Journal:Creating learning health systems and the emerging role of biomedical informatics

Full article title	Creating learning health systems and the emerging role of biomedical informatics
Journal	Learning Health Systems
Author(s)	Kohn, Martin S.; Topaloglu, Umit; Kirkendall, Eric S.; Dharod, Ahay; Wells, Brian J.; Gurcan, Metin
Primary contact	Email: mkohn at wakehealth dot edu
Editors	Wake Forest School of Medicine's Center for Biomedical Informatics
Year published	2022
Volume and issue	6(1)
Article #	e10259
DOI	10.1002/lrh2.10259
ISSN	2379-6146
Distribution license	Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International
Website	https://onlinelibrary.wiley.com/doi/10.1002/lrh2.10259
Download	https://onlinelibrary.wiley.com/doi/epdf/10.1002/lrh2.10259 (PDF)

Abstract

Introduction: The nature of information used in medicine has changed. In the past, we were limited to routine clinical data and published clinical trials. Today, we deal with massive, multiple data streams and easy access to new tests, ideas, and capabilities to process them. Whereas in the past getting information for decision-making was a challenge, today's clinicians have readily available access to information and data through the multitude of data-collecting devices, though it remains a challenge at times to analyze, evaluate, and prioritize it. As such, clinicians must become adept with the tools needed to deal with the era of big data, requiring a major change in how we learn to make decisions. Major change is often met with resistance and questions about value. A "learning health system" (LHS) is an enabler to encourage the development of such tools and demonstrate value in improved decision-making.

Methods: In this work, we describe how we are developing a biomedical informatics program to help our medical institution's evolution as an academic LHS, including strategy, training for house staff, and examples of the role of informatics from operations to research.

Results: We highlight an array of LHS implementations and educational programs to improve healthcare and prepare a cadre of physicians with basic information technology (IT) skills. The programs have been well accepted with, for example, increasing interest and enrollment in the educational programs.

Conclusions: We are now in an era when large volumes of a wide variety of data are readily available. The challenge is not so much in the acquisition of data, but in assessing the quality, relevance, and value of the data. The data we can get may not be the data we need. In the past, sources of data were limited, and trial results published in journals were the major source of evidence for decision making. The advent of powerful analytics systems has changed the concept of evidence. Clinicians will have to develop the skills necessary to work in the era of big data. It is not reasonable to expect that all clinicians will also be data scientists. However, understanding the role of artificial intelligence (AI) and predictive analytics, and how to apply them, will become progressively more important. Programs such as the one being implemented at Wake Forest fill that need.

Keywords: biomedical informatics, learning health systems, physician education

Introduction

The nature of data and evidence in healthcare is experiencing rapid change. Personalized healthcare—making decisions that are more likely to benefit the individual—is evolving and requires more kinds of data than has been used until recently. Personalized decision making requires moving beyond the traditional data sources to understand the individual more fully. [1] The techniques used in the past, such as the null hypothesis analysis in randomized controlled studies (RCTs), are still important but insufficient to achieve personalization. The myriad streams of data that may influence health (environmental, domestic, ethnic, cultural, access, political views, wearables, genomics, etc.) leave us in the position of not understanding which of the streams is more important or how they interact. The recent creation of the sub-specialty of clinical informatics confirms that more sophisticated approaches, such as artificial intelligence (AI) and machine learning (ML), are required to use both the usual clinical data and real-world data. Clinicians must become adept with the tools needed to deal with the era of big data, requiring a major change in how we learn to make decisions. [2]

Innovation in health information technology (HIT), such as the electronic health record (EHR), has often been met with resistance and frustration because it was perceived as requiring extensive training, not supporting clinical workflow, requiring additional time to use, or interfering with interactions with the patient, ultimately not providing sufficient value.

The implementation of AI faces similar challenges. Will it show enough value in achieving improved decision making to justify the time and energy necessary to convince stakeholders to use it? Some of the obstacles to adoption include being comfortable with the way things have always been done, having anxiety about learning new skills, and having concerns about the loss of autonomy or the “art” of medicine.

Physicians, after undergoing many years of training, may be particularly resistant to change and the adoption of new technologies. Resistance to the adoption of new HIT often results from a lack of perceived usefulness, perceived threats, and perceived inequities. If physicians believe that a new technology is not useful, especially if it results in more complicated workflow or diminishes their role, it will be difficult to facilitate adoption.

Common human psychological biases may also make it difficult for clinicians to use the information from personalized predictive analytics. This includes confirmation bias (the tendency to emphasize data that confirms one's prior beliefs), the availability heuristic (being more influenced by memorable cases in one's clinical experience), and the endowment effect (the sense of loss from losing what you have, e.g., a cherished belief or conviction). [3]

Overcoming these obstacles requires designs consistent with healthcare's broad workflow, education goals, and demonstration of value. The "learning health system" (LHS) provides the structure to meet those needs. Including concepts of the LHS early in the educational experience is valuable. The Center for Biomedical Informatics at the Wake Forest School of Medicine (WFBMI) is developing a research and educational program to support the inclusion of LHS concepts into mainstream healthcare.

Requirements of a learning health system=

In the past, the evidence for decision-making developed slowly and episodically. A topic was studied with a series of RCTs, with the result becoming the standard until the next study was published. The feedback loop to evaluate the impact could be months or years long. Such a lengthy process does not support personalized healthcare. With sophisticated analytics and rapid access to data, we can update the system with ongoing feedback. An LHS embodies such a loop. Data is converted to knowledge, knowledge is used to perform an intervention, the results of the intervention provide new data, and the process cycles continuously.

The elusive LHS is often described by its characteristics [4]:

1. Every patient's characteristics and experiences are securely available as data to learn from.
2. Practice knowledge derived from these data is immediately available to support health-related decisions by individual members of society, care providers, and managers and planners of health services.
3. Improvement is continuous, through ongoing study addressing multiple health-improvement and related goals.
4. A socio-technical infrastructure enables this to happen routinely, with a significant level of automation, and with economy of scale.
5. Stakeholders within the system view the above activities as part of their culture.

Working effectively in an LHS requires education and training. We are dealing with increasing amounts of data, and more kinds of data than ever before. Identifying the valuable data and discerning patterns that allow better decisions is a new skill. We are moving in that direction, and biomedical informatics, particularly clinical informatics, is key in those efforts. Clinical informatics is the newest clinical sub-specialty, and training programs and fellowships in informatics are becoming common. Creating the environment to support the development of an LHS is an important goal in which the WFBMI is an active participant.

A by-product of the widespread digitization of healthcare over the last two decades is the large amounts of “digital exhaust” (i.e., information collected from the internet by a person or an organization) from technology supporting the clinical and administrative process of delivering care. In fact, as in other systems outside of healthcare, the creation and existence of data is exponentially growing. [5] This data can be leveraged to the benefit of patients, families, providers and staff, as well as the healthcare system at large. One of the main challenges that now exists is how to process and analyze the data to gather actionable insight and knowledge from it. One can easily get lost in “big data” when mining it, hence the need for data scientists and domain experts well-versed in healthcare processes, i.e., clinical informaticians. Without being familiar with the clinical delivery system, the data will not be viewed in the right context to allow this synthesis to occur, and for intervention opportunities to become clear.

Examples of using data to create knowledge to improve practice and operations

Horwitz et al. [6] have published a prime example of how to systematically identify opportunities to apply LHS principles and create value from those activities. In their approach, they largely applied quality improvement and data-driven approaches to achieve favorable outcomes without expending a large number of resources. Much of their value came from “exnovating” or removing unproven activities that required resource spend for what was discovered to be no gain. We are following their lead and beginning our own process, as are many others.

EHRs are tremendous sources of data, and clinical decision support (CDS) in particular is a rich target for much of our work. In 2019 alone, there were 29 million best-practice advisories that were visible to providers and staff, many of which were not acted upon and only served to generate noise in our clinical care delivery processes if they were providing no value. We have begun to systematically and rigorously examine the data for the best opportunities and have already identified large sources of inefficiencies. For example, several of our ambulatory clinics are among the top locations where an influenza vaccination reminder alert has fired, yet several of these clinics do not even have the vaccination stocked and available. Exemption of one clinic alone from receiving the reminder should save about 20,000 alerts from being seen by providers who cannot act on them. Given a conservative estimate of an average alert interrogation time (time a provider spends reading an alert) of five seconds, this equates to about 28 hours of provider time in one flu season, for one clinic site. These alerts are largely targeting physicians and other high-cost staff. A simple configuration change to the alert would save thousands of dollars, and more importantly free up clinician time, in one single clinic.

In the example above, data that are readily available is being used to generate insight about practice patterns, which is then used to create practice efficiencies and optimize provider time. By utilizing data to identify instances where care is not being optimized—but is creating useless, inefficient “busy work”—we are replacing that work with the freedom to do other activities to improve patient care, while improving provider satisfaction and removing a source of ire of physicians. [7]

Developing a biomedical informatics program to help an institution's evolution as a learning health system

When we created the biomedical Informatics program at Wake Forest School of Medicine (WFSoM) in 2018, we designed its structure to help Wake Forest's evolution as an LHS. Wake Forest's vision is to be “a preeminent learning health system that promotes better health for all through collaboration, excellence and innovation.” This cross-disciplinary initiative is designed to integrate resources throughout WFSoM, Wake Forest Baptist Medical Center, and Wake Forest University while complementing the work of other research centers, including the Wake Forest Clinical and Translational Science Institute (CTSI). To achieve this vision, the program pursues three interrelated goals:

• Catalyze: Integrate resources for research informatics across Wake Forest Baptist Medical Center and Wake Forest University, and synergistically support the informatics needs of other research centers, departments, and institutes.
• Discover: Facilitate biomedical discovery across the spectrum of informatics, including bioinformatics, imaging, clinical, translational, and public health informatics, and develop a portfolio of externally funded research in the area of biomedical informatics.
• Educate: Support and inspire the training of the next generation of investigators in the principles and practice of biomedical informatics.

WFBMI's plans for innovation through AI [8] will take many forms, including building and implementing EHR-based clinical alerts, decision support tools, structured data capture, and order sets, as well as facilitating pragmatic clinical trials by being able to randomize by clinics/hospitals. Some driving projects include computer-assisted assessment of otoscopy imaging [9-11], grading of follicular lymphoma slides [12-14], chest pain risk stratification [15], prediction of glycated hemoglobin values [16], and development of an EHR-based frailty score. [17]

Operational informatics

A well-functioning LHS would need informatics science, as well as the capability of generating knowledge and improving care delivery. WFBMI strives to establish a data infrastructure that enables repeatable and cost-effective learning. Starting with the computational needs of the researchers and business operations at our institution, WFBMI has invested prudently in elastic computing infrastructures resulting from contractual engagements and business associate agreements (BAAs) with the Google Cloud platform (GCP) and Microsoft Azure separately. The GCP serves as the flexible, research computing framework, with available tools to address the needs of precision medicine and other large-scale computing endeavors. WFBMI also utilizes Microsoft Azure as a common computing model for several data sharing and federated learning projects with three other medical centers in the region.

While the reuse of clinical data is hampered by the still prevailing disconnect between the data standards used in patient care and the ones used in clinical research, some efforts to bridge this gap are finally reaping their benefits in the form of clinical research and EHR system integrations. [18]

Substitutable Medical Applications and Reusable Technologies (SMART) on Fast Healthcare Interoperability Resources (FHIR) is a web based standard platform that enables developing solutions to enhance interoperability with EHRs and other systems, including clinical research data capture (e.g., REDCap). We have implemented a SMART on FHIR tool called COMprehensive Post-Acute Stroke Services - Care Plan (COMPASS-CP). [19] COMPASS-CP is a patient-centered application for increased interoperability with any capable EHR. COMPASS-CP captures patient-reported measures of functional and social determinates of health (PROMs) including priorities from the patient and caregivers at the point of care. The COMPASS-CP algorithms evaluate the data captured in questionnaires and identify factors likely to influence recovery, health, and independence of the stroke survivor across each dimension of care and needed referrals for community-based resources. These are used to generate the patient-facing COMPASS-CP, entitled "Finding My Way Forward for Recovery, Health, and Independence." Care plans provide education, recommendations, and referrals across essential domains of self-management and care, anchored to the four cardinal directions of a compass:

1. Numbers: Know your blood pressure, hemoglobin A1c, cholesterol, etc.
2. Engage: Be active in mind, body, and spirit through physical, cognitive, and social activity.
3. Support: Seek support for your stress, family stress, finances for medications, and transportation.
4. Willingness: Be willing to manage your medications and lifestyle.

Results and details of the 40 hospitals and 8,000 patients who participated in the COMPASS-CP pragmatic trial can be found in Duncan et al. [19]

Another example of the use of FHIR technology is the Mobile Patient Technology for Health (mPATH) technology, designed to enhance patient involvement in decisions for colorectal cancer screening.

Traditionally, decision aids have been provider-focused, but they are increasingly becoming more patient-centric. Miller et al. developed the mPATH tablet computer-based tool, that provides educational material regarding colorectal cancer screening options. [20] After rooming, patients are given a tablet computer that provides educational material about the pros and cons of different screening options. While patients have the opportunity to discuss the options with the clinician during the visit, ultimately the patient “self-orders” the preferred screening test (e.g., fecal blood test or colonoscopy). Patients who self-order a colorectal cancer screening test receive text messages to remind them about the test, to provide encouragement, and to ask if patients have questions. Subsequent text messages are sent to patients in order to encourage completion of the screening test. A randomized controlled trial demonstrated that screening was twice as likely to be completed (30%) in the intervention group as compared to the control group (15%).

The design of this project was built on research showing that multi-level CDS tools are more likely to be effective. [20] For example, Roshanov showed that CDS systems were much more likely to succeed when the tools provided advice for patients in addition to practitioners. [21] The study design should be given significant credit for the success of this project, and it highlights the need for informatics professionals who are familiar with the literature to be involved in the implementation of new informatics tools. The WFBMI has built the Translational Data Warehouse (TDW) to facilitate a comprehensive and semantically rich data ecosystem. The development of our translational data warehouse preceded the formation of the WFBMI. The TDW serves as a central resource for a gamut of research and quality improvement needs in the institution (Figure 1). Data is mapped to the standards, including RxNorm, LOINC, NAACR, etc., and extracted from primary sources via the use of Extract, Transform, Load (ETL). Mapping to standard terminologies in the TDW provides a consistent means for researchers to collaborate with other institutions, and facilitates participation in several clinical data research networks (CDRNs) by conforming local data to these networks' Common Data Models (CDMs).

We are in the planning phase of utilizing Epic-supported FHIR resources as an additional mechanism for data transfer to the TDW. WFBMI is also committed to developing a semantic framework that will minimize data misinterpretation and discovery challenges. We maintain a terminology server to house Unified Medical Language System (UMLS) standard terminologies. It serves vocabularies and “ontologies” for multiple research processes for a cohesive interoperability SNOMED Clinical Terms (SNOMED CT), LOINC, and NCI Common Data Elements (CDEs). In addition, our NCI-trained staff creates CDEs in the Cancer Data Standards Registry and Repository (caDSR), which are used in case report forms (CRFs), as required for all cancer studies. To facilitate reuse, we currently use existing CRFs from our CRF library or create new CRFs that have been curated with CDEs.

Notes

This presentation is faithful to the original, with only a few minor changes to presentation, spelling, and grammar. In some cases important information was missing from the references, and that information was added. Several bits of text were inadvertently repeated, and those repetitions were removed. No citation was given for the Wake Forest quote, so one was added. Everything else remains true to the original article, per the "NoDerivatives" portion of the distribution license.