gms | German Medical Science

63. Jahrestagung der Deutschen Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie e. V. (GMDS)

Deutsche Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie

02. - 06.09.2018, Osnabrück

Enhanced cohort study analysis with multi-state models by example: effects of air pollution

Meeting Abstract

  • Henrik Rudolf - Ruhr-Universität Bochum, Bochum, Deutschland
  • Renate Klaaßen-Mielke - Ruhr-Universität Bochum, Bochum, Deutschland
  • Ulrike Trampisch
  • Dietmar Krause - Ruhr-Universität Bochum, Bochum, Deutschland

Deutsche Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie. 63. Jahrestagung der Deutschen Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie e.V. (GMDS). Osnabrück, 02.-06.09.2018. Düsseldorf: German Medical Science GMS Publishing House; 2018. DocAbstr. 192

doi: 10.3205/18gmds111, urn:nbn:de:0183-18gmds1117

Veröffentlicht: 27. August 2018

© 2018 Rudolf et al.
Dieser Artikel ist ein Open-Access-Artikel und steht unter den Lizenzbedingungen der Creative Commons Attribution 4.0 License (Namensnennung). Lizenz-Angaben siehe



Introduction: The lack of epidemiological studies for the long-term impact of air quality and traffic pollutants on health has recently been pronounced [1], [2]. Since pollutants like nitrogene dioxide and particulate matter are supposed to affect and modify several paths of disease progression, it might be possible that the application of multi-variable but univariate models does not sufficiently meet complexity of the situation. Multi-state models have been used for assessing the relationship between severity of a diseases and outcome(s) in epidemiological studies. We applied this to the data of getABI (German epidemiological trial on ankle brachial index) [3], [4]. This cohort consisted of 6880 aged (>65) patients and was followed over 7 years. For the assessment of air pollution effects on two endpoints (cardiovascular event (CVE) and death), results of an analysis of the study data extended by residential and emission data are presented here.

Methods: Accounting for losses and competing risks, the combined data was modelled as a five-state semi-Markov model. These states are healthy (no PAD) (A), PAD (peripheral artery disease) (B), CVE (C), loss (loss to follow-up) (D), and death (E). States were mainly defined according to getABI study records (case report form). Specifically for exposure to air pollution, particulate matter 10 (PM10) and nitrogen dioxide (NO2) data at patient home were used, which were set by interpolated annual background levels from a kriging model with a one square-kilometer resolution [5]. The multivariate structure of the model allowed us to check hypotheses on the influence of air pollution on the two endpoints (CVE and death) simultaneously, using the R-package msm [6]. Hereby, basic intensities to spontaneously change state were modelled by covariates. Selection of the potentially time dependent covariates was based on existing work of the getABI study group.

Results: With profound adjustments, it turned out that PM10 affected mortality of patients with present PAD (Hazard ratio=1.079, 95% confidence interval: 1.011-1.152), whereas NO2 elevated the risk of a cardiovascular event for healthy (no PAD) persons (HR=1.019, 95%CI: 1.003-1.035). With approximately 35.000 data points, these coefficients were estimated on a continuous scale (change per unit) to be interpreted as multiplicative risk, e.g. the risk of death for PAD patients (B→ E) with an exposure difference of say 5 [µg/m3] was increased by a factor of 1.46 (=1.0795). For NO2 a difference of 20 [µg/m3] raised the risk of CVE for healthy persons (A→ C) by 45%. Further, short distance (less than 50 meters) of residence to major roads increased mortality of healthy persons (A→ E) by 79% (HR=1.794, 95%CI: 1.057-3.046).

Discussion: Our results add to the growing wealth of knowledge concerning the impact of air pollutions on health. Results are based on the older general population from family practices spread over Germany. Biases could be reduced by accounting for losses and concurring risks. Results are applicable for individual risk assessment and hypotheses generation.

The authors declare that they have no competing interests.

The authors declare that a positive ethics committee vote has been obtained.


World Health Organization. Review of evidence on health aspects of air pollution – REVIHAAP Project. Copenhagen, Denmark; 2013.
U.S. EPA. Integrated Science Assessment for Oxides of Nitrogen. EPA/600/R-15/068. Washington, DC: U.S. Environmental Protection Agency; 2016.
Diehm C, et al. Mortality and vascular morbidity in older adults with asymptomatic versus symptomatic peripheral artery disease. Circulation. 2009;120(21):2053-2061.
Krause D, et al. The risk of peripheral artery disease in older adults–seven-year results of the getABI study. Vasa. 2016;45(5):403-410.
Beelen R, et al. Mapping of background air pollution at a fine spatial scale across the European Union. Science of the Total Environment. 2009;407(6):1852-1867.
Jackson CH. Multi-state models for panel data: the msm package for R. Journal of Statistical Software. 2011;38(8):1-29.