By streamlining processes, cutting costs and improving data quality, artificial intelligence (AI) and machine learning can reduce clinical trial spend and speed time to market. The technology is showing up in a number of applications, from guiding recruitment to developing biomarkers to determine who will respond to certain treatments and driving cost efficiencies.

By 2025, the AI market for health care is projected to hit $36.1 billion, up from the current $2.1 billion today, according to MarketsandMarkets. Companies from Google and IBM to digital health startups are vying for a piece of the pie.

Case in point: Scientists at Dana-Farber Cancer Institute reported that an AI tool performed as well as human reviewers, and much more quickly, in culling clinical signs of tumor changes from unstructured radiology reports of patients with lung cancer.

"More reliable, predictable and speedier trials have enormous cost savings implications for clients and, most importantly, have the ability to bring much-needed therapies to patients sooner," Ben Hughes, senior vice president at Iqvia, told BioWorld MedTech. The company's contract research organization has seen positive results applying AI to trial design and execution, including fewer study amendments, 40% faster site identification, 30% faster recruitment rates and major improvements in overall quality.

"Beyond improvements in study execution, the predictability that AI can bring has huge upside for R&D portfolio management," Hughes added. "We see important trends in in silico clinical trials. The ability of deep learning to predict study level and patient level clinical outcomes can transform R&D. Accurate predictions around clinical outcomes can de-risk clinical portfolios, and synthetic cohorts can reduce patient burden. Patient-level modeling can identify subtypes and improve protocol design. And most exciting, as the industry introduces molecular structure into these models, AI can create a new paradigm of integrated drug discovery and drug development."

What's out there

One doesn't have to search far for examples of how AI is being applied in clinical trials and research:

  • IBM partnered with the Juvenile Diabetes Research Foundation to develop and apply machine learning methods to analyze major academic type 1 diabetes datasets, with the goal of better understanding the factors that lead to disease onset. (See BioWorld MedTech, Aug. 22, 2017.) IBM's Watson Health is also applying AI to improve medication compliance.
  • Working with researchers at Moorfields Eye Hospital in the U.K., Google LLC demonstrated that its Deepmind AI system can detect and recommend how patients should be referred for more than 50 eye diseases with the same accuracy as a physician, but at a clip of 1,000 scans a day.
  • Allscripts Healthcare Solutions Inc.'s life sciences arm, Veradigm, is working with Microsoft Inc. to develop an integrated research model that will allow companies to collect trial data through point-of-care technology platforms, speeding time to market of new drugs while shaving R&D costs. The initial focus is on extending Allscripts' cloud-based electronic health record (EHR) platforms with features such as automated "matchmaking" between trial protocols and doctors and patients who qualify for the studies.
  • Berkeley, Calif.-based startup Eko's patient-facing app, Eko Home, enables remote monitoring of cardiac function using ECG and heart sounds. Its first use will be in a Mayo Clinic study monitoring breast cancer patients undergoing chemotherapy with trastuzumab, a drug known to cause a decline in heart function in a percentage of patients.










 

  • Biofourmis Singapore Pte. Ltd. is collaborating with the Yale University-Mayo Clinic Center of Excellence in Regulatory Science and Innovation to leverage its mobile platform, Biovitalshf, in a study of heart failure patients to monitor functional capacity and quality of life. The goal is to ascertain if these factors should carry more weight in the drug approval process.
  • Physiq Inc., of Naperville, Ill., is working with biopharma companies and payers to create novel datasets that support clinical trial endpoints using biosensors and cloud-based analytics. The company's line of FDA-cleared algorithms includes QRS detection, heart rate, heart rate variability, atrial fibrillation, continuous ambulatory respiratory rate algorithm and a personalized physiology change detection analytic.
  • Eli Lilly and Co., Evidation Health and Apple Inc. reported initial results from a feasibility study showing that an iPhone, Apple Watch, iPad and the Beddit sleep monitoring app, together with digital apps, may help to distinguish people with mild mental impairment from people with mild Alzheimer's disease symptoms.

Patient matching

One of the places AI is having an impact is in matching patients with clinical trials. With the push for precision medicines, companies and scientists are trading the one-size-fits-all approach for programs that target unique genomic footprints to develop drugs and diagnostics that will produce the best possible outcome for the patient. Meanwhile, they're using AI and machine learning to analyze huge amounts of unstructured data from patient questionnaires, information about the efficacy of a drug, its indications and contraindications, social media websites and more to find and recruit patients who fit a particular drug candidate's profile.

Last month, New York-based Massive Bio Inc. launched a global AI-enabled network aimed at connecting prominent Just-in-Time (JIT) vendors and clinical researchers in the area of precision oncology. The clinical trial "Hub" is aided by Massive Bio's Synergy-AI platform, which uses prescreening algorithms to capture and screen patients for clinical studies. The platform also serves as a portal for patients to apply for the Synergy AI clinical trial listed on clinicaltrials.gov.

"You basically structure the unstructured genomic and clinical information, you structure the clinical trials information in clinicaltrials.gov or a different mode that has been provided by the pharma and you do that prescreening at scale, because right now it's a fairly manual process," Selin Kurnaz, co-founder and CEO of Massive Bio, told BioWorld MedTech.

San Mateo, Calif.-based startup Notable recently raised $40 million in a series B round to advance development of its AI-powered platform aimed at identifying treatments for which patients with blood cancers are most likely to respond. "As the technology improves, we will gather the evidence for predicting responders and nonresponders to a given therapy, whether it is 'standard of care,' a repurposed drug or a novel investigational treatment," Hiroomi Tada, Notable's chief medical officer, told BioWorld MedTech. "Our goal is to change the way drugs are developed and how clinical trials are conducted, starting with innovative trials using our platform to inform treatment assignments based on predicted ex vivo responses."

Remote monitoring

One of the biggest opportunities for AI in research is in round-the-clock feedback, augmented with machine learning to interpret the data. With Eko Home, Mayo Clinic researchers are assessing different strategies of cardiovascular therapy with carvedilol with the aim of reducing the rate of heart failure in at-risk breast cancer patients who are on trastazumab therapy. "This particular study is looking at collecting data from patients at home over a period of time and then retrospectively looking back and saying how could Eko's algorithms have predicted patients' decompensation," Jason Belett, co-founder and chief commercial officer of Eko, told BioWorld MedTech.

The company has partnered with Mayo Clinic on an algorithm that uses one-lead ECG to screen for heart failure and is building it into Eko's digital stethoscope for use in the clinic or home. The plan is to eventually have the device screen for structural heart disease and atrial fibrillation, as well as heart failure. Such low-cost, noninvasive tools could help to paint a clearer picture of physiologic functions over time and provide more insights into the impact different therapeutics have on patient outcomes. "In the same way that a physician managing a heart failure patient will be able to better track their cardiac function, so, too, will a clinical researcher be able to better understand the impact of these drugs on cardiac function," Belett said.

Meanwhile, Physiq is tying AI to real-world data via wearable sensors to create proxies for long-standing measures of physical fitness, such as the six-minute walking and VO2max tests. The latter, which measures oxygen and CO2 levels at the point of total exhaustion, can be used to assess heart failure, but it's expensive and requires the patient to come into the clinic and engage in rigorous activity, something that may be beyond the capacity of someone who is seriously ill. For that reason, it's had limited use in clinical trials.

Physiq is attempting to create a better version of VO2max that achieves a maximum output using activities of daily living instead of total exhaustion, explained Chris Economos, chief commercial officer of Physiq. "What AI is really good at is looking at all the patterns within that physiological data and identifying features that would otherwise be impossible to quantify using ... an engineering-based approach," he told BioWorld MedTech.

The company sees other opportunities for biomarkers as well. A big one is cough detection. AI has the potential to extract physiological features of a coughing fit that could be used to measure things like severity and type of cough. This could be useful in clinical trials of chronic obstructive pulmonary disease (COPD) and other pulmonary issues. AI could also be used to augment current pain measures, which are inherently subjective and rely on self-reporting, by combining those with physiological data.

"Our focus is what are these biomarkers that are out there, or potentially out there, that can really transform pharma's ability to validate the safety and efficacy of their drugs, both in a clinical context as well as in a commercial context ­– this whole notion of real-world evidence and continuing to demonstrate value for different payers once a drug has been launched," Economos said.

Challenges

Despite its potential to increase efficiencies and track patient outcomes, AI is not a magic bullet. Even with good prescreening, enrollments are lost either because the site is not available for the patient, the patient's provider is not a research site, or there is an insurance-related issue. Massive Bio is attempting to ease the process with value-added services that make its AI-driven recommendations actionable for the cancer patient.

"When you are talking about AI in health care, everybody thinks about the data – Big Data. That is, of course, an important component, structuring the data, teaching the algorithms," Kurnaz said. "But that is a very limited component." Companies need to have good patient identification, good prescreening, operationalization and real-world data, she stressed. "If you can connect those dots, then you can have an artificial intelligence-based health care that is fully operational."

There is also the issue of interoperability. As long as EHRs sit on proprietary platforms, technology will have to grapple with challenges like unstructured data and disparate data sources that don't communicate with each other.

And there is the data itself. AI companies need the infrastructure to collect quality data and filter out that which is bad, said Economos. "It's not enough to have these awesome analytics, because the analytics you derive are only as good as the data that you're able to capture in a real-world challenging environment."

When it comes to clinical trials and especially patient enrollment, it's not enough to have good patient identification and prescreening. Companies also need to have operationalization – the ability to strictly define variables into measurable factors – and real-world data, Kurnaz added.

No Comments