In a presentation at the American Society of Clinical Oncology (ASCO) Annual Meeting (June 4, 2018; Chicago, IL), Robert S Miller, MD, FACP, FASCO, discussed the lessons learned from designing and improving CancerLinQ, a health information technology platform aimed at enhancing and improving the understanding and treatment of cancer through real-time data.
Journal of Clinical Pathways spoke with Dr Miller about CancerLinQ data curation, documentation management, phenotypic data, and the future of the quality improvement platform.
You mentioned in your presentation that improving data capture at the source is the key to curation for CancerLinQ. What strategies would you recommend for such improvement?
Dr Miller: Information in a medical record that exists as structured data is usually considered “computable” – its format (syntax) and meaning (semantics) can be understood by a computer, which in turn means it can be stored in a database for future retrieval to be manipulated, shared, or transformed via a set of rules or computer instructions. Many types of clinical data are inherently structured, such as cancer stage (often expressed as a single number or in some cases a short string of numbers and letters), Eastern Cooperative Oncology Group performance status (usually an integer from 0 to 5), or a lab value (eg, glucose = 83 mg/dl). However, in many cases, recording data in a medical record is done as free-text, particularly recording a patient’s history, since the “story” is what is important – what happened when, what symptoms the patient had, how the patient felt physically or emotionally, what external challenges were faced, and so forth. Consequently, the electronic health record (EHR) is full of unstructured, free text notes that may accurately reflect the patient journey but are not computable. This greatly impacts CancerLinQ because the platform requires someone to read the pathology report or clinic note to glean insights into the clinical events – a process which is time-consuming, expensive, and ultimately unsustainable. It would be far better if a greater percentage of medical information—especially data types like cancer stage—would remain as structured data. However, convincing doctors to do this is very difficult, since it is a much more time-intensive method of charting, despite the potential for downstream benefits.
Our strategy at CancerLinQ is to take a middle ground. We have begun a project at ASCO and CancerLinQ to bring together leading volunteer content experts among our members to define a core set of data elements for the EHR. These elements would essentially be in every oncology patient chart and would not require manual entry by the clinician. In addition, we expect that there will continue to be advances in machine learning algorithms and natural language processing to allow a computer to start to understand unstructured text and extract the meaning.
Can you explain the role and importance of unstructured content types in CancerLinQ? How do unstructured content types complicate documentation management?
Dr Miller: As aforementioned, many of the important oncologic data elements that are critical to understanding basic patient or tumor characteristics, treatment details, and patient outcomes are present in different types of unstructured documents in the EHR (eg, clinic notes, pathology reports, imaging reports). Our current solution, as discussed in my ASCO presentation, is to use teams of “human curators” or manual data abstractors. Generally speaking, the abstractors consist of nurses and others with some degree of clinical knowledge who are comfortable reading and extracting information from patient records. They are tasked with combing through these documents and manually pulling out the desired key elements. Their work is facilitated, but not replaced, by natural language processing technology. They focus on certain key concepts like cancer stage, biomarkers, tumor progression, and adverse events. This process also requires practices to allow CancerLinQ to retrieve their stored, scanned documents, often from a different database (which requires more work for the practice). In addition, there is a lot of local variation and a lack of quality control, rendering some documents unusable due to problems with run-on’s (two document erroneously combined into one) or pagination errors. CancerLinQ is working with the teams to try to minimize these challenges.
As more and more practices begin to use CancerLinQ, how is this quality improvement platform suited to meet the needs of increasing users?
Dr Miller: CancerLinQ was first developed as an extension of ASCO’s quality portfolio, most notably the Quality Oncology Practice Initiative (QOPI). The promise of CancerLinQ is to deliver an automated way to do QOPI, running directly against practice EHR data, therefore bypassing expensive and time-consuming manual data abstraction (as in the current QOPI program). While electronic QOPI is not fully developed at the moment, in 2018 we expect CancerLinQ to be able to deliver a set of quality measures that practices can use for submission of their required quality reporting to CMS under the MACRA/MIPS program. In addition, we are working on a related set of measures to be included in CancerLinQ that can be used to qualify practices to participate in the ASCO QOPI Certification program. We will continue to provide a set of tools and dashboards for practices to see their quality measure performance and how it compares to other practices in CancerLinQ.
How do you envision the future direction of CancerLinQ? What is the role of phenotypic data in its growth?
Dr Miller: CancerLinQ will continue to grow and develop in several ways in the coming years. We continue to onboard our signed practices and get them started on the basic quality improvement platform. We are looking to grow our overall base in the US and eventually internationally. As aforementioned, we are looking to deliver several specific tools related to clinical quality measures that can be used for Federal compliance reporting in the MIPS program and for ASCO’s QOPI Certification. Additionally, we intend on developing reports, dashboards, and other tools for improved user tracking of patients and real-time quality performance tracking. We are looking to develop a type of “app store” model where third-party developers can create specialized software applications that can run within the basic CancerLinQ platform. We will continue to deliver aggregated, de-identified “real-world evidence” datasets through our licensees Tempus and Concerto HealthAI for industry and life-sciences customers, as well as internally through ASCO for academic and community researchers who wish to use the CancerLinQ big data to answer specific queries.
The job of improving the basic CancerLinQ data quality never ends, and it is the joint responsibility of the CancerLinQ staff, our collaborators, and the participating practices. As noted, we will be working to develop a standard set of data elements to improve data capture from the EHR. We will be bringing in additional data sources such as tumor registry data and third-party obituary data to better characterize vital status. In addition, we are considering a number of potential solutions to bring in molecular data, especially somatic tumor genomics, that can be linked with phenotypic data to capture key precision medicine endpoints.
What is the outlook of CancerLinQ toward its collaborators and participating practices?
Dr Miller: CancerLinQ is rapidly evolving into the largest and most robust source of real-world evidence ever assembled in oncology for comparative effectiveness research and discovery. We expect that the data set will deliver multiple key insights into the many important clinical questions facing oncologists and patients today, and we look forward to presentations and publications with our many collaborators.
However, our first commitment remains to the practices that contribute data into the system. We have been gratified by the tremendous response of the oncology community, and honored by the trust of the nearly 100 institutions around the US that have agreed to be part of our network. To them, we look to continue to deliver real-time quality assessment and practice tools while optimizing the power of big data analytics as we build the rapid-learning health system in oncology.