Understanding the key factors influencing the clinical outcome of an infection is crucial for early diagnosis and optimised treatment. Despite widespread recognition of the importance of the genetics composition of pathogens, most efforts so far have focused on characterising disease and susceptibility genes in humans. Here, we propose a new flexible and powerful methodological framework to detect candidate genetic polymorphisms influencing clinical outcome from pathogen genomes. The rationale is to use well-supported clades in a phylogeny as statistical predictors for clinical outcomes rather than the individual polymorphisms themselves. This greatly increases the statistical power to detect candidate polymorphisms when analysing a large number of variable sites. In a second step, the candidate polymorphisms are recovered by characterising the polymorphisms that most strongly support the clades predicting the clinical outcome. The modelling approach further allows including host factors and testing for possible interactions between factors. We illustrate the approach by an application on a dataset of hepatitis B polymerase genes. The statistical model retains age at infection as well as six candidate polymorphisms as predictors for clinical outcome (acute, chronic and fulminant). The method is straightforward to apply and computationally effective. While the approach is focused on detecting candidate polymorphisms from pathogen genomes, the method might be more broadly applied for characterising the link between genotype and phenotype while statistically controlling for environmental factors.