Comments of Joseph I. Bormel, M.D., M.P.H.

Representing the Cerner Corporation

Washington, D.C.

March 30, 1999

PowerPoint Presentation

My name is Joseph Bormel. I am a physician and the Chief Architect for Medical Management at Cerner Corporation. Prior to joining Cerner, I had the opportunity to contribute to the National Library of Medicine’s Unified Medical Language System as an NIH fellow in Harvard’s Clinical Effectiveness program. My clinical practice and research experiences in Internal Medicine and Rheumatology have provided me first hand knowledge of the uses and challenges of diagnostic and classification systems relevant to the challenges of the Health Information Infrastructure.

The Cerner Corporation is a leading supplier of clinical and management information and knowledge systems for healthcare organizations in the United States and abroad. We have more than 1,000 clients in the U.S. and around the world. These clients include integrated health organizations, integrated delivery systems, community hospitals, ambulatory clinics, physician practices, management services organizations, blood banks, reference labs, and home health organizations.

Cerner Corporation and I very much appreciate this opportunity to testify and to provide our industry-oriented perspective to the National Committee on Vital and Health Statistics Hearings on Data Quality, Accountability, and Integrity.

(a) We define Patient Medical Record Information as the data required to deliver appropriate, effective and efficient care at the appropriate place and time to achieve the optimal health outcome. Bringing in the prevention aspects of the care mandate (and implicitly, the population focus), we believe it becomes necessary to make the distinction between "Patient Medical Record Information" and "Person Health Information" (PHI). The former having an acute-care, higher acuity, and probably disease connotation, and the later being more concordant with NCVHS’s defined computer-based health record (CHR).

In our experience, maintaining and attempting to integrate the three types of CHR (Personal, Patient, and Population) has huge not-so-hidden costs, and frequently leads inexorably to failures to operationalize basic business and clinical objectives. These objectives take the form of process automation and process-focused quality initiatives. Walking through the common processes of care will serve to both illustrate this point as well as to explicitly answer questions 1(a) and (b).

(b) Patient Medical Record Information is needed for the 8 process components (and underlying functions) of all health care delivery encounters:

1. Identify the person(s);
2. Assess their need(s);
3. Plan for care based on that assessment;
4. "Order" the components of that care;
5. Schedule what was ordered;
6. Carry out the order;
7. Document what was done; and
8. Account for the resources consumed, communicating that accounting to stake holders.

There are also two process components that transcend the bounds of typical health care encounters that involve monitoring a person’s health status:

1. Expectation tracking, and

2. Alert generating.

This model applies equally to traditional care delivery in ambulatory and acute visit-based encounters, as it does to prospective and retrospective care management. By that we mean that these steps and the functions served by the health information underlying these steps are vital to effective demand management, cross-continuum disease management, population health management, retrospective cost and quality profiling, and person self-education. These issues are reviewed in the supplementary slides.

Let’s quickly define the scope of the PMRI in terms of those process steps. Step one requires unambiguous person identification, both of person(s) receiving care and the associated providers of that care. Step two requires fully coded health status summary information. Traditionally, this has included active problems, allergies, current medications, and other information summarizing the current care context. Steps three and four require ways to make explicit references to guidelines and the associated orderable components of that specified care. This scoping exercise could be extended to all ten enumerated process steps. This would add process measures, outcome measures, discrete test assays, coded health maintenance goals, etc.

(c) Tremendous cost and quality penalties are paid when these eight process steps are each handled without the commonality of standards that the health information infrastructure (HII) has the potential to create. Without comparable PMRI, defects in identification, assessment, planning and all other subsequent processes are inevitable. The precision and accuracy or lack thereof has direct impact on clinical, financial, satisfaction, functional and process outcomes. Using ordering of drugs as an example, the consequences of inadequate information management contributes directly to "adverse drug events(ADE), a national problem that kills 60,000 to 140,000 Americans each year and costs the nation more than $76.6 billion annually. Much of that expense is shouldered by agencies of the Federal Government." The lack of a national standard to concretely identify drugs makes complex DUR impossible to do. (See the testimony before the National Committee on Vital and Health Statistics by Timothy McNamara, M.D., M.P.H.&T.M., representing Cerner Corporation and Multum Information Services, December 9, 1998.)

Comparable and structured, granular, codified process information constituting the PMRI serves another important function. It allows recording (capturing) of encounter data one time to serve all subsequent purposes of that health information. As outlined in the "Tasks for the Health Information Infrastructure" NCVHS concept paper of October 14, 1998, these purposes include population-data needs, transactional computer-based health record needs, integrated knowledge management and decision support, research, collaboration, and evolving consumer demands around health information.

This data must be encoded unambiguously and preferably by the provider him/herself, rather than by a post-hoc encoder. Substantial quality is lost when this is done due to both the introduction of another ‘receiver in the party line’ as well as the intrinsic decay that occurs in measurement accuracy with time.

Another significant problem occurs when the person doing the coding is forced to use a coding scheme that is designed for another purpose. This commonly occurs when a physician is forced to use billing codes to describe a clinical situation. This ensures a loss of quality and accuracy, as well as decreases the physician’s confidence in management initiatives derived from the "garbage" data.

Summarizing these points we conclude that having providers recording appropriately codified information would improve quality. What is less generally understood, however, is that process only becomes effective (and in fact possible) when the benefit exceeds the implementation burdens. In other words, if the recording process slows down the physician, they would be disinclined to use it. That necessitates removing steps from the new process through automation. For example, when a physician using a computer-based health record (CHR), identifies a patient and assesses that their most significant problem at the moment is sudden onset of partial blindness, huge automation opportunities and benefits are created if the infrastructure is integrative. The CHR system can suggest a plan, recommend orders and scheduling, facilitate activating that plan, its documentation and accounting. These steps can each speed up the physician, offsetting the burden of entering codified information.

Further, it can assure the data quality, accountability and integrity of the composite encounter information set, ensuring that blood pressure and other key observations are recorded. Please notice that this process required the physician to initiate two processes (one, identifying the patient and two, assessing their need) before facilitating completion of the remaining six process steps. The resulting facilitated total care process would exceed the performance of today’s care delivery system in that there would be less variation introduced by oversight and other task interfering causes.

(a) Yes, the current state of data is impairing our ability to perform the tasks of medical management (outcomes, quality and performance improvement). There is no assurance that classification systems are being applied consistently. For example, a patient with an identical description might be classified as having Undifferentiated Connective Tissue Disease with Arthritis at one institution, and be classified as having Rheumatoid Arthritis at another. These variations in the formality of application of classification systems would become less of an issue if we had a nomenclature that supported precise use of descriptive terms that were automatically "rolled up" into classification systems for billing and health management purposes. This is exactly the approach advocated by specialty societies, by research protocols and by utilization management programs.

We define the work of medical management as having 3 phases: (1) get the data; (2) find actionable opportunity; and (3) take action. The current incentive structure and nature of claims data combined with the relative absence of process data makes medical management initiatives challenging to the point of being impractical. Unless the capture of data is done by the provider, as a byproduct of the care process, and appropriately specific, each of the three phases are less likely to occur.

Most medical management today is done using claims data as the data source. Because the data was not collected for outcome, quality or performance improvement purposes, it does not accurately capture the care issues being addressed. This is a well understood problem with health care information. What is not well appreciated is that linking clinical and financial data from disparate sources cannot, according to Deming’s principles and common sense, create a closed-loop improvement system. This approach does not have the power to study clinical decision making, a requisite step in influencing subsequent decision making.

(b) The problems of data capture are also well described in the informatics literature. The tension between the ease and expressivity of free text versus the power/confines/risks/and appropriateness of structured template-based data capture is a core issue that does not go away with the application of advanced technology, standards, or applications. The advancing ease of use of voice recognition, like the introduction of facsimile technology, offers to grease the slippery path of easily capturing far less usable information. Without the ability to increase the probably of capturing ‘strongly suggested’ data elements and clarifying the content of dictation, voice recognition has the potential to short circuit this opportunity to improve the integrity and utility of health information, and the processes that use this information.

From our perspective, therefore, creating incentives, either through market-driven or government supported incentives (such as those outlined in the NCVHS concept paper) will be necessary to ensure capturing sufficiently codified information regarding assessment and plans as outlined above.

(c) and (d) The key to data capture that is robust for the broad purposes listed by NCVHS is a formal underpinning. Elements of this formal underpinnings for data capture include nomenclature, clinical grammar, canonical domains, and higher forms that contextualize the embedded knowledge. We believe that there is a great deal of value in contextualizing information, making it actionable or executable, rather than just viewable. Private sector efforts to create a de facto standard in this area have revealed two major findings. First, structured knowledge development is a very slow process with high domain dependencies. Secondly, the acceptability and perceived value of developing a largely complete set of templates was not an attractive differentiating market attribute to many buyers. There is a deep seated belief, sometimes justified that goes "Our needs are different".

(e) We believe that performance measurement standards are important and that the credentialing organizations (JCAHO, NCQA) provide an important first step toward raising data quality standards. We see a role for the professional societies (ACC, STS) to define performance measurement sets, and we recognize the role of other not-for-profit associations, such as disease specific entities in establishing specific data quality standards.

Within the next ten years, we expect to see and be supporting composite transactions that support richer, standardized data sets. Let’s return to the earlier example of the patient with new, sudden onset blindness. We expect that the patient’s first contact with a call center would initiate a codified assessment of ‘new-onset, partial blindness’, and that the care plan would guide the clinical and logistical process of ensuing care. As a by-product, the who-what-when-where-and-why of their health care delivery would be recorded with minimal ‘costs’ or interference with the human care givers tasks. Implicit in this vision is transferable, fully-coded, complex descriptors that accurately classify a person at a point in time. These descriptors would then "follow" the person through a complex system of care, allowing for best-effort decision support to be triggered at each step of the way, without forcing re-capture of the original data. Given that the ‘re-capture’ process guarantees a loss of data quality, frustrates and/or challenges the patient (decreased consume satisfaction), and decreases early intervention opportunity, we feel that a concerted effort to facilitate exchange of these complex descriptors is needed. Government leadership, government partnership with stakeholders in the private sector, and well-aligned incentives will be necessary to achieve this vision.