The medical sector gathers and digitizes a lot of quantitative information on patients. This quantitative information is used to describe health properties of patients, such as analysis samples and diagnoses. Doctors are known to utilize information from analysis samples to diagnose illnesses of patients, but this relationship is not preserved when the information is digitized. This project aims to test the existence of a relationship between analysis samples and illnesses. To test this existence, a non-parametric bayesian model is constructed, which aims to predict the illness of a diagnosis based on analysis samples. This model uses Kernel Density Estimation to estimate normality spaces for medical properties of analysis samples given illnesses of diagnoses. These normality spaces contain densities based on temporal data of analysis samples and are used for estimating likelihoods of illnesses through analysis samples. The model is evaluated on different sets of illnesses and compared to naive prediction approaches. For each set of illnesses the model outperformed the naive approaches. Based on this, it is assumed that there exists a relationship between analysis samples and diagnoses.