Background and Objective

According to the latest United Nationsprojection about two thirds of the world population will live in areas

classified as urban by the year 2050 [1].While accelerating growth of cities in terms of area and population

size will trigger broad changes insocial structure or occupational activities it also bears substantialchallenges

regarding social equity and thesustainability of public health systems [2, 3].Comparative analyses in rapidly

urbanizing countries suggest thateffects of crowding, pollution and the simultaneous adaption of unfavorable

life styles can lead to an increasedprevalence of diseases and health problems among the urban population

[4].

Although recently contested [5],also historical public health research often understands cities as high-risk

environments associated to the vastspread of infectious diseases and crime [6, 7].Hence, the urban penalty

hypothesis gained popularity in publichealth research, even though it was coined to describe vastly urbanizing

areas during the industrialization [8].In spite of the reduction of mortality through remarkable public health

interventions since the beginning ofthe century, there is consensus that the exposure of urban environment

affects health and wellbeing throughair and noise pollution [9, 10].

While analyzing direct effects ofpollution and access to green spaces on specific health outcomes i1s presumably

intuitive, the examinations of overallhealth in response to the exposure to an urban environment requires

linkable multi-level data and skillfulhandling of latent concepts. The absence of a universal definition of what

is rural and urban has often led tosimplifications in the analysis strategy, most commonly the lack of other

urban features than population density[11]. Furthermore, the examination of complex association between

health and residential environmentrequires access to rare linkable individual level information. With the help

of our project partner, the Instituteof Cartography and Statistics of Andalusia (ICEA), we aim to overcome

the aforementioned limitations. Theirlongitudinal individual level mortality data base allows a linkage to a

register-based residential informationand enables us to conduct a detailed analysis of mortality differentials

by environmental small area featuresfor Andalusia, the southernmost and most populated of the 17 Spanish

autonomous communities.

Data and Analysis Strategies

Following Vhalov and Galea (2002) [12],we generate a multidimensional indicator for our latent variable

urbanicity, whichrelates to the degree of urban characteristics including population density andartificial surface

area on census tract level. Theconcentration of urban areas in Andalusia is depicted in figure 1 by apreliminary

indicator. Through our project partnerwe are able to link the environmental information to the population and

housing census of 2001 and a 10 %mortality follow-up of the census population of Andalusia. The combination

of these information allow us to examinehow residential environment features affect individual level mortality

over a time period of 13 years.Considering the multilevel structure of our data, we aim to apply an adaptedCox

proportional hazards model with mixedeffects. To account for possible clustering in the baseline hazard, the

classic model is extended bystratum-specific term Wi = log !i whichcan be incorporated as in the following

equation.

h(ti;j ; xi;j) = h0(ti;j) exp(_T xi;j +Wi) (1)

where h0 isthe baseline hazard with multiplicative effects by the exponential terminvolving the covariates

and the Wi,the normally distributed stratum specific frailty term designed to capture thedifferences between

census tracts.

Preliminary results and Outlook

The preliminary results of our survival analysisindicate the existence of a small but highly significant urban

mortality penalty in modern day Andalusia. Thehazard ratios and goodness of fit parameters of two classic Cox

proportional hazard models are displayed in table1. To measure the impact of urbanicity, we use a preliminary

indicator and do not apply the aforementionedgeographical measures yet. Whereas urban environment

features do not seem to affect the risk of dyingin a model without further covariates, their effect sizes increase

and are strongly significant for the full modelcontaining socioeconomic and demographic control variables

(model 2 - shortened). The preliminary resultsindicate the existence of a mortality disadvantage for urban

dwellers in modern-day Andalusia.

In following steps, we aim to apply theaforementioned extension of the Cox model which allows for the

incorporation of spatial position of small areasand supports the detection of clustering and neighboring

effects. Furthermore, we plan to generate acomparative measure for urbanicity and control for theadditional

socio-spatial context variables which potentiallymitigate the effect of area deprivation on the urban health

penalty. If possible, we will also join theindividual information on household level.