As technology and cancer care continue to become intertwined at every level, there is an imperative need to effectively process and use vast amounts of data to provide the highest level of care for patients.
As technology and cancer care continue to become intertwined at every level, there is an imperative need to effectively process and use vast amounts of data to provide the highest level of care for patients. Learning health care systems represent an evolving model of cancer data integration and application that has the potential to positively affect care for many patients.
“What we try to do [with learning health care systems] is to make it easier for providers to take all the information at their disposal and use it in real time to come up with the best treatment option and care plan for their patient,” Duncan Allen, MHA, the senior vice president of clinical services at OneOncology in Nashville, Tennessee, said in an interview with OncologyLive®. “We need to make sure that the systems we’re [using] can ‘talk’ to each other. Do we have the right data, infrastructure, and interfacing in place, from when a new approval comes out to the time when providers are putting that new therapeutic regimen into practice?”
The Institute of Medicine describes the ideal learning health care system as one where, “…science, informatics, incentives, and culture are aligned for continuous improvement and innovation—with best practices seamlessly embedded in the delivery process and new knowledge captured as an integral by-product of the delivery experience.”1
The American Society of Clinical Oncology’s CancerLinQ (Learning Intelligence Network for Quality) is a learning health care system designed to collect clinical data from electronic medical records (EMRs) and provide clinical decision support based on available guidelines. The developers of CancerLinQ also designed the learning health system to capture processes and outcomes, allowing investigators to track trends and identify areas of unmet need (FIGURE).1
Results of a prototype leveraging CancerLinQ showed that the system was able to accumulate data from 130,000 cases of patients with previously treated breast cancer from the individual EMR systems of 5 large practices, surpassing the original target goal of 30,000 cases.1 The system then used the data to provide recommendations via a decision-support dialogue box within an open-source EMR that can be accessed by the clinician. Additionally, 10 Breast Cancer Quality Oncology Practice Initiative performance measures with compliance ratings were accessible. The system also included software that enabled the clinician to visualize the data set. The CancerLinQ prototype was assembled in 5 months.1
Nearly a decade after the initial design, investigators continue to use the CancerLinQ system. For example, in 2021, Ray et developed a prognostic tool using CancerLinQ data to predict risk of death within 30 days of a clinical encounter in patients with metastatic breast cancer (MBC). Data drawn from the EMR included age, vital signs, laboratory values, performance status, pain score, and time since chemotherapy, as well as estrogen receptor, progesterone receptor, or HER2 status from January 2000 to June 2020.2
The system identified 9270 patients, representing 586,801 encounters. The model had a prediction accuracy of 86%, positive predictive value of 50%, sensitivity of 70%, and specificity of 88%. Specifically, predictors of mortality within 30 days for patients with MBC included chemotherapy within 1 year but not the past 30 days (odds ratio [OR] 1.92; 95% CI, 1.67-2.20), opiate use (OR, 1.71; 95% CI, 1.17-2.52), and high pain score with no opiate use (OR, 1.27; 95% CI, 1.10-1.48).2
CancerLinQ is currently used by more than 100 oncology practices and contains real-world data representing more than 6 million patients.3
A major proposed addition to the learning health care system is the integration of genomic data. Targeted therapies that require precise knowledge to be effective—ie, mutations are being expressed in a disease—are becoming increasingly prevalent in cancer care. The addition of genomic profiling to the learning health care system will allow clinicians to quickly identify agents that will be effective for a patient depending on their specific mutations. Similarly, these data can be used to match patients with the appropriate clinical trials for their unique disease, Allen explained.
To achieve this, investigators have determined that the primary infrastructure elements of the learning health care system must include a common cancer genomic profiling (CGP) vendor, a common EMR system, integrated genomic results, and access to clinical trials. The CGP and EMR systems are then integrated to include clinical and genomic data together; these data are then moved into the point of care so to be easily accessed by the clinician.4
“The biggest elements are data and information consumption, making sure that you have solutions across the types of information that a provider requires to be practicing at the top of their license [and] then making sure you have the right connectivity built so it can be seamless for [clinicians] to figure out what treatment is best for their patients,” Allen said.
One of the biggest roadblocks to incorporating genomic profiling in a learning health care system is working to ensure that all patients who are eligible for genomic profiling receive it. Strategies to overcome this challenge include increased patient education efforts by clinicians to help patients understand all their treatment options.
In the hematology space, investigators who are part of the American Society of Hematology Research Collaborative, a nonprofit organization, are seeking build a data infrastructure platform designed to support learning health systems. The goal of the project is to provide the most effective treatments to improve the lives of patients by generating and applying clinical evidence from clinical, administrative, laboratory, and patient-reported sources. The organization is comprised of 2 primary initiatives: the data hub and the clinical trials network.5
The data hub draws on information from EMR integration, direct data entry via electronic data capture, and external data sources to better develop clinical care delivery sites for hematologic diseases. It optimizes the exchange of information via interconnected dashboards, queries, and research contributions, highlighting gaps in care and patient outcomes.
Multiple myeloma was chosen as an initial disease area of focus for the initiative because data sharing opportunities using real-world data are especially timely. Additionally, investigators note that multiple myeloma represents a prime opportunity for the learning health care system model because large data sets are required for the development and validation of prognostic biomarkers and because therapy sequencing decisions becoming more complex.
Investigators designed the data hub to encompass disease-specific end points, clinical guidelines, and other evidence-based documents to serve as core metrics. Baseline computable phenotypes (e-phenotypes) for various clinical concepts are formed using value sets from clinical terms drawn from EMR documentation codes. Data that may not be reliably recorded in individual patient EMRs are also sourced, leveraging local research data and/or disease-progression metrics. All data are funneled into the site dashboard, where trends may be referenced by the clinician.
The collective plans to develop a comprehensive e-phenotype knowledge base, with an emphasis on being able to easily share this information and having the e-phenotypes mature and inform treatment decisions over time. Developers plan to create algorithms based on an inclusive list of codes used to determine the sensitivity for capturing information to predict the positive value of the proposed concept. Machine learning and artificial intelligence can then be used to determine contributors to the e-phenotypes when rules do not exist or are not applicable.
“What we can do today that we weren’t able to do 15 to 20 years ago is take aggregated patient information that is deidentified and HIPAA [Health Insurance Portability and Accountability Act] compliant and utilize it in making the best treatment decisions for patients,” Allen said. “With all the information we’ve gathered, including clinical, genomic, financial, and operational data, we can help select the best therapeutic regimen for our patients. From a patient perspective, they significantly benefit from how the field of informatics has developed to a state where we can leverage big data to make much [better] decisions.”
It is important to note that learning health care systems are not perfect by any means and that constant work is being done to improve and optimize the model, Allen said. In many instances, connectivity issues impede fast and accurate communication between the EMR and CGP vendors. Fragmentation of the workflow can negate many advantages provided by the learning health care system model.
To address these shortcomings, firms such as OneOncology are using physician-driven councils to provide ongoing feedback on the systems they are building. These physician advisory boards can help inform the developers of learning health care systems what areas to focus on next, the biggest issues for practicing oncologists in the clinic, and how to continue to streamline the systems for ease and speed of use. Clinicians are also working to improve the interoperability between different EMR systems to better facilitate their integration into a learning health care system model.6
“Like everything [today], health care is getting smarter,” Allen said. “Much of that intelligence is driven from how you marry data together to extract more value. The future of learning health systems is that they’re much more automated. You’ll see more consolidation, especially in the oncology space, around partners where people are coming up with full suites of software that can help [further] unlock the learning health system approach.”