Digital intelligence blog

Pharma insight on digital marketing, social media, mobile apps, online video, websites and interactive healthcare tools

Artificial intelligence in healthcare

What are the benefits for pharma?

Tech

There seem to be almost endless ways artificial intelligence (AI) could transform healthcare, such as preventing disease and improving diagnoses and treatment, but for pharma there is a very real, near-term issue to solve – the escalating price and dwindling returns on R&D.

The AI revolution is being driven by massive leaps forward in computing power and an increase in access to health-related data sets that have made AI a powerful mining tool, able to identify patterns and relationships between data points that might never be identified by human researchers.

For pharma, much of the near-term excitement about AI stems from its potential to improve drug discovery and development, not least because using the technology promises to liberate researchers from the constraints of ‘perceived wisdom’ based on prior experience and expectations, unlocking new treatment pathways.

According to a Deloitte report, AI ‘can help analyse large data sets from sources such as clinical trials, health records, genetic profiles and preclinical studies; within this data, it can recognise patterns and trends and develop hypotheses at a much faster rate than researchers alone.’

Deloitte also suggests AI can help improve study design and decision-making in clinical trials, improving recruitment and enrolment, adherence and real-time clinical trial monitoring, which are all expensive parts of the drug development process. And applying AI to real-world evidence – another hot topic in pharma – could help companies predict performance of certain trial sites, anticipate drop outs and even predict outcomes.

With the cost of bringing a new drug to market now approaching $2bn – mainly due to a high attrition rate – anything that can improve the efficiency of R&D is being welcomed by the pharma industry. Pfizer is using IBM’s Watson machine-learning platform to help find new immuno-oncology targets and drugs, for example, while Sanofi is using a rival system from UK firm Exscientia in a metabolic disease programme. Roche’s Genentech unit is collaborating with GNS Healthcare on a cancer project, and GlaxoSmithKline recently signed an AI-based drug discovery alliance with Cloud Pharmaceuticals.

Investments like these are helping to drive explosive growth in the health AI sector. Recent market research from Global Market Insights (GMI) suggests that from a fairly modest level of $760m in 2016, the market will surge in the following eight years, growing at more than 40% per annum to top $10bn in 2024.

Pharma researchers are using the technology more and more for labour-intensive tasks like target identification, drug discovery and design and compound screening. And the technology is allowing the discovery process to be turned on its head, using patient-driven biological data to uncover new targets, rather than relying on the conventional trial-and-error approach.

The first AI-developed drugs are already starting clinical testing. For instance, a US biotech called Berg is comparing the profiles of cells in healthy and disease states to discover new drug targets based on improved understanding of diseases, including cancer. Phase 1 results with its lead candidate BPM 31510 were reported at this year’s American Society of Clinical Oncology (ASCO) annual meeting.

NHS data a key resource

In Europe, the UK has emerged as something of a hot spot in AI, largely because of the potential within the huge quantity of NHS health data, and spurred in part by the government selecting AI as one of four ‘Grand Challenges’ in its industrial strategy.

Access to centralised data – sometimes spanning clinical, genomic, social and environmental and behavioural factors offers the opportunity to identify at-risk individuals before they develop full-blown disease, and reduce healthcare costs. That’s an approach already being explored by IBM’s Watson for Health and Google’s DeepMind Health in partnership with pharma companies and clinical institutions, along with tools to improve clinical decision-making in order to provide more efficient, cheaper care.

The UK is home to some of the industry’s biggest names like DeepMind, SwiftKey and Babylon, as well as a large number of emerging start-ups. Anxious to maintain its lead in the face of emerging competition in the US, Israel and South Korea, among others, the government has agreed a £1bn deal with industry to advance the technology as well as a £7bn R&D budget funding around a 1,000 AI-related PhDs.

One project benefiting from that investment is a new AI facility that opened at the Rosalind Franklin Institute (RFI) in Harwell, Oxfordshire in June. It is investigating how AI can be combined with robotics to tease out new biological pathways and discover new diagnostics and drugs. The ‘hands-free’ technology will allow hundreds of thousands of molecules to be investigated at once, as well as the direct observation of interactions between drug candidates and their targets, and could speed up the process of drug development tenfold, according to the RFI.

The number of start-ups pledging to deploy AI in healthcare is far too large to cover in a relatively short article, but it’s a measure of the rapid maturing of the sector that some are now achieving impressive private fundraising rounds and even public listings on stock exchanges.

Just a few weeks ago, for example, UK-based Sensyne Health made its debut on the AIM, raising £60m to advance a business model based on sifting through anonymised NHS data to find clinical leads and guide the development of new drugs. The company has already forged agreements with a handful of NHS Trusts and says it intends to act as a bridge between the pharma industry and NHS to ‘unlock the value’ of patient data.

Meanwhile, another UK player – BenevolentAI – raised $115m in a private round in April that gave it a valuation of more than $2bn and additional cash to advance the more than 20 drug candidate programmes it has in development, including a Johnson & Johnson compound that was dropped for attention-deficit hyperactivity disorder but, thanks to the company’s AI screening programme, could be reborn as an insomnia therapy. The company’s CEO, Ken Mulvany, has said he
thinks the AI platform can cut early-stage drug discovery by four years and potentially deliver efficiencies in the entire drug development process of 60% against pharma industry averages.

The benefits may be large, but there are some fundamental issues threatening to hold back the deployment of AI in the drug industry, not least access to funding for new companies, the capital investments in IT infrastructure needed to handle enormous quantities of data, and being able to hire top-flight expertise – all of which are problematic, particularly for smaller companies.

Moreover, a recent KPMG report looking at the UK’s position in health-related AI concluded that while access to NHS data could be a competitive advantage for the nation, concern among the public on sharing health data with third parties is high – with just 15% willing to share data with pharma. It also pointed to a digital skills gap in the UK workforce that needs to be addressed.

The potential pitfalls of tapping into public data were exemplified last year when an independent panel set up to oversee the activities of Google’s DeepMind ruled that an agreement with the Royal Free NHS Foundation Trust for a kidney disease app was illegal – because it did not sufficiently safeguard data from 1.6 million people that was shared with the company.

Beyond R&D

Looking at the broader healthcare arena, AI is already being used to check data to discover patterns that can be used to improve analyses and, ultimately, provide better care for patients while also reducing costs. One example is a Novartis project with Watson to combine AI with real-world data to provide better insights on the expected outcomes of breast cancer treatment options.

Another is Google-backed company Forward, based in California, which launched a service using AI and connected tools such as body scanners last year that promises to keep tabs on its members’ health and wellness – at a cost of around $150 per month – in combination with bricks-and-mortar clinics.

AI can also help people stay well so they don’t need a doctor, via apps – sometimes linked to digital wearables – that encourage healthy lifestyles and help doctors understand the day-to-day patterns and needs of their patients. AI is already being used to identify women whose Twitter posts indicate may have increased risk of developing ovarian and cervical cancer, and earlier this year the UK launched a project with clinical centres, pharma and medical charities to develop algorithms that can mine patient data and lifestyle information to warn GPs when a patient should be referred to an oncologist. The aim is to spot at least 50,000 cases of prostate, ovarian, lung or bowel cancer earlier each year.

In another development, start-up company Medopad has developed a platform for creating modular apps that can be applied across multiple disease states and is already deploying AI to map patients’ progress in selected diseases such as rare paediatric cancers and spot early signs of deterioration. In January, the company signed a £100m deal with China’s Tencent focused on developing AI algorithms that could predict treatment complications.

Meanwhile, there is a growing body of research showing that AI can improve ophthalmic diagnoses and other procedures such as CT and MRI scans, with studies suggesting accuracy can be increased by 60-70% versus human operators. And last year, a Google AI system called Show and Tell was able to detect 90% of benign skin blemishes, outperforming dermatologists who were correct 76% of the time.

The next step is for AI systems to perform autonomously, delivering diagnoses without the need for human intervention. Among the front runners in this area is an AI system that scans patients at risk of diabetic retinopathy, which performed well in early clinical testing reported in the journal Nature Digital Medicine. The system – developed by IDx – was as reliable as traditional diagnostic imaging techniques in testing.

The ability of AI to decipher medical images has even led some to suggest that some professions such as radiologist might disappear altogether, although many believe human beings will remain indispensable as it will be many years before artificial intelligence is capable of the nuanced, contextual insights of experienced professionals.