AI Applications for Clinical Trials Increase, Refining Endpoints, Quantifying Pain, & More
Artificial intelligence (AI) and machine learning (ML) are enabling clinical trial sponsors to fill in the gaps when real world evidence is incomplete or inconsistent. Trials suggest it can do even more, detecting early indicators of disease, and even quantifying pain.
“Real world evidence (RWE) rarely has the same rigor as research data. RWE is collected through electronic health records (EHRs) as well as through clinical trials. Although agencies want real world evidence, but its quality isn’t yet geared to deliver meaningful insights. Machine learning can derive those insights,” Jaydev Thakkar, COO, Biofourmis, told BioSpace.
One example comes from the Ministry of Health in Singapore. With data from thousands of COVID-19 patients, the Ministry is using machine learning to detect patients who have COVID-19 before clear symptoms are detectable without performing a PCR-based test. Data analysis also can detect worsening patients, Thakkar said. He said he expects that data to be published in the near future.
“When applying AI/ML to clinical trials, there’s still hesitation,” Thakkar said. Aside from the perception of novelty, the pharmaceutical industry is concerned about regulators’ comfort levels when AI/ML is used. “That’s a big factor holding back AI’s use, but the mindset is shifting,” Thakkar said.
That’s partially because COVID-19 is making it difficult to conduct clinical trials traditionally. “Managing patient burden and site burden for clinical trials are key concerns for sponsors. Consider the number of times patients are asked to fill out surveys or evaluations on-site,” he began. With COVID-19, many patient are reluctant to expose themselves to the risks of going to a healthcare facility, so they drop out of the trial or don’t enroll. “Much of that data can be collected digitally, off-site, to reduce the burden on the patient and therefore address delays in recruitment,” Thakkar explained. AI/ML can help bridge data gaps and inconsistencies inherent in RWE by pulling data from large pools of patients.
AI can have a deeper role, though. Biofourmis is working with Chugai Pharmaceutical in Japan to objectively determine patients’ levels of pain in endometriosis using its Biovitals® platform. “Traditional subjective assessments have significant variability, so they’re not an optimal way to identify efficacy of a drug or to perform drug titrations to deliver pain medications to patients. Instead, we can objectively identify the pain level by collecting data through sensors and quantifying that information,” he said.
The Biovitals platform senses more than 20 physiological changes, such as skin temperature, heart rate, or electrodermal activities that suggest, for example, frustration. Feeding that data into an algorithm compares the data to that of millions of other patients. In the Chugai study, this is providing a way to objectively quantify pain. That project is in the clinical trial phase, “evaluating the clinical research and enhancing the algorithm.” It’s too early in the trials to share results, but initial data is fairly accurate, Thakkar said.
In another example, AI can enhance patient adherence. Remote sensors or wearables are an increasingly common element of clinical trials. By determining whether patients actually are wearing them or which are completing the necessary online surveys, AI analysis can determine their level of engagement and predict which patients may drop out of trials.
When clinical trials consider patient burden by selecting devices that fit well into patients’ daily lives, patients are more likely to remain engaged. Consequently, pharmaceutical developers are gaining meaningful insights using patient-friendly devices and, in the process, discovering novel endpoints and signals that may be missed when patients are seen only episodically.
As clinical trial sponsors begin to use AI/ML in clinical trials, Thakkar said, “Primarily, we’re seeing a shift toward endpoint design leveraging RWE by applying AI/ML to more effectively design protocols and outcomes to identify the change introduced by the therapy,” Thakkar said. For example, trying to measure effectiveness of a heart failure drug by recording patients’ daily step counts may be less effective than measuring their sedentary hours or the intensity of their walks. By applying data collected from many trials, you can build more relevant endpoints.”
Many algorithms for biology are applicable for clinical research, but are not yet commercially available. The FDA, however, is amenable to their use, Thakkar said. “We’re working with the FDA, and it’s very welcoming. For trial sponsors, the agency encourages early engagement, with an explanation of how you plan to use AI/ML in clinical research.” By engaging early, companies can incorporate FDA feedback into their clinical trial design. Biofourmis is working with at least eight pharmaceutical companies to apply the AI tools.
“Continuous data collection is critical, but it’s hard for the human eye to pick out subtle, yet meaningful, signals in high volume data. AI/ML, however, can find signals that can be proven statistically,” Thakkar said. The insights that tool can reveal are an important addition not only to drug development, but to the advancement of predictive and personalized medicine.