Invited Sessions Details

Methodological Challenges in Observational Studies – Current Perspectives and Future Directions in Function Form, Measurement Error and Causal Inference

Presenter: Michal Abrahamowicz

Session Synopsis:

Unresolved issues in modeling of functional forms for continuous variables in multivariable analyses

On behalf of STRATOS Topic Group 2: Selection of variables and functional forms in multivariable analysis. The ultimate objective of the STRATOS Initiative is to improve the applications of statistical methodology used in real-life observational studies. Initial stages focus on selected �generic� issues, encountered in a broad range of observational studies. Task Group 2 (TG2) addresses analytical challenges related to two inter-related problems, typically encountered when building multivariable models (1) selection of independent variables, and (2) choice of functional forms for continuous variables. First, some of the challenges, and the need to simultaneously address issues related to both variables selection and modeling of their effects, will be illustrated using survival data. Then, a brief review of recent progress in statistical research on both topics will be presented, with main focus on the methods developed and/or validated by TG2 members. For variables selection, we will advocate use of the computationally intensive methods, that rely on general cross-validation and shrinkage. For modeling of the effects of continuous variables, we will briefly review different recent developments in the flexible modeling using various types of splines and fractional polynomials. Next, the main results of a few published simulation studies, which attempt to compare alternative methods, will be summarized. Based on the above review, we will try to briefly summarize the current state of research in our field and identify some important challenges that remain to be addressed.

Methodological Challenges in Observational Studies – Current Perspectives and Future Directions in Function Form, Measurement Error and Causal Inference

Presenter: Laurence Freedman

Session Synopsis:

Barriers and challenges to the use of statistical methods for addressing errors in the measurement and classification of outcome and explanatory variables in observational studies

Errors commonly occur in the measurement and classification of many variables that are used in epidemiological observational studies. However, in many fields of epidemiological research the impact of such errors is either not appreciated or is ignored. The STRATOS Task Group on measurement error and misclassification is currently engaged in several approaches to increasing the awareness of this problem among biostatisticians and epidemiologists and pointing to methods of addressing it. This endeavor includes (i) conducting literature searches of current practice in nutritional cohort studies, nutritional surveys, physical activity cohorts, and air pollution studies, (ii) preparing a general guidance document on the impact of measurement error and misclassification on study results, and methods of design and analysis that are needed to yield more reliable results, and (iii) preparing a guidance document focusing specifically on nutritional cohort studies. In this presentation we will describe the latest results and ideas that are emerging from our work. The general guidance document includes explanations regarding errors in explanatory variables versus outcome variables in regression analyses, classical measurement error versus Berkson error, misclassification of categorical variables and misclassification of continuous variables into categories. It points to options for software that has been developed to allow analyses that adjust for such error and misclassification. However, use of these methods depends on the availability of data that allows estimation of the magnitude and nature of the errors, and currently there is a great need to incorporate the collection of such data within study designs.

Methodological Challenges in Observational Studies – Current Perspectives and Future Directions in Function Form, Measurement Error and Causal Inference

Presenter: Els Goetghebeur

Session Synopsis: