Technology continues to impact healthcare, no more so than in the area of clinical trials. Artificial intelligence and machine learning show promise in bringing some order to the current chaos and inefficiency around matching patients to the right clinical trial.
AI also is helping clinical researchers do a better job of monitoring patients during the clinical trial itself, giving them more opportunities to coach patients. That, in turn, leads to more accurate clinical trial results.
Taken together, AI may help researchers avoid the high failure rates that have plagued clinical trials for years.
AI and Cohort Selection for Clinical Trials
Despite an increase in the amount of money invested into the research and development of pharmaceuticals, the failure rate of so many clinical trials has led to an inefficient drug development cycle, according to Trends in Pharmaceutical Science magazine.
The magazine places blame for this on two main areas:
- Suboptimal patient selection and recruiting techniques
- The inability to monitor and coach patients effectively during clinical trials
This had led to a situation where many clinical trials never get off the ground. Wired magazine reported that of the 19,816 clinical trials planned at the beginning of 2018, 18,000 of them will never get started because not enough patients are recruited.
The problem matching patients with trials comes from how the information is displayed online by the government at ClinicalTrials.gov, according to Wired. Eligibility requirements are entered in a free text field, making them impossible to search.
Companies are putting AI to use to solve this problem. Pablo Graiver, who worked as a vice president at online airfare aggregator Kayak, talked with Wired about his new company, Antidote. The company, with the help of clinicians, has taken the unstructured, jargon-filled language on clinical trial eligibility and put it in a structured language that search engines understand.
Antidote has now annotated more than 14,000 trials. That’s about half of what’s listed on ClinicalTrials.gov, according to Wired.
How AI Can Help with Drug Development
It takes 10 to 15 years and about $1.5 billion to $2 billion to bring a new drug to market, according to Trends in Pharmaceutical Research. About half that time and money is spent on clinical trials. The other half is in drug development. There is potential for AI in this area, as well.
One of the main issues with drug development is during the trial itself. Patients are given the experimental drugs, which they then take home. They are told to record when they use the drug, what other drugs they might have been taking, as well as recording any side effects.
But many trials still use paper documents, according to CB Insights. And in many cases the clinicians are relying on a patient’s memory and their accuracy in recording information on paper. Non-adherence to taking the drugs as directed can lead to problems with results from the trial.
According to CB Insights, methods for dealing with this issue include using AI to send reminders on when to take drugs, visual confirmation that drugs have been taken through face-tracking software (which requires the patient to take a video of themselves taking the drug) and even sensors that can be swallowed (Pfizer and Novartis have invested in this technology).
This fine tuning of the data collection process is vital, as USF professor Christina Eldredge noted in a recent Faculty Spotlight.
“AI is going to depend on the quality of the data it receives,” Eldredge said. “If the data going in isn’t good quality and clean, AI can’t actually tell you anything. What the future really comes down to is fine tuning that and making it become an aid to healthcare providers rather than a burden. We need to fine tune it, but I do think it’s going to be big in cancer research because we haven’t conquered that area yet and AI can help us target specific types of cancer to provide more of a precision medicine approach.”
Driving Better Success Rates
There are still many hurdles to clear even as companies such as Antidote take steps to improve the process. One of the biggest continues to be interoperability, which refers to the inability of computer systems at different medical facilities to share information.
While health care operations have moved into using electronic health records (EHR), there is no one, central repository for the information. Also, a patient seeing a doctor who uses one software system and a specialist who uses another will have difficulty getting their information shared between the two.
Another key factor is the development of standardized methods of recording adverse events in clinical trials. In order to tailor the trial to the person, this needs to be normalized and better understood.
“What I’m realizing from my own research is that there is such a diversity in how we record adverse events in clinical trials that it’s really hard to study them on a population level,” Eldredge said. “It’s comparing apples to oranges. It needs improvement if we want to push population health.”
The hope is that these issues and other roadblocks will be overcome, allowing AI to improve on the high failure rates of clinical trials and instead support trials that result in more accurate findings.
As computer scientist Stefan Harrer, a researcher at IBM Research-Australia, told Health IT Analytics, “AI is not a magic bullet and is very much a work in progress, yet it holds much promise for the future of healthcare and drug development.”