Imagine how many pieces of data must be viewed, analyzed, and tabulated to accurately track the actions of the roughly 1,148,000 patients who participate in the more than 6,000 biopharmaceutical clinical trials in the United States in a year. On average, each trial includes around 200 participants whose actions must be recorded in detail to maintain study quality, accuracy, regulatory compliance, and the patients’ own safety.
Clinical trial data is collected from a variety of sources and formats, and data is increasing in volume and complexity by the day. Trial sponsors and contract research organizations are using more complex treatment protocols and larger digitized data sets in search of the next great medical breakthrough. Studies are becoming more geographically expansive, spreading across multiple sites and countries to target just the right patient populations.
With clinical research complexity and costs soaring, humans can do only so much to keep up.
Enter the machines
We are at the early stages of a revolution in machine learning-based clinical trial management. Machine learning — a type of artificial intelligence concerned with development of algorithms that infer knowledge from raw data and provide actionable recommendations — is already helping researchers better flag and predict clinical research risks. This includes site management, patient adherence, and how diseases and treatments will impact patients participating in clinical trials.
Already, machine learning is being used to detect health risks in premature newborns, verify that clinical trial participants are taking their medications, and identify the presence of sepsis in patients before doctors may see the physical signs.
Remarque Systems is using machine learning to improve the quality, reliability, and safety of clinical trials by generating the knowledge needed for better decision-making during clinical trial monitoring. To this end, we have developed a system-agnostic risk-based monitoring (RBM) platform that uses data from any source in any format to predict, detect, analyze, and manage risk in clinical trials — without the challenges and cost of maintaining or upgrading a data warehouse.
Our customizable platform can consume and analyze data from electronic health records, electronic data capture systems, central lab data, safety data, electronic patient-reported outcomes, quality management systems, issue management systems, electronic trial master files, eConsent and other eSource technologies, mHealth wearable data, and other formats.
Our RBM platform has several machine learning algorithms that help users better understand the data, identify common and specific risks that could surface in clinical trial management, and provide timely recommendations. The algorithms, customizable through the user interface to fit individual study needs, run on demand or on a schedule and follow a strict recommendation criterion set by the user.
The machine learning algorithms follow a passive approach when generating recommendations. We do not allow algorithms to act or change any data on their own, but only to make recommendations on risks they identify to human associates who are authorized to act or make changes at the patient, site, or protocol levels. This eventually may change as we collect more historical data and evaluate interactions between users and the machine. Moreover, all actions pertaining to these algorithms — such as setting up algorithm parameters, storing results, and recommended actions — are audited.
We are currently building huge machine learning-based data libraries focused on various drug classes, therapeutic areas, and chemical structures that risk-based monitoring systems can draw upon to help humans make smarter, faster, and safer quality decisions in clinical trials.
Patterns and trends
Unlike human eyes and brains alone, a machine-learning platform can analyze large data sets from multiple sources simultaneously and recognize patterns and trends measured against a predetermined set of parameters or rules. It then provides one or more actionable hypotheses or recommendations faster than dozens of researchers could recognize or handle alone. Culling from millions of data points in near real time, machines can flag missing informed consent documents of trial participants, patterns of trial participants routinely missing visits or medications, and potential errors or outliers in key clinical data. They can even identify potential fraud by specific research sites. And unlike trying to track and record all this in separate paper files or computer systems, RBM technology provides a comprehensive electronic audit trail available at any time for internal or regulatory review.
This added machine-based risk monitoring firepower could not come at a better time. In 2016, the International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use Guideline for Good Clinical Practice was updated for the first time in 20 years. The most significant change came in what is known as ICH E6 (R2), guidance requiring sponsors and those who support them to take a more structured approach to risk management in clinical trials.
Achieving the new best practice in clinical trial risk-based monitoring is difficult or impossible with outdated paper-based or multiple-spreadsheet trial management systems. Errors contained in paper-based monthly reports issued on a weekly, monthly, or longer schedule may escape detection for long periods or altogether. Worse yet, errors or missing data may proliferate before the study team recognizes that there is a problem. Periodic trial reports often provide data tables that are only a starting point for analysis, not a basis for action.
Closing the gap
Predictive machine learning helps close this gap using statistical computation to see issues and risks humans cannot. A machine program can then provide near-real-time analysis and alerts to clinical trial associates so the CRAs can take appropriate action before it’s impractical or too late.
What this work can lead to is potentially huge. The McKinsey Global Institute estimates that such data-driven solutions to improve research, clinical trial efficiency, and other healthcare decision-making could add up to $100 billion in value annually across the U.S. healthcare system alone.
Multi-site clinical research studies cost millions of dollars, often take years, and involve server-loads of data. Odds of a drug development project ever making it from Phase 1 through regulatory approval currently stand at less than 10 percent. Improving already challenging prospects of success can come down to making sure all clinical trial sites are achieving what they are supposed to sooner rather than later. Any error or delay may be a multimillion-dollar misstep or threaten the safety of those whose lives clinical trials seek to improve.