Reliable research with virtual patients
The right medicine in the right dosage for every patient, that is something medical science aspires. 'Personalised medicine' is the term for that. But it does require a lot of research with patient data and that can be hard to get due to privacy legislation. Researchers Laura Zwep and Coen van Hasselt have found a solution.
Zwep and Van Hasselt, both working at the Leiden Academic Centre for Drug Research (LACDR), are studying the use of statistical models called copulas. After a copula model is 'trained' to describe individual, privacy-sensitive patient data, it can be used to create virtual patients. Researchers can then freely use and share the data of these ‘patients’.
‘With a copula, you can very well describe the distribution and interrelation of patient characteristics,’ say Zwep and Van Hasselt. ‘This is important because there are often complex relationships between characteristics such as age and body weight and the function of the liver, kidneys and concentrations of various molecules in the blood for example. With a copula, we can map these relationships and then use them to generate new virtual patients.’
Simulating datasets without protected patient data
To develop better treatment methods, you first need a lot of data from patients. After all, the same drug does not have the same effect on everyone. There is a difference between men and women, but the ideal dose also differs per person. This has to do with age and body weight or other ailments, for instance. Pathways to obtain such information from hospitals are complicated, especially if you do not work there yourself. Moreover, the use of this patient data is subject to many regulations. Research into optimising doses and discovering new relationships would go much faster if researchers had easier access to medical data.
'Virtual patients behave as complexly as real people.'
With the new, virtual data from copulas, researchers can simulate new datasets without the need for any of the original, protected patient data. ‘The real data remain properly stored in the hospital, but the so-called ‘probability distribution’ remains the same. With a copula, researchers can work systematically and the outcomes of the studies are identical because the virtual data are statistically exactly the same.’
Open science: create your own virtual patients
The Netherlands Organisation for Scientific Research (NWO) awarded an Open Science grant to this research project. Zwep and Van Hasselt want to use it to train hospitals to convert their databases into anonymous copulas, which can then be used for further research.
‘The grant for this project allows us to continue developing software and training materials to enable the use of copulas in an open science environment. We want to train hospitals and other data owners to easily create their own copulas that others can then use for further research. By creating infinitely many virtual patients who behave just as complexly as real people, we can take science another step forward.’
Text: Willemien Timmers
Image: Pexels