Real-World Evidence Meets Machine Learning: What It Takes to Future-Proof Evidence Generation

Real-world evidence (RWE) has become a crucial driver of the pharmaceutical and biotechnology industries in recent years. In fact, RWE is now included in 70 percent of new drug and biologic regulatory submissions to the U.S. Food and Drug Administration (FDA).

As RWE continues to drive breakthroughs, machine learning (ML) has emerged as a must-have tool for future-proofing evidence generation. Life sciences organizations now have access to more real-world data (RWD) than ever before, making a manual, human-driven approach to analysis through SAS scripting and ad-hoc queries unrealistic and unsustainable.

Leveraging ML models is the most effective way to wrangle and make sense of RWD at scale. It empowers life sciences organizations to analyze more data from a greater variety of sources, conduct ongoing analyses, and carry out more complex analyses with the help of high-performance computing (HPC) and cloud compute.

Regulatory agencies have taken notice. The FDA recently released guidance regarding the use of artificial intelligence (AI) in support of regulatory approvals, and updated a discussion paper earlier this year to cover the role of AI and ML in the drug development lifecycle.

It’s no wonder regulators are paying close attention — the promise of RWE paired with ML is tremendous. With ML-powered RWE, scientists can create synthetic control arms that accelerate research timelines, drastically reduce costs, and cut patient recruitment demands by 20-50 percent.

Furthermore, ML models can identify patient subpopulations most likely to benefit from particular treatments, or those at highest risk for adverse events. They can also predict how certain patients will respond to various therapies based on their unique clinical and genomic profiles. Additionally, ML models can continuously trawl the FDA Sentinel system to uncover potential adverse events exponentially faster than manual queries.

Every life sciences organization needs to consider how best to apply ML to RWE to support better patient outcomes. Here’s how they can ensure RWE is “ML-ready” and operationalize ML for use in GxP-compliant environments. 

Getting RWE ML-ready

To get RWE ML-ready, life sciences organizations should treat it like a product: planning for it, deploying it, and taking the appropriate steps to ensure “customer satisfaction.”

Here’s what that looks like: 

1. Planning

Organizations need to create a clear data governance framework for RWD that defines ownership, access controls, and usage policies. This fosters trust and helps uphold compliance with regulations like HIPAA and GDPR. They should also create a data infrastructure by defining common data models (e.g., OMOP, FHIR, SDTM)and data quality frameworks (e.g., completeness, conformance, plausibility, timeliness).

2. Deployment

To successfully deploy RWE, organizations need an environment that enables analysis the right way (more on the tools that enable this shortly). First and foremost, this environment should be standardized to jumpstart validation. It should support reproducibility and have built-in compliance and governance protocols. Importantly, it must also easily connect to and transform RWD for efficient and reliable analysis, without compromising data integrity or regulatory compliance. Finally, this environment must allow for seamless collaboration between stakeholders (e.g., quality assurance, data science, and clinical teams) during the build process. 

3. Ensuring “customer satisfaction”

Ensuring “customer satisfaction” in this context means the environment should be built not only for data scientists, but for the RWE and business teams that will make important business and clinical decisions based on ML models. These teams will need mechanisms to view the models and their outputs, provide insights and feedback, and govern the models to prevent drift, bias, etc., as the data feeding them inevitably evolves over time.

Operationalizing ML for RWE

Historically, operationalizing ML for RWE has caused issues when life sciences organizations attempt to use legacy approaches within GxP environments. This is because teams often work in non-validated environments with uncontrolled codebases, and attempt to “wrap” validation around the final output after the fact. This approach doesn’t hold up to GxP standards or offer reproducibility.

Today, GxP-ready data science platforms have compliance and governance baked directly into the platform itself to meet regulatory expectations. They support reproducibility, governance and audit trails, and model monitoring and management so life sciences organizations can stand up to regulatory scrutiny. These platforms support the entire model lifecycle, from exploratory analysis to deployment and monitoring. They can also connect to external data sources, like APIs, without requiring as much pipeline engineering as traditional analysis.

Additionally, they can build visualizations or applications that allow business users to see and understand the insights their ML models are unearthing. The result? Life sciences organizations can safely and effectively apply ML to RWE to unlock deeper insights while maintaining compliance and control over their models. 

ML-powered RWE is no longer just an interesting experiment — it’s a necessity driving critical advancements. But before life sciences organizations can harness the full power of ML, they need to get their RWE ML-ready and adopt tools that support compliance and reproducibility in GxP environments. By prioritizing these initiatives, they can future-proof evidence generation and pave the way for the next wave of ML-driven innovation in life sciences.

Photo: claudenakagawa, Getty Images

This post appears through the MedCity Influencers program. Anyone can publish their perspective on business and innovation in healthcare on MedCity News through MedCity Influencers. Click here to find out how.