Cloud Hosting

Data source and study population

Data came from the CLHLS, a widely representative cohort survey conducted by the center for healthy ageing and family studies at Peking University and the Chinese center for disease control and prevention (http://opendata.pku.edu.cn/dataverse/CHADS). This survey was first instituted in 1998 and is conducted at roughly three-year intervals to gather information on health status, socioeconomic characteristics, lifestyle, psychological attitudes, and the accessibility of healthcare service. All information was obtained through face-to-face interviews by specially trained interviewers from the local centers for disease prevention and control. In cases where participants were unable to answer questions, a proxy respondent (usually a spouse or a close relative) was interviewed, but questions about mood were answered by the participants themselves. The CLHLS study was approved by the Research Ethics Committee of Peking University (IRB0000105213074), and all participants or their proxy respondents provided written informed consent.

The samples were selected from 23 of 34 provinces in China. Detailed sampling procedures are available in elsewhere [18]. The most recent survey (2018) was used to explore the multimorbidity patterns in this study and involved interviews with 15,874 participants. After excluding 599 of these participants because they were under the age of 65, multiple imputation was used to deal with missing data, a valid sample of 15,275 were analyzed in this study.

Eighteen chronic noncommunicable diseases or conditions in CLHLS (see Table S1 in Additional file 1) and two affective disorders were included in our study. Most chronic diseases or conditions listed in the CLHLS are determined by answering the question: Are you currently suffering from any of the following chronic diseases? For the purpose of our analysis, heart disease and stroke/CVD were merged into cardiovascular disease; bronchitis, emphysema, pneumonia, and asthma were merged into respiratory disease; cataract and glaucoma were merged into vision impairment; Parkinsons disease, dementia, and epilepsy were merged into nervous system disease; arthritis and rheumatism/rheumatoid disease were merged into rheumatoid arthritis; and cholecystitis and cholelithiasis were merged into biliary disease.

Participants were identified as having an affective disorder if they had an anxiety disorder and/or depression. The Generalized Anxiety Disorder Scale (GAD-7) was used to assess anxiety symptoms and consists of seven negative-oriented questions. All responses were coded on four scales ranging from none (coded as 0), a few days (coded as 1), more than half the time (coded as 2), and almost every day (coded as 3). Total scores ranged from 0 to 21, with >4 being diagnosed as having anxiety disorder [20]. Depression disorder was accessed by the 10-item Center for Epidemiologic Studies Depression (CES-D) scale [21], which includes eight negative-oriented questions and two positive-oriented questions. We recoded all responses in a four-scale metric, ranging from always (coded as 0), often (coded as 1), sometimes (coded as 2), and seldom or never (coded as 3). Furthermore, the positive-oriented questions, including are you full of hope for future life? and do you feel as happy as you were when you are young? were reverse coded before the summary (seldom or never coded as 0, sometimes coded as 1, often coded as 2, and always coded as 3). The depression score ranges from 0 to 30, with a higher score suggesting a greater degree of depressive symptoms. A cutoff of 10 was defined as diagnosed with depression disorder. Multimorbidity was defined as an individual who has two or more chronic diseases or conditions and was divided into two types for analysis: 2 (MM2+) and3 (MM3+).

Personal information was collected about the respondents, including age, sex, and region. Age was classified into three categories: 6579years, 8089years, and90years. Residences were divided into urban (city and town) and rural on the basis of their geographical location. We categorized the regions into East (Shanghai, Jiangsu, Zhejiang, Anhui, Fujian, Jiangxi, and Shandong), West (Chongqing, Sichuan, and Shaanxi), South (Guangdong, Guangxi, and Hainan), North (Beijing, Tianjin, Hebei, Shanxi, Liaoning, Jilin, and Heilongjiang) and Central (Henan, Hubei, and Hunan).

ARM is the process of exploring associations between data items. Frequent itemset and association rules are mined according to the Apriori algorithm approach to discover combinations of variables that are associated in large databases [22, 23]. ARM consists of two steps: first, it involves listing all high-frequency items in the set; second, it generates frequent association rules based on the high-frequency items [24]. In the Apriori algorithm, Support, Confidence, and Lift are the measures of the degree of association. An association rule is an implication of the form {X}{Y}, where {X} and {Y} are disjointed, non-empty sets of codes. Code sets of {X} and {Y} are the antecedents and the consequents, respectively. The strength of an association rule {X}{Y} can be measured by Support (the prevalence of both X and Y co-occurring.) and Confidence (the probability that Y occurs given that X is already present.). Lift refers to the deviation of the support parameter from what would be expected if X and Y were independent; Lift (XY)=P(X, Y) / P(X)P(Y). When the Lift value is >1, it implies X and Y are positively correlated. A higher Lift value indicates a stronger association between X and Y.

Considering the CLHLS dataset, for example, we used the rule hypertension, rheumatoid arthritis=>vision impairment. This is expressed as follows: if a participant has both hypertension and rheumatoid arthritis (antecedent), this will lead to vision impairment (consequent). To simplify the analysis, the following minimum thresholds were used to define the degree of interest: support 3.0% and confidence >30.0%, lift >1. This means that for all association rules shown here ({X}{Y}), the joint set {X, Y} emerges more frequently than would be expected under statistical independence, the consequent ({Y}) occurring in at least 30% of all cases that show the morbidity in the antecedent ({X}).

In this study, missing data existed for all chronic disease variables at random, ranging from 653 in hypertension to 1097 in cancer. Direct deletion of missing data cases may cause a significant information loss, we performed multiple imputation by chained equations (MICE), using the mi package developed by Gelman in R studio [25]. Descriptive statistics were used to show sociodemographic characteristics. Analyses were stratified by age, sex, residence, and region. Categorical variables were expressed as frequencies. Frequency and percentages were reported for qualitative variables and assessed by the Chi-square test or Fishers exact test, as deemed appropriate. Findings at corrected P-values of <0.05 were considered significant. To visualize the epidemiological trends of multimorbidity, the distribution of the study population, the prevalence of MM2+ and MM3+ was mapped using ArcMap. We also plotted percentage stacked bar charts to reflect multimorbidity. Finally, we use the arules package in R studio for ARM to identify multiple patterns of chronic diseases. Sensitive analyses were conducted for association between ADL disability and chronic conditions. Statistical analyses were conducted using SPSS, version 22 for Windows (SPSS Inc., Chicago, IL, USA), R studio, version 4.1.2 (R Foundation for Statistical Computing, Vienna, Austria), and ArcGIS, version 10.3 (ArcMap, ESRI Inc., Redlands, CA, USA).

More:

Age, sex, residence, and region-specific differences in prevalence and patterns of multimorbidity among older Chinese: evidence from Chinese...