This cross-sectional study was conducted among rural older adults in Sichuan Province, China, using a multi-stage stratified convenience sampling approach. Data were collected from January to June 2020. Stratified convenience sampling was used to facilitate fieldwork while capturing socioeconomic diversity within the rural population. Stratification was based on county and per capita gross domestic product (GDP), as reported in the 2018 Sichuan Statistical Yearbook, to represent different socioeconomic subgroups. Sichuan was selected because it has a large, diverse rural population with substantial variation in geography, socioeconomic conditions, and health care access. Variation in local infrastructure and health care services also supports its suitability for examining factors related to self-neglect among older adults.

Participants were eligible if they (1) were aged 60 years or older, (2) had a clinical diagnosis of at least one chronic condition, including diabetes, hypertension, coronary heart disease, stroke, chronic obstructive pulmonary disease (COPD), or cancer, and (3) provided written informed consent after a full explanation of the study. Exclusion criteria were (1) withdrawal before study completion, (2) terminal illness, or (3) severe communication disorders that precluded participation.

According to Nylund-Gibson et al. (26), at least 300 participants are required to ensure stable parameter estimation in LPA. On the basis of this recommendation and the study design, the required sample size was calculated using the following formula:

In this calculation, Z = 1.96 for a 95% confidence interval (CI), the expected prevalence (p) was 27% (27), and the allowable margin of error (d) was 0.035. The minimum required sample size was 619. Allowing for an anticipated 10% non-response, the target sample size was increased to 688 participants.

Field surveys were conducted from January to June 2020 in six regions of Sichuan Province, China: Deyang, Chengdu, Luzhou, Neijiang, Dazhou, and the Aba Tibetan and Qiang Autonomous Prefecture. These sites were selected using a multi-stage, stratified convenience sampling approach based on per capita GDP levels reported in the 2018 Sichuan Statistical Yearbook. Investigators from medical universities received standardized training to ensure consistent questionnaire administration and reduce interviewer bias. With support from local village doctors and community staff, they carried out community-based household surveys. Before data collection, investigators explained the study objectives, procedures, and participants’ right to refuse participation or withdraw at any time without consequence. Surveys were administered only after written informed consent was obtained. Given limited literacy among many rural older adults, trained interviewers conducted face-to-face interviews, read each question aloud, and recorded responses on paper questionnaires, which were later entered into an electronic database. Interviews lasted approximately 20 to 30 min, depending on response pace and the need for clarification. After each interview, questionnaires were reviewed for completeness and internal consistency. Incomplete questionnaires were addressed on site by recontacting participants when needed, and questionnaires with substantial missing data, defined as more than 10% of items unanswered, were excluded from analysis. Owing to on-site review and clarification, missing data in the final dataset were minimal, at less than 5% across variables. Additional quality-control procedures included logical checks of responses and, when applicable, cross-verification with community staff. All data were anonymized and used solely for academic research.

Analyses were performed using IBM SPSS Statistics version 24.0 (IBM Corp., Armonk, NY, United States) and Mplus version 8.3 (Muthén and Muthén, Los Angeles, CA, USA). LPA was used to identify subgroups of ESN among older adults with chronic diseases, using the 14 SESN items as indicators. The dependent variable was latent profile membership. Model fit was assessed using the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), adjusted BIC (aBIC), entropy, the Lo–Mendell–Rubin adjusted likelihood ratio test (LMRT), and the bootstrap likelihood ratio test (BLRT). Lower AIC, BIC, and aBIC values indicated better fit. Entropy ranges from 0 to 1, with values closer to 1 indicating higher classification accuracy. A significant LMRT or BLRT result (p ≤ 0.05) supported a k-class model over a k−1 class model. The final number of classes also considered interpretability and adequate class sizes. Continuous variables are presented as medians with interquartile ranges (IQR), and categorical variables as frequencies and percentages. Between-class comparisons used chi-square (χ²) tests or Kruskal–Wallis H tests, as appropriate. Multinomial logistic regression was conducted to examine factors associated with profile membership. Predictors included age, sex, education level, living alone, widowed status, self-rated economic status, grandchild caregiving, frequency of communication with children, perceived loneliness, personality, physical examination status, mobile phone use, sedentary behavior, sleep quality, self-rated health, cognitive impairment, smoking, alcohol consumption, pain, comorbidity status, depression, and social support score. Variables significant in univariate analyses were entered into multivariable logistic regression models to identify predictors of latent profile membership. Statistical significance was set at a two-tailed p < 0.05.

The study protocol was reviewed and approved by the Ethics Committee of the Affiliated Hospital of Southwest Medical University (Approval No. KY2019274) and conducted in accordance with the Declaration of Helsinki. All participants provided written informed consent after a detailed explanation of the study objectives. Data were handled confidentially, and all identifying information was removed to protect participant privacy.

A total of 785 rural older adults with chronic diseases were invited to participate in this cross-sectional study. After exclusion of 66 questionnaires with incomplete or erroneous responses, 719 participants (91.6%) were included in the analysis. Participant sociodemographic characteristics were summarized in Supplementary Table 1. Most participants were female (n = 392, 54.5%), and most had a junior high school education or less (n = 624, 86.8%). The median age was 70.0 years (IQR, 66 to 76). The prevalence of chronic diseases was as follows: hypertension (n = 554, 77.1%), diabetes (n = 154, 21.4%), coronary heart disease (n = 119, 16.6%), COPD (n = 79, 11.0%), stroke (n = 68, 9.5%), and cancer (n = 36, 5.0%). Univariate analyses (Supplementary Table 2) indicated that ESN profile distribution differed by chronic disease type. COPD was most frequently observed in the moderate neglect profile (18.9%), whereas cancer was more frequently observed in the severe neglect profile (11.1%; both p < 0.05). ESN profiles did not differ significantly by hypertension, diabetes, coronary heart disease, or stroke (all p > 0.05).

The median total ESN score was 9 (5 to 13). Median dimension scores were 2 (1 to 4) for medical health and care, 2 (0 to 3) for environmental sanitation and personal hygiene, 2 (1 to 3) for mental health, 1 (0 to 3) for safety, and 1 (0 to 2) for social communication.

LPA was conducted to identify subgroups of ESN among rural older adults with chronic diseases. As shown in Table 1, model fit was evaluated using the AIC, BIC, aBIC, the entropy, the LMRT, and the BLRT. The four-class model showed the best fit, with the lowest AIC (20,622.627), BIC (20,956.811), and aBIC (20,725.015), and high entropy (0.962), indicating strong classification accuracy. The LMRT (p = 0.0057) and BLRT (p < 0.001) also favored the four-class model over the three-class solution. Accordingly, the four-class model was retained for subsequent analyses.

Profiles were labeled to reflect an overall severity gradient and domain-specific scoring patterns on the ESN scale, aligned with the Biopsychosocial Model (36) and the medical, hygiene, psychological, safety, and social domains assessed by the instrument (Figures 1, 2). Class 1, low-level neglect (n = 252, 35.0%), showed consistently low scores across all 14 items. The selective mild neglect profile (n = 271, 37.7%) had relatively low overall scores, with comparatively higher scores in the medical and hygiene domains, indicating a selective pattern. The moderate neglect profile (n = 106, 14.7%) showed higher scores across domains, with particularly elevated scores in the safety, psychological, and medical domains relative to the selective mild neglect profile. The severe neglect profile (n = 90, 12.5%) had the highest scores across all domains. Classification quality was further examined using average latent class probabilities for most likely class membership (Supplementary Table 3). Diagonal probabilities were high (Class 1 to Class 4: 1.000, 0.916, 0.980, and 1.000), and off-diagonal probabilities were low.

Median (IQR) total ESN scores differed across profiles: 5 (2 to 7) for low-level neglect, 8 (6 to 11) for selective mild neglect, 16 (14 to 19) for moderate neglect, and 21 (14 to 25) for severe neglect (H = 422.741, p < 0.001). The four profiles also differed in demographic, behavioral, and psychosocial characteristics, including age, living arrangement, marital status, self-rated economic status, grandchild caregiving, frequency of communication with children, perceived loneliness, personality traits, physical examination status, mobile phone use, sedentary behavior, sleep quality, self-rated health, cognitive impairment, smoking, alcohol use, pain, depression, and social support (all p < 0.05; Supplementary Table 1). Participants in the moderate neglect group were older, with a median age of 74 years compared with 69 years in the low-level neglect group (p = 0.001). The proportion living alone increased across profiles, from 11.9% in the low-level neglect group to 30.0% in the severe neglect group (p < 0.001). Psychosocial vulnerability also increased, with perceived loneliness rising from 25.4% in the low-level group to 64.4% in the severe neglect group and depressive symptoms increasing from 13.5 to 54.4% (both p < 0.001). Poor self-rated economic status was more common in higher-risk profiles, reported by 37.8% of the severe neglect group versus 9.5% of the low-level group. Social support differed by profile (p < 0.001), with lower median scores in the moderate and severe neglect groups (both 37) than in the low-level neglect group.

No multicollinearity was detected among the independent variables (all tolerance values > 0.60; all VIFs < 2.0). The low-level neglect group (Class 1) was used as the reference category, and results are shown in Table 2. Compared with the low-level neglect group, participants in the moderate neglect group (Class 3) were older (OR = 1.046, p = 0.025) and had lower social support scores (OR = 0.924, p < 0.001). For family interaction, communicating with children 1–3 times a week (OR = 3.382, p = 0.011) or once a month (OR = 2.068, p = 0.016) was associated with higher odds of Class 3 membership than communicating less than once per month, whereas frequent communication was not significant. Annual physical examinations were inversely associated with moderate neglect relative to no regular examinations (OR = 0.431, p = 0.047). Absence of cognitive impairment, non-smoking, and absence of pain were each associated with lower odds of Class 3 membership (all p < 0.05), consistent with a comparatively healthier profile among those remaining in the low-level neglect group. Compared with the low-level neglect group, the severe neglect group (Class 4) showed a distinct psychosocial and behavioral pattern. Fair self-rated economic status was associated with lower odds of severe neglect relative to poor economic status (OR = 0.466, p = 0.047). Not providing grandchild caregiving was associated with higher odds of Class 4 membership (OR = 2.147, p = 0.012). Absence of perceived loneliness (OR = 0.472, p = 0.031), non-smoking (OR = 0.424, p = 0.013), and absence of depressive symptoms (OR = 0.340, p = 0.002) were each associated with lower odds of severe neglect.

Additional pairwise comparisons of adjacent profiles (Class 3 vs. Class 2, Class 4 vs. Class 2, and Class 4 vs. Class 3) identified stage-dependent associations for several factors (Supplementary Table 4). Social support reduced the odds of moderate neglect relative to selective mild neglect (OR = 0.929, p < 0.001) but was associated with higher odds of severe neglect relative to moderate neglect (OR = 1.060, p = 0.021), indicating a reversal at higher severity. Absence of pain was associated with lower odds of moderate neglect relative to selective mild neglect (OR = 0.474, p = 0.018) yet higher odds of severe neglect relative to moderate neglect (OR = 2.811, p = 0.009), consistent with a non-linear pattern.

Guided by the Risk and Vulnerability Model of self-neglect (37), interaction terms were added to the multivariable multinomial logistic regression models to assess whether internal vulnerabilities and external risk factors jointly influenced ESN latent profile membership. Five interactions were examined: living alone × cognitive impairment, social support × depression, age × pain, age × social support score, and age × self-rated health. Among these, only the age × pain interaction reached statistical significance. Relative to the low-level neglect group (Class 1), the age × pain interaction was significant for selective mild neglect (OR = 1.063, p = 0.043) and was stronger for moderate neglect (OR = 1.124, p = 0.016). These findings indicate that the relationship between physical pain and ESN profile membership varies by age. No other interaction terms were significant across group comparisons (all p > 0.05). Detailed results are provided in Supplementary Table 5.

LPA identified four subtypes of ESN among rural Chinese older adults with chronic diseases, supporting the heterogeneity of ESN. The four profiles, low-level (35.0%), selective mild (37.7%), moderate (14.7%), and severe (12.5%), differed in overall severity and in domain-specific patterns across the five SESN domains (medical health and care, environmental sanitation and personal hygiene, mental health, safety, and social communication; Figures 1, 2). The low-level neglect profile showed consistently low scores across domains, suggesting strong engagement in self-care and relatively intact biopsychosocial resources. Individuals in this group may have sufficient agency to meet daily and health-related demands. In the context of Orem’s Self-Care Deficit Nursing Theory (38), this profile may represent a domain-specific self-care deficit in which therapeutic self-care demands related to chronic illness, such as medication management, diet, and hygiene routines, exceed self-care agency in selected areas, while other domains remain preserved or compensated. The moderate neglect profile showed broader elevations, particularly in the safety, psychological, and medical domains, consistent with progressive accumulation of deficits in the presence of older age, pain, cognitive impairment, and lower social support. The severe neglect profile showed the highest scores across all domains, indicating pervasive disruption of self-care systems and the greatest health vulnerability.

This typology aligns in part with the work of Burnett et al. (25), who applied latent class analysis to U.S. APS data and identified four subtypes: predominant physical and medical neglect (48%), environmental neglect (22%), global or multidimensional neglect (21%), and financial neglect (9%). Both studies indicate substantial heterogeneity in ESN and a large subgroup with prominent physical and medical impairments. The present findings also add context-specific detail. The selective mild profile emphasizes discrete difficulties in chronic disease self-management, including medical compliance and hygiene, rather than financial neglect or severe environmental problems described in APS-referred samples. These differences may reflect the community-based rural Chinese sample with diagnosed chronic conditions, where disease-related demands and limited access to care may heighten specific vulnerabilities. Together, the profiles suggest that ESN follows distinct patterns, offering a context-relevant classification to inform targeted interventions.

The heterogeneous ESN profiles identified in this study suggest a tiered, community-based approach in rural China. Class 4 represents the highest-risk group, with pervasive deficits and strong associations with depressive symptoms, loneliness, smoking, economic hardship, and lack of grandchild caregiving roles. Class 3 represents the second-highest risk group and shows broader impairments, particularly in the safety, psychological, and medical domains; it was associated with older age, pain, cognitive impairment, lower social support, and infrequent physical examinations. Prioritizing these high-risk subgroups may support more efficient targeting in resource-constrained rural settings.

Selective mild neglect may serve as an early focus, with attention to modifiable lifestyle and resource vulnerabilities. Village-level coordination could include strengthened financial protection for chronic disease management and targeted health education addressing sleep problems and alcohol use. For moderate neglect, township health centers could implement routine monitoring through home visits or telephone follow-ups that include basic health checks, pain assessment, brief cognitive screening, and chronic disease self-management support. For severe neglect, a multidisciplinary approach may be considered, with emphasis on scalable psychosocial support and loneliness reduction, along with caregiver support and respite services where available. Screening for depressive symptoms, loneliness, and cognitive impairment could be incorporated into routine chronic disease management to identify older adults more likely to be classified in the severe ESN profile, while noting that longitudinal studies are needed to confirm transitions between profiles.

At the policy level, these findings may inform national rural health initiatives, including the National Basic Public Health Service Program (NBPHSP) and the Healthy China 2030 strategy, by incorporating ESN risk stratification and profile-informed service packages. Directing services toward depressed and lonely individuals, those with cognitive impairment or pain, smokers, economically disadvantaged older adults, and those with limited engagement in family roles may support more targeted allocation within constrained rural health infrastructure.

This study has several limitations. First, the cross-sectional design captures associations at a single time point and does not allow causal inference. Second, self-reported questionnaires may be subject to recall or social desirability bias. In addition, although the Chinese versions of the UCLALS-3, GDS-5, and SPMSQ have been widely used in older Chinese adults with acceptable reliability, including in the current sample, psychometric validation specifically in rural Chinese older populations is limited. Lower literacy and differences in cultural expression of emotional distress may affect measurement accuracy for loneliness, depression, and cognitive impairment. Third, the study did not include several environmental and systemic factors, including family caregiving intensity, access to community health resources, and medical insurance coverage. In rural China, these factors influence individuals’ ability to manage health and seek care, and their omission may limit the completeness of the identified profiles and their predictors. Data collection occurred from January to June 2020, during the early stage of the COVID-19 pandemic; although study areas were not high-risk regions, the pandemic may have influenced health care seeking, social support, and psychological status. Finally, convenience sampling within one rural province and the focus on six chronic diseases may limit generalizability to other regions or broader disease groups.

Future studies should use longitudinal designs to examine transitions between ESN profiles over time, clarify temporal relationships, and identify time points for earlier intervention. Qualitative interviews with individuals in each profile and their caregivers could provide insight into lived experience, barriers to self-care, and contextual factors related to heterogeneity. Randomized controlled trials tailored to each profile, including economic and lifestyle support for selective mild neglect, regular monitoring and health literacy programs for moderate neglect, and intensive psychosocial interventions for severe neglect, are needed to evaluate feasibility and efficacy in rural settings. Incorporating biological markers, such as inflammatory indicators and cortisol levels, or functional measures may further clarify mechanisms related to profile stability.