Machine learning’s impact on clinical trials

Machine learning is starting to show its potential in multiple fields. According to Gareth Baxendale, head of technology for the NIHR Clinical Research Network, clinical trials are no exception.

Could machine learning be used to enhance or improve the success of a clinical trial? There is much excitement around machine learning and the opportunities it can provide, and it’s already beginning to permeate every area of our digital life. In fact, you’re likely to make use of machine learning almost every day without even knowing it.

For example, Amazon’s Echo and Apple’s Siri use machine learning for speech recognition. Google’s image search, meanwhile, uses machine learning to ‘understand’ the components that make up a picture. It does a pretty good job too, spawning the meme ‘Chihuahua or blueberry muffin?’ Feel free to Google it.

‘New knowledge’

Machine learning is a branch of the more commonly understood field of artificial intelligence, the preserve of many Hollywood ‘rise-of-the-machines’ dystopian movie story lines. In short, artificial intelligence attempts to mimic human intelligence or behaviours. Machine learning, on the other hand, attempts to analyse, map and associate ‘patterns’ and ‘behaviours’ in multiple data sets. In so doing, it supports intelligent, data-driven, decision making based on ‘new’ knowledge and understanding.

It’s this ‘new’ knowledge that is the exciting part. It could be used to predict, for example, a patient’s diagnosis, best course of treatment, or even their level of risk. And it has the added advantage of potentially reducing human error.

Given the obvious possible benefits, it’s well worth exploring the opportunities machine learning has to offer – including to clinical trials.

The clinical trial

At its heart, a clinical trial is a set of questions that need answering to determine the efficacy and safety of a particular biomedical, pharmaceutical or behavioural intervention.

Some trials are focused on developing new treatments, some consider new combinations, others apply existing treatments in a different therapy area. Many undertake comprehensive reviews of existing treatments, considering their longer term efficacy and safety.

Significant amounts of data will be collected during a trial so as to provide robust and reliable answers to the questions posed. And herein lies the first opportunity for the application of machine learning in this field.

Trial data

During a clinical trial, various data sets will be generated and collected by the investigator and their study staff. The patient may also generate data, by filling in a questionnaire, maintaining a diary or using a custom app. Information collected could include results from ECGs, MRIs or blood tests. Machine learning can be applied to this data to surface ‘new’ information that otherwise may not be found.

Take for example BERG Health, a Boston-based biopharma company. They are using a machine learning platform they’ve named Interrogative Biology, which allows them to identify biomarkers for drug discovery and monitor patient responses during a clinical trial. They state: “We can build models with the platform using the patient’s own biology in order to stratify the population by response to the trial drug as well as monitor patient response over time at a biological level, which may lead to more successful trials.”

In the area of cancer trials, UK firm ImageAnalysis are taking advantage of machine learning to analyse thousands of images. It is hoped this analysis can then be used to identify or predict early signs of cancer and potentially personalise a course of treatment for the patient.

Recently the NIHR School for Primary Care Research carried out a study that used electronic medical records from 378,000 patients in general practices across England, taken from the Clinical Practice Research Datalink (CPRD). Data on key risk factors, such as smoking status and blood pressure, was used to develop and test four different machine learning algorithms for predicting cardiovascular risk.

The report suggests that “these algorithms were better than existing medical risk models at both predicting the number of people who would develop cardiovascular disease and excluding people who would not get heart problems.”

Recruitment

Another key area for clinical trials is recruitment and the identification of suitable and willing patients to participate and complete the study.

Cincinnati Children’s Hospital Medical Center is using machine learning to understand why people accept or decline an invitation to participate in a clinical trial. Recruiting sufficient numbers of participants to answer the research question is a challenge in medical studies. In their research, 60% of patients approached with traditional recruitment methods agreed to participate. Researchers are predicting that their new automated algorithm could help push acceptance levels up to about 72%.

Publication

When a clinical trial is completed the outcomes of the trial are published. Even in this final activity machine learning may be able to help.

King’s College London is running a machine learning project called Robot Reviewer. Funded by the Medical Research Council (MRC) and others, the aim is to develop a system that will automate bias assessment in systematic reviews. These syntheses will enable decision makers to consider the entirety of the relevant published evidence.

The future

It will likely be some time before humans would take an unsupervised approach to decision making based on machine learning. However it’s almost certain that, in the near future, all medical diagnostics, monitoring and treatment plans will come in part from knowledge derived from and recommended by machine learning platforms. It’s also the case that many are already confident such platforms offer very real and tangible applications for clinical trials – which can only be a good thing for health research and patients the world over.