Are we on the cusp of an AI revolution?
pharmafile | April 27, 2021 | Feature | Business Services, Manufacturing and Production, Medical Communications, Research and Development, Sales and Marketing |
The pharmaceutical industry has historically been slow to adapt to change, with high regulatory hurdles and widespread scepticism facing new technologies attempting to make waves in pharma. But, has a year battling a pandemic opened the door to change? Kat Jenkins takes a look at the AI companies hoping to revolutionise the healthcare landscape
|
Last year the COVID-19 pandemic hit the world like a wrecking ball, disrupting traditional working models across all areas of the healthcare industry. Companies suffered financial losses, hospitals faced staffing shortages while struggling with an immense influx of patients; all while the pharmaceutical industry worked tirelessly to create and supply an arsenal of vaccines to battle the virus and its subsequent variants. |
As often happens in times of great turbulence, development trends have rapidly been brought forward, this has notably been seen in the healthcare industry. The digital landscape has particularly been impacted, with NHS app registrations up 111% this year, the creation of online infusion-site locators in the US, and the introduction of virtual clinical trials across the globe, being just a few examples. It is evident that both professionals and patients are ready for a technological overhaul, and many believe artificial intelligence (AI) will be at the forefront of this.
Looking at recent developments in the field of AI, this is hard to dispute. A GlobalData survey in 2020 predicted that AI will be the most popular form of technology to influence medicine in 2021, and found that 90% of large pharma companies initiated AI projects in 2020 – including all of the top ten drug makers in the world. Almost 100 partnerships have been struck between AI specialists and pharma companies since 2015, and last September GSK opened a £10 million AI research base in Kings Cross, London. GSK, alongside AstraZeneca, also committed to a five-year partnership with Cambridge University to fund the Cambridge Centre for AI in Medicine. The centre has set its sights on the advancement of clinical trials, personalised medicines, and drug discovery.
AI in the Healthcare Industry
In short, AI uses computer programmes to do things that traditionally require human intelligence. This means creating algorithms to classify, analyse, and draw predictions from data. It also involves acting on data, learning from new data, and evolving over time. Just like we humans do as we grow up, AI takes in data given and learns from it, working at a rapid pace whilst constantly improving its understanding. AI can work with vast data and has hundreds of potential applications.
Pharmafocus tracked down three professionals working at the forefront of AI in healthcare: Simon Coulthard of Pangea Data, Amanda Schierz of DataRobot, and Andrew Hopkins of Exscientia.
At Pangea Data, the main focus is upon utilising unstructured, textual data to help improve diagnosis for clinicians, among many other things. Unstructured data, such as doctors’ notes, make up 80% of healthcare data and is crucial for patient information, but this has so far been overlooked by pharma and healthcare companies. Their focus has traditionally been upon structured data, such as International Classification of Disease codes, gender, and age.
Speaking to Pharmafocus, Coulthard, Senior Financial Officer of Pangea Data, explained how their AI programme has been working with unstructured data: “What the software is trying to do is to improve patient outcomes through looking through the vast and complex datasets that are predominately unstructured data. 80% of the records are actually freehand notes, written by the doctor or somebody at the hospital, in the discharge notes, for example, and that’s obviously a lot harder to spot patterns in.
“Machine learning and artificial intelligence has to be able to identify and separate out all these different complex elements, such as the way that individuals write – and everybody obviously writes in a different fashion. What we’ve got is the machine learning working its way through all of these unstructured elements to access previously unavailable information.”
Mining unstructured data is a complex process, but without it there is an increased risk of misdiagnosis and significantly reduced success rates at screenings and in clinical trials. Pangea’s algorithms can mine unstructured data 50% faster than the traditional keyword and neuro-linguistic programming approaches, with an accuracy of 87.5%. From this data, Pangea have created a library of AI disease models that have been used by scientists, biopharma companies, as well as clinicians.
Another key application of unstructured data mining AI is the identification of participants for clinical trials. Coulthard explains: “Often trials fail because they just can’t find sufficient cohorts, especially as we’re moving to looking at rare diseases. By definition it means that there are very few people that are diagnosed because the rarer a disease is, the harder it is to diagnose, so you actually find it a smaller percentage of that small percentage.
“If we can run through and find [the cohorts] 80% faster than a manual process with greater accuracy, then fewer trials fail and each trial can have more patients or can get to its cohorts, much more quickly.”
More Than Mining
AI has also been developed as a scalable digital workforce, helping to streamline processes that don’t require human interaction. This allows clinicians to focus on higher value, human-led decision making, diagnosis, and treatment. DataRobot began as a vendor of automated machine learning, extracting important intelligence from datasets, and has expanded its work into three main areas: medical imaging, patient management, and patient treatments.
Speaking to Pharmafocus Amanda Schierz, Principal Data Scientist at DataRobot, said: “Any of us who have had an X-ray, ultrasound, or MRI scan know that it requires an expert to translate the imaging results into a diagnosis. This process can take quite a few weeks and patients can get increasingly nervous during this time. DataRobot works closely with the radiology experts to classify each image with a medical diagnosis and acts as a triage and decision support tool which helps to reduce the turnaround time for the patient and frees up the radiologist to treat more patients.”
The company’s AI technology is also currently being used by the NHS to help predict demand in their A&E departments, Schierz explains: “DataRobot has been forecasting A&E demand based on historical A&E data to ensure all the necessary resources, including ambulances, and staff are available at the busiest times. The same historical data is used to predict whether a patient should be admitted or not and, if so, their predicted length of stay.
“For hospital in-patients, the risk of discharging a patient too early, which could lead to the patient being readmitted again in the near future, can be averted by predicting if the patient is at risk of being readmitted before the decision is made to discharge the patient.”
The work carried out here by DataRobot has a ripple effect that can encompass the entire healthcare industry. With increased automation in labs, more data can be linked back into manufacturing and other centralised data repositories, improving visibility of trends which, in turn, aid scale manufacturing and increase the agility of supply chains.
Stepping up Drug Discovery
DataRobot is also involved in the drug manufacturing process, Schierz comments: “The speed and productivity of machine learning, and more specifically automated machine learning, align with the drug discovery philosophy of ‘fail fast’ and helps to reduce the number of failures further down the pipeline.
“With public interest focusing on – and becoming more knowledgeable on – the drug discovery process in relation to COVID-19, it’s time for companies to rely more on machine learning where they can both fail fast and innovate, rather than putting all their resources and money into one metaphorical basket.”
At the forefront of this is Exscientia, the first company in the world to submit an AI-designed molecule for human trials. Typically, it takes around five years to select a molecule successfully and get it to human trials – Exscientia did this in just 12 months. Pharmafocus spoke to Andrew Hopkins, CEO and founder of Exscientia, about his work. He said: “It is Exscientia’s ambition to find a quicker, smarter, more precise way of discovering and developing new drugs so that we can meet the medical needs of more people living longer lives.
“Drug design is precision engineering at the molecular scale. The position of every atom in a drug molecule determines how the drug is distributed around the body and what proteins it interacts with. Whether a molecule becomes a transformative new medicine is determined by its molecular design.
“The goal of AI design is to improve the probability of success of the drug in the clinic, at its moment of creation and at the design stage, by bringing to bear all the data in the form of machine learning models into the design of the drug.”
A mere one in 1,000 molecules identified as potential treatments make it to human trials, and 90% of chemical candidates are eliminated in trials over an arduous five-year process. Usually, it takes over a decade to get a product to market, tallying up costs of around $2.16 billion in R&D in that time. This is why AI in drug development is arguably the most lucrative area of the technology.
Hopkins has set his sights for Exscientia even higher, with hopes of moving from predicting effective molecules to predicting drug toxicity in humans. This would further speed up the clinical trial process, as toxicity can often surprise researchers and halt trials.
A Brave New World
So, what is in store for the future of AI in the healthcare industry? Coulthard believes AI will become more common but may face regulatory pushback: “I think prevalence will definitely increase, I think the question mark will be about the speed and how it will fall into the regulatory framework.
“If you’re starting to move towards where you’re actually claiming a diagnosis, then they’ll definitely be quite a hard regulatory hurdle that you’ll have to overcome. If it stays on the sort of clinical support side, and I think it will accelerate quite quickly.”
Amanda Schierz also sees a bright future for AI in healthcare, she said: “My biggest hope for AI in healthcare is in early diagnosis systems and in increasing the efficacy of cheap, easily available tests like lateral flow assays or pinprick tests. The area of bioinformatics uses AI models to discover blood and saliva biomarkers of diseases, including cancer. Combining the AI model with diagnostic tests could provide a cheap, over-the-counter, and reliable test for the early diagnosis of several diseases.”
Hopkins believes AI in drug development will become standard industry practice in the future, commenting: “By the end of this decade, we expect all drugs entering into the clinic will be designed through AI. With these benefits of efficiency and precision, we can bring innovative breakthroughs more rapidly to the patient.”
Through data mining, automation, or precision engineering, there is a wealth of untapped potential in the realm of AI for the healthcare industry. The question then remains, not of how the industry can benefit from this technology, but when will it be ready for it?
