Bernd Jäger

Different risk factors in basaloid and common squamous head and neck cancer.

Objectives/Hypothesis: The prevalence of human papillomavirus (HPV), herpes simplex virus (HSV), cigarette smoking and alcohol abuse was compared between two histological subgroups of head and neck cancer.

Data Mining and Metrics on Data Sets

One needs suitable metrics for topologically based methods of the data analysis. Examples of distances and metrics are given. A general method for the generation of metrics is developed. Its usefulness is demonstrated in support of classification methods for mixed type data

Cluster- and Discriminant Analysis With Both Metric and Categorial Data

Questions in medicine and biological sciences which require a cluster or discriminant analysis frequently appear. If there isn’t clarity concerning a probability theoretical data model, topologically based methods should be preferred. In this paper, a topological data model is defined for both metric and categorial variables. Every categorial character can be transformed into dichotomous characters. Therefore, we restrict the description to continuous and binary variables. A cluster analysis regarding this data space topology is illustrated on hand of a data set from forensic medicine.

Transformations of Confidence Intervals for Risk Measures

The relative risk and the odds ratio are discussed briefly in connection with their point estimations and confidence estimations. At an example is proved that the numeric conversion of the confidence limits of the odds ratio does not yield a confidence interval for the relative risk necessarily.

How Exactly Do We Know Inheritance Parameters

There are different concepts to quantify the information contained in a data set. The classic result is from R.A.Fisher: Regarding a sample over a random variable. The Fisher-Information is defined under certain assumptions as the inverse of the Rao-Cramer barrier in the well known inequality of the same name. The Fisher-Informations are considered for the simplest genetic models of intheritance. This applies to medically relevant ranges of the inheritance model parameter and large data sets. The results are compared with exact calculations.