Professor in Epidemiology
Division of occupational and environmental medicine
A complete publication list can be found here.
Current methodological projects
Longitudinal data analysis
Results of observational studies are often used as a basis for public health interventions, e.g. investments in neighborhood facilities aiming at reducing social inequalities in well-being and health. It is essential that the scientific basis for such decisions is sound. In this project we will explore methodological issues in design and analysis of longitudinal studies with a series of simultaneous assessments of both exposure and outcome. Such designs are often survey-based and are increasingly used in order to overcome the well-known limitations of cross-sectional studies, but remaining validity issues related to self-selection, reversed causality and time-varying confounding deserve careful attention. Moreover, representation and analysis of longitudinal data differ noticeably across studies, often with unknown consequences for the validity of the results.
It is often of special interest to identify subgroups, defined by e.g. genetic, environmental and/or social factors, where the exposure under investigations is particularly predictive of health and well-being. Analyzing such heterogeneity in registry-based studies is challenging, since the number of individual trajectories that could potentially modify the effect is likely to be numerous. Propensity scores techniques have since their introduction evolved into a formal step-by-step procedure for confounding control, through stratification (or similar types of adjustments) based on the exposure propensity. In on-going work we study stratification for outcome propensity (also known as a prognostic score or as risk stratification) as a means of investigating heterogeneity of exposure effects while keeping the number of subgroups limited. This approach is potentially useful for assessing individual susceptibility to environmental exposures, but practical concerns with outcome propensity modeling as well as the theoretical properties of such modeling warrant further investigation.
Spatial reliability analysis
Comparisons of perceived and objectively measured neighborhood attributes or exposures show associations (correlations) but generally low agreement. The low agreement has been explained as a mismatch between perceptions and objective facts, but may also suggest that perceived and objective measures are capturing different aspects of the local environment. Additionally, the instruments used to capture perception may lack inter- or intra-reliability, and thereby contributing to the low agreement with objectively measured attributes. In this project, we develop the use of kappa and colocation coefficients as tools to investigate the reliability of environmental ratings among nearest neighbors in a spatial context.
Björk J. Praktisk statistik för medicin och hälsa. Liber 2011. (Practical statistics for medicine and health)
- Assistant coordinator for SIMSAM Lund with grants from the Swedish Research Council (VR).
Member of the SBU-expert panel "Metoder för att skatta njurfunktionen" (Endogenous markers for estimation of kidney function), 2009 - 2011
The report is available here.
Member of steering committee for the Swedish Epidemiological Society
(Svensk Epidemiologisk förening; SVEP), 2007-
Registry-based research in Sweden
Seminars about methodological issues in registry-based research
Equator Network - guidelines for transparent reporting of health research
Points of significance in Nature
Statistics notes in BMJ
Free software for sample size calculations
Verktyg för klinisk forskning i Läkartidningen (Swedish)