Bayer, Aalto and HUS expand collaboration – artificial intelligence to support clinical drug trials

Drug development is a laborious and time-consuming process. Developing a new drug takes 10 years on average, with the share of clinical research amounting to even seven years. The development of a prescription drug takes an average of two billion euros.

Clinical drug trials may also involve some ethical issues, as one group invited to trials receives a new drug and the other a placebo, or the best treatment currently known. This is a problem, especially in rare diseases with no established, effective medication.

‘The solution could be a virtual control group shaped based on medical databases using artificial intelligence. This way, the control group does not need to recruit patients, which also increases the cost-efficiency of drug development,’ says Professor Harri Lähdesmäki from Aalto University and the Finnish Center for Artificial Intelligence (FCAI).

AI methods are also being explored as a means to support the quality and reliability of clinical trials. In the best scenario, this research can improve drug trial practices worldwide.

‘The goal is to have AI-based algorithms identify even rare signals related to the adverse effects of drugs with increased sensitivity. We are also looking for possibilities to reduce the number of patients needed for clinical trials while ensuring the safety and reliability of the trials,’ says Jussi Leinonen, Principal Clinical Data Scientist at Bayer.

Finland as number one in health databases

The international Future Clinical Trials project is led by Bayer Finland’s clinical development and operations unit. The three-year project supported by Business Finland seeks to build an innovation ecosystem of companies, universities and representatives of the health sector, making Finland a top country in clinical research.

‘Clinical research has been advanced on a relatively small scale in the Nordics. Especially Finland has faced the challenge of having a small population. Increasing the utilisation of health data with the help of algorithm development and AI solutions would make up for this and mark a significant victory for research. Finland has high-quality health data and leading practices in sharing it,’ Leinonen says. 

As part of FCAI, Aalto University develops statistical machine learning models in projects to model health data collected about patients during different periods. The problem with analysing the patient group and control group consisting of health data is the comparability of the measurement data. The data may overlap only partially and have different measured variables.

‘The statistical harmonisation of different datasets requires a new probabilistic machine learning machinery that enables reliable comparison of the efficiency of a new drug between datasets,’ Lähdesmäki points out.

Many techniques with AI applications are known to work when supported by massive measurement datasets. Yet measurement datasets related to health and medicine are often limited. ‘Artificial intelligence methods must be developed so that they offer reliable conclusions, also in analyses made on limited patient numbers. On the other hand, artificial intelligence methods should be able to tell a user if they are no longer reliable,’ Lähdesmäki continues.

Open publication at the core

The starting point for clinical research is openness. Close collaboration with and surveillance by relevant authorities require that all processes can be audited and verified. The openness of research is also important for AI researchers at FCAI.

‘Aalto University’s top expertise is exemplified by prestigious scientific publications. Open publication and the openness of algorithms’ software implementations guarantee reliable research from both the patients' and citizens' perspectives. The research community’s first publications related to AI methods have already been submitted for publication,’ Leinonen says.

The collaboration between Bayer and Aalto will continue in the research starting in February as AI methods and algorithms are applied in collaboration with the Helsinki University Hospital (HUS) to its patient data.

‘Combining the real-world data and clinical research data involves several challenges. Doing this manually is very laborious. With AI, it can be done much faster, more efficiently and also more reliably. A tripartite, AI focused research partnership between a pharmaceutical company, university and hospital is unique, even on a global scale,’ Leinonen notes.

Utilisation of AI on the rise

Expectations in clinical research regarding the use of AI are high. One of FCAI’s current research areas, in addition to those mentioned before, is the development of AI methods to find molecules suitable for drug ingredients. ‘Researchers are trying to find a solution for having to go through millions of different drug molecules experimentally and instead have a machine learning method to suggest certain molecules worth exploring more deeply in a laboratory setting. This would enable drug development to be done faster and with less resources,’ Lähdesmäki states. 

Analysing large data masses and utilising the data in decision-making will increase. Data can be used on the individual level in healthcare as well as in the effective targeting of healthcare resources.

‘Artificial intelligence will not replace humans in clinical trials.  AI in itself, in the near future, will not make independent decisions but recognises situations that require a reaction or guides reactions in a certain direction. The same applies to AI used by doctors in patient work,’ Leinonen says.

Text: Marjukka Puolakka