The primary data used in this study were derived from the Sixth (2018) National Health Service Survey (NHSS) in Yunnan province. According to the unified requirements and deployment of the National Health and Wellness Commission, the survey samples followed the principle of economic and effective sampling, and adopted the method of multistage stratified cluster random sampling. A total of 10 sample counties (cities and districts) were selected: five urban sample areas and five rural sample counties. Five townships are selected from each county (city, district), and the quota (economy, region) is used for simple random sampling. Two villages (neighborhood committees) are selected from each township, and the quota (economy, region) is used for simple random sampling. Each village (neighborhood committee) selects 60 households and adopts systematic sampling. A total of 50 townships and 100 villages (neighborhood committees) were selected in the province, including 6,000 households (actually 6,010 households) with 19,490 individuals. Data is gathered by trained and qualified investigators, using the unified mobile terminal to carry out the household inquiry investigation; quality control is carried out in all aspects of the investigation. The survey results show that the answer rate of people over 15 is 83.0%. Our research sample comprises 11,304 respondents aged 35 and above from this health service survey (Figure 1).

Economic burden and socio-demographic characteristics were analyzed with descriptive statistics. The Chi-square test was used to explore the associated factors of MCCs. Associations among MCCs and factors were estimated using GEE models which further controlled for possible influencing factors of the internal connection among family members with households as the sampling unit. The GEE model assumes that the probability of parameters is binomial distribution with a logit link function. Therefore, due to the nature of the complex data structure of NHSS, the GEE model was used to adjust the relevant personal responses (18). At the recommendation of Liang and Zeger (19), we are interested in OR values in this study, which should be analyzed using GEE2. The SPSS software (Version 26) was used to carry out all statistical analyses. The significance level was set at less than 0.05, two-tailed, except for GEE. Considering that the dependent variables in GEE involve multiple comparisons, we adjusted the significance level according to Bonferroni method, and the adjusted significance level is 0.05/2 = 0.025.

The calculation formula of GEE2 model is as follows:

β : Coevariate coefficient

ϕ : Dispersion parameter

α : Related parameters

Both ϕ and α are functions of β . Only given the estimated values of ϕ and α can the solution of β be obtained.

Table 1 presents the major characteristics of all 11,304 respondents aged 35 years old and over. The prevalence of chronic diseases was 35.93% and the prevalence of MCCs was 10.12%, increased with age. In urban areas, the prevalence of one chronic condition in participants was 24.56, and 9.23% of the study population suffered from two or more chronic diseases. The prevalence rates of one chronic disease and two or more chronic diseases were 27.01 and 10.99% in rural areas, respectively. Among the age groups, the highest prevalence rates of multiple chronic diseases were 14.00% (55–64 years old) and 19.28% (65 years old or over). As age increased, the prevalence of one chronic disease and two or more chronic diseases significantly increased (p < 0.001). The prevalence of one chronic disease were 24.23% (male) and 27.34% (female), respectively, and the comorbidity rate of chronic diseases in females (10.82%) is higher than that in males (9.40%) (p < 0.001). The prevalence rates of one chronic disease in Han and other minority groups were 26.57 and 23.58%, respectively, and the comorbidity rates of chronic diseases were 11.13 and 7.18%, respectively. The remaining characteristics of study population with MCCs are shown in Table 1.

The top 8 chronic disease prevalence rates among males from high to low were hypertension (14.89%), intervertebral disc disease (3.84%), diabetes (3.81%), acute and chronic gastroenteritis (2.35%), cerebrovascular diseases (2.29%), rheumatoid arthritis (1.77%), other chronic obstructive pulmonary diseases (1.65%), other sports diseases (1.61%) respectively. The top 10 chronic disease prevalence rates among females from high to low were hypertension (18.30%), intervertebral disc disease (5.35%), diabetes (4.58%), rheumatoid arthritis (3.02%), acute and chronic gastroenteritis (2.94%), other sports diseases (1.64%), cerebrovascular diseases (1.59%), other types of heart disease (1.15%), cholelithiasis and cholecystitis (1.12), other chronic obstructive pulmonary diseases (1.03%) respectively. The prevalence of other chronic diseases in males and females were less than 1% (see Table 2).

Table 3 indicates the results of GEE models by taking MCCs occurrence as the dependent variable. First, we screened the noticeable demographic covariates of MCCs at a lower significance level of 0.025; all demographic covariates, such as area of residence, age group, sex, marital status, and education were included in the classified variables analysis. Second, GEE models 1 showed the adjusted associations between no chronic condition versus two or more chronic conditions and demographic covariates. Finally, in model 2, we further incorporated the product term of two or more chronic vs. one chronic condition and demographic covariates.

From the results of model 2: who residents lived in rural areas were more likely to report MCCs than those who lived in urban areas (adjusted OR = 1.347, 97.5% CI: 1.116–1.626); compared to people 35–44 years old, those with 55–64 and 65 years old and above were more likely to report MCCs (OR = 2.231, 97.5% CI: 1.584–3.144; OR = 2.540, 97.5% CI: 1.786–3.612, respectively). People of other minorities were less likely to suffer from MCCs than Han (OR = 0.752, 97.5% CI: 0.601–0.942). Compared with those with primary school education and below, people with junior high school education were more likely to suffer from MCCs (OR = 1.274, 97.5% CI: 1.010–1.607). People with an annual household income of ¥100,000 were more likely to report MCCs than those less than ¥10,000 (OR = 1.358, 97.5% CI: 1.024–1.802). People who exercise five or more times a week were more likely to report MCCs than those who did not exercise (OR = 1.571, 97.5% CI: 1.275–1.935); those who were in good and better health were less likely to suffer from MCCs than those in poor health (OR = 0.432, 97.5% CI: 0.331–0.563; OR = 0.225, 97.5% CI: 0.152–0.333, respectively). Overweight or obese people were more likely to report MCCs than normal shape people (OR = 1.317, 97.5% CI: 1.099–1.579).

Table 4 presents a total of 243 participants who reported hypertension and diabetes, accounting for 21.24% of 1,144 patients with MCCs. There were 503 (43.97%) respondents with hypertension and non-diabetic chronic diseases. The number of patients with diabetes and non-hypertensive chronic diseases and non-hypertensive and non-diabetic chronic diseases were 49 (4.28%) and 349 (30.51%), respectively. In urban area, there were 149 (13.02%) participants with hypertension and diabetes. There were 94 (8.22%) participants with hypertension and diabetes in rural area. The number of patients with hypertension and diabetes were 99 (8.65%) in males and 144 (12.59%) in females. Among the age groups, the largest number of patients with hypertension and diabetes was 77 (6.73%) (55–64 years old) and 131 (11.45%) (65 years old or over). The percentage of hypertension complicated with diabetes in the Han minority and others were 17.05 and 4.20%, respectively.

As shown in Table 5, a total of per capita expenses of 2 weeks’ illness of patients with MCCs was ¥216.05 (±1067.50) while per capita hospitalization expenses, annual household income, annual household expenditure, annual household medical expenses of participants were ¥4163.34 (±11518.77), ¥38918.28 (±40613.14), ¥35624.96 (±36720.35) and ¥10647.42 (±12556.92), respectively. The per capita expense of 2 weeks’ illness of hypertensive co-diabetic patients was ¥292.90 (±1427.80), while per capita hospitalization expenses, annual household income, annual household expenditure, and annual household medical expenses of participants were ¥4804.22 (±11851.63), ¥51064.77 (±52158.76), ¥41933.50 (±39940.02) and ¥11724.94 (±11642.74), respectively. The per capita expenses of 2 weeks’ illness, per capita hospitalization expenses, annual household income, annual household expenditure, and annual household medical expenses of hypertensive co-diabetic patients were more compared to the patients with the other three comorbidity modes (see Table 5).

In the current study, we analyzed the aggregated cross-sectional data from NHSS in Yunnan Province in 2018 to determine the economic burden of MCCs and associated factors specific to multimorbidity among middle-aged and older individuals. The prevalence of chronic diseases was 35.93% in 11,304 participants; the prevalence of MCCs was 10.12%. The prevalence of MCCs increased with age, and this finding has been consistently reported in previous studies (20). A reasonable explanation for the increasing prevalence of MCCs with age is that it may be associated with many age-related changes in the individual’s physiological state, such as changes in metabolism, immune response, and organ function (21). Compared to other Chinese studies, the prevalence of chronic diseases is higher than 30.3% while the prevalence of MCCs is lower than 12.3% reported by Wong’s team (22), based on a sample of respondents over 18 years old. The difference in estimated values among different studies was caused primarily by researchers’ choice of research group/population, the number of chronic diseases considered in the study, or both (23). For example, some studies defined MCCs differently as three or more chronic diseases.

This study evaluated the effects of eight socioeconomic factors (area of residence, age, ethnicity, educational level, household income, physical exercise, state of health and body mass index) on MCCs in model 2 of Table 3. Physical state of health is a positive factor for MCCs. The prevalence of MCCs in rural residents is higher than that in urban residents. People living in rural areas may have less access to health care and other services than those in urban areas (24). While compared with urban residents, people living in rural areas are often poorer and less educated (25, 26). People of other minorities were less likely to experience MCCs than Han, which may be caused by the association between ethnicity and BMI. Previous studies in China have found that the risk of obesity and central obesity among ethnic minorities is reduced (27, 28), which result from genetic and environmental heterogeneity (29). Compared to those with a primary school education or below, those with a junior high school education had a much higher risk of experiencing MCCs. The reason behind this inconsistency (29) with other studies needs to be further studied. Respondents with an annual household income of ¥50,000–100,000 had much higher odds of experiencing MCCs compared to those less than ¥10,000. However, behavioral interpretation is not possible because the link between income and education is significant even after controlling for various healthy behaviors.

Surprisingly, individuals who exercised ≥6 times per week had a higher risk to occur MCCs than those who did not exercise. This finding is consistent with recent studies in high-income, middle-income and low-income countries, which reported that higher levels of physical activity were associated with the prevalence of MCCs (30, 31). We propose that individuals with MCCs started to exercise more because of preexisting illnesses. Overweight and obese individuals are more likely to have MCCs than those with normal weight. This result is consistent with other studies around the world (30, 32). Studies have shown that both underweight and normal body weight can help reduce the incidence of chronic diseases (33, 34). Therefore, the inclusion of body mass index to identify and monitor obesity should be a health care priority, along with weight management and targeted control of other risk factors such as MCCs (35). The current perspective is that any movement from unhealthy to healthy behaviors has significant health benefits (36).

The results of model 1 in Table 3 demonstrated that, in addition to the above socio-economic factors, sex and alcohol consumption also affected the development of MCCs. People who did not drink alcohol were more likely to develop MCCs than those who drank alcohol. This is most likely because patients are forced to abstain from alcohol after learning of their illness. Meanwhile, female had a lower prevalence of MCCs than male. This finding contradicts the previous research (37–39), which showed that MCCs tend to be more common in female than male. However, sex was not independently associated with MCCs. Men were more likely to smoke and drink alcohol than women (40), which may explain why men are more prone to experience MCCs. These findings can provide added value in designing guidelines for different chronic conditions that can be assessed in groups or individually. These predictions can be useful for frontline medical staff seeking the best approach to care for multi-disease patients.

Table 5 indicates that the per capita expenses of 2 weeks’ illness, per capita hospitalization expenses, annual household income, annual household expenditure, and annual household medical expenses of hypertensive co-diabetic patients were more compared to the patients with the other three comorbidity modes. Most previous studies have looked at diabetes and high blood pressure separately (41–43), not in combination. The results of this study showed that hypertension co-diabetes has brought severe economic burdens. We should not only pay attention to the respective diagnosis, management, treatment, and economic burden of hypertension and diabetes, but also pay close attention to a series of studies on hypertension co-diabetes. Chronic diseases usually do not appear in isolation. Most patients with chronic diseases have multiple diseases (44). Different chronic diseases may influence the development of each other, and the control and treatment of one disease may adversely affect another. Therefore, the management of multiple chronic diseases is a complex public health problem.

This research is meaningful in health promotion, for the following reasons. This study can provide a broader view of MCCs and its associated factors, by including a larger population, adjusting personal responses, and analyzing related social demographic factors. Firstly, this is a large population-based study on the economic burden of MCCs and associated factors. The data is derived from the Sixth (2018) National Health Service Survey in Yunnan Province, a cross-sectional study. According to the unified requirements and deployment of the National Health and Wellness Commission, the survey samples followed the principle of economic and effective sampling, and adopted the method of multistage stratified cluster random sampling. Secondly, due to the nature of the complex data structure of NHSS, the GEE model was used to adjust the relevant personal responses in this study. Thirdly, we analyzed the influence of urban and rural, ethnic, gender and other social demographic factors on MCCs, which helped us to understand the distribution characteristics of MCCs and may bring policy makers closer to narrowing the gap of observed health inequality. Finally, this study combined diabetes and hypertension to investigate their prevalence and economic burden, which can provide a basis for further research on MCCs, and encourage policymakers and implementers to shift from managing a single chronic disease to the co-management of MCCs.

Our study also has some limitations. Firstly, this study is a cross-sectional analysis, therefore, cannot address a causal (bidirectional) relationship between determinants and MCCs. Secondly, we only analyzed the NHSS data of Yunnan Province, which may limit the generalization of the observed influencing factors of MCCs and their economic burden on a wider population. Additionally, we use the generalized estimation equations model (including behavioral factors, social and economic status factors, health factors, etc.) to control the influence of the confounding factors. However, despite many variables, confounding factors cannot be completely avoided. Finally, there is no standard method to measure MCCs.