Modelling biomarkers for clinical and anti-doping purposes

This project will develop models for monitoring the urinary levels of endogenous anabolic androgenic steroids (EAAS) for clinical and anti-doping purposes. EAAS are popular as doping agents since the distinction between their endogenous (natural) production and exogenous administration is still difficult to ascertain. Moreover, EAAS can be used to identify metabolic imbalance and pathological conditions such as benign prostatic hyperplasia and prostatic carcinoma. Current methods for monitoring EAAS are based on univariate modelling of each biomarker in the steroidal profile. The models developed in this project will be able to simultaneously consider more than one biomarker.

In this project we aim to answer two types of questions:

1. Given a history of measurements, is there evidence of a change (suggesting doping or the development of an abnormality such as prostate cancer)?

2. Does the complete history of measurements, including the current observation, indicate that the individual in question is outside the reference distribution for the population?

Both questions can be answered by the same hierarchical model. The first question relates to the conditional distribution of the observations given the parameters of the individual, whereas the second question relates to the parameters of the individual in comparison to the population.

The model should be able to adapt in a consistent way as new data becomes available (more measurements for existing individuals as well as for new individuals). This can be achieved by employing a Bayesian framework.

As recent research suggests that the biomarkers are affected by genetic differences, it is unlikely that the reference distribution in the population can be captured by a single normal distribution. For this reason, different types of models will be considered including nonparametric models for the parameters of the individual (i.e. the random effects) and the distribution of the observations (i.e. the error).

The student will work on a Bayesian computational implementation of the above aspects so that the updating, as new measurements become available, can be done efficiently.

Though the project is mostly focusing on nonparametric multivariate hierarchical modelling, it also incorporates aspects of changepoint and anomaly detection.