A MULTILEVEL HEALTH PROFILE OF MOSCOW

Irina Campbell, PhD, MPH ivm1@columbia.edu

www.CampbellHealthAssociates.com

Objectives• 1.) identify macro and micro risk factors for poor physical

health in Moscow;

• 2.) assess the effect of two dimensions of micro determinants – personal health habits and social connectivity, such as social cohesion, social support, and social networks;

• 3.) examine the hypothesis that relative social inequality is a significant structural condition at the community level which influences the physical health of individuals, as a main and as a joint effect with psychosocial behaviors.

Results of this study demonstrate that the social

context in a community affects the health of people living there independently from the effects of individual health lifestyle or

social connectivity.

INTRODUCTIONThe objective of this paper is to describe a cross-sectional multilevel health profile of the city of

Moscow, which was obtained before implementation of macro economic changes of January, 1992, in a

social epidemiological survey. The development of a multilevel theory and model of health was

undertaken in keeping with the WHO Healthy City Program and policy for the twenty-first century of

Health For All: “by the year 2000, the actual difference in health status between…groups…should

be reduced…by improving the level of health of disadvantaged…groups” (WHO, 1985).

Social epidemiology has traditionally been concerned with the distribution of morbidity or mortality in relation to a causal triad: personal

characteristics, geographical or community determinants, and change in occurrence over time.

These parameters were included in the design of the health profile, which examined the differential

effect of community level social inequality, a characteristic of the environment which was

hypothesized to increase vulnerability to poor health-related quality of life (HRQOL) in the individual host, in addition to the individual

psychosocial risk factors of the host.

The 3 research questions addressed

in this paper are:• 1.) to identify the macro and micro level and array of risks for poor physical health among individuals in the city of Moscow;

• 2.) to assess the additive or interactive effects on physical health of two dimensions of micro level risks - personal health habits and psychosocial behaviors, such as social connections in the form of cohesion, support, formal and informal networks;

• 3.) to examine the hypothesis that the distribution of social inequality at the community level influences the physical health of individuals, as a main and joint effect with personal health habits and psychosocial behaviors.

Multilevel modelsA multilevel theoretical perspective of health provides

explanations for multidimensional problems such as the health patterns among individuals in groups as a

consequence of social relationships between groups and among individuals within groups. Multilevel models may explain the variation in physical health by apportioning the effect directly to characteristics of the individual, to community contexts, or to the interaction between the

individual and community context. Multilevel models are thus able to provide a robust statistical analysis of

clustered, hierarchical data, such as individuals within groups or multistage sampling designs, without losing information about the independent effect of groups or

strata on individuals.

Multilevel models of health can analyze the emergent properties of social structure, such as

social inequality or relative income inequality, in conjunction with micro level properties, such as

smoking, drinking, distress, gender, or educational level. Context or the emergent properties of

structure at each level refer to those characteristics which exemplify aspects of the whole unit of

analysis and not the separate components of that unit (Blau, 1980). Contextual analysis can explain

the influences which the structure of a unit has within a hierarchy and upon its individual

components.

Macro determinants of health in Moscow

Reduction of inequalities in health has become a major concern of both national and

international public health policy (Kaplan, 1997; WHO, 1994). There has been some debate on the lack of standard definitions and measurement of

health-related inequality as a risk factor or outcome, as a micro and macro level indicator,

or as a relative versus average indicator. Absolute standards of living as well as income

distributions have become conventional determinants of public health.

Inequality in health has been successfully related to multiple dimensions of socioeconomic position: occupational status and prestige, education, and

income or access to resources (Siegrist, 1995). Each dimension of social inequality may not only have a unique distribution in a community, but

be related to different sets of health determinants. The theoretical contribution of the relative definition of social inequality addresses

the structural issue that an individual has a variety of social relations which are associated

with a variety of social positions within an array of social units (Blau, 1980).

The health patterns of East European countries have followed the deterioration of sociopolitical

structure with the ideological and market transformations of the 1980’s. A similar dynamic

operated in Perestroika Russia prior to the collapse of the Soviet Union, when widening income differentials within the country were due to

exogenous changes set in motion by fiscal policies. Many of these policies cut back the communist welfare state to stimulate economic growth and privatization, changing the relative and average

distributions of social status and health.

Three dimensions of social inequality, occupational status and prestige,

education, and income, were measured by relative indicators as:

1.) occupational status and prestige - the ratio of blue-collar to white-collar residents within areas;2.) income - the ratio of below average to above average apartment size or per capita living space in areas; 3.) education - the ratio of lower to higher educated residents in areas.

Average Inequality / 1000 persons within urban area N ed high - number of students in higher educational institutions N ed gen - number of students in general basic education, grades 1-10 N food - square meter space in food trade stores N nonfood - square meter space in nonfood trade stores N govt cafes - number of places in government cafeterias/ restaurants N public cafes - number of places in public cafeterias/ restaurants N hospital - number of hospital beds/10000 persons in each area N clinic visit - number of ambulatory polyclinic visits/shift/10000 persons N doctors - number of medical physicians all specialties/10000 persons N clubs - number of cultural clubs N libraries - number of public libraries N new moves - number of residence changes for better housing N new trade - square meter space of newly constructed trade stores N new houses - square meter space of newly constructed residential houses N preschools - number of students in kindergartens and nursery schools

Relative Inequality in urban areaBlue collar – ratio of %blue-collar to %white-collar workers living in areaApt size – ratio of above average apts. (10 sq.m. per capita) to below average apts.(9 sq.m. per capita) in areaEducation level – ratio of %low ed (secondary general, incomplete secondary and lower)to %hi ed (higher, incomplete higher, and secondary technical) residents in area

Click for larger picture

Micro determinants of health in Moscow

There were three dimensions of micro determinants of physical health which were

included in the Moscow health profile: • 1.) age, gender, education, marital status,

• 2.) the personal health habits of smoking, drinking alcohol, exercise, as well as the body mass index (the ratio of body weight to height-squared) as an indicator of diet quality,

• 3.) psychosocial factors, such as social cohesion, social support, formal networks of group memberships and informal networks of friends and family who would provide help when needed.

Social connectivity has been hypothesized as sustaining individual well-being or physical

health through the integration of the public and private spheres. The lack of formal networks,

such as participation in religious and community groups, the lack of informal networks, such as close friends and family, and the lack of social

cohesion have been associated with greater mortality from cardiovascular diseases (Bruhn, 1979), declines in life expectancy (House et al.,

1988), increases in homicides, the infant mortality rate (Kawachi et al., 1997), and crime (Wilkinson

et al., 1998).

METHODS

A random sample of Muscovites with telephones was collected, September 15-17, 1991. Only adults

18 years and older were interviewed. The total sample size of nearly 2000 telephone numbers

(n=1991) had a completed interview rate of 81.8% (n=1629). There was a two-stage sample selection

of respondents. The first stage was a random sample of telephone numbers within the 33

Moscow administrative districts; the second stage was the random selection of one respondent using

Kish probability tables.

The Physical Health Profile is constructed from a series of questions concerning disability, 13 specific chronic conditions, 11 specific symptoms, and three energy levels. The four dimensions were combined

into a mutually exclusive seven-point spectrum, based on frequency of conditions within the past 12 months: from optimum health of having 1) high energy; to 2)

low/medium energy levels; 3) one or more symptoms; 4) one chronic condition or impairment; 5) two or

more chronic conditions or impairments; 6) restricting activities, type or hours of work for 6

months or longer; and 7) severe disability, reported as difficulty with feeding, dressing, mobility, or

inability to work for 6 months or longer.

Social inequality indicators were derived from the 1989 City of Moscow census. Average inequality was measured by two factors extracted by varimax rotation: access to

material resources (eigenvalue= 8.64) and new development of resources (eigenvalue=2.51).

The two factors had an inverse relationship and varied with geographic location: centrally

located areas with access to resources around the Kremlin and peripherally located areas with

less access but greater new development of resources on the outer boundaries of the city.

The multilevel model was estimated in stages. Initially the null model was

estimated to derive the intraclass correlation coefficient (ICC): the proportion of variance in physical

health that is due to the variation of physical health between areas as a

portion of the total variance: = 00/(00 + 2).

RESULTSGeographic variation

Average inequality varied by geographic location. Most areas scored consistently as centrally located

near the Kremlin with high access/low new development, or peripherally located with high new development/low access. Areas with a larger ratio

of big families (5 or more members) were correlated with areas which had greater ratios of smaller than average apartments, lower educated and blue collar residents, and were located in the

periphery of Moscow.

Logistic regressionThe array of factors which predicted poor physical

health at the individual level did not vary by the average inequality within areas. Average inequality

was not a significant predictor for the fully adjusted model. Almost identical models were

significant for the sample as a whole, and within both high access and new development urban

areas. There was a slight effect of living in areas which had high access to material resources as

compared to areas of new development areas on

the poor physical health of women (Table 2)

Alameda Physical Health Profile (low)Total sample Average Inequality n=1629

Hi access area n=392

Hi develop area n=689

OR (95% CI)

OR (95%CI)

OR (95%CI)

Sex (F=1) 1.77 (1.35-2.32)

2.97 (1.62-5.43)

1.80 (1.21-2.67)

Age (per yr) 1.06 (1.05-1.07))

1.05 (1.03-1.08)

1.06 (1.04-1.08)

Personal habit: Overwt =1

Smokes any=1 AnySport (per n) Alcohol (per gms/past mos)

0.53 (0.38-0.75)

0.38^^ (0.17-0.85)

0.54^ (0.30-0.96)

Social cohesion: Normless (per n)

1.25 (1.11-1.40)

1.45 (1.11-1.88)

1.28^^ (1.06-1.57)

Meaningless (per n) Social support: Poor marriage (per n)

Alameda Physical Health Profile (low)Total sample Average Inequality n=1629

Hi access area n=392

Hi develop area n=689

OR (95% CI)

OR (95%CI)

OR (95%CI)

Social networks: Family (per n)

Friends (per n) Union grp =1 Prof/trade grp =1 Child/social grp =1 2.63

(1.56-4.45) 3.27* (0.87-12.30)

3.05^ (1.00-9.31)

Religious grp =1 1.72 (1.38-2.15)

2.08 (1.21-3.56)

1.54^ (1.04-2.27)

HRQOL low self-rated health =1

3.84 (3.01-4.90)

2.58 (1.46-4.57)

3.54 (2.32-5.40)

low life satisfaction =1

1.49* (.92-2.43)

low life happiness =1

1.34^(1.004-1.79)

Click for larger picture

Hierarchical Linear Regression

The multilevel model of physical health

is shown in Table 3 . The coefficients may be contrasted to the base intercept category of a 45.22 year old man, with better than a secondary/technical level education, and who consumed about

0.26 liters of alcohol per month.

Alameda Physical Health Profile(scoring from high energy to severe disability)

Null model

Model 1

Bivariate model controlling covariatesModel 2

Est. (SE) prob est. (SE) probFixed effects physical health γ 00 4.18(.04) p<.000 3.89(.09) p<.000

Covariates

Sex (fem=1) γ 10 .46(.06) p<.000

Age γ 20 .03(.002) p<.000

Educ γ 30 -.05(.02) p<.022

Single/div=1 γ 40-.02 (.07) p<.731

Married=1 γ 40 .09 (.06) p<.145Level 1HabitsObese =1 γ 40

Smokes =1 γ 50

Any sport γ 60

Etoh γ 70

.26 (.09) p<.005-.05 (.07) p<.464-.005 (.03) p<.856-.15 (.09) p<.095

Social cohesionNormless γ 90 .16 (.03) p<.000Social supportPoor marriage γ 100 .04 (.02) p<.015

Click for larger picture

Alameda Physical Health Profile(scoring from high energy to severe disability)

Null model

Model 1

Bivariate model controlling covariatesModel 2

Est. (SE) prob est. (SE) probSocial networksNfamily γ 110

Nfriends γ 120

Union grp =1 γ 130

Soc/child grp =1 γ 140

Prof grp =1 γ 150

Relig grp =1 γ 160

.001 (.006) p<.829

.005 (.006) p<.448

.03 (.07) p<.636

.39 (.12) p<.004

.24 (.14) p<.098

.28 (.07) p<.001Level 2Ratio Family size γ 01

Ratio Blue collar γ 02

Alcohol, area γ 03

Ratio Apt size γ 04

Ratio Low Ed γ 05

HiAccess, area γ 06

.02 (.09) p<.814-.13 (.12) p<.271.05 (.78) p<.919.07 (.11) p<.523.10 (.12) p<.410.01 (.04) p<.822

Random effectsLevel 1σ 2 (sd) 1.75 (1.33) 1.37 (1.17)Level 2intercept σ 2 (sd) (df=28)

0.014 (.12)(χ 2=44.97 p<.06)

0.014 (.12) (χ 2=42.06 p<.04)

Level 2slope σ 2 (df=32)

prof grp{.16 p<.061}relig grp{.043; p<.088}

Deviance 5554.78 (df= 2)

Click for larger picture

Model 2, in Table 3 , illustrates that gender, age, and education had fixed main effects in a bivariate model of physical health, which varied

significantly between individuals but not between urban areas. Individual level education

was a significant predictor, in contrast to the absence of this expected relationship in the

logistic regression. Neither marital status nor informal networks were significant predictors of

physical health, consistent with the logistic model. Lack of social cohesion and social

support, as well as membership in either social or child related groups, also had significant

fixed effects on poor physical health.

None of the level 2 macro indicators varied randomly or were significantly related to

physical health outcome in a bivariate model. Several alternative variables were

included in the model as possible explanations of the significant contextual effect shown by model 1. An interaction

between level 1 and level 2 variables may still be significant even if individual slopes

are not random because the test for detecting an interaction has a higher power than the test for detecting a random slope.

Macro-micro interactionsThe multilevel model explains the change in the

intercept of physical health by the main effect of level 1 or level 2 variables, as outlined above. It also explains the effect of individual level variables on the intercept of physical health by urban area variables through an

interaction effect. The cross-level model formally addresses the hypothesis of the third research question. This posits that physical health for

individuals varies across Moscow due not only to gender, age, and education groups with various

psychosocial factors, but also to the moderating effect of relative social inequality in the urban areas in which

they live (Table 4) .

Alameda Physical Health Profile (scoring from high energy to severe disability)Fixed effects estimate γ (se) L2 Main effects model 3 model 4 model 5 model 6 model 7 model 8 model 9 Physical Health γ 00 3.72(.12)* 3.70(.11)* 3.79(.10)* 3.73(.10)* 3.68(.11)* 3.68(.10)* 3.63(.11)*Family size γ 01 ns ns ns ns ns ns nsBlue collar low ed γ 02 ns ns ns ns ns ns nsAlcohol, area γ 03 ns ns ns ns ns ns nsApt size γ 04 .40(.17)̂ .33(.14)# .26(.14)̂ ^ .30(.14)̂ .31(.14)̂ .30(.14)̂ .30(.14)̂L1 Main effects (bold) and slope interaction effects (italics)Sex (fem=1) γ 10 .70(.13)* .68(.11)* .63(.11)* .69(.11)* .73(.11)* ..72(.11)* .64(.11)* Apt size γ 14 .52(.19)# .37(.13)# .32(.13)̂ .35(.13)# .37(.13)# .36(.13)# .37(.13)#

Age γ 20 .03(.003)* .03(.002)* .03(.003)* .04(.003)* .03(.002)* .04(.003)* .03(.003)* Family size γ 21 -.008(.004)̂ ^ -.006(.004)̂ ^ -.006(.004)̂ ^ -.009(.005)̂ ^

Education γ 30 -.05(.02)# -.05(.02)# -.05(.02)# -.05(.02)# Alcohol, area γ 33 -.91(.43)̂ -.86(.33)# -.86(.33)# -.90(.33)# -.80(.33)# -.87(.33)# -.89(.33)#HabitsObese =1 γ 40 .28(.09) #

Click for larger picture

Click for larger picture

Alameda Physical Health Profile (scoring from high energy to severe disability)Fixed effects estimate γ (se) L2 Main effects model 3 model 4 model 5 model 6 model 7 model 8 model 9

Social cohesionNormless γ 80 .13(.03)*Social SupportPoor marriage γ 90 .05(.02)#Social networksProf grp=1 γ 100

Child grp=1 γ 111

Relig grp=1 γ 121

.24(.14)̂ ^.40(.12) #

.28(.06)*Random effectsLevel 1 var σ 2 (sd) 1.36 (1.17) 1.35 (1.16) 1.34 (1.16) 1.36 (1.17) 1.35 (1.16) 1.35 (1.16) 1.34 (1.16)Level 2 varIntercept, σ 2 (sd) df=28 .014 (.11)̂ .014(.12)# .014 (.12)̂ .013 (.12)̂ .016(.12)̂ .013 (.11)̂ .013 (.11)̂Level 2 varSlope, σ 2 (sd) df=32

age (u2j) ̂ ^.00003(.005)

profgr(u100j).16(.40)̂

Deviance 5176.01*(df=4)2=378.68

5181.84*(df=4)2=372.94

In the first interaction, poor physical health was predicted by living in areas with a greater poverty risk (ratio of large families to all families) for individuals with poor social

support (model 6), or membership in child/social (model 8), or religious groups (model 9). While

lack of social support, or greater involvement in religious activities or other social groups had a negative effect on physical health, being older

and living in urban areas with a greater ratio of larger families had a beneficial effect on

physical health while being younger compounded the negative effect.

The second interaction involved gender and relative income inequality. While poor physical health was significantly greater among women

than men, it was even worse if women resided in areas with greater inequality of apartment sizes (Ymen 4.12; Ywomen 5.34). Obesity, lack of social cohesion and social support increased the risk for poor physical health significantly

more among women living in areas with greater inequality than men. This relationship also held

for those women with an array of formal networks.

In the third interaction, the expected positive association between physical health status and education was significantly moderated by the

contextual effect of alcohol consumption level in urban areas when social support was lacking, or there was participation in professional or child-related groups. The main effect of education on

physical health was positive for individuals living in areas with mean alcohol consumption,

as was the main effect of urban area alcohol consumption levels for individuals with a

secondary/technical level of education.

This interaction is partially due to average inequality and relative inequality being

geographically related to mean alcohol consumption in urban areas. Greater access to material resources

and lower ratios of inequality in education were found in central areas, which had average alcohol

consumption levels. Individuals living in such areas with lower education had better physical health than if they lived in other urban areas. About half of the peripheral urban areas with low access to material

resources and higher ratios of inequality in education were also areas with higher than mean

alcohol consumption.

DISCUSSIONIn this cross-sectional multilevel study of the city of

Moscow, the context of social inequality characterizing the urban area in which individuals lived was found to have significant main, additive,

and interactive effects on individual physical health, controlling for gender, age, educational

level, personal health habits, and social connectivity. Although proximal individual lifestyle behaviors have been most often examined as causes

of poor health, the structural effects of social context have not been systematically addressed in

the same model.

Variation in physical health was due to gender, age, education, lack of social cohesion, and involvement in

two types of formal networks: religious groups and child related or other social groups. Hierarchical

linear regressions indicated that physical health was also due to the relative social inequality in urban areas, regardless of which psychosocial factors

influenced health. In addition, the random effect of formal networks supported the hypothesis that the

distribution of physical health was significantly different between urban areas due to the distribution

of professional group membership between urban areas, as well as social inequality.

However, education, poor diet, professional group membership and lack of social support in the form of

poor marital relations were found to have direct effects on the physical health of individuals by the multilevel

model. Although individual educational status and increased alcohol use were not related to better

physical health in the expected direction in the logistic regression, a similar relation was not replicated by the hierarchical regression. The multilevel model indicated

that a contextual effect of area level alcohol consumption was significant in moderating the effect

of education on physical health, while individual alcohol consumption did not have a significant main

effect, accounting for the unexpectedly disparate finding of the logistic model.

The identification of conditions which increase the health disadvantage of some social groups is important for defining the targets of preventive health policy.

The multilevel city health profile of Moscow demonstrated which specific structural conditions at the community level and which specific psychosocial factors at the individual level could be

improved by health policy.

CONCLUSIONThe Moscow City Health Profile found that

individual physical health depended upon macro indicators of relative social inequality, and micro

indicators of social connectivity and personal health habits. There was support for the hypothesis that the contextual effects of relative social inequality acted

upon physical health independently from psychosocial factors. The structural conditions in

Moscow which increased the vulnerability of specific social groups for poor physical health were identified

for health policy as relative income inequality, poverty risks, and mean levels of alcohol

consumption in urban areas.

Although political liberty and economic prosperity were low in Soviet Russia relative to western democracies, the

centralized planning within Perestroika Russia nevertheless distributed economic

and social assets more evenly than the transitional market of today. An increase

in relative social inequality, as a contextual precursor to individual lifestyles for

example, may be a fundamental structural condition underlying the current health

crisis in Russia.