12 - AI: Trials of the century

The ability to remotely collect big data from clinical trial participants, using wearables and other devices, promises to improve efficiency and success, while also providing a deeper understanding of new drug efficacy.

Decentralised trials can be designed around the volunteers, which removes some of the limitations pharma companies have faced, particularly when it comes to recruitment of a broad sample. It gives them a bigger pool of participants, from more diverse backgrounds, who can commit to the trials for longer, reducing the risk of dropouts. It also helps to overcome the challenge of finding suitable people to test medicines for rare diseases.

Using standard devices like fitness trackers and smartphones, it is possible to generate data continuously. This data taken alone, however, won’t enhance clinical trial methods – which though resource-intensive, have the advantage of a clinician being close at hand to rigorously monitor the drug’s effect within a strict regulatory framework.

To achieve not just comparable but better results you need to apply AI models to the streams of real-world data coming from the devices.

It enables you to see clearer, see further, think smarter and act faster. Potential issues would be identified quickly, and trials adapted, so that new drugs can be launched sooner and more safely. Another advantage is the ability to continue collecting data beyond approval.

You’d see whether it’s more or less effective in different subsets of the population, especially those who tend to be under-represented in clinical trials. It would also enable quick withdrawal of any drugs that had harmful side effects.

The BBC recently reported that genetic testing for drugs might be available on the NHS as early as next year, providing yet more data that could predict the impact of drugs on individual groups with similar genetic characteristics.

The big question is not whether we can apply AI to clinical trials data generated remotely (we clearly can!). Rather, it’s whether the AI models are robust and can satisfy the requirements of the regulators, while reassuring prescribers and patients.

While AI can identify patterns and answer questions humans can’t, it’s critical that pharma companies are able to explain the steps in the decision-making process.

They need to be able to demonstrate that the methods used will extract meaningful insights from a stream of data. Then they need to package them in a way that allows regulators to gain confidence in the legitimacy of the data.

Clinicians don’t have to be experts in the technology itself but as part of their due diligence, the onus is on them to choose a partner who gives them control of the data and models.

It’s this illumination of AI that will bridge the gap between where we are now to a near-future state, where we have a clear understanding of the data streaming from multiple devices.

Finally, it’s about recognising that the models themselves can be continually improved. This is possible through the use of ‘challenger models’ that run in parallel, meaning clinical trial leaders can be agile in switching to another model if it proves more capable of optimising analysis.

Trials of the century

The I in AI

Convincing case