In clinical research, access to data — and the software to make sense of it — means researchers are better equipped to define their research questions and align various study requirements to optimize results.
By providing improved site selection and recruitment, along with a greater insight into patient populations and their shared traits, AI and machine learning turn data into a valuable tool to define clinical study questions.
Big data and machine learning are making enormous inroads into the pharmaceutical and medical research markets. They increase the revenue streams of pharmaceutical companies through improved decision making, innovation, boosting the efficiency of trials, and new tools for everyone involved.
Consider how AI-powered algorithms are gathering and making sense of valuable data and what this means for the researchers’ areas of study.
AI to Trawl Data Records
The first step in the digital revolution in healthcare was creating electronic health records (EHR), but possessing the data only gets researchers so far. They need the capability to analyze that data. This is what software company CBInsights posits as an ideal and achievable AI solution. Sophisticated software scans and gathers relevant information from a patient’s EHR, and matches it against ongoing and upcoming trials.
One example of a company using AI to improve trial research is Antidote, which functions in the clinical trial space the same way travel aggregation sites like Kayak do. Antidote uses machine learning to translate inclusion/exclusion criteria trial jargon on ClinicalTrials.gov into language a search engine can understand.
The result for patients is that, after entering their condition along with their age, sex, and geographical location, Antidote returns a list of trials that meet those basic criteria, along with further prompts to determine eligibility for existing trials.
AI-Powered Cognitive Assistants
Across the medical field, the doctor shortage is a real concern.
According to “The Complexities of Physician Supply and Demand: Projections From 2019 to 2034,” a report by the Association of American Medical Colleges, physician demand will grow faster than supply in the coming decade. By 2034, there will be a deficit of between 37,800 and 124,000 physicians in the United States. There will also be a shortage of nurses, midwives, dentistry professionals, and pharmacists.
Some experts hope that artificial intelligence will ease this crisis on multiple levels, from big data analytics to AI-powered cognitive assistants that will help clinical researchers do their jobs better.
Consider IBM’s Medical Sieve project. Its aim is to build a cognitive assistant system for radiologists and cardiologists with reasoning capabilities and clinical knowledge that can analyze radiology images to detect medical issues. Then there’s Deep Genomics, an AI-powered discovery platform that trawls large data sets of medical records to determine links between diseases and genetic information.
Atomwise uses supercomputers in pharmaceutical research to find new therapies. In 2015, the company found two drugs in less than a day that may lower Ebola infectivity rates. A traditional approach to research would have taken years to find similar results.
While AI-based services can assist with diagnostics, decision making, and administration, limitations exist. Researchers still need human experts to guide and use the technology. AI is not a replacement for working professionals.
The Next Step in Clinical Trial Development
With artificial intelligence, machine learning, big data, and even social media being used in clinical research, pharmaceutical researchers are entering a new era of clinical research. Specifically, AI will enable researchers to tap into previously unavailable genomic data, allowing AI-personalized, real-world studies.
The big shift here is how research questions are formulated. Traditionally this was done by asking a specific question and testing it against a hypothesis in a controlled environment. AI will grant researchers the capability of asking (and answering) additional questions by analyzing real-world data from many thousands of patients recruitable through tech such as Apple’s ResearchKit.
This might include data from EHRs, registries, hospital records, and health insurance data alongside biobank, genomic and digital phenotyping information, and wearables. It’s a boon for recruitment as millions of patients’ data is available without patients required to submit that information themselves as trial applicants.
With this approach, research questions can be changed and improved upon as more data is collected and analyzed with a flexibility that was previously impossible with a traditional study.
Teaching AI Isn’t Easy
While many researchers are eager to reap the benefits of AI and machine learning (ML) for their clinical trials, technology developers are still working to create effective models that pharmaceutical companies can use.
One group of MIT researchers studied questions that physicians ask when reviewing EHRs and then fed 2,000 questions into an AI. The team found that machine-learning systems asked high-quality, authentic questions 60 percent of the time compared to actual physicians.
“Two thousand questions may sound like a lot, but when you look at machine-learning models being trained nowadays, they have so much data, maybe billions of data points,” says the study’s lead author Eric Lehman, a graduate research assistant at Massachusetts Institute of Technology. “When you train machine-learning models to work in health care settings, you have to be really creative because there is such a lack of data.”
While clinical researchers might get excited about the potential of AI and ML for clinical question development, the technology isn’t readily available just yet.
The goal of many ML systems in clinical trials is to teach AI tools to “read” information and generate appropriate questions. These questions can range from standard queries for building patient profiles to highlighting issues due to incomplete data.
“The machine learning algorithm in smart querying reads trial data and determines possible queries that can be raised for different field items,” according to Clinical Research News Online. “If the algorithm identifies an error, it will raise a query. The query is then reviewed by a data manager, who either qualifies it as a valid query or discards it.”
In this way, AI tools can be incorporated into the beginning of clinical trials (basic screening) through data management, cleaning, and analysis, leading to publication.
There’s No Need to Rush Into Technology
The COVID-19 pandemic turned clinical research on its head. Existing trials scrambled to offer decentralized models while the COVID vaccine itself was tested and developed at an extraordinarily rapid pace. Patients now look at other trials and expect the same speed of development, despite the fact that most trials take several years to complete.
“Perhaps the greatest challenge in a clinical study is developing a robust and flexible protocol and study design,” write Jennifer Duff and Jessica Schell, at Merative, formerly IBM Watson Health.
This is another reason why researchers are so eager to embrace technology. The need to stay competitive and meet development expectations is driving trial teams to look for AI tools. However, rushing anything can lead to mistakes. Investing in poor tech — or using good tech poorly — will do more harm than good.
Technology itself isn’t inherently good or evil, says Eric Perakslis, chief science and digital officer at Duke Clinical Research Institute. It depends on how you use it. A poorly designed data set will generate poor results. Unfortunately, in the world of clinical trials, patients don’t have the luxury of working with bad AI. Incorrectly worded or unasked questions could move unqualified patients through to a trial and potentially put their lives at risk.
“It’s very easy to get excited about what technologies can do,” says Perakslis. “It can take much longer to figure out what harms they cause.”
Are the worlds of machine learning and artificial intelligence for clinical trials promising? Absolutely. These tools can sort through endless data sets, provide valuable questions, and check information for problems before it gets published. However, most AI systems are still in the development stages and need to fully form before researchers can use them.