Learn to adapt

There is increasing discussion about the potential uses for technologies like AI, automation and machine learning in the industry. However, this has primarily related to the supply chain, with less attention paid to their use within clinical trials. This is set to change over the coming years as the benefits of these tools become more well known.

Yuri Martina, vice-president of clinical operations at Grünenthal, is passionate about the use of machine learning in clinical trials. With more than 18 years’ experience in the industry and with an impressive 13 new drugs having been brought to market, his credentials are impressive. For Martina, there are two key applications of this technology: strategic input and operational input. Both rely upon predictive analytics, where past data is used to inform future decisions in a logical and systematic fashion.

“With machine learning and analytics, you can improve trial design by creating better development plans to include, for example, the right population, and achieve the end point and drug approval faster,” Martina says. “A series of strategic inputs translates into a more targeted trial design, which ultimately means more efficient and effective execution.”

This is not only beneficial for the company. “It also leads to more streamlined and logical clinical developments,” says Martina. This is an advantage for the patients and the payers because clearly the drug development time would be reduced, and the drug could be on the market sooner to serve the right patients.”

There is also huge value to be gained from exploring machine learning from an operational perspective. “Machine learning can also have an impact on cost and time in terms of site selection because, clearly, you could use past performance and other data available as a measurement of predicting future performance,” explains Martina. “This can improve site selection and patient recruitment.”

This could be particularly valuable when attempting to access hard-to-reach patient populations. “Often patients have limited knowledge and access to trials, and machine learning could help connect them to the right clinical trial,” explains Martina.

Machine learning can also be used to predict what will happen over the course of a trial. “You could also go down specific applications, like predicting where a patient is most likely to drop out and, ultimately, increasing the quality of your trial,” says Martina. “The aim is to have better data, faster data, more reliable data and being able to speed up your clinical development to serve patients best.”

Sifting through the information

Although this sounds straightforward, in practice it is a lot more complicated. One of the reasons for this is the wealth of data that is available to be analysed in clinical trials. “There is a lot of talk about big data,” says Martina. “The only way to make sense of a large amount of data is machine learning and analytics. You will not be able to do this manually.”

This volume of data is also increasing, further heightening the challenge of managing it. “According to estimates, we produce as much data per minute today as in the past 20 years put together,” says Martina. “As you can imagine, that is a lot of data to work with.”

Central to using machine learning effectively is having a clear goal right from the outset. “The key is to ask the right questions, and then apply good machine learning and analytics to answer those questions.”

It is tempting to think of these technologies as being somewhat ‘magical’ in their ability to make sense of data, but they require direction to be effective. “Machine learning will not provide an answer if you don’t have the right question,” says Martina. “You also need to be able to translate that question into something that can be used by analytics. It’s a multistep process.”

Although there is a lot that can be analysed with machine learning, more data isn’t always better. “The more data you have, the more conclusions you can make, usually,” explains Martina. “It needs to be categorised, organised in a certain manner in order for machine learning or analytics to make sense of it. Just having unstructured or unrelated data does not necessarily bring you anything new or any added value.”

Knowing what data to use is the central challenge of using technologies such as machine learning. “Looking at everything doesn’t necessarily bring you an advantage and it might be very expensive,” says Martina. “Furthermore, the technology needs to be trained with the proper data set. Otherwise, it will not give you the advantage that you were originally seeking.” There are a number of best practices to maximise the usefulness of machine learning. After establishing the question, this involves ensuring that the right individuals are recruited to different tasks. “Scientists can be very creative but are not necessarily business-minded,” says Martina. “Many companies will have a business translator, which makes that question into a use case, and that is then applied through machine learning or other analytics to get an answer.”

Never be afraid to look around – the data is out there

Effectively using the technology also involves thinking outside the box when it comes to data. “Don’t limit yourself to the data that you have internally,” says Martina. “There is a wider array of external data that is either freely available or that can be purchased.” This might involve investigating the competitiveness of a particular drug compared with another, for example.

Collaboration is also key, which might involve looking outside the company. “Don’t do it all by yourself,” warns Martina. “That is self-limiting. There are some good examples of companies that have successfully used crowd thinking or have leveraged external experts in terms of machine learning and analytics.”

Although these technologies can save time in the long run, it is important not to rush the process. “Be patient,” says Martina. “The technology requires time to be trained. It might sound strange because it is an analytic tool, but it needs time to become effective. You need to run a certain amount of data through it in order to make it a useful tool for generating predictive information.”

It’s also important to be realistic about what they can offer. “We cannot answer everything with machine learning,” says Martina. “So focus on the scope.”

It also needs supervision to be effective and cannot simply be left to run of its own accord.

“It is important to state that the starting point and the interpretation of results are still managed by human intervention,” says Martina. “As I mentioned before, you need a good question to start with and a good interpretation of what has been found.”

Looking to the future, machine learning and related technologies look set to be increasingly used, which offers numerous benefits, not just to those in the industry. “I predict better, more useful design, particularly when shifting the paradigm of healthcare or medicine to a more personalised approach,” says Martina. “A better execution means delivering medicine to patients faster and at a better value for the whole healthcare system.”