Health Lifestyles and Their Inuence on Chinese Oldest-old’s Health Outcomes —Evidence From a Latent Class Analysis

A strong association between individual health behaviors and health outcomes has been emphasized by previous analyses. However, how individual health behaviors can be classied into health lifestyles and the manner in which health lifestyles have impacted Chinese oldest-old’s health status are largely unknown. Four distinct classes representing health lifestyles emerged. Health lifestyles were found to be strongly associated with Chinese oldest-old’s health outcomes which were measured by self-rated health, functional independence, cognitive function and chronic diseases, even after controlling for demographic features as well as individual and parental socioeconomic disadvantage. Findings also showed a convergence of health disparities caused by demographic and SES characteristics in very old age.


Background
In recent years, researchers have begun to use clustered health lifestyles to explain the health disparities among individuals [1][2][3][4]. The bene t of this perspective is that it has extended the scopes of existing analyses on individual health behaviors to classi ed health lifestyles. Individual or single health behaviors that have been commonly used in prior analyses included poor dietary habits, cigarette smoking, excessive alcohol consumption et al. [5][6][7][8][9]. Scholars who promoted the health lifestyle approach argued that health behaviors tend to cluster in ways that re ect the social and structural contexts of individuals, which in turn affects individual health status [10]. This is because behaviors are not isolative, but co-occur with another [4]. Health lifestyle theories therefore contended that concentrating on single behaviors or small subsets of risky behaviors provides limited insight into health behavior patterns [11]. Thus, considering multiple behaviors simultaneously is a more appropriate strategy that creates larger and more enduring behavior change to improve individual health [12].
As far as studies on health status of Chinese elders, abundant analyses have documented a strong link between health behaviors and health outcomes among Chinese older adults [13][14][15][16]. Nevertheless, as studies focusing on other social contexts, most research on Chinese oldest-old also focused on single health behaviors. Since the sub-population of oldest-old is growing at extraordinary speed in China, it is important to explore potential factors that may improve the oldest-old's health status to alleviate the burden of the society as well as the family caregivers. Under such a proceeding, this research intended to take the health lifestyle approach, i.e., a combination of multiple health-related behaviors, to attain a better understanding of health-related practices and their relationship with Chinese oldest-old's health outcomes. Relying on latent class analysis strategy, the study used 2014 wave of the Chinese Longitudinal Health and Longevity Survey (CLHLS 2011), a nationally presentative data, to include health behaviors from multiple domains to present a relatively more comprehensive picture of health behaviors among Chinese oldest-old. It also aimed to elucidate how health lifestyles have shaped Chinese oldest-old's health outcomes. Findings based on analyzing nationally representative data in China are valuable to address disease prevention and health promotion related issues among the oldest-old in world countries. Exploring how clustered health behaviors in uence the oldestold's health outcomes can also expand theories explaining health disparities among elders in general.
The health lifestyle approach and prior literature The health lifestyle approach can be considered as a theoretical development in research of health disparities. The concept of health lifestyle was derived from Weber's idea of lifestyles as the interaction of life choices and life chances.
Weber reasoned that lifestyles are not associated with individuals but groups of people with similar social status and backgrounds. Such a de nition has been further expanded to include factors such as understandings of what good health means, health norms, policy environments et al. (Krueger et al., 2009). Bourdieu [17] further treated health lifestyles as broad and potentially unobservable orientations that organize patterns of behaviors. Health lifestyle perspectives emphasized more on patterns of behaviors rather than single behaviors. The perspectives highlighted social, cultural and economic forces on individual choices of health behaviors [10].
Some pioneer studies using the health lifestyle approach have been conducted to examine the general population.
These studies can be classi ed as the following groups: First, linking personal characteristics, such as gender and age, to individual health lifestyles [10,18]. Second, demonstrating a strong positive association between SES and clustered health behaviors among adults in different social contexts [19][20][21][22]. Third, exploring determinants of health lifestyle behaviors in adolescence and revealing how early age health lifestyle behaviors had imprints on one's health behaviors in adulthood [23,24]. Fourth, documenting signi cant in uence of health lifestyles on individual health outcomes, including mental health, self-rated health (SRH) and alike [2,25,26] and underlining the positive effects of health lifestyle behaviors on diseases prevention [27,28].
Health lifestyle approach has also been found useful in epidemiological studies examining health and mortality among older adults in a variety of countries. By operationalizing healthy lifestyle behaviors as physical activities, consumption of fruits and vegetables, and whether smoking, Martin-Maria and colleagues' [29] study showed signi cantly positive effect of healthy lifestyle behaviors on subjective well-being among Spain sample aged 65 and over. Through studying multiple lifestyle behaviors of older persons in Korea and Amsterdam, scholars highlighted that participation in healthy lifestyles contributed to the maintenance of functional independence (measured as ADL and IADL) and cognitive function in later life [30,31]. The study based on examining lifestyle behaviors including non-smoking and physical activity among elders in Sweden revealed that a low risk health behavior pro le could add ve years to women's lives and six years to men's after age 75 [32].
The above reviewed analyses have provided guidance to this current research investigating the link between health lifestyles to Chinese oldest-old's health outcomes. The selection of health lifestyles as well as health status measures was based on the commonly used measures in previous studies. The analysis answered two main questions: First, what are predominant health lifestyles of the Chinese oldest-old? Second, how have these main health lifestyles shaped Chinese oldest-old's health outcomes? Findings of this study were expected to ll the voids of prior literature by investigating Chinese oldest-old's health disparities from single health behaviors. Results based on analyzing the China data were also supposed to enrich health lifestyle theories on the whole. Below the paper moved to an introduction of data, measures and methods used in the study.

Data
Data came from the 2014 Chinese Longitudinal Healthy Longevity Survey (CLHLS) which was conducted in randomly selected half of the counties/cities in 22 provinces of China. Until now, 7 waves (1998, 2000, 2002, 2005, 2008, 2011-12, and 2014) of survey data have been collected. The survey was initially launched to meet the needs for scienti c research on the oldest-old. Thus, the dataset provided an excellent source for studying the oldest-old in China. Previous literature showed that persons who reported age 106 or higher were considered as invalid cases [33]. Therefore, persons aged 106 and higher were excluded from this study due to insu cient information to validate their reported extremely high age. The study eventually obtained 3,416 oldest-old aged 85 to 105, with 2,025 males and 1,391 females.

Health lifestyle indicators
Health lifestyles measures used in previous analyses can be classi ed as the following categories: (1) dietary patterns (including eating fruits, vegetables, breakfast et al.), (2) smoking, alcohol consumption, (3) sleep, (4) obesity and physical activity, (5) seat belt wearing and media use, (6) body mass index (BMI), and (7) regular physical examination [34][35][36][37][38][39][40][41]. The selection of health lifestyle indicators in this research has been largely guided by prior studies and four key domains were applied, including dietary behaviors, smoking and alcohol use, sleep, and physical and leisure activities.
The rst domain was dietary behaviors. In the CLHLS survey, the respondent was asked the frequency of eating or drinking fresh fruit, fresh vegetables and tea. The study coded these three variables as dichotomous ones with labeling respondents answering "almost everyday" and as "1" and "0" if otherwise. Tea consumption was considered because previous research pointed out that tea drinking related to longevity and reduced risk of mortality and death from cardiovascular diseases [42]. Tea consumption was thus used as an important health lifestyle behavior in this study.
The second domain related to smoking and alcohol use. Since the variables measuring the respondent's exact amount of cigarette or alcohol consumption had an extremely large amount of missing values with responding rates lower than 20.0% of the total sample, the research applied other measures. Those measures relied on CLHLS survey questions asking the respondent whether he or she smoked or drank alcohol "in the past" and "at present". The respondent who never smoked in the past or at present was coded as "0" and "1" if otherwise. It was assumed that for those individuals who smoked in the past and was still smoking when the survey was conducted was a heavy smoker; the same rationale and coding strategy were also applied to the alcohol consumption variable.
Sleep was the third domain which was represented by two indicators: sleeping duration and sleep quality. The study dichotomized the sleep duration variable as "1" indicating having 8 hours or more sleep each day and "0" as having less than 8 hours sleep. The sleep quality variable was dichotomized with those who reported their sleep quality as "good" and "very good" as "1" and poor sleep quality as "0" (including the categories that were originally coded in the survey as 'so so', 'bad' and 'very bad').
The fourth domain was physical and leisure activities. The research relied on two survey questions asking whether the respondent exercised regularly in the past and at present to determine if he or she was physically active. Those who exercised regularly both at present and during the past were coded as "1", and "0" if otherwise. The research also classi ed leisure activities into sedentary actives and active activities. Sedentary activities were such as reading newspapers/books, playing cards and /or mah-jong, and watching TV and/or listening to radio. Active activities included raising domestic animals, doing gardening work et al. For those who participated in leisure activities almost everyday were coded as "1" and "0" if otherwise.

Health outcome measures
The health outcome measures used in this research were consistent with measures used in previous research, including self-rated health (SRH) [43,44], cognitive function [45][46][47], chronic diseases [13,48] and activity of daily living (ADL) [49,50]. The respondent's SRH was coded as a continuous variable (1=very bad, 5=very good). Chronic disease variable was measured by whether the respondent reported any chronic diseases (1=yes, 0=no). The CLHLS survey asked the respondent whether he or she was suffering from 24 types of chronic diseases, including: hypertension, diabetes, heart disease, stroke/ cerebrovascular disease, bronchitis/emphysema/asthma/pneumonia, pulmonary tuberculosis, cataracts, glaucoma, cancer, prostate tumor, gastric or duodenal, Parkinson's disease, bedsore, arthritis, dementia, epilepsy, cholecystitis/cholelith disease, blood disease, rheumatism or rheumatoid disease, chronic nephritis, galactophore disease, uterine tumor, hyperplasia of prostate, and hepatitis. Since the missing values for prostate tumor, chronic nephritis, galactophore disease and hyperplasia of prostate exceeded half of the respondents, these 4 types of chronic diseases were dropped from the analysis. As a result, the study included the rest 20 types of chronic diseases. If the respondent answered he or she was suffering from at least one type of the 20 types of chronic diseases, then the respondent was coded as "1" for the chronic disease variable, and "0" if otherwise. Cognitive function of the respondent was measured by using the Chinese version of the Mini-Mental State Examination (MMSE). The MMSE was adapted from Folstein, Folstein, and McHugh [51] and tested four aspects of cognitive functioning: orientation, calculation, recall, and language. The total possible score on the MMSE is 30, with lower scores indicating poor cognitive ability.
Based on recommendations in the literature, responses of ''unable to answer'' were coded as incorrect answers [52]. Activity of daily living (ADL) disability was de ned as self-reported di culty with any of the following ADLs items: (a) bathing, (b) dressing, (c) eating, (d) indoor transferring, (e) toileting, and (f) continence. To avoid problems of complications and small sub-sample sizes in model estimation, the ADL functional capacity was dichotomized into "0" (meaning no ADL limitation) and "1" (meaning at least one ADL limitation).

Control variables
The analysis also controlled for the respondent's demographic characteristics such as age, gender, rural and urban residence. Respondents who lived in cities and towns were classi ed as urban residents. The respondent's socioeconomic status was also controlled, including years of schooling, per capita household income, and occupation before age 60. The occupation variable was coded as "1" if the respondent held a professional or administrative occupation and "0" if otherwise. Since socioeconomic condition in early childhood has been documented to have a cumulative effect on one's later life health status and mortality [53,54], the early childhood (or parental) SES was controlled as well. These measures included whether the respondent frequently went to bed hungry as a child, education of the respondent's father and the respondent's father's occupation before age 60 (1=professional or administrative job, 0=otherwise). Although the percentages of respondents and respondents' fathers who held professional or administrative jobs were low, the occupation measure has been repeatedly used as indicators of one's SES [55][56][57]. Thus, the validity of the occupation measure representing SES has been proved by previous analyses. Table 1 showed descriptive statistics for all variables.

Latent class analysis
Latent class analysis (LCA) was used to predict membership in latent or unobserved groups. LCA is a statistical method that nds subtypes of related cases (latent classes) from multivariate categorical data. It can be used to nd distinct categories considering presence/absence of several symptoms. The rationale of LCA is that given a sample of cases (subjects, objects, respondents, patients, et al.) measured on several variables, one wishes to know if there is a small number of basic groups into which cases fall. The results of LCA can be used to classify cases to their most likely latent class. Within each latent class, each variable is statistically independent of every other variable. If one removes the effect of latent class membership on the data, all that remains is randomness. In this study, respondents in a certain latent class shared similar health lifestyle patterns. Each case was assigned a probability of membership in each class.
LCA uses observed data to estimate parameter values for the model. The model parameters are the prevalence of each of C case subpopulations or latent classes and conditional response probabilities. A randomly selected member of that class will make that response to that item/variable. Parameters are estimated by the maximum likelihood (ML) criterion. The ML estimates are those most likely to account for the observed results. Estimation is usually done with simple iterative numerical methods. The traditional forms of LCA used complicated estimation methods based on matrix manipulation and simultaneous linear equations. Later on, simple iterative proportional tting was used to nd ML parameter values. In this analysis, since the exact number of health behavior typologies is unknown, an explanatory approach was used. It started with the most parsimonious 1-class model and tted successive models with increasing numbers of classes. Each latent class solution was replicated 20 times beginning at random starting values. This method included a close examination of item loadings and model t indices for estimating latent classes [58]. The nal number of classes was determined by the conceptual meaning, and commonly used t measures such as the Akaike Information Criterion (AIC), the Bayesian Information Criterion (BIC) and the value of entropy. The values of these indices for different class categories were shown in Table 2.
When running LCA, the Stata software showed that convergence was not achieved when constructing 5 classes.
Considering that the four-class solution provides the most conceptually coherent description of health lifestyles, the four-class solution was chosen as the most appropriate solution that represented health lifestyles among Chinese oldest-old. Since smaller values of AIC and BIC are better and the four-class model had both the smallest AIC and BIC (see Table 2), the statistical data proved that the four-class solution is preferred. As Table 2 presented, the entropy for the four-class model (0.732) was also beyond the criteria for good class separation cutoff point of 0.60 [59]. Thus, the four-class solution was determined to be the best classi cation type to represent health lifestyles among Chinese oldest-old.
Based on the results of LCA, the paper presented item response probabilities and shares for the analyzed sample for each class in Table 3. Such information clearly showed the four predominant health lifestyles among Chinese oldestold and the percentage distribution of sample among the four classes. Meanwhile, the table also showed the percentage distribution of the respondents for each health behavior, which helped us to describe the patterns of the four classes that represented Chinese oldest-old's health lifestyles.

Descriptive and regression analyses
The analytical part started with descriptive analysis to report means and percentage distributions of all variables (see Table 1). Multiple regression models were then constructed to predict Chinese oldest-old's health status on the basis of their health lifestyles, controlling for the respondent's demographic and socioeconomic characteristics. Since the health outcome measures of ADL disability and chronic diseases were coded as dichotomous variables, logistic regressions were used to perform the analyses. The other two measures of health status, namely, SRH and cognitive function scores, were continuous variables, ordinary least squared (OLS) regression were therefore applied to show how health lifestyles predict health status of the oldest-old.

Descriptive statistics
Descriptive statistical results for all variables were presented in Table 1. Among the 3,416 respondents aged 85 to 105, there was a higher percentage share of rural than urban respondents in the sample (57.3% and 42.7%, respectively) and females outnumbered males (59.3% vs. 40.7%). The mean age of the sample was 93.1 with a standard deviation of 5.7. As to SES of the respondents, the average reported years of schooling among the oldest-old was 1.5 with a standard deviation of 2.8. And the average schooling year for the respondent's father was 0.5 with a standard deviation of 1.8. The mean household per capita income for the year before the survey was 15,832.8 RMB (which is equivalent to 2,261.8 USD with 1 USD =7 RMB), with a standard deviation of 17,233.8. The percentage of the respondents and their fathers who had professional or administrative jobs before retirement was 5.7% and 2.6%, respectively. About 77.4% of the studied sample reported being hungry when going to bed in childhood.
The health outcome pro le among Chinese oldest-old was fairly good. To illustrate, the average SRH score was 3.3 (between fair and good). About 57.1% and 34.8% of the respondents reported having at least one type of chronic diseases and ADL disabilities, respectively. The mean cognitive function score was 22.5, suggesting a relatively good cognitive function status of Chinese oldest-old.
The health lifestyle patterns can be described as fairly healthy. The studied oldest-old seemed to be frequent fruit and vegetable eaters, with 13.5% and 49.1% of them eating fresh fruit and vegetables almost everyday. Tea was also preferred by some oldest-old. Nearly one fth of them had the habit of drinking tea almost everyday. About 24.9% and 21.3% of the studied sample reported themselves as smokers and drinkers, respectively. For three fth of the sample, sleeping was not an issue since they reported good sleep quality. A slightly over 55.0% of the sample had 8 or more hours of sleep each day. Considering the very old age of studied sample, the lifestyle of the respondents tended to be more sedentary than active. About 26.8% of them reporting doing physical exercise before age 60 and were still exercising when surveyed. And 48.4% and 41.7% of the oldest-old individuals reported that almost everyday they participated in at least one physical and sedentary type of leisure activity, respectively.

Health lifestyles among Chinese oldest-old
After choosing the 4-class model as the best tted latent class model, the study estimated item probabilities for the four identi ed latent classes. The four predominant Chinese oldest-old's healthy lifestyles (latent classes) and their share of the sample were presented in Table 3. Class 1 can be described as less healthy diet, not smoking, not drinking, poor sleep, low engagement in physical exercise and leisure activities, which contained 32.4% of the total sample. The in uence of health lifestyles on Chinese oldest-old's health OLS regression models were constructed to predict the respondent's health status that was measured by continuous variables (i.e. cognitive function and self-rated health) and logistic regression analyses were conducted to estimate how health lifestyles predicted the oldest-old's reported ADL disabilities and chronic diseases. Tables 4 and 5 presented OLS and logistic regression results when controlling for the respondent's demographic and socioeconomic factors, respectively. In both Tables 4 and 5, two models were constructed. The rst model included the health lifestyle classes as well as the respondent's demographic and SES characteristics; the second model further added the respondent's parental SES variables since prior research showed that parental SES had signi cant in uence on one's later life health status [49,53]. Table 4 showed that including parental SES variables did not signi cantly change the statistical results. As compared to class 3, SRH scores for individuals in class 1 (less healthy diet, not smoking, not drinking, poor sleep, low physical exercise & leisure activities) were 0.56 lower than the scores of those in class 3 (consistent engagement in healthy behaviors). SRH scores for older adults in class 2 (less healthy diet, not smoking, not drinking, good sleep, lowest physical exercise & leisure activities) and class 4 (moderate diet, smoking and drinking, moderate sleep, moderate exercise and leisure activity engagement) were 0.21 and 0.19 lower as compared to SRH scores reported by members in class 3. These results suggested that less healthy lifestyles led to worse self-rated health.

Models 1 and 2 in
Regarding the control variables, males and individuals with higher income tended to reported better SRH.
Regression results of using health lifestyle measures as well as control variables to predict cognitive function status of Chinese oldest-old were showed in Models 3 and 4 in Table 4. Similarly, adding parental SES covariates did not signi cantly change the statistical results except that the effect of one's education on cognitive function score turned to be nonsigni cant. Model 4 showed that as compared to the references group, class 3, cognitive function scores for the oldest-old in class 1, 2, and 4 decreased by 2.98, 3.45 and 1.37, respectively. The ndings again highlighted that less healthy lifestyles linked to worse health outcome that was measured by cognitional function among the oldest-old. As to the covariates, results showed that an increasing age related to lower cognitive function score, whereas being male, having higher education and higher income showed signi cantly positive effects on cognitive function scores of seniors. Going to bed hungry in childhood had signi cantly negative effect on the oldestold's cognitive function scores, supporting the cumulative disadvantage theories that childhood disadvantage was still able to explain part of the health disparities in older ages.
When predicting ADL disabilities and chronic disease status, two models were constructed with models 2 and 4 adding parental SES controls. Similar to results showed in Table 4, the results in Table 5 showed that adding parental SES covariates did not signi cantly change the regression results presented in Models 1 and 3. The odds of having ADL disabilities among oldest-old in classes 1, 2, 4 were all about 2.8 times of the odds for members in class 3 (see Model 2). The ndings indicated that all other 3 lifestyle classes had higher risks of reporting ADL disability as compared to the consistently positive class (class 3). Except for class 2 (sedentary group), classes 1 and 4 also showed signi cantly higher odds of having chronic disease(s) than class 3. As compared to the respondents in class 3, sample in class 1 and class 4 were 1.3 and 1.5 times more likely to have chronic disease(s), respectively, when controlling for the covariates. These ndings implied that health lifestyles can explain the health disparities among Chinese oldest-old.
The health differentials among Chinese oldest-old can also be explained by the respondent's demographic and socioeconomic characteristics. An increasing age and having professional or administrative jobs before age 60 increased the risks of elders experiencing ADL disabilities. Whereas being urban decreased the odds of the oldest-old having ADL disabilities. An increasing age lowered the odds of reporting chronic diseases; and higher family income and holding professional or administrative jobs before age 60 increased the likelihood of oldest-old's reporting chronic disease(s). These results seemed to be contradictory to ndings of prior research based on the Western society that higher SES led to better health condition. The paper offered possible explanations for such contradiction in the discussion section. In sum, ndings of this research proved signi cant in uence of health lifestyles on Chinese oldestold's health status, after controlling for covariates.

Discussion
With the trend of population aging, the oldest-old has becoming a fast growing group in China. Among the oldest-old, some of them are able to live longer and healthier whereas others suffer long-term disabilities and chronic diseases, which brings a heavy burden to the society as well as their family members. Therefore, a striking array of studies has been conducted discussing the in uential factors that have caused health disparities among the oldest-old.
Nevertheless, most of the existing studies examined elderly health from the perspective of considering single behaviors. The research also showed that healthier lifestyles resulted in better health outcomes. The ndings highlighted that consistent engagement in healthy behaviors linked to better SRH, higher cognitive function scores and a lower likelihood of being functional dependent and suffering chronic conditions. These results suggested that practicing healthier lifestyles can be an effective way to improve Chinese oldest-old's health status and postpone long-term disabilities. In this sense, the research results echoed arguments of researchers that multiple health behavior change interventions outperformed single-behavior interventions in health promotion [60, 61]. Findings of this study certainly provided strong proof that applying an integrative approach rather than individual health behavior perspective can be a better way to achieve a more effective health promotion. Healthy lifestyles were proved to be an important tool to prevent chronic diseases and long term disabilities among the oldest-old. Findings based on analyzing the China data also provided valuable implications to address disease prevention and health promotion related issues among older adults in other countries. Caregivers, clinicians and professionals can educate the elderly and their family to form healthier lifestyles in order to improve the oldest-old's health status and longevity.
The signi cant impacts of the covariates on Chinese oldest-old's health outcomes also had important implications.
Gender only showed signi cant effects on cognitive function scores and ADL disabilities, with males doing better on these two dimensions than females. Age generally showed signi cantly negative effects on health of the oldest-old, except for chronic conditions. The exception may be explained by the survival selection theory that the oldest-old with severe chronic illnesses had already died or been censored. Thus, older ages showed a negative effect on chronic conditions among survived individuals. Higher education and income was associated with better cognitive function.
Higher income and holding professional jobs also linked to greater risks of reporting chronic disease(s). This nding seemed to be incongruent with results documented in prior literature that higher SES led to better health outcomes.
Such inconsistency may be explained by underreporting of chronic illnesses among disadvantaged groups in China due to limited access to medical services and diagnoses. It can also be caused by the fact that people with higher SES in China have a more sedentary lifestyle and consume more high-fat and energy-condensed food, which resulted in a prevalence of chronic conditions. Another issue that worth mentioning is the urban and rural divide that has been documented repeatedly in prior literature [62,63]. Nevertheless, ndings of this research did not show signi cant health differentials among the oldest-old in these two spheres. The health disparities caused by residence only showed on the measure of ADL disabilities. That is, as compared to rural residents, urban oldest-old had a signi cantly higher likelihood of having ADL disabilities. The signi cant health differentials showed in ADL disabilities may be explained

Declarations
Ethics approval and consent to participate Not applicable.

Consent for publication
Not applicable.

Availability of data and materials
This article is based on a publicly available dataset derived from the Chinese Longitudinal Healthy Longevity Survey (CLHLS). The dataset can be obtained after sending a data user agreement to the data team.

Competing interests
The authors declare that they have no con icts of interest. Authors' contributions LZ was a major contributor in designing the research, writing the manuscript, conducting literature review and analyzing data. BXY and ZHD analyzed and interpreted the data; they also revised the earlier version of the paper. All authors read and approved the nal manuscript.    Note: *<..05, **<.01, ***<.001. R represents the respondent. OR: odds ratio; CI: con dence interval.