Personalised health and patient safety


The health system is producing data at an unrestrainable speed. Data that can mean personalized therapy, patient safety, personalized cancer prognoses, better prevention and monitoring of epidemics. We show how such data can be exploited, with a series of innovative prototype projects.


National population based cancer registries publish survival statistics by cancer site, stage, gender and time period, using relative survival methods. As new clinical registries are established, more data on treatment and later events become available, in addition to information on comorbidity or income and educational level.

To become more relevant both to the clinician and the patient, the survival statistics will be tailored to encompass more detailed information, in line with the tradition of risk prediction models.

We have investigated a number of different methods for relative survival and formulated them in terms of differential equations. Based on such a formulation, we can compare the behavior of the methods. As a by-product, we are developing a new method for measuring the probability of being "statistically cured", i.e. having the same risk of disease as the general population.


Personalized cancer treatments

We have developed a new multivariate penalized regression method (IPF-tree-lasso) that improves prediction of drug sensitivity in large-scale screening experiments based on molecular characterization of cancer cell lines in two ways: (i) by a more efficient combination of several sources of molecular data using the Integrative lasso with Penalty Factors, which we extended to the IPF-elastic-net and (ii) by borrowing information from available data on similar drugs through a hierarchical (tree) structure in the penalty terms. In simulation studies and application to the Genomics of Drug Sensitivity in Cancer (GDSC) data, multivariate IPF methods outperformed others, and IPF-tree-lasso obtained the best prediction.

Different measures for synergistic (or antagonistic) effects in drug combinations exist based on different definitions of non-interaction. We propose a fully statistical approach to model drug interactions based on a flexible Bayesian regression model. This allows us to include relevant information into the model specification, as well as to perform prediction for quantities of interest (e.g., for predicting the effect of those combinations of drugs that have not yet been tested). Furthermore, we link the proposed model with standard methodologies in quantitative drug interaction studies, and explore their performance on both simulated and real-life datasets.

Healthcare safety management

We are developing a two-step lasso-type model for prediction of hospital acquired infections (HAI) based on time series data. The goal is to use high-dimensional data from health records and administrative databases, routinely acquired in hospitals and health institutions, to predict the occurrence of HAI, while at the same time being able to identify factors to intervene on, in order to prevent HAI from happening. Simulations indicate very interesting properties of the method. Subsequently, the method will be applied to data from Oslo University Hospital.

network theory for health

We are interested in the spread of infectious diseases, with the goal of planning and prevention. We have developed a new model that allows the simulation of varying degree of population clusters within a country. The specific goal of this study is to understand the effect of restricting long distance travelling as a function of the level of clustering of the population, which can be seen as a level of urbanization of a country.

In a different project, together with the Norwegian Institute of Public Health, we have investigated to what extent pain tolerance, as measured by three different tests, is correlated among friends in a social network. Our study exploits data from Tromsø on friendships among all high school students and their pain tolerance. We find a significant correlation between cold-pressor pain tolerance of an individual and the pain tolerance of the individual’s friends.


Mathematical modelling and simulation have emerged as a potentially powerful, time- and cost effective approach to personalised cancer treatment. In order to predict the effect of a therapeutic regimen for an individual patient, it is necessary to initialize and to parametrize the model so to mirror exactly this patient’s tumor. We are developing a comprehensive approach to model and simulate a breast tumor treated by two different chemotherapies in combination or not. In the multiscale model we represent individual tumor and normal cells, with their cell cycle and others intracellular processes (depending on key molecular characteristics), the formation of blood vessels and their disruption, extracellular processes, as the diffusion of oxygen, drugs and important molecules (including VEGF which modulates vascular dynamics). The model is informed by data estimated from routinely acquired measurements of the patient’s tumor, including histopathology, imaging, and molecular profiling. We implemented a computer system which simulates a cross-section of the tumor under a 12 weeks therapy regimen. We showed how the model is able to reproduce patients from a clinical trial of OUS, both responders and not. We show that other drug regimens might have led to a different outcome.

Principal Investigator  Magne Thoresen

Principal Investigator
Magne Thoresen

co-Principal Investigator  Clara Cecilie Günther    

co-Principal Investigator
Clara Cecilie Günther