing literature often discusses this group as a social issue, linking their living arrangements to the negative consequences of aging—such as a decline in quality of life, deteriorating mental health, and increased consumption of social resources^2–4. Such framing commonly reflects a monolithic perspective that classifies a heterogeneous group of older adults living alone into a single vulnerable category. Such a view may contribute to their stigmatization, framing them not only as a group placing a burden on social resources, but also as passive recipients of support rather than active agents in their own lives⁵.

However, not all older adults living alone form a uniformly vulnerable group; rather, they exist along a broad and varied spectrum. Some older adults are enjoying a high quality of life, drawing on emotional depth, spiritual insight, and the wisdom gained from life experience, and actively contributing to society as mentors^6–8. This differentiated perspective allows for a more precise and dynamic assessment of quality of life among older adults. Moreover, it can help diversify the direction of prevention and health promotion services and policies targeting the entire population of older adults living alone. In other words, by identifying high-risk groups and allocating appropriate medical services to them, while also recognizing and promoting factors that may enhance well-being, programs can be implemented more effectively even with limited resources.

In this regard, South Korea is one of the countries that urgently require attention in light of this paradigm shift. It has been experiencing an unprecedentedly rapid demographic transition toward an aging society⁹ with adults aged 65 and older accounting for 19.2% of the population in 2024¹⁰. Due to population aging combined with gender disparities in life expectancy, the proportion of single-person households has been rapidly increasing in South Korea¹⁰. In response, the proportion of the national budget al.located to elderly welfare has risen to 25% of total social welfare spending¹¹, and along with this increase, funding specifically targeting older adults living alone has also grown accordingly¹². At first glance, the increased budget share may appear positive, as it reflects growing social interest. However, in the absence of clearly defined target groups and policy objectives, support programs risk becoming fragmented and inefficient, ultimately leading to suboptimal use of limited resources.

Meanwhile, as noted earlier, older adults living alone represent a heterogeneous group, encompassing a broad spectrum. This means that individuals with both low and high quality of life coexist within this population. Numerous studies have shown that the quality of life in older adults is influenced by a range of factors, including personal attributes, social participation, and environmental conditions. Depending on whether these factors are facilitated or constrained, quality of life may either improve or deteriorate^13–15. However, although there are various scales for measuring quality of life, widely used instruments such as the Short From-36 (SF-36) and EuroQol 5-Dimension (EQ-5D) were originally developed as clinical tools focusing on health status. As such, they have limitations in fully capturing the multidimensional nature of overall quality of life in older adults^16–18. In addition, the assumption that older adults in poor health inherently have a low quality of life may underestimate their potential for a positive and fulfilling life¹⁹.

Quality of life is a comprehensive concept that encompasses both objective and subjective aspects of individuals²⁰. The Older People’s Quality of Life (OPQOL) scale, developed by Bowling²¹, is well-suited to this comprehensive nature, as it measures nine major domains: social relationships, family and neighbors, psychological well-being, health, leisure and activities, religion and culture, economic situation, independence, and overall life satisfaction. Notably, it offers the advantage of more accurately assessing the quality of life in older age by incorporating experiences and environmental factors unique to older adults—elements that are often overlooked in conventional scales.

However, the OPQOL scale does not include suicidal ideation, which may reflect the lowest level of quality of life. While earlier studies have typically treated suicidal ideation as a predictor or proxy preceding a suicide attempt²², more recent research suggests that it can represent quality of life in and of itself. For example, Jung and Jang²³, using an ecological model, identified socioeconomic status, residential type, physical health, health behaviors, and mental health as influential factors associated with suicidal ideation in older adults. Additionally, Jung and Park²⁴ argued that both household and regional-level variables can significantly influence individual-level suicidal ideation. These studies are meaningful in that they examined suicidal ideation from multidimensional perspectives. However, they focused on the overall older adult population rather than those living alone and had limitations in that they also considered those elderlies living alone as a homogeneous group.

In response to the findings and limitations of previous research, this study was designed to classify the psychosocial well-being of older adults living alone through a non-hierarchical cluster analysis based on two axes: quality of life and suicidal ideation. The central objective is to identify heterogeneous clusters within this population based on psychosocial well-being, which is a novel approach that, to the best of our knowledge, has not yet been applied in this context. By uncovering distinct cluster types, the study explores the factors influencing each cluster and suggests policy directions for both the prevention of suicidal ideation and the promotion of quality of life. Given the prevailing social perception that older adults living alone tend to have a low quality of life, the group characterized by low quality of life and minimal suicidal ideation was designated as the baseline cluster. This approach starts from the perspective of a typical older adult living alone—one with a low quality of life but without suicidal ideation—and enables an understanding of the different factors that lead either to elevated suicidal ideation or to improved well-being.

The study hypothesizes that a range of factors, including socioeconomic status, physical health, health behaviors, mental health, social networks, and social support, affect psychosocial well-being, in line with the WHOQOL framework. However, it is also hypothesized that the influence of these factors may differ depending on the specific cluster type.

Study data and methods

Research data

The data for this study were derived from the IMCELA program, which has been operated by the Dementia Care Center (DCC) in Chungcheongnam-do since 2015. The program recruited older adults living alone who were receiving Basic Care Services from the Cheonan Senior Welfare Center and were simultaneously registered for case management at the DCC. Individuals who met these criteria were informed about the IMCELA program, and those who expressed willingness to participate were further reviewed. Final eligibility and enrollment were determined by an internal assessment committee composed of the DCC chief psychiatrist and three researchers, based on clinical and psychosocial needs. Data were collected from participants who voluntarily enrolled and provided informed consent for the use of their information in research between July 2015 and March 2018. Of the 1,427 recipients of Basic Care Services, 1,099 individuals enrolled in the IMCELA program. For the present analysis, we used data from 1,060 participants, excluding 27 participants with missing Severe Cognitive Impairment Rating Scale (SCIRS) scores and 12 participants with missing loneliness scores (complete-case analysis). A comparison of included and excluded participants on age, gender, and insurance status showed no statistically significant differences, suggesting minimal risk of selection bias due to missingness. Because the IMCELA program was delivered through community-based care services for socially vulnerable older adults living alone, the study population includes a comparatively higher proportion of women and medical-aid recipients. This contextual characteristic should be taken into account when interpreting the generalizability of the findings. This study was reviewed and approved by the Institutional Review Board of Dankook University based on the Helsinki Declaration, which granted a waiver of written informed consent (Approval Number: 1041849-202207-SB-123-01).

Model framework

This study employed the WHOQOL framework, which includes several dimensions such as socioeconomic factors, physical health, health behavior, mental health, and social relationships to explain psychosocial well-being. In the case of social relationships, we separated them into two dimensions—social networking and social support—in accordance with Berkman and Glass²⁵. Social networking, which reflects the structural and quantitative aspects of social relationships, can be directly measured through variables such as years of living alone, time spent with family per month, and time spent in social gatherings per month. Meanwhile, social support represents the functional and qualitative aspects of relationships, including emotional support, informational support, tangible support, and affective support.

Dependent variable

Psychosocial well-being was categorized into three clusters based on non-hierarchical (K-means) cluster analysis of the SIS²⁶ and OPQOL²¹ scores. Prior to clustering, both variables were standardized using z-scores, and the optimal number of clusters (k = 3) was determined using the elbow method. The resulting clusters were: (1) Low SIS and Low OPQOL (hereafter, Baseline cluster) (n = 389, 36.7%), (2) High Suicidal Ideation and Low OPQOL (hereafter, High Suicidal Ideation cluster) (n = 138, 13.0%), and (3) High OPQOL and Low SIS (hereafter, High OPQOL cluster) (n = 533, 50.3%). These labels are used descriptively to summarize observed score patterns rather than as diagnostic classifications. The Baseline cluster was designated as the reference category because it represented the largest and most typical score pattern in the data, making it the most statistically stable comparison cluster. This reflects a statistical rather than conceptual choice. To assess the stability of the clustering solution, we reran the K-means algorithm with multiple random starts and observed no changes in cluster membership patterns, indicating acceptable internal stability. Full-scale questionnaires and detailed clustering procedures are provided in the Supplementary Materials.

Independent variables

Socioeconomic factors

This study considered gender, age, years of education, and National Health Insurance (NHI) status as socioeconomic factors associated with psychosocial well-being. Among these, NHI status was employed as a proxy for income level. In Korea, the NHI system classifies individuals as either NHI-insured or medical aid recipients, offering a practical means of distinguishing economic status. Among adults aged 65 and older, reported income may be unreliable due to limited or irregular employment. Therefore, the dataset used in this study did not include income information. Given that medical aid recipients—representing the most economically disadvantaged group—are distinctly identified within the NHI system, this classification has been regarded as an appropriate and effective proxy for economic status in relevant study²⁷. Furthermore, years of education can serve as another proxy for income. The strong correlation between education and income has been well established²⁸, and many previous studies have utilized education as a single representative indicator of socioeconomic status in place of income²⁹. Accordingly, in this study, years of education were considered an important socioeconomic status factor alongside NHI status, reflecting both individual economic resources and social positioning.

Physical health

The total score of the Cumulative Illness Rating Scale (CIRS), originally developed by Linn et al.³⁰, was used to represent physical health in this study. The CIRS is one of the most widely used tools for assessing comorbidity, alongside the Charlson Comorbidity Index (CCI) (ref). Unlike the CCI, which assigns weighted scores based on the presence of specific diseases, the CIRS-G (Geriatric version) registers all coexisting conditions and rates the severity of each across major organ systems using a 5-point scale (0–4). This system allows for a more nuanced assessment of the overall illness burden. Comparative studies have shown that the CIRS-G is more sensitive than the CCI and may provide superior prognostic information^31,32. The validity of the CIRS-G has been supported by numerous studies across various populations, particularly among older adults, including geriatric outpatients and acutely hospitalized elderly patients^33–37.

Health behavior

Many theorists have suggested that unhealthy behaviors, such as smoking and alcohol consumption, may represent a form of suicidality rooted in self-destructive tendencies³⁸. At the same time, health-promoting behaviors—including regular exercise, a balanced diet, smoking cessation, and reduced alcohol intake—have been shown to be positively associated with quality of life³⁹. To capture health behavior in this study, we included four variables: meal frequency, weekly exercise time, alcohol consumption, and lifetime smoking exposure. Among these, meal patterns have been considered a significant factor influencing mental health in older adults, particularly in Korea⁴⁰. Although dietary patterns such as meal frequency were not included in previous OPQOL studies—likely due to the predominance of Western-based research cohorts—they have been recognized as important determinants of mental health and quality of life in other studies involving older populations^41,42.

Mental health

For measuring mental health, we utilized three scales: the De Jong-Gierveld Loneliness Scale⁴³, the SCIRS, and the Short Form of the Geriatric Depression Scale (SGDS). Although De Jong-Gierveld and Van Tilburg published the manual for the full 11-item Loneliness Scale in 1999 they also developed a shorter 6-item version to improve feasibility in large-scale surveys⁴⁴. The short form was validated in multiple studies and has been widely used across diverse older populations^45,46. This study employed the 6-item De Jong-Gierveld Loneliness Scale (hereafter, the loneliness scale), which includes three items each for emotional and social loneliness.

The SCIRS is a clinician-rated instrument developed by Rabins and colleagues to assess cognitive functioning in institutionalized elderly populations⁴⁷. It was designed to complement objective cognitive screening tools by evaluating functional cognitive impairment across several domains, such as memory, orientation, comprehension, and executive function. The SCIRS has demonstrated good inter-rater reliability and clinical validity in geriatric settings, and its structured format allows for a nuanced assessment based on clinical judgment and observational data. It consists of 15 items scored from 0 to 3, yielding a total score ranging from 0 to 15, with higher scores indicating more severe cognitive impairment and lower scores reflecting better cognitive functioning. For analytic consistency across variables, the SCIRS scores were reverse-coded in this study, such that higher values represent better cognitive functioning.

The SGDS is a 15-item version derived from the original 30-item Geriatric Depression Scale, which was developed by Yesavage and colleagues⁴⁸. The SGDS was created to provide a quicker yet reliable alternative for screening depressive symptoms in older adults. Numerous validation studies have confirmed its sensitivity and specificity in both community and clinical populations⁴⁹. In this study, the SGDS was used to assess depressive symptoms, with higher scores indicating more severe depressive mood.

Social networking

For measuring social networking, three variables were included: years of living alone, monthly family gathering time, and monthly social gathering time. These indicators reflect the quantity and frequency of interpersonal interactions, which have been shown to be closely associated with both suicidal ideation and quality of life in older adults⁵⁰.

Social support

To assess social support, this study utilized overall scores from the Medical Outcomes Study Social Support Survey (MOS-SSS). Developed by Sherbourne and Stewart⁴⁹, the MOS-SSS is a widely validated instrument that measures perceived availability of social support across multiple functional domains, including emotional/informational, tangible, affectionate, and positive social interaction^51,52. It consists of 19 items, each rated on a 5-point Likert scale ranging from “never” to “always.” Total scores range from 19 to 95, with higher scores indicating greater perceived social support (see Supplementary Material).

Statistical analysis

Descriptive statistics were first conducted to examine the characteristics of the study population. Analysis of variance (ANOVA) and chi-square tests were performed to compare differences across psychosocial well-being clusters. For multivariate analysis, multinomial logistic regression (MLR) was conducted using the mlogit command in STATA program to evaluate factors associated with psychosocial well-being. All predictors were entered simultaneously to allow for examination of the unique contribution of each variable. Compared to conducting multiple binary logistic regressions, MLR offers the advantage of simultaneously modeling all outcome categories within a unified framework. Meanwhile, continuous predictors (e.g., exercise time and social gathering hours) were modeled as linear terms, and potential nonlinear relationships were not tested. Prior to estimation, multicollinearity was assessed using variance inflation factors (VIF), and no issues were detected (see Supplementary Material). This allows for consistent adjustment of covariates, reduces the risk of Type I error due to multiple comparisons. All analyses were performed using Stata/MP version 18.0 (StataCorp LP, TX, USA).

Study results

Descriptive statistics of sample based on social determinants of health

Table 1 presents the descriptive statistics of the sample, reflecting the characteristics of elderly participants in the IMCELA program. Regarding socioeconomic factors, females comprised 87.83% of the total sample, and 22.08% were medical aid recipients. The average age was 79.02 years (range: 64–98), and the average years of education was 3.22 years (range: 0–18). In terms of physical health, the mean total CIRS-G score was 5.41, ranging from 0 to 19. With respect to social networking, participants had lived alone for an average of 21 years (range: 0.08–66 years). The average time spent on family gatherings per month was 7.46 h, while time spent on social gatherings averaged 36.48 h. Regarding health behaviors, the average meal frequency was 1.75 meals per day. Participants exercised an average of 155.67 min per week, with a range of 0 to 2,100 min. The average weekly alcohol consumption was 1.34 standard drinks. Lifetime smoking exposure averaged 4.47 packs of cigarettes, with a range of 0 to 200 packs. In terms of mental health, the mean (range) scores for total loneliness, SCIRS, and SGDS were 3.8 (0–6), 28.53 (9–30), and 6.58 (0–15), respectively. Regarding social support, the average overall social support score was 0.47, ranging from 0 to 1.

Table 1.

Descriptive statistics of samples.

Individual characteristics			N(%)	Mean(S.D.)	Range
Socioeconomic	Gender	Male	129 (12.17)
		Female	931 (87.83)
	NHI status	NHI	826 (77.92)
		Medical aid	234 (22.08)
	Age			79.02 (6.00)	64–98
	Years of education			3.22 (3.97)	0–18
Physical health	Total score of CIRS-G			5.41 (3.02)	0–19
Health behavior	Meal frequency			1.75 (0.46)	0–4
	Weekly exercise time			155.67 (243.49)	0–2100
	Weekly alcohol consumption			1.34 (12.88)	0–63
	Lifetime smoking exposure			4.47 (15.47)	0–200
Mental health	Total loneliness score			3.8 (1.68)	0–6
	Total SCIRS score			28.53 (2.08)	9–30
	Total SGDS score			6.58 (4.31)	0–15
Social networking	Years of living alone			21.00 (14.31)	0.08–66
	Monthly family gathering time			7.46 (16.82)	0–240
	Monthly social gathering time			36.48 (51.34)	0–270
Social support	Total MOS-SSS score			0.47 (0.24)	0–1
	Total		1,060 (100.0)

Note: NHI: National Health Insurance; CIRS-G: Cumulative Illness Rating Scale-Geriatric version; SCIRS: Severe Cognitive Impairment Rating Scale; SGDS: Short form of the Geriatric Depression Scale; MOS-SSS: Medical Outcomes Study Social Support Survey.

Associated factors with psychosocial well-being clusters

Table 2 presents the factors associated with psychosocial well-being clusters among elderly participants. Significant differences were observed across the three clusters—Baseline, Suicidal Ideation, and High-QPQOL—in various socioeconomic, physical health, social networking, health behavior, mental health, and social support factors. For socioeconomic factors, NHI status was significantly associated with psychosocial well-being (p < 0.001). Medical aid recipients were more likely to be included in the High Suicidal Ideation cluster and less likely to be in the High-QPQOL cluster than NHI subscribers. Years of education also significantly differed among the clusters (p = 0.038), with participants in the High-QPQOL cluster having the highest mean years of education (3.57 ± 4.17) compared to the other clusters. Regarding physical health, the Suicidal Ideation cluster had the highest total CIRS scores (6.4 ± 5.3) compared to the other clusters (p = 0.003). With respect to health behaviors, meal frequency significantly varied across clusters (p = 0.016), with the High Suicidal Ideation cluster reporting the lowest frequency (1.56 ± 0.51) compared to the Baseline (1.76 ± 0.46) and High-QPQOL clusters (1.8 ± 0.42). Weekly exercise time also differed significantly (p = 0.016), with the High-QPQOL cluster reporting the highest average exercise time (177.8 ± 257.7 min).

Table 2.

Factors associated with cluster membership.

			Psychosocial Well-being Clusters
Independent Variables			Baseline	High Suicidal Ideation	High OPQOL	P-value
Socioeconomic	Gender	Male	345 (37.06)	116 (12.46)	470 (50.48)	0.338
		Female	44 (34.11)	22 (17.05)	63 (48.84)
	NHI status	NHI	291(35.23)	94 (11.38)	441 (53.39)	< 0.001
		Medical aid	98 (41.88)	44 (18.8)	92 (39.32)
	Age		79.36 (6.15)	78.0 (6.34)	79.04 (5.77)	0.236
	Years of education		2.83 (3.75)	2.86 (3.7)	3.57 (4.17)	0.038
Physical health	Total score of CIRS		5.4(2.8)	6.4 (3.53)	5.16 (2.99)	0.003
Health behavior	Meal frequency		1.76 (0.46)	1.56 (0.51)	1.8 (0.42)	0.016
	Weekly exercise time		131.4 (218.7)	138.58(246.6)	177.8 (257.7)	0.003
	Weekly alcohol consumption		1.05 (5.62)	1.69 (7.56)	1.46 (17.1)	< 0.001
	Lifetime smoking exposure		4.9 (14.7)	5.96 (14.49)	3.77 (16.24)	0.06
Mental health	Total loneliness score		4.46 (1.34)	4.97 (1.03)	3.00 (1.66)	< 0.001
	Total SCIRS score		28.12 (2.38)	28.28 (2.78)	28.88 (1.48)	< 0.001
	Total SGDS score		8.41 (3.61)	11.39 (2.92)	4.00 (3.2)	0.004
Social networking	Years of living alone		21.74 (14.41)	23.8 (15.69)	19.74 (13.7)	0.119
	Monthly family gathering time		5.72 (17.73)	4.83 (9.67)	9.41 (17.37)	< 0.001
	Monthly social gathering time		28.79 (45.79)	23.83 (41.49)	45.38 (55.7)	< 0.001
Social support	Total MOS-SSS score		0.37 (0.2)	0.34 (0.23)	0.57 (0.23)	0.017

Note: NHI: National Health Insurance; CIRS-G: Cumulative Illness Rating Scale-Geriatric version; SCIRS: Severe Cognitive Impairment Rating Scale; SGDS: Short form of the Geriatric Depression Scale; MOS-SSS: Medical Outcomes Study Social Support Survey; Continuous variables are presented as mean (SD) and analyzed using ANOVA; Categorical variables are presented as n (%) and tested using the Chi-square test; Other categorical variables are presented by n (%) and tested by Chi-square test.

Additionally, lifetime smoking exposure was highest in the High Suicidal Ideation cluster (5.96 ± 14.49 packs) and lowest in the High-QPQOL cluster (3.77 ± 12.46 packs) (p = 0.012). In terms of mental health, significant differences were found in total loneliness scores (p = 0.007), SCIRS scores (p = 0.048), and SGDS scores (p = 0.002). The High Suicidal Ideation cluster reported the highest levels of loneliness (4.97 ± 1.03) and depressive symptoms (SGDS: 11.39 ± 2.92). In terms of social networking, participants in the High-QPQOL cluster spent the most time on monthly family gatherings (9.41 ± 17.37 h), while those in the High Suicidal Ideation cluster spent the least time (4.83 ± 9.67 h), compared to the Baseline cluster (5.72 ± 17.73 h) (p = 0.006). Similarly, time spent on monthly social gatherings was highest in the High-QPQOL cluster (45.38 ± 55.76 h) and lowest in the High Suicidal Ideation cluster (23.83 ± 41.49 h) (p < 0.001). Lastly, regarding social support, the High-QPQOL cluster had the highest overall social support score (0.5 ± 0.3), compared to the other clusters. (p = 0.046).

Effect of associated factors on psychosocial Well-Being using MLR

We conducted a MLR analysis to identify factors associated with the psychosocial well-being clusters. The MLR results are presented as two separate models comparing each cluster to the baseline: (1) High Suicidal Ideation cluster vs. baseline, and (2) High OPQOL cluster vs. baseline. Before presenting the specific results, it is worth noting that the model demonstrated moderate explanatory power, with a pseudo-R² value of 0.311. This suggests that the model explains a meaningful portion of the variation in psychosocial well-being, while substantial unexplained variance remains.

MLR results on the high suicidal ideation cluster

Table 3 presents the results of the MLR, which estimated the factors associated with the relative likelihood of belonging to the High Suicidal Ideation cluster in comparison to the baseline cluster. Only two variables were statistically significantly associated with this likelihood. Meal frequency showed an odds ratio of 0.608 (95% CI = 0.400–0.923), indicating indicating an association whereby lower meal frequency was associated with a greater likelihood of belonging to the High Suicidal Ideation cluster. In contrast, the total SGDS score showed an odds ratio of 1.29 (95% CI = 1.194–1.393), suggesting that higher depression scores were associated with a greater likelihood of being in the High Suicidal Ideation cluster.

Table 3.

MLR results for the high suicidal ideation cluster compared to the baseline Cluster.

	Variables	Adjusted OR	95% CI		P-value
Socioeconomic	Gender (Female)	1.546 (0.559)	0.761	3.142	0.229
	NHI status (NHI)	1.059 (0.279)	0.632	1.776	0.827
	Age	0.981 (0.019)	0.944	1.020	0.337
	Years of education	0.975 (0.315)	0.915	1.039	0.439
Physical health	Total score of CIRS	1.038 (0.038)	0.966	1.114	0.311
Health behavior	Meal frequency	0.608 (0.130)	0.400	0.923	0.020
	Weekly exercise time	1.005 (0.003)	0.999	1.011	0.081
	Weekly alcohol consumption	1.009 (0.012)	0.986	1.032	0.457
	Lifetime smoking exposure	0.619 (0.360)	0.198	1.934	0.409
Mental health	Total loneliness score	1.188 (0.129)	0.961	1.469	0.112
	Total SCIRS score	1.018 (0.049)	0.926	1.118	0.717
	Total SGDS score	1.290 (0.051)	1.194	1.393	< 0.001
Social networking	Years of living alone	1.011 (0.008)	0.996	1.026	0.162
	Monthly family gathering time	1.031 (0.041)	0.954	1.113	0.443
	Monthly social gathering time	0.997 (0.049)	0.906	1.098	0.957
Social support	Total MOS-SSS score	1.355 (0.785)	0.435	4.215	0.600
Constant		0.037 (0.091)	0.000	4.686	0.182

Note: NHI: National Health Insurance; CIRS-G: Cumulative Illness Rating Scale-Geriatric version; SCIRS: Severe Cognitive Impairment Rating Scale; SGDS: Short form of the Geriatric Depression Scale; MOS-SSS: Medical Outcomes Study Social Support Survey; =0.311.

MLR results on the high OPQOL cluster

In contrast to the High Suicidal Ideation cluster, the High OPQOL cluster showed associations across all dimensions except for physical health, although not all factors within each dimension were consistently significant. In terms of socioeconomic factors, older adults who were male and living alone were significantly more likely to be included in the High OPQOL cluster (adjusted OR = 1.954, 95% CI = 1.078–3.543), and longer years of education were associated with a higher likelihood of belonging to this cluster (adjusted OR = 1.049, 95% CI = 1.001–1.100). With respect to mental health, all variables showed statistical significance. Higher loneliness scores were associated with a lower likelihood of membership in the High OPQOL cluster (adjusted OR = 0.844, 95% CI = 0.738–0.965). Older adults living alone with higher SCIRS scores were more likely to be classified into the High OPQOL cluster–a noteworthy association, given that prior literature has primarily linked cognitive decline to reduced quality of life. Meanwhile, higher SGDS scores were associated with a lower likelihood of belonging to the High OPQOL cluster (adjusted OR = 0.741, 95% CI = 0.702–0.782). Regarding social networking, only time spent in social gatherings was significantly associated with the outcome (adjusted OR = 1.106, 95% CI = 1.028–1.190). For health behavior, higher meal frequency was associated with a lower likelihood of belonging to the High OPQOL cluster. Lastly, in terms of social support, the MOS-SSS score showed the strongest association, with the highest adjusted odds ratio (OR = 13.793, 95% CI = 5.562–34.203).

Discussions

This study aimed to identify distinct clusters representing psychosocial well-being among older adults living alone, using non-hierarchical cluster analysis based on SIS and OPQOL scores. Building on this, it sought to examine the determinants of psychosocial well-being within the WHOQOL framework. More specifically, psychosocial well-being was categorized into three clusters: (1) Low SIS and Low OPQOL, which was used as the reference group; (2) the High Suicidal Ideation cluster; and (3) the High OPQOL cluster. The optimal number of clusters was determined using the elbow method.

The data sample largely reflects the characteristics of older adults living alone participating in the IMCELA program (Table 1). Due to gender-based disparities in life expectancy, the majority of older adults living alone are female, accounting for 87.83% of the sample. Furthermore, the proportion of medical aid recipients was 22.08%, which is substantially higher than the national average of 2.9% in South Korea⁵³. This pattern may reflect the high elderly poverty rate in South Korea, reported to be 43.2%⁵⁴. Moreover, the average years of education among participants was only 3.22 out of a possible 18 years, which may be related to limited educational infrastructure available during their youth, particularly in the post-Korean War period⁵⁵. In addition, the mean meal frequency was considerably lower than the traditional three-meal pattern, which is consistent with findings from previous studies⁵⁶. Furthermore, the amount of time spent on monthly family gatherings was five times shorter than that spent on social gatherings, which may reflect the rise of nuclear families and weakening intergenerational ties^57,58. Meanwhile, the duration of living alone ranged widely, from 0.08 to 66 years.

When it comes to the MLR results for psychosocial well-being, two separate models were presented (Tables 3 and 4). The first model estimated factors associated with the likelihood of belonging to the High Suicidal Ideation cluster, with the Low SIS and Low OPQOL cluster serving as the reference group. As previously noted, MLR simultaneously models all outcome categories and covariates within a unified framework, enabling consistent covariate adjustment and direct comparisons across outcome categories. In this regard, the first model showed that only meal frequency and total SGDS scores were statistically significant.

Table 4.

MLR results for the high OPQOL cluster compared to baseline Cluster.

	Variables	Adjusted OR	95% CI		P-value
Socioeconomic	Gender (Female)	1.954 (0.593)	1.078	3.543	0.027
	NHI status (NHI)	0.730 (0.158)	0.477	1.116	0.146
	Age	1.000 (0.016)	0.970	1.031	0.983
	Years of education	1.049 (0.025)	1.001	1.100	0.047
Physical health	Total score of CIRS-G	1.029 (0.032)	0.969	1.094	0.352
Health behavior	Meal frequency	0.614 (0.122)	0.416	0.906	0.014
	Weekly exercise time	1.003 (0.002)	0.998	1.007	0.202
	Weekly alcohol consumption	0.993 (0.008)	0.978	1.009	0.416
	Lifetime smoking exposure	0.665 (0.290)	0.283	1.564	0.349
Mental health	Total loneliness score	0.844 (0.058)	0.738	0.965	0.013
	Total SCIRS score	1.190 (0.060)	1.078	1.312	< 0.001
	Total SGDS score	0.741 (0.021)	0.702	0.782	< 0.001
Social networking	Years of living alone	0.997 (0.006)	0.985	1.009	0.625
	Monthly family gathering time	0.999 (0.027)	0.949	1.053	0.985
	Monthly social gathering time	1.106 (0.041)	1.028	1.190	0.007
Social support	Total MOS-SSS score	13.793 (6.391)	5.562	34.203	< 0.001
Constant		0.044 (0.093)	0.001	2.702	0.137

Note: NHI: National Health Insurance; CIRS-G: Cumulative Illness Rating Scale-Geriatric version; SCIRS: Severe Cognitive Impairment Rating Scale; SGDS: Short form of the Geriatric Depression Scale; MOS-SSS: Medical Outcomes Study Social Support Survey; =0.311.

Specifically, higher meal frequency was associated with a lower likelihood of belonging to the High Suicidal Ideation cluster. Existing literature on the eating behaviors of older adults has primarily focused on the biological and physiological changes that occur in later life, and how these changes affect food intake⁵⁹. A large body of research has also identified a significant association between psychosocial factors and eating behaviors^60–63, although most studies have emphasized the effect of eating behaviors on mental health⁵⁹. For example, Kwak and Kim⁴⁰reported that older adults who skipped breakfast had greater odds of experiencing depression and suicidal ideation compared to those who ate breakfast. Stroebele-Benschop et al.⁶⁴, argued that living with others could provide opportunities for social modeling during mealtimes, thereby increasing food intake. Based on these findings, encouraging older adults to eat with others, for example through community-based congregate meal programs (part of the Older Americans Act Nutrition Program)⁶⁵, may support improvements in psychosocial well-being.

With respect to other significant findings from the first model, a higher SGDS score was associated with an increased likelihood of belonging to the High Suicidal Ideation cluster. This association is well-documented in vast of literature, including several systematic reviews that require a substantial accumulation of evidence^66–69. These studies commonly reported that the deterioration of physical health—such as frailty, falls, agitation, fatigue, increased risk of disease, and limitations in activities of daily living—contributes to depression and may even lead to suicidal ideation in older adults⁷⁰. From a cognitive perspective, depression often leads to increased hopelessness, self-blame, and feelings of worthlessness, all of which can distort one’s perception of future prospects and personal value⁷¹. Beck’s cognitive theory highlights that such negative automatic thoughts and dysfunctional beliefs contribute significantly to the emergence of suicidal ideation⁷². Furthermore, Shneidman’s theory conceptualizes suicide as an escape from unbearable psychological pain, which is often exacerbated by the emotional distress inherent in depressive episodes⁷³. In addition, depression impairs cognitive flexibility and problem-solving ability, limiting the individual’s capacity to perceive alternative coping strategies⁷⁴.

The second model estimates the factors associated with the likelihood of membership in the High OPQOL cluster, relative to the reference group. Compared to the first model, more significant factors were identified across six dimensions. Among socioeconomic variables, males were more likely to belong to the High OPQOL cluster. This aligns with Ko et al.⁷⁵, which found that older men living alone tended to report better quality of life, supported by stronger physical and cognitive functioning, than women. However, Ko et al.⁷⁵ also noted that men were more vulnerable to mental health challenges, including emotional stress, loneliness, depressive symptoms, and suicidal ideation. These seemingly contradictory findings may reflect the heterogeneity within the male population, as highlighted in the present study. While some older men enjoy a high quality of life, others struggle with poor mental well-being and reduced life satisfaction. By employing a cluster analytic approach, this study identified such internal stratification within the male group. This methodological strategy helps reconcile inconsistencies in previous research that treated gender as a homogeneous category.

In the case of educational level, longer years of education were associated with a greater likelihood of belonging to the High OPQOL cluster. Indeed, the positive impact of education on quality of life has been well established. Edgerton et al.⁷⁶ highlighted its influence on various life domains, including achievement, material well-being/standard of living, emotional well-being/resilience, physical health, community engagement, intimate relationships, and personal safety/future security.

However, the effect of education has not been consistent among older adults living alone. For instance, while Kim and Lee⁷⁷ reported no significant association between educational attainment and Health-related quality of life (HRQoL) in this population, Yahaya⁷⁸, on the contrary, confirmed a positive influence of education level. This study supports the findings of Yahaya⁷⁸, contrasting with the results of Kim and Lee⁷⁷, which may be attributed to differences in the quality of life scales used—HRQoL and OPQOL. While HRQoL focuses solely on health-related factors, OPQOL encompasses broader societal aspects, including social relationships, community participation, physical and psychological health, and overall life satisfaction. In this regard, we argue that education may be more closely related to these broader dimensions of life rather than on physical health alone. As a source of cognitive resources—such as conceptual understanding, contextual awareness, and problem-solving ability—education may be associated with a greater capacity to navigate daily circumstances, which may be linked to higher quality of life.

Among the health behaviors, meal frequency was significantly associated with a lower likelihood of being included in the High OPQOL cluster. Taken together with the results of Model 1, increasing meal frequency may be linked to both reduced likelihood of belonging to the High Suicidal Ideation cluster and, paradoxically, the High OPQOL cluster. This finding suggests that the role of meal frequency may differ depending on the psychosocial profile of older adults. For individuals at higher risk of suicidal ideation, more frequent eating may be associated with better basic functioning and psychological stabilization. However, among those not at high risk, meal frequency may not directly translate into greater well-being, and in some cases, may even reflect routine rather than satisfaction. These results indicate that efforts to increase meal frequency may be relevant for suicide risk mitigation among high-risk groups, whereas alternative strategies may be needed to support life satisfaction in the broader population of older adults living alone. Such differentiated interpretation was made possible by the classification derived from the cluster analytic approach. Importantly, these associations should not be interpreted as causal. Rather, they reflect exploratory subgroup patterns that require validation in longitudinal or intervention-based studies.

Regarding mental health, all variables showed significant effects on the likelihood of belonging to the High OPQOL cluster. Firstly, as scores on the De Jong-Gierveld Loneliness score increased, the likelihood of membership in the High OPQOL cluster significantly decreased. The negative association between feelings of loneliness and quality of life has been well documented in previous studies^79–81. Secondly, a higher SCIRS score, indicating more robust cognitive functioning, was associated with an increased likelihood of belonging to the High OPQOL cluster. Although no prior study has directly examined the relationship between SCIRS and OPQOL, substantial evidence supports the association between cognitive impairment and quality of life in older adults⁸². Thirdly, a higher SGDS score was associated with a decreased likelihood of belonging to the High OPQOL cluster. This finding reinforces the well-established link between depressive symptoms and reduced quality of life in older adults. This association may be partly explained by emotional burden—such as persistent sadness, hopelessness, and social withdrawal—and functional decline, including reduced motivation and activity levels, both of which have been associated with lower subjective well-being^83,84.

Lastly, the MOS-SSS scores emerged as the most influential variable, showing the highest odds ratio among all predictors. The comparatively large effect size (OR = 13.793) and the limited number of significant variables may partly reflect the analytic structure of the study. Because cluster analysis groups individuals with similar psychosocial profiles together, the variability of predictors within each cluster becomes reduced. Consequently, in the multinomial logistic regression stage, only variables that clearly distinguish one cluster from another tend to remain significant, which can produce larger effect sizes for those differentiating variables. Even so, the finding is consistent with Schulz and Waldinger (2024), which draws on findings from the Harvard Study of Adult Development. They emphasize that emotionally supportive and trusting relationships are central to sustained happiness and well-being, a perspective that aligns with the present results. This suggests that it is not merely the frequency of social encounters but rather the quality of social support that may plays a more central role.

Notably, the association of MOS-SSS in Model 2 differed from that in Model 1. Although many previous studies have examined social support as a moderator in relation to suicidal ideation^85–88, the present study did not observe a statistically significant association between social support and suicidal ideation in the High Suicidal Ideation cluster. Given that individuals experiencing suicidal ideation often carry emotional distress or interpersonal strain, efforts to enhance social support may require careful and individualized approaches.

Consequently, this study suggests potential directions rather than definitive recommendations. Clinical strategies for suicide risk mitigation among older adults living alone may benefit from approaches that address both meal frequency and depressive symptoms. From a welfare policy perspective, community-based programs where older adults can dine together may help support nutritional intake and facilitate meaningful social interaction. Additionally, programs that aim to strengthen emotionally supportive and trusting relationships may be relevant for improving quality of life among older adults living alone.

However, it is important to note that the temporal direction of the associations identified in this study cannot be inferred. In particular, the relationship between depressive symptoms and suicidal ideation may be reciprocal, in that depressive mood can intensify suicidal thoughts, while persistent suicidal ideation may further exacerbate depressive affect. These bidirectional dynamics highlight the need for caution when interpreting the mechanisms implied in the present findings. Taken together, these interpretations should be considered tentative. Future longitudinal or intervention studies are required to further clarify these associations and to determine whether the patterns observed here hold over time and across different contexts.

This study is subject to a number of limitations. To begin with, the generalizability of the results is constrained by the fact that the sample was drawn exclusively from participants in the IMCELA program, which targets socially vulnerable older adults living alone. Consequently, the sample includes a comparatively higher proportion of women and medical-aid recipients than the general older adult population in Korea. This structural sampling characteristic should be taken into account when interpreting the results. Moreover, the small number of male participants in particular limits the statistical power of gender-specific comparisons and warrants caution in interpreting gender-related effects.

Additionally, participants with missing values on key variables were excluded. Although this procedure did not substantially affect the basic demographic characteristics of the analytic sample, the possibility of bias due to complete-case analysis cannot be fully ruled out. Moreover, because this study focused on a particularly vulnerable group of older adults receiving Basic Care Services, reliable income data were not available, and thus the income effect could not be assessed in the present analysis. It is reasonable to expect that, if the sample were expanded to include the broader population of older adults living alone, income would play a more substantial role. Future research should therefore widen the sampling frame and examine the impact of income more explicitly.

Furthermore, the cross-sectional nature of the study design prevents any inference about causal relationships and does not allow for control over endogeneity concerns such as potential reverse causality. Lastly, multinomial logistic regression relies on the independence-of-irrelevant-alternatives assumption, and this assumption was not formally tested in the present analysis; therefore, the findings should be interpreted with appropriate caution. Future research should aim to include more diverse and representative samples to enhance generalizability. Longitudinal or experimental study designs would also be beneficial in establishing causal pathways and better addressing endogeneity issues.

Conclusions

This study identified key factors influencing psychosocial well-being among older adults living alone. Depression and meal frequency were significantly associated with suicidal ideation, while perceived social support was the strongest predictor of high quality of life. These findings highlight the importance of not only managing mental health and nutrition but also fostering meaningful social connections. Although the study’s scope was limited by its cross-sectional design and localized sample, it offers valuable insights for future research and practical interventions targeting this vulnerable population.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Hyun Woo Jung: Writing-original draft, formal analysis; Do Yeon Kim: data curation; Ilju Lee: formal analysis; OK Kim: Writing-review & editing; Seungjin Lee: data curation; Sujin Lee: Writing-review & editing; Un Sun Chung: Writing-review & editing; Jae-Hyun Kim: Conceptualization, Methodology, Writing-review & editing; Kwang-Soo Lee: Methodology, Writing-review & editing; Jung Jae Lee: Conceptualization, supervision, Investigation, funding acquisition.

Funding

This study was funded through a grant from the Korea Health Technology R&D Project, supported by the Korea Health Industry Development Institute (KHIDI) and the Ministry of Health & Welfare, Republic of Korea (Grant No. RS-2024-00438829).

Data availability

The authors will provide the raw data supporting the conclusions of this article upon reasonable request, with no unnecessary restrictions. Requests for the data should be directed to the corresponding author, Jung Jae Lee, at mdjjlee@dankook.ac.kr.

Declarations

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

The studies involving human participants were reviewed and approved by the Institutional Review Board of Dankook University (DKU 2021-05-038). The patients/participants provided their written informed consent to participate in this study.

Role of the funding source

The funding source had no role in the study design, data collection, analysis, interpretation of data, writing of the manuscript, or the decision to submit the manuscript for publication.