Best Practices for Incorporating Clinical Imaging Data into RWE Programs
January 28, 2019

Best Practices for Incorporating Clinical Imaging Data into RWE Programs

In our last post, we talked about the current course of real world evidence (RWE) and the recently announced FDA framework for its RWE program. In this post, we will identify the best practices for incorporating imaging into RWE to accelerate the drug development cycle and improve the confidence in the final clinical arguments made against safety and efficacy.

The FDA framework states that there are significant data gaps in current clinical trial designs as electronic health records (EHRs) and medical claims data–the predominate sources of analytics–do not accurately capture a patient’s clinical circumstance. These platforms were predominantly designed to accurately facilitate and manage billing and payment activities, leaving material holes in accessing clinical data for decision making. For example, an EHR or billing system might indicate that a CT scan was performed and how much the diagnostic exam cost, but it will not readily identify the tumor that that was found nor its size and composition. This is why, for many therapies and clinical trials, medical imaging is an essential component of RWE. While access, aggregation and normalization of medical imaging has been challenging in the past, it is now becoming more available. To be truly useful, medical imaging must be integrated with related data sets and optimized on a broad scale. This process must meet five requirements:


1. Derive evidence from a broad base of sources
On a population basis, data from a single health system or group of facilities tends to cluster around reasonably consistent treatment patterns, thus creating institutional bias relative to specific patient outcome profiles. This bias can materially affect the evaluation of a therapy, but can be overcome with a conscious effort to produce evidence diversity.

As an example, imaging for a trial conducted in a leading academic medical center (AMC) where the institution has the resources to upgrade their machinery routinely, maintains state-of-the-art training for technicians, implements new software versions immediately and has access to a highly educated labor force will likely produce results on a population basis different from the results from an institution with fewer resources, older machines, minimal maintenance and junior technicians.

Thus the inclusion of diverse data is important to credibly evaluate trial results. The variables that contribute to defining a diverse data set include: influences pocketed by geography such as diet, environment, education, culture and the availability of healthcare resources; the influences of payer mix, demographics, benefit plans, and even the influence of medical economics that can define the currency of medical tools, training, availability and the skills of medical personnel.

To make RWE most effective, imaging and other medical data must be drawn from varied populations that include research and non-research settings, AMCs and community hospitals, publicly and privately funded institutions, highly insured and uninsured patients, and eventually be inclusive of patients across the globe.

The general rule: work with partners who have access to broad networks of providers to diversify evidence and minimize bias.


2. Tackle volume, veracity and velocity of linked data assets

While it is important to note that the quality of data is far more important than the quantity of data pulled into statistical models, the nature of our healthcare data today is siloed and disconnected. To deal with this problem, very large data sets must be accessible to match against other large data sets in order to locate and normalize enough information about enough patients to make RWE effective.

As an example, a trial for a new drug might be looking for 300 patients, but for each patient it needs five years of prescription information, four years of blood tests, five years of imaging data, a genomics test and other demographics such as age. The problem, however, is that patients have used multiple pharmacies, may have had multiple health plans over time, received imaging from multiple institutions and had blood tests performed at outside facilities that appear in two different EHRs and even scanned fax formats. Due to the difficulties in acquiring data in the first place, the RWE need might have to be addressed by combining data from 100 million claims across three different payers, with 60 million pharmacy benefit manager records and query 30 million patients from different EHR systems while sending manually managed CDs just to find 300 patients that meet the inclusion and exclusion criteria. The state of the art today, therefore, requires large data sets.

Interestingly, this phenomenon is correlated to the development of artificial (or augmented) intelligence as machine learning models require large, varied examples to operate. Indeed, access to massive context-sensitive imaging and other medical data from heterogeneous populations has challenged the effectiveness and slowed the development of both life science and AI advancements.

The general rule: work with partners who have access to large data sets and have the ability to combine their data with other diverse data sets.


3. Include the clinical context and align with patient-specific clinical decision-making
The power of RWE is that it more accurately represents how healthcare is delivered and experienced in reality. The art and science of medicine, however, is so highly influenced by many factors from biology to human behavior, that the performance of a static drug compound, physical device or even a computerized diagnosis then affects people differently. The only way to improve the statistical models used to determine safety and effectiveness is to see how the test performs against populations in the actual context as executed in the clinical workflow at the point of care.

By definition then, RWE needs to come from the context of actual medical decisions and activity, as opposed to being influenced by how healthcare is paid for or tracked to accomplish billing, marketing or other requirements of the system. The continuing challenge of RWE is in the creation and expansion of tools to make the unstructured data populated by clinicians for clinicians as represented in imaging, notes, reports, texts, voice and even fax available, accessible and interoperable.

The general rule: work with partners who at the core of their activities work inside clinical workflows and participate in or influence actual clinical decision-making. The data will therefore more accurately represent reality.


4. Be device, standard, manufacturer and software version agnostic and fully interoperable

In particular, imaging must be easily accessible from any capture device, software version and modality in a standard format. Technical standards for images from devices or storage for visualization can differ greatly from manufacturer to manufacturer, therefore the data must be standardized, normalized and interoperable to be useful. The ability to pull data from disparate picture archiving and communications systems (PACS), dermatology cameras, PET scans or elastography indifferently is crucial to support the needs for data breadth, diversity and quantity.

The general rule: avoid working with partners who rely on proprietary workflows or utilize proprietary software and technology architectures. The lack of interoperability will at best slow processes and at worst bias data.


5. Be unambiguously indexed to a specific patient

In medical imaging specifically, but also with other types of data, each device type, storage system institution, EHR, PACS, modality and third-party vendor use different information to reconcile patients. Master patient indexes exist and will continue to evolve, but since comprehensive multi-storage patient indexing technology has not been historically applied to imaging, organizations looking to enhance their data sets with imaging RWE must work with partners who have enabled this technology and have the ability to normalize and combine imaging with diverse data sets.

The general rule: work with partners who have invested in and understand how to digitally recognize and index patients across providers, geographies and institutions across their complex patient journey.


Imaging data holds the biggest potential to transform research

  • Imaging is the most mature diagnostic evidence for many diseases
  • Innovations in AI are largely being applied directly to image analysis first, promising results earlier in the research lifecycle
  • Imaging can uniformly demonstrate disease progression and drug impact in many therapeutic areas
  • Imaging is instrumental in helping find and confirm molecular biomarkers
  • Pathology information is used for key decisions in oncology treatment selection


Incorporating image data into the entire evidence cycle

At Life Image, we have been developing solutions, like our Interoperability Suite, that make a patient’s imaging and clinical data fully interoperable and accessible regardless of file type or EHR vendor. Our network connects more than 1,500 facilities in the U.S. and 8,000 affiliated sites, including eight of the top 10 U.S. hospitals, with 150,000 U.S. providers and 58,000 clinics globally, which can finally provide the long-desired integration of clinical imaging into RWE.

Let us know what you think about the new FDA framework and we are happy to talk with you more about how our solutions are working to incorporate imaging data into RWE. Be sure to follow us on Twitter and LinkedIn, and like us on Facebook to stay up to date as we look to help solve some of bigger and most complex challenges in health care.


Matthew A. Michela, President and CEO

Matt Michela

President and CEO