By Joseph Zabinski, PhD, Senior Director of AI & Precision Medicine, OM1
As innovation in devices and diagnostics accelerates, research approaches and outcomes programs also need to advance to support these developments with high-quality evidence. At the same time, demonstrating the effectiveness of a device can be challenging because outcomes depend highly on clinician training and vary in different care settings. While critical to development, traditional clinical trials and registries are often too impractical, too time consuming, or too costly to use to evaluate all safety, effectiveness, and value questions for devices and diagnostics.
Real-world data (RWD) and artificial intelligence (AI) are playing growing roles in generating evidence and improving decision-making based on outcomes. By definition, RWD is collected in the real world from patients with broad ranges of characteristics and heterogenous disease, and from many different care settings and geographies. All these data are ‘imperfect’ in some way – again, reflecting the real world – but by bringing them together, they allow patterns to be explored and outcomes to be evaluated in ways that aren’t possible in clinical trials, and even in traditional registries.
While RWD and AI are growing in use and applications, understanding what constitutes high-quality RWD and AI model outputs to meet different stakeholder needs can seem confusing. Even the choice of where to start with these tools can be overwhelming, and hype without practical follow-through doesn’t help clarify much. While there are many different use cases for RWD and AI, we’ll highlight a few areas where success has already been demonstrated: meeting regulatory requirements, finding patients, and enhancing decision-making in clinical workflows.
Regulatory: Expand indications and meet post-marketing commitments (PMCs)
RWD is being used more and more for regulatory decision making. Between the 21st Century Cures Act and the FDA guidance document Use of Real-World Evidence to Support Regulatory Decision-Making for Medical Devices, regulators are building clarity around what RWD are and how they can used to generate real-world evidence. You can find more information about how the FDA defines RWD and RWE here.
Transparency and traceability are the keys to success when using RWD to satisfy regulatory requirements. Quality begins with intentionality – it can’t be baked in later. As with traditional data collection in clinical trials, protocols should guide data collection when working with real-world data sources. Think about what data should be collected upfront. Trace every data element from its origins through transformations to the final variables that will be used for analysis. Assess quality along the way, considering characteristics of each unique data source: a global assessment of a variable’s completeness, for example, could miss specific sources with particularly rich data availability for that variable. Use technology to tackle the enormity and chaos that comes with dealing with multiple data sources. Finally: missingness is inevitable even from the ‘best’ sources, and certain measures of interest may not be routinely collected in clinical practice. Fill in the gaps by tapping into unstructured data, such as physician notes, in a reliable and validated way.
Some companies, like OM1, do much of the complex work of sourcing data and linking, cleaning, standardizing, and normalizing them to prepare for research and regulatory use. Consider the case of a manufacturer of an oncology device used for gastrointestinal cancer. Looking to demonstrate patterns of real-world use and safety and effectiveness for an expanded indication, OM1 helped prospectively collect and transform disparate data, gathered from a range of EHR sources and including patient reported outcomes (PROs), into a single usable data source that ultimately included 35 sites and approximately 600 patients. In addition to reducing the burden of monitoring, reducing bias (since data for all qualifying patients were automatically ingested), and improving data completeness for untreated patients, the manufacturer was also able to use the data to construct comparator cohorts.
The regulatory use cases for RWD and subsequent real-world evidence (RWE) span the device product lifecycle, including pre-market submissions, de novo applications, humanitarian device exemptions, and post-marketing monitoring. The FDA has collated a growing list of regulatory decisions supported by real-world evidence between 2012-2019, which can be found here.
Find Patients: Enhance clinical trial recruitment
RWD and AI together are a powerful combination for finding patients. Device manufacturers have enormous unmet need in this area, including for trial recruitment and feasibility analyses. Covering millions of patients, real-world datasets can accelerate access to relevant cohorts and enable analysis of gaps in care, development of protocols, and evaluation of new uses for existing devices.
One use case where RWD and AI truly shine is in finding un- or mis-diagnosed patients, including those at higher risk of needing some sort of intervention, such as surgery. Using large multi-source datasets that pull together more complete patient journeys, and then applying AI modeling to those data to look for common themes, researchers can discover patterns that would not be found through manual analysis or traditional techniques alone. Sup-populations of patients with distinct disease or condition characteristics and treatment responses frequently emerge. This technology can also be tuned to find specific patient characteristics, symptoms, labs and other parameters flagging patients who are more likely to have an undiagnosed condition or may benefit from an intervention.
At OM1, we’ve built a patient finder tool that can help with these types of scenarios. For example, consider abdominal aortic aneurysms (AAA). Screening guidelines to identify AAA patients capture a minority of cases, leading to a large number of patients left undiagnosed. Working with a device manufacturer, we used both our RWD cloud and AI platform to address this gap and find undiagnosed patients. We identified approximately 500,000 AAA patients in the OM1 RWD Cloud. We used a targeted AI modeling approach to identify twice as many patients as would’ve been found through screening guidelines alone. These higher-risk patients can be flagged for referral to diagnostic screening, including those who fall outside current screening guidelines, to arrive at a potentially life-saving diagnosis more quickly.
Impact Care & Outcomes: Improve clinical decision-making and value-based care models
Beyond clinical development and regulatory applications, RWD and AI can be used to differentiate products for reimbursement and for personalizing treatment options. In addition, providers are entering value-based contracts with both payers and employer groups more and more often, with some degree of ‘payment for value’ in these arrangements. These relationships are driving secondary pressure on manufacturers to provide more evidence of both clinical and cost-effectiveness. RWD and AI are helping meet these needs, too. For example, manufacturers and providers can use predictive models to identify likely ‘bundle breakers’ in bundled payment surgical settings, before surgery actually occurs, to manage risk. At OM1, we’ve used our AI platform to tackle this problem in spine surgery applications. The platform uses RWD to generate preoperative OM1 Medical Burden Index (OMBI) scores – customized for spine patients – that help identify which patients are at higher risk of greatest postsurgical resource utilization, so care teams can help reduce this risk.
Shared decision-making is another use case where manufacturers and providers can work collaboratively through technology to deliver more personalized care with a focus on improving outcomes. Shared decision-making happens when clinicians and patients use clinical evidence to make decisions together on treatments, procedures, and care. Predictive models can be built leveraging real-world data, including PROs, to determine a personalized risk-benefit profile and help drive the conversation around what outcomes could look like following certain treatments, such as surgery.
For example, at OM1, we’ve built a shared decision-making tool called OM1 Joint Insights™, in collaboration with Dell Medical School. Joint Insights combines PROs, patient medical history data, education, and predictive modeling to deliver personalized risk-benefit predictions that better inform clinical decision making and appropriateness of surgery for total joint replacements. Clinical trial results, published in the Journal of the American Medical Association (JAMA), demonstrate that Joint Insights positively impacts decision quality, shared decision making, patient experience, and functional outcomes in adults with knee osteoarthritis (OA) considering joint replacement.
By sponsoring these tools and programs, device manufacturers can support evidence-based care and help improve clinical outcomes, while also having access to critical data that can further inform research and development and value-based initiatives.
In all of these use cases and examples, RWD is the foundation on which high-performance, reliable predictive models are generated. Once AI models are validated, they can be used to optimize diagnostics, generate personalized predictions for candidate for treatments, improve how interventions are targeted, and drive how patients with specific risk factors are identified and helped prior to undergoing procedures.
In summary: Explore the possibilities
The terms ‘RWD’ and ‘AI’ have become watered down from overuse, and yet hyped up in ways that detract away from the real value that both are providing today. RWD and AI are transforming how clinical outcomes are being evaluated and measured, and how care is being delivered. Their use cases continue to expand as healthcare stakeholders become more comfortable with the use of both, as real-world datasets become more accessible, and as AI tools evolve to deliver meaningful insights into care pathways. While it is critical that healthcare stakeholders continue to insist on high standards with RWD and AI, we shouldn’t be afraid to push towards delivering on the promise of better and more precise treatments and care through these tools.
