Bringing a drug to market is no easy feat, and that’s putting it mildly. Traditional clinical trials are lengthy — taking nearly ten years on average,1 expensive — with an average cost of around $1 billion,1 and inefficient — with around a 10% success rate.1 Despite the level of scientific skill exhibited by some of the most adroit professionals in the world — leveraging every tool at their disposal to perform their due diligence and bring safe and efficacious products to market — the simple fact remains that most clinical trials are “busts.” For most pharma startups, failure to attain U.S. FDA approval on a drug or treatment spells the end of the road, as only established companies with deep pockets can withstand the negative return on investment (ROI) of a failed trial and continue day-to-day operations and future R&D.
Given the myriad of challenges that coincide with bringing a novel drug or treatment to market, companies are continuously looking for ways to do things better, cheaper, and safer — and utilizing new technologies can result in greater efficiency across all facets of the clinical trial process. Specifically, artificial intelligence (AI) could allow for data to be aggregated in a way that makes recruitment, adherence, and data analysis more streamlined, shortening trial cycles and improving patient outcomes.
Traditional Clinical Trial Timeline and Process
Linear and sequential clinical trials are the benchmark for ensuring safety and efficacy in the development of new medicines. They fail for a variety of reasons, including under-enrollment, mid-trial attrition, unexpected side effects, and inconsistent data. After discovery and pre-clinical research is conducted, there are four phases to a clinical trial:
- Phase I – researchers test a small group of people (20–80) for the first time to learn about drug/treatment safety and identify side effects. This typically takes about 3–6 months, and around 70% of participants move on to phase II.2,3
- Phase II – The drug or treatment is administered to a larger group (100–300) to determine its effectiveness and further assess safety. This typically takes about 1–2 years, and around 33% of participants move on to phase III.2,3
- Phase III – The test group expands tenfold (1,000–3,000) to further confirm effectiveness, discover and monitor side effects, and compare observations with similar treatments on the market. This typically takes between one and four years, and only around 25–30% of participants move to the next phase.2,3
- Phase IV – This phase comes after a drug has been approved by the FDA and becomes available to the public. Researchers continue to track its safety in the general population to better learn about its benefits and optimal use. This takes one or more years and has around a 70–90% success rate, which is to be expected, given the regulatory rigors a drug or treatment undergoes to this point.2,3
Some trials optionally include an early phase I that tests a subtherapeutic dose to very small group of people (10–15) to test the body’s responses.2
Where Promise Meets Real-World Challenges
High clinical trial failure rates have been normalized for decades and have contributed substantially to the overall inefficiency of the drug development cycle, as fewer new drugs are reaching the market despite increasing pharma R&D investment.4 Traditional clinical trials lack the analytical sophistication, flexibility, and speed needed to develop complex new therapies — especially those that target smaller, heterogeneous patient populations.2 Suboptimal patient selection and recruiting techniques, paired with challenges in adherence monitoring, are two of the main causes for high trial failure rates and rising costs of research and development.4
Establishing the medical criteria for patient eligibility is one of the most important steps in clinical trial design. The participant pool is created using far-ranging data that may first and foremost include whether an individual has the disease or condition for which researchers are trying to treat. Other indicators, such as levels of bilirubin, hemoglobin, platelet counts, blood pressure, red and white blood cells, and many others contained in medical records can be and are used to determine patient eligibility. Should patients fail to meet the thresholds set forth in the study design, they do not qualify for the trial.5 Setting restrictive criteria is believed to protect vulnerable patients from the potential side effects of untested drugs, but researchers have found that, whether a trial used relatively aggressive or weak restrictions, patient withdrawal rates due to adverse side effects can be virtually the same, depending on the drug and condition that is being tested.5 In many cases, restrictions have little bearing in preventing the adverse events they are designed to mitigate. Additionally, criteria used to exclude patients can be inconsistent from trial to trial and are sometimes set arbitrarily.5 Many failed trials may have succeeded if their respective studies implemented tools for modeling/data aggregation and extrapolation from rich databases to determine the optimum participant pool to reduce attrition and unsubstantiated disqualification. Failure to bring the best-suited patients to a trial in time with the technical infrastructure to cope with the complexity of running a trial — especially in phase III trials, which are the most complex and most expensive4 — results in billions of wasted dollars and no new drugs or treatments to show for it.
Adherence — or lack thereof — is a major roadblock to the success of a clinical trial. The use of rudimentary data collection and verification methods to monitor and coach often puts the onus on patients to ensure that treatment is properly received and data are accurately collected. Wearable devices powered by AI technology can help with improving adherence and, by extension, improving clinical outcomes. Archaic practices, such as sending patient medical records via fax, manually counting leftover pills in bottles, and relying on patients’ diary entries to determine medication adherence, are all too common.1 Non-adherence can have adverse effects on patient health, increase costs associated with recruiting new patients, and interfere with the accuracy of study outcomes. Generally, adherence rates of 80% or more are required for therapeutic efficacy, but up to 50% of medications prescribed in the United States are taken incorrectly.1
Artificial Intelligence – The Future is Now
Enter artificial intelligence — what once seemed like the domain of science fiction movies is now already common in many aspects of our lives. Manufacturing robots, self-driving cars, smart assistants, automated investing, social media, digital advertising, robot vacuums, smartphones and smartwatches, GPS navigation, facial recognition, smart thermostats, streaming platforms, and ecommerce sites are just some of the everyday applications of AI to which we rarely give a second thought. If AI is powering so many of the routine and often mundane tasks associated with making our lives more convenient and efficient, it would stand to reason that it should be leveraged to improve clinical trial outcomes and help bring more lifesaving drugs to market. AI was first described in 1955 as “the science and engineering of making intelligent computer programs.”6 It can further be described as “an entity or collective set of cooperative entities able to receive inputs from the environment, interpret and learn from such inputs, and exhibit related and flexible behaviors and actions that help the entity achieve a particular goal or objective over a period of time.”6 The ultimate goal of AI is to use machine simulation of human intelligence processes, such as learning, reasoning, and self-correction, to mimic human decision processes.6 AI encompasses a variety of techniques, including machine learning (ML), deep learning, (DL) natural language processing (NLP), and optical character recognition (OCR). ML is widely leveraged in the pharma industry, because it creates data analytical algorithms and mathematical models to extract features from sample data with the objective of making predictions or decisions.6
How AI Could Transform Clinical Trials
There are a multitude of potential use cases for leveraging the benefits AI brings to the clinical trial process. Trial design is one of the most important aspects of a clinical trial. Having enough of the right data is key, but knowing how to organize and analyze them in a manner that allows researchers to extract meaningful patterns to inform design can cut time and costs and increase chances of success. AI can also help with patient enrichment, recruitment, and enrollment. Through data mining, analysis, and interpretation of multiple data sources — particularly electronic health records (EHRs) that also include medical imaging — researchers can more easily verify biomarkers, select patients who are more likely to have measurable clinical endpoints, and reduce variability by identifying patients most likely to respond to the treatment.2
One branch of AI called natural language processing (NLP) enables computers to analyze both written and spoken words. When applied to clinical trials, such techniques could allow algorithms to search doctors’ notes and pathology reports for trials recruitment. Text in these types of documents is often free flowing and unstructured, and valuable information might only be implicit, requiring some background knowledge or context for it to make sense.7 Doctors have several ways of describing the same concept — a heart attack might be referred to as a myocardial infarction, a myocardial infarct, or even just MI7 — but an NLP algorithm can be trained to identify and group these descriptions in annotated medical records, and then apply ML to interpret unannotated records.7
AI can also help patients find clinical trials on their own. Patients typically rely on their doctors to inform them about suitable studies, because scouring sites like ClinicalTrials.gov, which lists more than 300,000 studies that are being conducted in the United States and 209 other countries,7 is daunting, to say the least. AI has the potential to notify interested patients when a suitable trial is available for their participation.
AI can also help biopharma companies better identify target locations, qualified investigators, and priority candidates while simultaneously collecting and collating evidence to submit to regulators to prove that the trial process follows Good Clinical Practice requirements. AI can help shape administrative procedures, evaluate resource availability, and locate clinicians with in-depth experience and understanding of the disease or condition being studied, which can positively impact trial timelines, data quality, and integrity.2
AI algorithms can help monitor and manage patients by automating data capture, digitizing standard clinical assessments, and sharing data across systems. When paired with wearable technologies, AI can also allow for round-the-clock patient monitoring, providing real-time insights into the safety and efficacy of a treatment while also predicting attrition, which allows pharma companies to make optimizations to recruitment more quickly.2
Trials generate a ton of data, but they're often siloed from disparate systems that prevent researchers from being able to aggregate them in a manner that allows for holistic analysis across multiple global sites. The ability to consolidate data from all sources onto a shared analytics platform that is supported by open-data standards can foster collaboration and integration and provide insights across vital metrics.2 Beyond aggregating and organizing data in a way that makes it more easily consumed for analysis, incorporating an AI system that “self-learns” can help with predictive analytics, improving efficiency, and reliability.
Future Outlook
Biopharma companies are looking to forge strategic and operational relationships based on outsourcing and partnership models.2 They are increasingly forming strategic partnerships with contract research organizations (CROs) that have invested in data science, which provides access not only to specialized expertise but to a wide range of potential trial participants as well.2 Biopharma companies have also attracted the attention of tech giants, as they present both an opportunity and a threat as they disrupt specific areas of the industry.2 In tandem, an increasing number of digital technology startups are now working in the clinical trials space and partnering or contracting with biopharma companies.2
The FDA considers AI- and ML-based software to be medical devices, and expect companies to comply the requirements of clinical, analytical, and technical validation, as well as quality systems, good machine learning practice, assurance of safety and effectiveness, transparency, and real-world performance monitoring.6 Any new AI tech that claims to improve the efficiency in clinical trials should be validated by testing alongside the current operational standards. Regulatory agencies and end users should expect AI insights to be understandable, ethical, replicable, and scalable.6 Similar to other advances in automation that have replaced human workers with computer technology, there are also concerns about the potential for job loss, which may delay the adoption of AI tech in clinical trials.
Privacy concerns will also drive the adoption rate, as maintaining the privacy of a large volume of patient data must be of utmost importance. Advances in blockchain and tokenization allow for global collaboration across a digital ledger that can be universally seen and verified, allowing for various levels of access among stakeholders with full transparency.
There are still learning curves to overcome before being AI garners the full trust of professionals in the clinical trials space. However, as AI continues to serve a more prominent role in nearly every facet of our lives, it stands to reason that its ubiquitous adoption isn’t a matter of “if” but merely a matter of “when.” Continuous data collection is critical, but it is difficult to find subtle yet meaningful signals in large data sets. The adoption of AI could be a gamechanger in personalized medicine, and the potential gains to be made in the creation of novel drugs and treatments, as well as the revenue-generating potential for AI software companies, will likely converge to create the perfect cocktail for AI’s continued integration into clinical trials.
References
- “The Future of Clinical Trials: How AI Big Tech & Covid-19 Could Make Drug Development Cheaper, Faster, & More Effective.” CB Insights. Web
- Taylor, Karen; Properzi, Francesca; Cruz, Maria Joao. “Intelligent Clinical Trials.” 10 Feb. 2020. Web.
- “NIH Clinical Research Trials and You.” National Institutes of Health. 10 Nov. 2021. Web.
- Harrer, Stefan; Shah, Pratik; Anthony, Bhavna; Hu, Jianying. “Artificial Intelligence for Clinical Trial Design.” Science Direct. 19 Jul. 2019. Web.
- Myers, Andrew. “AI Expands the Reach of Clinical Trials, Broadening Access to more Women, Minority, and Older Patients.” Stanford University
- Bhatt, Arun. “Artificial Intelligence in Managing Clinical Trial Design and Conduct: Man and Machine still on the Learning Curve?” National Institutes of Health. 19 Jan. 2021. Web.
- Woo, Marcus. “An AI Boost for Clinical Trials.” Springer Nature. 25 Sep. 2019. Web.