Statistical Model Building Strategies for Cardiology

In all medical fields, the correct evaluation of disease progression and treatment response are essential for the judgment and the improvement of therapies. Regression models with many risk factors are particularly important in the context of observational studies where groups of patients are likely to show structural inequalities. There are several distinct aims of such models : 1) to identify risk factors, which explain differences in the outcome of interest, 2) to describe the association between risk factors and the outcome of interest, and 3) to predict the outcome of interest. The statistical challenges for these three aims are different. Generally, the development of a valid descriptive model relies on two main steps: a) the identification of a meaningful number of risk factors, and b) the specification of the functional form of the association between these risk factors and the outcome of interest. Intensive statistical research on both aspects has been performed for decades. However, the results of this statistical research are only poorly incorporated into clinical research. The project Statistical Model Building Strategies for Cardiology intends to build a bridge between statistical research on model building and implementation of these methods into actual medical research by means of four typical research questions from cardiology. This transdisciplinary project aims at 1. identifying deficiencies in current cardiovascular applications with respect to statistical model building, 2. building advanced statistical models for the four research questions by applying state-of-the- art methods, 3. developing and evaluating new methods to correct the typical overestimation error arising in data-driven model building, and 4. providing guidance for model building strategies, which are understandable for applied researchers. From a statistical point of view, the aim is to identify, discuss and improve the current standards applied in clinical research with respect to model building. To guarantee that our methodologic results have a true impact in medical application, we develop and test our methods with four well defined research questions from cardiology. Our defined medical starting point is both as concrete as possible, but also has a broad potential for more general transferability. From a medical point of view, the aim of this project is to gain new medical insights from statistical models, which are built by employing better methodology. As a comprehensive result, we expect to deduce statistically improved and valid models for each of the four research questions. For this purpose, we will use several data sources of cardiologic studies and combine them with results from the corresponding medical literature.

In all medical fields, the correct evaluation of disease progression and treatment response are essential for the judgment and the improvement of therapies. Regression models with many prognostic factors are particularly important in the context of observational studies where different groups of patients are likely to show structural inequalities. There are several distinct aims of such models: 1) to identify prognostic factors, which explain differences in the outcome of interest, 2) to quantify the association between prognostic factors and the outcome of interest, and/or 3) to predict the outcome of interest. The statistical challenges for these three aims are different. Generally, the development of a valid descriptive model relies on two main steps: a) the identification of a meaningful number of prognostic factors, and b) the specification of the functional form of the association between these prognostic factors and the outcome of interest. Statistical research on both aspects has been performed for decades. However, the results of this statistical research are only poorly incorporated into clinical research. The project 'model building strategies in medical applications' intends to build a bridge between statistical research on model building and implementation of these methods into actual medical research in the field of cardiology. This transdisciplinary project aimed at 1.) identifying deficiencies in current cardiovascular applications with respect to statistical model building, 2.) developing new methods and evaluating already available methods of statistical model building focusing on the identification of prognostic factors for a final model and the specification of the functional form of continuous prognostic factors with the outcome, and 3.) providing guidance for model building strategies, which are understandable and applicable for applied researchers. From a statistical point of view, the aim was to identify, discuss and improve the current standards applied in clinical research with respect to model building. To guarantee that our methodologic results have a true impact in medical application, we developed and tested our methods with well-defined research questions from cardiology. Our defined medical starting point was both as concrete as possible, but also has a broad potential for more general transferability. From a medical point of view, the aim of this project was to develop guidelines aimed at analysts for modeling with all the special features of real data. In addition, we showed how new medical insights from statistical models, which are built by employing better methodology, can be gained. For this purpose, we used several data sources of cardiologic studies.