Data Sources and Study Population
The data utilized for this study were derived from the China Health and Retirement Longitudinal Study (CHARLS) 2015. CHARLS adopts Probability Proportional to Size sampling, and aims to collect a high-quality nationally representative sample of Chinese residents aged 45 and older to serve the needs of scientific research on older adults. Detailed information about the CHARLS design strategy has been described elsewhere, the ethical approval of data collection was from the Biomedical Ethics Review Committee of Pecking University (IRB00001052-11015) . Variable screening and data cleaning were performed according to the needs of the study, and after removing the non-elderly sample and the sample with missing key variables, 4190 older adults ³60 years old were included as the study population for analysis. Among them, 1962 (46.8%) were male and 2228 (53.2%) were female; 2724 (65.0%) were aged 60-69 years, 1211 (28.9%) were aged 70-79 years, and 255 (6.1%) were aged ≥80 years; 672 (16.0%) without spouse, 3518 (84.0%) with spouse; 997 people (23.8%) had no formal education, 1995 people (47.6%) had graduated from elementary school, 745 people (17.8%) had graduated from middle school, 390 people (9.3%) had graduated from high school(including vocational school), 63 people (1.5%) had graduated from college and above; 3006 people (71.7%) had agricultural household registration, 1184 people (28.3%) had non-agricultural household registration.
In this study, six indicators were selected from the CHARLS data for identifying latent classes of multidimensional health in older adults, including self-reported health status, the number of chronic diseases, activity of daily living (ADL), depressive symptoms, cognitive ability, and social activities participation. Among them, self-reported health status, number of chronic diseases and social activities participation were self-reported. Self-assessed health status was evaluated using 5 categories (1=Very good, 2=Good, 3=Fair, 4=Poor, 5=Very poor). The number of chronic diseases was integrated into a multi categorical variable based on the number of chronic diseases self-reported by respondents, with 1 indicating no chronic diseases, 2 indicating having 1 chronic disease, and 3 indicating having 2 or more chronic diseases. The notion of social activities participation was operationalized by asking the question "Have you done any of these activities in the last month?" with 0 indicating no social activities participation and 1 indicating having social activities participation.
ADLs were assessed with the ADL scale. The scale includes 6 items such as bathing, eating, getting in or out of bed, dressing, using the toilet, and controlling urination and defecation. The response scale contained 4 options: 1) No, I don’t have any difficulty; 2) I have difficulty but can still do it; 3) Yes, I have difficulty and need help; and 4) I cannot do it. The respondents were classified as ADL independent if they do not need help with all items, whereas those who needed help with at least one item were regarded as ADL impairment.
Depressive symptoms were assessed using the 10-item Center for Epidemiological Studies Depression Scale (CES-D-10) . The survey respondents were asked about their feelings and behaviors over the past week. According to the level of depressive symptoms reflected by the items, they were assigned a score of 0–3 points, giving a total score of 0–30 points. The higher the score, the more severe the depressive symptoms. Generally, a cutoff score ≥ 10 is considered to indicate the respondent having depressive symptoms .
Cognitive ability was measured using the brief Community Screening Instrument for Dementia (CSI-D) . The scale consists of two parts: the interviewee part (7 items with a score of 0-9) and the informant part (6 items with a score of 0-6). The scale score is the interviewee part score minus the informant part score and ranges from -6 to 9, with scores of 4 or less highly suggestive of dementia.
Demographic variables such as gender, age, marital status, education level, and household registration were included as control variables.
The Latent Class Analysis (LCA) method is used to deal with the classification of the multi-dimensional health groups of the elderly. As a person-centered approach, the goal of LCA is to reveal a minimal number of unobserved groups of individuals in terms of health latent classes to fully explain the associations between observed health dimensions, ultimately grouping individuals into categories, each containing individuals that are similar to each other but different from other categories . The Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC) ,and adjusted BIC (aBIC) are used as tests of model fit for latent class analysis, and a smaller index indicates a better model fit . Entropy values were used to assess the accuracy of the model classification, with Entropy values ranging from 0 to 1. Lubke and Muthén stated that Entropy < 0.60 corresponds to more than 20% of individuals with classification errors; Entropy = 0.80 indicates a classification accuracy of more than 90% . The difference in fit between k and k-1 profile models was compared using Lo–Mendell–Rubin adjusted likelihood ratio test (LMR) and the bootstrapped likelihood ratio test (BLRT), and the significant p-value(p<0.05) indicating that the k profile model significantly outperformed the k-1 profile model. Data were analyzed using Mplus8.3.
Then, analysis of variance (ANOVA) was performed to test whether there were differences in demographic characteristics across classes, and multivariable logistic regression was used to analyze the predictive effect of demographic characteristics variables on different health classes of older adults.