Transforming Clinical Treatment Paradigms through AI & Molecular Profiling
This year, over 1.7 million individuals will be diagnosed with cancer in the United States. Add to this the estimated 8 million people living with and undergoing treatment for cancer, and it is perhaps unsurprising that oncology drug development is big business. Indeed, a best-selling blockbuster drug can be expected to generate $2.7Bn per year.
New drugs progress from the experimental stage to being approved medicines through a series of clinical trials. There are an estimated 18,000 clinical trials currently recruiting patients in the US — 1,000 of these are in breast cancer alone. Each can cost upwards of a billion dollars and takes an average of 7.5 years to complete.
And yet, according to the White House, only 3% of cancer patients ever enrol in a clinical trial. For some patients this is understandable: in certain jurisdictions, patients may only enrol in a drug trial when existing forms of treatments have already failed. On top of that, not all patients diagnosed with cancer are eligible to participate in any one trial.
But for some patients, the barriers to finding and enrolling in a suitable clinical study are artificial and completely avoidable. In some cases, for example, the patients themselves have to manually search the thousands of recruiting trials indexed in the jargon-heavy ClinicalTrials.gov. For patients who, when all else is equal, may do better when enrolled on a clinical trial than on standard care, this is an obstacle to getting the healthcare they deserve. For pharmaceutical companies, enrolment issues lead to a delay in 70% of trials; this costs an estimated $8m per day. When both sides of the market are underserved, disruption is the only course of action: the $65Bn clinical trial market needs to be dragged into the 21st century.
Several emerging technologies have the potential to streamline the clunky clinical trial process, from Internet of Things (IoT) for remote monitoring, to Machine Learning (ML) for Electronic Health Record (EHR) processing, to blockchain-based cybersecurity for data protection. In this article, we consider an evolving clinical trial space in oncology, and how Artificial Intelligence (AI) is making an impact here and elsewhere in the treatment of cancer.
AI & ML are (relatively) old technologies that have been boosted in recent years thanks to hardware improvements and increased access to cloud computing. Broadly, their aim is to mimic human cognitive functions in algorithms that ‘learn’, reinforce or iterate based on a cycle of feedback. Although shrouded in hype, these techniques have the potential to transform the clinical trial space and bring a paradigm shift to cancer care as a whole.
The potential for these technologies to disrupt all sorts of markets has led to some big investments. In April 2017, the European Commission announced a €20Bn AI strategy for Europe. France also launched its own €1.5Bn program, which was followed by the opening of new R&D facilities by companies like Fujitsu, Facebook and Google DeepMind.
Recognising the specific advances that could be seen in oncology, many government bodies have begun encouraging a convergence of AI and cancer research. In the UK, the BBSRC (Biotechnology and Biological Sciences Research Council) has recently awarded the University of Cambridge $5m, closely followed by the Mark Foundation’s $5m pledge to the same University. Cancer Research UK, the world’s largest independent funder of cancer research, even has a £20m “Grand Challenge” dedicated to using AI to fight cancer. Such investments are surely to encourage exciting developments at the interface of AI and Cancer Research.
Whilst many startups and big tech companies are striving for a true “generalised artificial intelligence” this isn’t likely to impact healthcare in the near term. IBM Watson is the closest we have come and has been marred by poor overall performance. ML is likely to have much more of a near term impact on healthcare. This group of technologies excel at finding trends and associations in large structured and unstructured datasets, making them perfect for making sense of healthcare data.
At an ever-increasing rate, oncologists are being swamped with data from imaging, genomics, co-morbidities and previous treatments. ML has the potential to crunch the data to predict the prognosis of the patient and, additionally, to advise doctors with different options available, including personalised medicine and clinical trials with experimental therapies.
Nowhere is the potential of ML greater than in the process of taking new cancer drugs to market. The current drug development pipeline is long and arduous, averaging 7.5 years, and costing anywhere between $161M and $2B per drug. On top of this, 90% of drugs entering Phase I clinical studies will fail by Phase III. The high costs associated with R&D ultimately affects healthcare prices for doctors and consumers. Development costs of unsuccessful drug programs are absorbed into the costs of successful drugs; an oft-mentioned reason for the high costs of cancer drugs.
For most patients, finding an eligible clinical trial is a herculean task and, even if they manage this, participating in the trial is cost- and time-intensive. Ultimately, AI and related technologies have the potential to improve the clinical trial and cancer therapy space for both patients and pharmaceutical companies. However, for this to become a reality, modernisation is key. Currently, clinical studies often rely on rudimentary and outdated methods for data collection and verification: for example, sending patient medical records via fax, manually counting leftover tablets in bottles, and relying on patients’ diary entries to determine adherence.
Precision oncology is the process of tailoring drug treatments to an individual’s specific cancer. In some cases, such as with precision oncology software from Cambridge Cancer Genomics, this extends to accounting for tumour heterogeneity and evolution so that not only the drug, but the specific treatment regimen, is tailored for maximum effectiveness. With a market size projected to reach $87 billion by 2023, precision oncology is likely to become an ever more important part of routine cancer care.
The potential benefits of precision oncology in both improving patient outcomes and reducing overall healthcare costs are substantial. However, a major bottleneck is that the field still lacks enough molecularly targeted drugs to treat all tumours. This shouldn’t be the case for long, according to Sam Natapoff, analyst at Bloomberg:
“drug development is made for AI applications.”
This opportunity has attracted large AI developers, big pharma and a huge number of startups. If the champions of these technologies are right, AI and ML will usher in an era of quicker, cheaper and more-effective drug discovery.
Some are sceptical, but most experts do expect these tools to become increasingly important. For example, pharmaceutical giant Pfizer is using IBM Watson to power its search for immuno-oncology drugs. Sanofi has signed a $300m deal to use UK start-up Exscientia’s AI platform to hunt for metabolic-disease therapies, and Roche subsidiary Genentech is using an AI system from GNS Healthcare to help drive the search for new cancer treatments.
Most sizeable biopharma players have similar collaborations or internal programmes. However, this field is still at a relatively early stage. Only the British company BenevolentAI, in partnership with Janssen, has shown concrete results, which have led to a drug candidate moving to a Phase II clinical trial.
Currently, there are two main barriers to greater implementation of personalised cancer medicine: high costs and technological limitations. Genomics is closely related to precision medicine, as sequencing technologies are needed to identify molecular driver events fuelling a tumours growth. For a long time, the cost of sequencing each tumour and analysing its genetic fingerprint was a major limiting factor in personalised oncology. Fortunately, the cost of sequencing a genome continues to drop year-over-year, far outpacing Moore’s Law, making a future when each patient is given truly personalised cancer care a very real possibility. As Roland Kanaar, Professor of Molecular Radiation Genetics at Erasmus University puts it:
“Technological advances have paved the way for a precision treatment tailored to both patient and tumour. DNA is the key to targeted anti-cancer therapies. We want to use our knowledge to prevent people from going into treatments they won’t respond to”.
The major barrier now, is in data analytics. To tackle the vast amount of patient data that must be collected and analysed, and to help cut down on costs, many researchers are implementing machine learning techniques to help truly personalise cancer treatments.
Companies like Deep Genomics use machine learning to help researchers interpret genetic variation. Specifically, algorithms are designed based on patterns identified in large genetic data sets which are then used to help interpret how genetic variation affects crucial cellular processes — including metabolism, DNA repair, and cell growth — that can cause diseases such as cancer.
Cambridge Cancer Genomics, a Y Combinator backed precision oncology startup, has built an interface for oncologists to integrate and analyse exactly what genomic features are driving a tumour’s growth. Predictive ML processes helps guide clinicians down the most effective therapeutic route whilst analysis of patient blood samples gives them feedback on whether treatment is working or not.
Matching the right trial with the right patient is a time-consuming and challenging process for both the clinical study team and the patient. Enrolment difficulties, in particular, account for approximately one-third of Phase III clinical study terminations and thus a huge amount of money, almost 30% of a Phase III trial’s budget, is spent on trial recruitment, to counter this. Roughly 80% of clinical trials fail to meet enrolment timelines, and approximately one-third of Phase III clinical study terminations are due to enrolment difficulties. Recruitment, as a whole, accounts for almost 30% of a Phase III trial’s budget. Delays don’t only impact the patient; for every day a blockbuster drug is delayed in clinical trials, the sponsoring company can say goodbye to $8m in lost revenues.
Patient recruitment is fundamentally a matching process. Clinical trials publish inclusion and exclusion criteria, this is compared with patient medical records and if there is a good match, the patient may be a candidate for trial inclusion. As you can imagine, significant time is typically required to manually evaluate available clinical trials and determine a patient’s eligibility for trials based on all the relevant information in the medical record. Manual screening for trial eligibility can take nearly two hours per patient.
Occasionally, patients may get trial recommendations from their doctors, but this tends to be when a physician is aware of an ongoing trial. Otherwise, the onus of scouring through ClinicalTrials.Gov — a comprehensive federal database of past and ongoing clinical trials — falls on the patient. Theoretically, this arduous process could be reduced to nearly instantaneous results when leveraging an ideal AI platform.
In practice this has proved more difficult than initially thought. In 2013, the M.D. Anderson Cancer Centre launched a $62M program with IBM Watson to speed up the process of matching patients with clinical trials. The program struggled with the unstructured nature of healthcare data and disparate data sources that didn’t communicate with each other — the program was ultimately deemed not to be cost-effective. One of the leaders in the field, Flatiron Health, explains:
“Structured data can become unstructured due to transmission methods. For example, a spreadsheet that is faxed or turned into a read-only document (such as PDF) loses much of its structure.”
California-based startup Mendel.ai tries to solve the challenge of piecing together a patient’s medical history by allowing cancer patients to submit their medical records to its platform. Alternatively, patients can give Mendel permission to collect all medical records from doctors on their behalf. The startup is developing ML based approaches to extract information from digital records and to match patients with ongoing trials that best suited their needs. However, there are challenges here also, as one study from John Hopkin’s highlights:
“Standard natural language processing tasks such as sentiment analysis and word sense disambiguation are difficult in clinical notes, which are misspelled, acronym-laden, and copy-paste heavy.”
Other startups, like Antidote.me, are using ML to simplify the jargon in the “inclusion/exclusion” criteria listed in trials on ClinicalTrials.Gov. On the B2B side, startups are now using deep learning and natural language processing to automate clinical trial matching by directly partnering with health institutions. For example, Deep 6 AI works with clients like Cedars-Sinai Medical Centre to improve the efficiency of finding and recruiting patients to clinical trials.
Despite the importance of ensuring that experimental medications are taken exactly as prescribed, many clinical studies still rely on patient self-reporting to ensure compliance. Patients are expected to note when they took the study drug, what other medications were taken on those days, and any adverse reactions (including headache, stomach ache, muscle aches). Relying on unverifiable sources (like patients’ memories and paper journals) is plagued with inefficiencies.
In an effort to reduce these inefficiencies and drag treatment compliance into the 21st century, pharmaceutical companies Pfizer and Novartis have been investing in IoT and “smart tablets” to track drug intake. These ingestible sensors can automatically report compliance directly to supervising clinicians who can remind patients to take medication, inform nursing staff or remove the patient from a study arm. Not to be left out, Merck Ventures has recently participated in a $14.5M Series B to Medisafe, a startup developing wireless pill bottles for a similar purpose.
Some startups are going one step further. New York-based AiCure uses facial recognition to track adherence. Patients use their phones to take a video of themselves swallowing a pill, and AiCure confirms that the right person took the right pill. Catalia Health, is developing a healthcare companion and coach using AI. This Khosla Ventures backed startup hopes to enforce behavioural changes in patients by asking specific questions and setting reminders. However, an AI assistant’s ability to successfully enforce lifestyle changes largely depends on patients’ willingness to interact with it on a daily basis.
Healthcare, as a whole, lacks in AI adoption. Despite the richness of the data generated by the healthcare industry, and the global importance of that data, several limitations in data recording are restricting the effectiveness of AI and ML.
AI adoption in the actual clinical trial process is still in its early stages. Compared to other areas of healthcare, fewer startups are directly targeting clients in the clinical trials space. In many aspects of clinical trials, there’s a need for digitisation that precedes the need for AI. As discussed, many parts of clinical trials and routine cancer care still rely on physical clinical notes and diaries. Diaries are stored digitally, often in difficult-to-search formats, while handwritten clinical notes pose unique challenges for natural language processing algorithms to extract information.
Several startups are experimenting with applications ranging from machine-assisted diagnostics to extracting information from electronic health records. In particular, using AI software to design new drugs has gained momentum, with pharma giant Merck partnering with startup Atomwise and GlaxoSmithKline partnering with Insilico Medicine, among others.
As oncologists are exposed to ever more sources of data on patient stratification, predicted outcome and potential therapies, they will have to develop skills more traditionally seen in data scientists. Integration of ML and AI technologies to help in the analysis of this data is inevitable for the future of cancer care. However, specific challenges in modernisation of healthcare practices and data collection need to be overcome before these technologies can reach their full potential for cancer patients.