Precision Medicine

Development of novel prediction and stratification tools 

Research Question

To develop a tool which can predict whether or not a person with cardiovascular disease is likely to develop a mental health condition, and how old they will be when this occurs.
(and vice versa: onset of mental ill-health in people with cardiovascular disease).

Principal Investigator: Lili Milani


Risk of comorbidity

Currently, if a patient is diagnosed with mental ill-health, we have no reliable way to predict whether or not they are likely to develop cardiovascular disease. And the same uncertainly exists the other way around, when cardiovascular disease is diagnosed first.


Researchers have identified some genes that appear to increase a person’s risk of developing a mental health condition. Similarly, some rare genes have been identified that increase a patient’s risk of developing type 2 diabetes and obesity.
Nevertheless, these existing genetic tools do not have the power to say when this might occur (age of onset) or what the outcome might be (life-time trajectory).

Medication side effects

In a further twist, we know that some mental health medications can increase a patient’s chance of developing cardiovascular disease in the future, but we currently have no way to predict which patients are most at risk.

Our Research

Work package 5 will make use of the tools and infrastructure created by work package 1, the prescription data from work package 2, and gene discoveries from work package 1 and 2, to develop their prediction tools.

A key aspect of the new CoMorMent dataset from a work package 5 perspective is the timing of each health event/diagnosis/prescription (epidemiology); e.g. how old are people at the point of first diagnosis? How long is the gap between the first condition being diagnosed and the comorbidity occurring?

Age of onset of disease

We have recently developed tools that can predict the likely age-of-onset for various diseases of the brain and cardiovascular system. These tools look for clues (biomarkers) in our genetics, our existing health conditions, our lifestyles and our scans (MRI).

We will adapt these tools so that they can be used to predict probable age-of-onset for cardiovascular disease in patients with severe mental health disorders (comorbidity).
A significant part of this adaptation will involve looking at the cardiovascular-related side effects of mental health medications (e.g. antipsychotics/SSRIs) and assessing what additional risks these may confer.

Multiple Gene Effects

We expect that a lot of our predictive power will come from analysing groups or networks of genes that act together and influence one-another. As the other work packages discover additional genes, so the accuracy of our model is likely to increase.

By studying many hundreds of genes and other biomarkers simultaneously, we hope to find patterns that will help us split people into different ‘treatment/risk groups’ (stratification).
This would help doctors to prescribe the most appropriate medications and recommend other interventions to improve future health (such as dietary interventions or physical activity).


Our methods are based on polygenic hazard scores (PHS) which takes into account of both genes and your current age. For example, in people who take anti-psychotics, the age-of-onset of cardiovascular disease may be lower than that of the general population (e.g. CVD at 45 rather than 55 years old).

Using novel machine learning tools, we will identify which genes put people with mental ill-health at additional risk of developing cardiovascular diseases, and critically when this is likely to occur (age-of-onset).

To examine the side effects of antipsychotic drugs we will study both the genes (GWAS) and the proteins that they produce. How do these proteins interact in the cell (protein-protein interaction algorithms)?


Validation is an important step. It is where we check that our results hold true, with a different sample of people.

  • To train our machine learning tools we will use a Nordic sample.
  • To test that our model is accurate, we will see if it can accurately predict what happened next in a clinical sample (The model will only be fed the early data, but the outcomes have already occurred, so prediction accuracy can be assessed)
  • Once validated, we will consider how to move the tool from the research lab to the doctor’s surgery. 


1. Desikan, R. S., Fan, C. C., Wang, Y., Schork, A. J., Cabral, H. J., Cupples, L. A., ... & Chen, C. H. (2017). Genetic assessment of age-associated Alzheimer disease risk: Development and validation of a polygenic hazard score. PLoS medicine14(3).

2. Khera, A. V., Chaffin, M., Aragam, K. G., Haas, M. E., Roselli, C., Choi, S. H., ... & Kathiresan, S. (2018). Genome-wide polygenic scores for common diseases identify individuals with risk equivalent to monogenic mutations. Nature genetics50(9), 1219-1224.

Useful links 

Institute of Genomics, University of Tartu 

Psychiatric Genomics Consortium 


Collaboration for sensitive data

Staff involved

Kelli Lehto

Kristi Krebs

Tuuli Jürgenson

Hanna Maria Kariis

Published May 15, 2020 5:45 PM - Last modified June 8, 2020 4:03 PM