How AI Can Deliver Frictionless Patient Matching for Clinical Trials

The pandemic has highlighted inefficiencies within many parts of our healthcare system. Initially, the virus called attention to the national shortage of ventilators and personal protective equipment at our hospitals. It then spotlighted how racial minorities and lower-income Americans are more vulnerable to public health threats. Now COVID-19 is highlighting the shortcomings in our vaccination distribution.

In responding to each of these crises, the medical community has turned to Big Data-related technologies – artificial intelligence technologies, such as natural language processing, as well as data management. Researchers and physicians have deployed these tools to target testing toward those at highest risk of contracting COVID-19; reallocate resources to the most vulnerable patients; accelerate the search for a vaccine; and most recently, coordinate the rollout of approved vaccines.

While these advanced technologies may be a reality for world-class research laboratories and elite medical institutions, they are not yet standard features in the tool belt for most hospitals and community cancer centers caring for patients around the country — for COVID-19 or other leading life-threatening diseases, including cancer.

Digital healthcare technologies that scale in cancer centers and elsewhere are now widely available, though. But while this technology can make our healthcare systems more resilient and effective, it can only do so if it can be folded into already overstressed clinical settings. As we look to restore faith in our US health systems post-COVID-19, we need better technology, but only if clinicians can use it.  Utility — not computer power or fancy features — should be our guiding light.

Catching Up on Cancer Trials

Oncology has also suffered serious setbacks due to the pandemic. Cancer trials, often the last and best hope for patients, have faced the challenge of identifying and enrolling participants when hospitals are at limited capacity.

Even before the pandemic, trials often failed due to the difficulty of finding enough eligible patients. Then in April, enrollment dropped by 70 percent at 1,500 trials monitored by healthcare data services firm Medidata. That number began to rebound in June but has likely dipped again as COVID-19 surges across the country.

Thousands of lives now depend on oncologists and researchers to make up for the lost time. As with the fight against COVID-19, there’s a tremendous opportunity to use available technology to support and accelerate the work of physicians and scientists.

Despite the hype around many AI solutions, however, the challenge of using, say, machine learning or natural language processing, is not whether one can secure the computing power or fancy algorithms. With rapid advances in cloud computing and software, that’s the easy part. Our focus as oncology innovators needs to be on solutions that fit seamlessly into the clinical setting — tools that physicians and their staff will use. That’s the key to generating patient data, which is the engine that will drive life-saving technology across the health care system.

The Data Is at Our Fingertips

Solutions that succeed in connecting patients to clinical trials will integrate into existing systems and workflows without generating friction or too many new tasks for overworked physicians or their support staff. Many AI solutions seek to expand on electronic medical records, requiring physicians to make additional inputs to create data. The reality is, doctors and patient-facing staff don’t have the bandwidth to do it. They shouldn’t have to.

The majority of oncologists report spending 13 to 24 minutes with patients per visit. At the same time, physicians spend only around 15 percent of their time with patients. The last thing they need is to spend more time staring at computers. Those of us developing these tools should strive to give time back to physicians, not make additional demands on them. We have designed our AI-enabled platform to mine existing sources of structured data — EMRs and tumor registries, for example. But, we’ve also focused on extracting other clinically relevant data necessary for those conducting trials to find possible matches. This includes unstructured data such as doctor’s notes from patient reports, lab reports from external sources, or faxes and scans that may include key information needed to match to a complex clinical trial’s inclusion or exclusion criteria.

Unlike many products being piloted in the oncology space, we have already seen positive results on the ground. From March to June, as many clinical trials were on pause and as enrollment was plummeting, a Cancer Center using our technology enrolled 10 patients, a threefold increase on previous quarters.

The extraction process uses Natural Language Processing to unlock data hidden in unstructured sources, finding the histology and behavior of patient cancers, their genetic makeup, their stages, whether or not patients have undergone a procedure and their potential Gleason scores.

It’s work that medical professionals could and would be doing if only they had the time. Thanks to tremendous advancements in cloud-based computing and data processing, the technology sifts through tens of thousands of permutations in the criteria for thousands of clinical trials, a task that would literally take weeks per patient in the hands of a human.

While the technology can’t make the final decision about whether patients are a perfect fit, importantly it significantly narrows down the field, giving physicians a better chance of finding a clinical trial that matches their patients’ needs.

Research has shown that only one in 20 adult cancer patients enroll in clinical trials, for reasons ranging from their own hesitancy to geographic or financial impediments to access. But perhaps the most important factor in that shocking statistic is a lack of information or referrals coming from physicians. And that’s not because doctors don’t think clinical trials are the right course of action — the vast majority believe trials provide high quality care and benefits for enrolled patients.

Ultimately, the difficulty of enrolling patients comes down to l*********** and cumbersome logistics at the clinical level. We need technology that recognizes this reality and helps physicians overcome it.