Using genetic variants as a treatment decision aid for the optimization of antipsychotic treatments: a critical appraisal of the literature.

Antipsychotic drugs are widely used to treat schizophrenia. However, they only work in about 60% of patients, and many people who take these drugs suffer from side-effects. These range from mild ones, like headaches and dizziness, to serious and life-threatening. For example, the antipsychotic clozapine carries a risk of myocarditis, a heart condition that can be fatal.

Presently, it is very hard to tell which patients will suffer from which side-effects for a particular drug. Recent research has focussed on trying to find things we can measure within individuals to predict how they will react to a drug. These things are known as 'biomarkers' which can be genetic mutations, proteins in the blood, or other measurements.

In order to prove there is an association between a biomarker, a drug, and its side-effects, there needs to be a lot of high-quality evidence. The problem is that there are many different antipsychotic drugs, with hundreds of known side-effects. Some of these associations have been shown in small studies, but to really understand these findings and ensure they are reliable, it is necessary to combine these data in larger analyses. These analyses involve methods called 'systematic reviews' and 'meta-analyses' and our group at the University of Liverpool are very experienced with these methods, in other areas of medicine. Meanwhile, the group in Canada has done lots of work on specific genetic mutations linked to antipsychotic drug responses.

While these larger analyses are needed, they are time-consuming and labour-intensive. They need multiple researchers to locate, assess, and analyse hundreds (sometimes thousands) of scientific papers. A new approach has been pioneered by a group in the USA using machine learning to automate this process. This is very new, and while it is potentially useful, we need to check and make sure it works as well as the traditional approaches. That will be the main goal of this proposed internship. We will compare the results of the machine learning approach to the traditional literature searching approach for identifying biomarkers of antipsychotic drug response. This will help us evaluate the machine learning method, as well as provide a comprehensive overview of the field. We will then examine the evidence produced and evaluate how reliable it is by using formal criteria drawn up for this purpose. By doing this, we hope that we can identify biomarkers capable of predicting drug response, with strong and reliable evidence behind them. We will then be able to make recommendations for these biomarkers to be used in clinical practice.

This project utilises the strengths of both the University of Liverpool and the Canadian group. It will facilitate knowledge transfer between the countries and create ongoing and future collaboration opportunities.