The Opportunities and Challenges of Machine Learning in Conducting Clinical Trial Data Analysis

Clinical trial research is experiencing a groundbreaking revolution. Trial data has always been essential for testing drugs, devices and treatment protocols. Common trial challenges such as timelines, trial costs and compliance have traditionally impacted the amount of data that was collected. Now, thanks to artificial intelligence (AI) and machine learning (ML), researchers can access, manage and analyze data faster, cheaper and more compliantly than ever before. With the aid of ML advancements, trials yield actionable data in days or weeks instead of months or years. For example, in a recent sleep study, researchers found that an ML-based monitoring approach is faster and more convenient for patients and achieved higher accuracy in classifying people with chronic insomnia than the currently popular support vector machine (SVM) method. With results like this, it is little wonder that in an analysis of the future of sleep health, researchers said that AI-based advances like wearable sensors and ML data processing are already taking sleep monitoring out of the laboratory and providing access to more comprehensive sleep data than has ever been available before.

This is exciting news for research organizations intensely focused on trial speed and cost cutting since the COVID-19 pandemic. ML systems feature a combination of statistical methodologies that imitate how humans learn, gradually improving accuracy and achieving outcomes faster while cutting worker-hours and other operational expenses. In one recent case, researchers reported a 90% time savings on trial survey completions owed to ML. The technology poses several challenges that need to be addressed so organizations can maximize the many benefits of machine learning. It is critical for organizations to pay close attention to the interpretability and transparency of ML algorithms while taking steps to protect patient data. Those who take time to tackle the challenges and embrace the many opportunities of machine learning will create a harmonious environment that harnesses the strengths of ML and traditional statistical methods to ensure the most accurate outcomes.

ML Opportunities

The integration of ML into clinical trial data analysis presents unparalleled opportunities. ML has the potential to enhance efficiency and accuracy in large-scale trials by swiftly processing vast data sets, revealing patterns and better predicting outcomes. The following are some benefits that ML is already providing in clinical trials:

• Saves time. Research from the Deloitte Centre for Health Solutions indicates that ML can reduce the time it takes to bring a new drug to market from 10-12 years to 5 years. A key reason for that reduction is the improved decision-making that ML unlocks. Researchers can make more informed decisions that reduce rework and inefficiency, resulting in faster, more accurate outcomes. One biopharma company used AI/ML to select higher-frequency study endpoints and cut trial length by 15%-30%.
• Increases enrollment. Enrollment continues to be a major challenge in clinical trials. One group of researchers recently solved this problem by creating an AI/ML tool that allowed them to double the number of eligible patients for the study. Other studies found that using AI/ML on multimodal imaging markers improved ideal patient selection for clinical trials and reduced sample size while maintaining high statistical power. One company increased enrollment for a breast cancer study by 200% with its ML platform.
• Creates targeted drugs. Genetic Engineering & Biotechnology News reports that ML models provide scientists with the data they need to more accurately identify precise structures that make a drug more effective. In one study, researchers used an ML-based approach to predict drug-binding targets with up to 90% accuracy on small molecules. The technology also identifies connections between different drug classes, indicating potential new uses for existing drugs. Johnson & Johnson uses ML algorithms to help identify tumor differences to develop medicines specifically for those subsets of patients.

With ML’s unique ability to analyze large amounts of complex data, researchers can design trials with expedited timelines, reduced costs and increased odds for success. By leveraging ML, these trials also lower patient burden and improve retention. According to Jeff Headd, vice president of commercial data science for Janssen North America Business Technology, “Using the latest innovations in AI and machine learning (ML), we are able to quickly analyze these vast datasets (including electronic medical records, lab results or even medical imaging like X-rays, MRIs and CT scans), uncover new insights and then drive actions with real potential to improve patient outcomes.”

AstraZeneca reports that implementing an industrialized ML platform allowed it to “streamline pharmaceutical drug discovery, clinical trials, and patient safety for hundreds of scientists.” Meanwhile, a group of researchers reported they used ML to identify eight current drugs that “hold great potential” for being beneficial to patients with Alzheimer’s disease. Finally, the makers of a cancer drug reported that they saw drug outcome review shorten from 20 months to 20 days, and the drug development process was reduced from an estimated 10-18 years to an expected 4 years by incorporating ML into their trial. 

ML Challenges

One significant concern about the use of ML algorithms is that they are created by humans, which means there could be bias and human error built into the models. In turn, that can cause the models to produce inaccurate data. Poor data can then lead to bad decision-making and unreliable trial outcomes. Additional concerns about ML include compliance, as it may put patient data at greater risk of exposure and increase the risk of data generation bias. For example, an ML system gave Black patients a lower risk score compared with white patients because it misinterpreted an algorithm meant to indicate that Black patients had less access to healthcare (rather than needing less healthcare than equally sick white patients). Other common ML challenges include:

• Poor quality data. Without high-quality data, the performance of ML systems suffers. Many organizations currently do not have this type of data.
• Research gaps. Research gaps can lead an ML system to value existing data insufficiently or incorrectly fill in data, negatively impacting analysis.
• Data access and sharing obstacles. Many organizations have obstacles that may prevent ML systems from correctly accessing or sharing vital information.
• Human capital shortage. Companies need knowledgeable, highly trained staff to optimize ML performance, but unfortunately, many lack these staff members.
• Regulatory uncertainty. Some organizations hesitate to embrace AI and ML in handling patient information and drug development.

Overcoming these challenges through proper planning and ML system oversight is key to taking full advantage of ML. Companies can then leverage ML to enter a new era of healthcare innovation in which drug development is accelerated, clinical trials are optimized and patient outcomes are enhanced.

Harnessing the True Power of ML

Randomized clinical trials have been the gold standard for medical research since the 1940s. Over the years, these trials have become expensive and time-consuming. It is now estimated that the average cost of a three-phase trial is $37 million (Phase 1 = $4 million, Phase 2 = $13 million, Phase 3 = $20 million), and the trials typically take several years to complete. Organizations now have a way to reverse this trend and run faster and cheaper trials – provided they take steps to maximize their use of ML. A recent study offered tips for using ML in clinical trials. The first is to use intuitive metrics beyond technical accuracy and include quality of care and patient outcomes. Second, identify themes of algorithmic bias and unfairness while developing mitigations to address these themes. Finally, “reduce brittleness and improve generalizability” and develop methods for improved interpretability of ML predictions.

Another study found that the optimum application of ML requires attention to data security and privacy and seamless integration with established medical practice knowledge. By taking the time to carefully implement ML, organizations can unlock advantages that improve trial accuracy and slash their trial expenditures. 

ML Offers Boundless Possibilities

Although there are challenges with using ML in clinical trials, including human error, poor quality data, human capital shortages and regulatory uncertainty, the positives far exceed the negatives. As ML continues to evolve, its integration into clinical trials will likely introduce a more efficient, patient-centric approach to research and drug development. ML makes trial participation easier by allowing patients to complete tasks remotely, such as expressing informed consent and completing questionnaires. It is also used to streamline enrollment and find more eligible study participants. One future advantage of ML is that it may help alleviate racial disparities in clinical trials. Previously, 90% of study participants in clinical trials were white. Now, ML classifiers and automated methods can be used to achieve greater representation. “Machine learning is going to play an increasingly important role in finding patients and filling in information that is not recorded in EHRs [electronic health records],” says Adam Dunn, head of biomedical informatics and digital health at the University of Sydney School of Medical Sciences.

What other changes could ML introduce to clinical trials in the future? Experts believe ML will continue to increase the efficiency of clinical trials and decrease the cost and time associated with bringing drugs to market. These improvements could be highly impactful in saving lives and reducing suffering. It is important to remember that ML systems are only as effective as the data they are based on and the algorithms that direct their actions. By optimizing algorithms and protecting patient data, organizations will be more likely to produce accurate, unbiased clinical study data analysis, deliver faster analyses, conduct more efficient trials and reduce costs. Savvy organizations will take a balanced approach that considers both the opportunities and challenges of ML and be better positioned to benefit from the technology while upholding the rigorous standards of statistical analysis in clinical research.