Socioeconomic Determinants of Multimorbidity: A Population-Based Household Survey of Hong Kong Chinese

Introduction Multimorbidity has been well researched in terms of consequences and healthcare implications. Nevertheless, its risk factors and determinants, especially in the Asian context, remain understudied. We tested the hypothesis of a negative relationship between socioeconomic status and multimorbidity, with contextually different patterns from those observed in the West. Methods We conducted our study in the general Hong Kong (HK) population. Data on current health conditions, health behaviours, socio-demographic and socioeconomic characteristics was obtained from HK Government’s Thematic Household Survey. 25,780 individuals aged 15 or above were sampled. Binary logistic and negative binomial regression analyses were conducted to identify risk factors for presence of multimorbidity and number of chronic conditions, respectively. Sub-analysis of possible mediation effect through financial burden borne by private housing residents on multimorbidity was also conducted. Results Unadjusted and adjusted models showed that being female, being 25 years or above, having an education level of primary schooling or below, having less than HK$15,000 monthly household income, being jobless or retired, and being past daily smoker were significant risk factors for the presence of multimorbidity and increased number of chronic diseases. Living in private housing was significantly associated with higher chance of multimorbidity and increased number of chronic diseases only after adjustments. Conclusions Less advantaged people tend to have higher risks of multimorbidity and utilize healthcare from the public sector with poorer primary healthcare experience. Moreover, middle-class people who are not eligible for government subsidized public housing may be of higher risk of multimorbidity due to psychosocial stress from paying for the severely unaffordable private housing.


Introduction
Multimorbidity has been well researched in terms of consequences and healthcare implications. Nevertheless, its risk factors and determinants, especially in the Asian context, remain understudied. We tested the hypothesis of a negative relationship between socioeconomic status and multimorbidity, with contextually different patterns from those observed in the West.

Methods
We conducted our study in the general Hong Kong (HK) population. Data on current health conditions, health behaviours, socio-demographic and socioeconomic characteristics was obtained from HK Government's Thematic Household Survey. 25,780 individuals aged 15 or above were sampled. Binary logistic and negative binomial regression analyses were conducted to identify risk factors for presence of multimorbidity and number of chronic conditions, respectively. Sub-analysis of possible mediation effect through financial burden borne by private housing residents on multimorbidity was also conducted.

Results
Unadjusted and adjusted models showed that being female, being 25 years or above, having an education level of primary schooling or below, having less than HK$15,000 monthly household income, being jobless or retired, and being past daily smoker were significant risk factors for the presence of multimorbidity and increased number of chronic diseases. Living in private housing was significantly associated with higher chance of multimorbidity and increased number of chronic diseases only after adjustments.
Introduction experienced much faster epidemiologic transition than other Western populations [23]. Therefore, in the present study, we aimed to test how different socioeconomic status indicators were associated with multimorbidity as well as the number of chronic health conditions within the Hong Kong population. We hypothesized that while lower socioeconomic status is generally associated with higher chance of multimorbidity, the patterns of association in Hong Kong may differ from those observed in other long-term developed Western populations.

Data Collection and Study Population
We used data from a large population based survey, the Thematic Household Survey (THS) conducted by the Census and Statistics Department (C&SD) of the Hong Kong SAR Government for the current study. THS provided information on both socioeconomic status indicators and multimorbidity of the population. It is a regular series of territory-wide cross-sectional household surveys covering different social issues since 1999, including health-related topics. The data in the present study was obtained from the October 2011 to January 2012 round of the THS, which collected data on various health-related topics including current health conditions, medical insurance coverage, health care utilization and hospitalization, and also sociodemographic and socioeconomic characteristics [24].
The target population of the THS is the Hong Kong land-based population, excluding institutionalized residents, board vessel residents, foreign domestic helpers and hotel transients. The sampling frame of the survey consisted of all addresses of permanent quarters in built-up areas and records of area segments in non-built-up areas. The latter is deemed necessary because the quarters in these areas may not have clear addresses and cannot readily be identified. The survey covered around 95% of the Hong Kong resident population including both usual and mobile residents.
In this study, 13,411 households were randomly sampled from the discussed sampling frame stratified by district and housing type, within which 10,065 were successfully interviewed face-to-face by trained staff sent to the quarters with structured questionnaires. A total of 29,187 individuals were interviewed. An overall response rate of 75% was achieved. After excluding 3,407 individuals aged under 15 years who were not requested to answer any healthrelated questions, 25,780 subjects were included for statistical analyses in the study.

Independent Variables
As recommended by Braveman and colleagues [19], all available and feasible individual indicators of socioeconomic status in relation to the differences in health should be examined so as to reflect the multidimensional nature of socioeconomic status and avoid limitations of standardized measures.
Socioeconomic variables. In this study, we included household income, educational attainment, employment status and type of housing as socioeconomic variables, since they were available variables typically indicative of socioeconomic status from the THS. On the other hand, ethnicity, though usually viewed as an important factor in studies of other populations, was not included as one of the independent variables because the majority (around 94%) of the Hong Kong residential population was ethnic Han Chinese, while other ethnic minorities constitute only less than 2% of the population [25]. It is therefore less of a relevant determinant in Hong Kong than elsewhere.
Education attainment achieved by the individual was classified into 'post-secondary,' 'secondary,' 'primary,' and 'kindergarten or below.' Employment status: Employment status was divided into four categories: 'employed,' 'jobless,' 'student,' and 'retired.' Individuals who responded that they had a part-time or full-time job in the past seven days were regarded as 'employed,' and those who responded otherwise were regarded as 'jobless,' excluding 'students' who had to attend educational institutions and 'retired' persons.
Type of housing consisted of five categories: 'public,' 'subsidized,' 'private (tenant),' 'private (owner),' and 'others.' According to the latest Census [27] in Hong Kong, 31% resided in public rental housing, 17.8% resided in subsidized sale flats provided by Housing Authority and Housing Society of the Hong Kong Government, 49.3% resided in private housing and 1.9% resided in other types of housing including temporary housing, rooftop structures, mobile dwellings, institutional housing, dormitories, and staff quarters. Ever since the 1950s, public housing has been offered to provide affordable housing for households of low incomes. In other words, residents of public housing do not bear the heavy financial burden from rental expenses or mortgage loan when compared to residents of subsidized or private housing. According to the latest Household Expenditure Survey [28] conducted by the Census and Statistics Department in 2010, an average of 32.8% of the average household expenditure were spent on housing in Hong Kong.
Lifestyle factor. Smoking history was included in the analysis and was categorized into two groups: 'non-smoker,' and smoker (including 'occasional smoker,' 'past daily smoker,' 'past weekly smoker,' and 'daily smoker'). Due to data unavailability, we did not include other lifestyle factors such as alcohol consumption as independent factors. Nevertheless, the level of alcohol consumption is traditionally known to be low compared to other countries [29].
Outcome Measures. In this study, we used two outcome variables to measure the level of multimorbidity to test for consistency-1) the presence of two or more chronic health conditions (i.e., the presence of multimorbidity) and 2) the number of chronic health conditions.
Based on self-reported responses, information regarding chronic health conditions as diagnosed by western medical practitioners was collected. Subsequently, the reported health conditions were coded according to a comprehensive list of chronic conditions developed and used by a previous large-scale German study on patterns of multimorbidity where reported diagnoses under the same category were coded as one single condition [30]. Also, we included both concurrent as well as past conditions, because past chronic conditions are highly indicative of later poorer health, and this goes in line with the common view that multimorbidity should be seen as an indicator of health [2,5,31]. In our analysis, multimorbidity was defined as suffering from two or more chronic health conditions as in previous similar studies [32]. We used this definition to code a dichotomous variable of the presence of multimorbidity, where only two groups existed, namely: 'multimorbid' and 'non-multimorbid'. On the other hand, the number of chronic health conditions was calculated using the same aforementioned definition and coding of chronic health conditions. Please refer to S1 Appendix for the list of chronic diseases included in this study.

Statistical Analysis
Associations among independent variables. Before examining the association between the socioeconomic variables and multimorbidity, the pairwise bivariate association between independent variables were investigated by constructing a matrix of bivariate relationships. The Cramer's V statistic was computed for pairs with one or more nominal variables, and the Gamma statistic for pairs of ordinal variables. In addition, generalized variance inflation factor (GVIF), a well-accepted collinearity statistic [33], was calculated to detect the problem of multicollinearity in subsequent multivariable analyses. Also, log-likelihood ratio test were conducted to examine whether the models had any improvement with the inclusion of each additional variables.
Presence of multimorbidity. Unadjusted logistic regression analysis was first conducted for each independent variable for an examination of apparent association with multimorbidity, without adjustments for other co-variables. Unadjusted odds ratios and their corresponding 95% confidence intervals were calculated for an illustrative interpretation of results.
We then conducted multivariable logistic regression to examine the association between each independent variable with multimorbidity, with adjustments for other variables. Adjusted odds ratios and their corresponding 95% confidence intervals were calculated to investigate the adjusted association between each independent variable and multimorbidity.
Number of chronic health conditions. In addition to the presence of multimorbidity, we also investigated whether the independent variables were associated with the number of chronic conditions. An unadjusted negative binomial regression analysis was conducted to calculate the relative risk and their corresponding 95% confidence intervals for each independent variable.
As for the analysis of the presence of multimorbidity, a further analysis was conducted by constructing an adjusted model with all potential independent variables included altogether, to examine the adjusted association between socioeconomic indicators and number of chronic conditions. Again, relative risk and their corresponding 95% confidence intervals were calculated for an illustrative interpretation of the results.
Sub-analysis of possible mediation effect through financial burden. An additional subanalysis testing possible mediation effect through financial burden on the association between type of housing and outcome (multimorbidity and number of chronic health conditions) was also conducted. While residing in private housing estates is generally viewed as having a better quality of life or having a higher socioeconomic status than residing in subsidized and public housing estates, it is possible that its association with multimorbidity may be mediated by heavier financial burden borne by being private housing residents, which lead to higher burden on paying for the housing and hence more chronic diseases. In order to estimate the financial burden, we used the ratio of total household monthly expenditure over total household monthly income to generate a new proxy variable of financial burden. For both household monthly expenditure and household monthly income, we assigned a continuous score to each categorical level since only categorical data of both variables were collected. A higher score represented a higher income or higher expenditure. By dividing the score on expenditure by that on income, the expenditure-income ratio ranges from 0 (lower financial burden) to 1 (higher financial burden). Subsequently, model-based mediation analysis was conducted to examine the significance of financial burden as a mediator between housing and multimorbidity, and also to provide estimates and confidence intervals of those effects on multimorbidity [34].
Statistical analyses were performed using the computing environment R 3.1.1 [35] and IBM SPSS 21. Multivariate regression analyses were conducted using the R package of 'mass' [36] and mediation analysis was applied using another R package, 'mediation' [37].

Results
Descriptive Statistics (Table 1) In the present study, 13.5% of the respondents had two or more chronic health conditions (i.e., multimorbidity). More than three percent had three chronic health conditions, and nearly another three percent had four or more. Please refer to Table 1 for more detailed descriptive statistics of the respondents.
Associations among independent variables (Table 2) The Cramer's V statistics showed that the pairs of 'education attainment-employment status', 'age-employment status' and 'gender-smoking status' were strongly associated (larger than 0.30). However, age and gender are important covariables in any association model and thus should be included in our regression analyses. Smoking status was the only lifestyle factor in the regression analyses, and thus it would be reasonable to keep it. Also, since the aim of this study was to study the association between different socioeconomic indicators and multimorbidity, and education attainment and employment status could not fully represent each other (Cramer's V only slightly over 0.30), it was justifiable to keep both variables in our regression analyses. On the other hand, the computed Gamma statistics did not further reveal any strong association between ordinal variables (all below 0.60), and GVIF calculated for each variable were all within an acceptable range, i.e. smaller than 10, implicating that the problem of multicollinearity was unlikely in the subsequent multivariable models. Also, log-likelihood ratio test was statistically significant (p<0.01) at each step of additional variable inclusion. (Table 3) Table 3 showed the odds ratios and the corresponding 95% confidence intervals of the bivariate and multivariable analysis on the presence of multimorbidity as well as the relative risks and corresponding 95% confidence intervals of the bivariate and multivariable analysis on the number of chronic health conditions.

Presence of Multimorbidity and Number of Chronic Health Conditions
In the bivariate analysis for the presence of multimorbidity, being female, being 25 years old or above, having less than post-secondary education level, having less than HK$40,000 monthly household income, being jobless or retired, and smoking were statistically significant factors associated with the presence of multimorbidity. There were also apparent doseresponse relationships between multimorbidity with age, education level, and household income. Residing in non-public housing and being a student were negatively associated with the presence of multimorbidity. The results for the number of chronic health conditions were generally consistent with those for the presence of multimorbidity-being female, being 25 years or above, having less than post-secondary education level, having less than HK$25,000 monthly household income, being jobless or retired, and smoking were all statistically significant risk factors for increased number of chronic conditions. Similar to the presence of multimorbidity, dose-response relationships with increased number of chronic conditions were also observed for age, education level and household income. Also consistent with the results for the presence of multimorbidity, residing in non-public housing and being a student were negatively associated with the increased number of chronic conditions.
In the analysis for the presence of multimorbidity after adjustments for other co-variables, the direction of association remained the same for most factors-i.e., being female, being 25 years or above, having an education level of primary schooling or below, having less than HK $15,000 monthly household income, being jobless or retired, and smoking remained to be significant risk factors for multimorbidity. However, the direction of association with multimorbidity became reversed for housing types. In particular, residing in private owned or rental housing became significant risk factors for multimorbidity after adjusting for other variables. After adjustments, only age still showed significant dose-response relationship with multimorbidity. Similarly in the analysis for the number of chornic conditions after adjustments for other covariables, the direction of association remained similar for most factors-i.e., being 25 years or above, having an education level of secondary schooling or below, having less than HK $15,000 monthly household income, being jobless or retired, and smoking remained to be significant risk factors for increased number of chronic diseases. Also consistent with the results for the presence of multimorbidity, the direction of association with increased number of chronic diseases was reversed for housing types after adjustments. Residing in private owned or rental housing again became significant risk factors for increased number of chronic diseases after adjustments. Also, age remained to show significant dose-response relationship with increased number of chronic diseases after adjustments. Both logistic and negative binomial regression analyses were also conducted separately for male and female, as well as for the four different age groups. Similar results were identified.
Possible mediation effect of financial burden (Table 4) As shown in Table 4, statistically significant, albeit small, mediation effect of financial burden (proxied by the simulated expenditure/income ratio) acting between housing type and multimorbidity was evidently supported by the data. It was shown that 1.36% (1.46%) of the association of non-private housing (private housing) with the presence of multimorbidity was significantly mediated through increasing financial burden. Similarly, for number of chronic diseases, 1.15% (1.31%) were the proportion mediated through increasing financial burden. On average, around 1.41% of the association of housing type with the presence of multimorbidity was significantly mediated through increasing financial burden, while 1.24% of that association with the number of chronic health conditions was mediated similarly.

Discussion Summary
In general, our hypothesis that socioeconomic status is inversely related to multimorbidity and number of chronic health conditions is empirically supported by the results. Among the four indicators of socioeconomic status, household income, educational attainment and employment status were shown to have negative associations with the outcome variables (i.e. the presence of multimorbidity and the number of chronic diseases). On the contrary, we also consistently found that living in private housing was significantly associated with higher chance of multimorbidity and increased number of chronic diseases, with slight mediation effect through financial burden. Limitations Nevertheless, notwithstanding the population representativeness of the THS survey, there are limitations. First, all data, including the health conditions, were self-reported, which might lead to bias and inaccuracy in recalling and reporting. The prevalence of multimorbidity is low in the present study compared to other studies. In this case, it is likely that the less advantaged might have generally use western medical practitioners in the outpatient clinics less often, since most people visit private western medical practitioners for outpatient consultations in Hong Kong. Thus, the levels of multimorbidity might have been underestimated. Also, as all data were self-reported, there could also be an association between one's socioeconomic status and the number of diseases a person can remember or recall. We should be aware of this limitation when interpreting our results. Second, the definition of multimorbidity, as previously mentioned, has been under ongoing debates and modifications; hence, using different definitions might give different results in the analyses. The version we adopted in the present study has also been criticized as being overly simplistic [3,5]. However, many previous studies have adopted this definition, and the use of it could enhance the comparability and interpretability of the results [5]. Third, the cross-sectional nature of the dataset renders any causal inferences implausible. It is nevertheless an important first step in the exploration of social determinants of multimorbidity in Hong Kong which could shed light on future studies of similar kinds and help identify entry points for public health policies for the government. Social strata that are in need of additional resources could also be easily recognized according to the results of this study; thus, a general social pattern of multimorbidity is identified. Fourth, due to data availability, we did not include other determining factors of multimorbidity in our models, such as other lifestyle factors other than smoking, family history of chronic conditions, the use of medications and clinical history. Nevertheless, there is no reason why these factors would be unevenly distributed across the socioeconomic spectrum. In addition, we did not use all available socioeconomic indicators for the analyses. Although information on respondents' occupational position is available in the THS, the classification is too ambiguous for analysis. For example, the options 'professional' and 'associate professional' could be difficult for respondents to distinguish. Besides, the results of the rest of the models seem indifferent to the presence of occupational position in numerous trials. We therefore decided not to include occupational position. Last, there may also be bias in the sample towards the younger age groups who tend to reside in relatively newly developed districts of the city. Apart from the age effects, the district they lived in might have played an important role in determining their health via the variation in quality of the environment. Nevertheless, due to the absence of available district data, we did not include that for adjustment. Future studies should place more emphasis on that as a determinant of multimorbidity or even overall health status.

Relationship with Existing Literature
The present study obtained similar results with the existing literature [16,17,[38][39][40][41][42][43][44][45]. For instance, educational attainment and income were consistently found to be negatively related to the presence of multimorbidity in various populations, mostly European [39][40][41][42][43]46]. Nonetheless, this study is the first of its kind in a Chinese population. On the other hand, residing in private housing, albeit a conventionally indicator of a higher SES [47], appeared to be positively associated with multimorbidity and increased number of chronic diseases after adjustments for other variables. This differed from the results of previous studies [17,32,43,48] where residents in deprived areas were found to be associated with multimorbidity and more chronic health conditions.

Explanation of Findings
Household income, education attainment and employment status still remained to be significant, albeit weaker, risk factors of multimorbidity after adjustments for other variables, implying that they were individually associated with multimorbidity and number of chronic health conditions. On the other hand, the different patterns of association between housing type and multimorbidity observed in Hong Kong versus in previous studies can be explained by several possible reasons. First, previous studies [32,48] used neighborhood-level deprivation index as the independent variable, which was a more composite measure that consists of several socioeconomic factors, and did not distinguish the association between residence area and multimorbidity, as independent to that of income, age and education level with multimorbidity. Second, residing in public housing in Hong Kong is a completely different experience from residing in deprived areas in the United Kingdom. Public housing do not necessarily situate in the more socially deprived areas in Hong Kong. In fact, in Hong Kong, public housing estates are rather evenly distributed throughout the land, and some public housing estates situate themselves in the richest areas of Hong Kong. On other hand, as mentioned earlier, residents of public housing in general do not bear the heavy financial burden from rental expenses or mortgage loan when compared to residents of subsidized or private housing. Hong Kong ranks number one in the world for most unaffordable housing-it would take 14.9 years for a household earning median household income to pay for the median housing price, which is well above the second-place Vancouver (10.3 years) and other developed cities including London (7.7 years) and Singapore (5.1 years) [49]. This may have significant psychosocial impact on health, especially for the private housing residents, leading to a higher chance of multimorbidity [50,51]. Also, the follow-up analysis we conducted on the mediation effect of the financial burden between private housing resident status and multimorbidity served as additional empirical ground on which we based our speculation.
Nevertheless, this warrants further studies to verify, partly because our analysis was limited in the sense that the proxy of financial burden was calculated using total household expenditure instead of housing expenditure, and was therefore not an exact measure of the housing stress. Moreover, the proxy of financial burden only represented the current financial situation instead of the accumulated burden throughout the years. Both limitations may contribute to why the proportion of mediated effect was low. It is therefore recommended that future studies to include variables on financial situation and patterns of expenditures and savings of individuals so that the adjusted association between the type of housing and multimorbidity could be further investigated.

Public Health Implications
Our findings showed that people in lower socioeconomic position (as proxied by monthly household income, education attainment and employment status) tended to have higher risks for multimorbidity and increased number of chronic diseases and therefore were in need of medical care more than their counterparts with higher socioeconomic position. Hong Kong has a mixed healthcare system with both public and private sectors, where the public healthcare services are largely paid by the general taxation with additional low fees at the point-of-care, while the private healthcare service are largely paid out-of-pocket in addition to private medical insurance and employer-provided medical benefits if available [52]. This design of the healthcare system is in line with Hong Kong's general healthcare principle that "no one should be denied adequate medical treatment through lack of means" [53] so that the less advantaged individuals can access affordable healthcare from the public sector. Nevertheless, a previous study in Hong Kong [54] showed that patients of higher social status has had better primary care experiences which included better access and continuity of care when compared to those with lower social status. As previous research has suggested the important role of primary care in the management of multimorbidity [3], findings from the present study and the study by Wong et al [54] suggest that people with lower socio-economic status, although having a higher burden of multimorbidity, may receive less access and continuity of care as compared to those of higher status but with less burden of diseases.

Conclusion
In this study, we found that less advantaged people tend to have higher risks of multimorbidity and utilize healthcare from the public sector with poorer primary healthcare experience in Hong Kong. This warrants future healthcare practice to pay more attention to the social determinants of health. Moreover, middle-class people who are not eligible for government subsidized public housing may be of higher risk of multimorbidity possibly due to psychosocial stress from paying for the severely unaffordable private housing. Since little is known about the specific mechanism by which socioeconomic determinants affects multimorbidity, more studies on the mediation by intermediary (e.g. psychosocial, lifestyle and materialistic determinants) on the association between socioeconomic status and multimorbidity are needed to gain a better understanding of social production of health. Longitudinal studies are also needed to establish temporality for the causal relationship between socioeconomic factors and multimorbidity. Studies should also be conducted in other parts of China to test for external validity.
Supporting Information S1 Appendix. List of the 46 chronic conditions used in this study and their corresponding ICD codes. ICD = International Classification of Diseases (10th edition). (DOCX)