Modelling Health Data on a Small Urban Scale Using Deterministic Iterative Proportional Fitting : A Contribution to Setting up Citywide Health Monitoring Systems

Abstract

Big cities are often regarded as unhealthy places. Poor air quality, noise pollution, and scarce public green space are usually part of everyday life in metropolises. Yet, the physical attributes of the urban environment can vary significantly across neighbourhoods, thus leading to unequal living conditions. This is often the result of social deprivation – while some have the luxury of choosing where in the city to live, others are left only with the affordable options. Over time, the continuous exposure to adverse factors of the living environment may trigger chronic illnesses such as hypertension or asthma. Nevertheless, health-related dynamics at the neighbourhood scale usually remain hidden. In Germany, health data is published exclusively in aggregated form for spatial units encompassing tens of thousands of inhabitants. As each health insurance fund manages its own data, there is currently no centralised point of contact for health data acquisition. This situation poses a challenge for researchers, who strive to quickly gain insight into ongoing health processes and their manifestation within the spatial realm of the city. In this context, the appreciation for small-scale data is growing – especially in light of the COVID-19 pandemic. In North America, such data is used to estimate local risk of infection and its potential severity, and to prioritise vulnerable populations for vaccination. Against this background, the dissertation proposes a spatial microsimulation method for generating individual health data on a small urban scale – one corresponding to the perception of neighbourhoods. To that end, the city of Hamburg serves as case study. Prior to introducing the chosen modelling method, the advantages and limitations of the main established approaches to creating a synthetic population are discussed. The perspective of several public health researchers, based in Germany, on the necessity of small-scale health data and the implementation of modelling strategies for its generation is also brought to attention. The generated spatial microsimulation model is product of a well-established data modelling technique, which is being implemented for decades in countries such as the UK, Australia, and the USA. The data modelling process is divided into several main stages: input data selection and preprocessing, population synthesis using deterministic iterative proportional fitting, and model evaluation with sample survey data and data from three of Hamburg’s health insurance funds – AOK Rheinland/Hamburg, BKK Mobil Oil, and DAK-Gesundheit. The biggest asset of the introduced approach is the proposed two-tier modelling strategy, which allows constraining the synthetic population at two of Hamburg’s spatial scales simultaneously – that of the city quarters and the statistical areas. Thus, available health data aggregated at the level of the city quarters is used to optimise the reliability of the modelled disease prevalence at the smaller, neighbourhood scale. The synthetic population allows combining individual health data with available social, economic, and geodata to put all the pieces of the puzzle together. It thus lays a foundation for further analysis of a broad range of topics regarding cities and health. In this regard, two application examples for the generated individual health data are introduced: examining the small-scale distribution of people with hypertension and their exposure to excessive noise from road and/or air traffic; and identifying hotspots of individuals at elevated risk of developing severe symptoms of COVID-19. The limitations of the model and the necessary prerequisites for transferring the proposed method to other cities are also discussed.