Clinical Intelligence
Access to information is at the very core of medical practice in the modern age. Ironically, medicine has lagged significantly behind most other fields in the type of data awareness and evidence-based feedback by which other industries live and die. And patient-specific information that individual providers need in order to provide quality care is only a part of what is needed.
Pressure from quality-oriented groups such as Leapfrog and IHI, as well as impending payer initiatives on pay-for-performance, have begun to highlight the serious drawbacks in most provider organizations ability to access, analyze and implement change driven by clinical information. Many studies of outcomes have used billing data, always a crude tool at best and rarely an accurate reflection of what care has been provided. Adhoc studies of limited duration and scope in real clinical data are either prohibitively expensive, or so narrow as to be difficult to generalize.
Gaps in Information and Knowledge
Even in organizations that have robust clinical information systems in place, extracting meaningful data from them for examining the quality of care is extremely difficult; these systems are designed to be patient- and provider-centric the architecture of a true reporting tool that gets at process information and aggregate data was never part of their initial design. One can easily ask the question, What was Mrs. Jones potassium over the past month? but getting to, How many patients with a potassium of 6.0 received a repeat lab and/or timely treatment in the past six months? is another type of question entirely one that few clinical systems are designed to answer without an extensive adhoc query which, if even technically possible, would have a significant impact on the systems realtime transactional performance.
It has become clear that there are not only a great many unanswered questions to address, but an even greater number of unasked questions. One of the hallmarks of clinical intelligence (essentially what other industries call business intelligence, as applied to clinical data and processes), is that there is often the unanticipated result of finding answers to questions that we didnt even know enough to ask, sometimes referred to as knowledge discovery.
Foundations of Clinical Intelligence
Continuous quality improvement through clinical intelligence is not merely about finding answers, however. The amount of effort and expense required to implement such a system is only rewarded when the information gathered will have an impact on the processes that drive the organization. Using clinical intelligence to drive quality improvement is a circular process which rests on three critical elements:
- Identification of the proper measures to apply
Using real data on outcomes, rather than surrogate process measures, where possible;
Center for Medicare and Medicaid Services and private payfor- performance indicators provide a starting point; and
Reiteration: As answers come back, using this information to ask questions with more precision and power.
- Data
The quality of data collection, its reliability and granularity, will be the cornerstone of any clinical intelligence initiative;
The breadth and depth of the data gathered will have an exponential effect on the quality of information generated.
- Action
Implementing clinical intelligence will be an initially expensive effort, both in financial and political resources. Organizational leadership must see the clear benefits and make a commitment to an evidence-based approach to patient care at the management level; and
Technology is merely an enabler of the processes in any healthcare organization. Information generated by data warehousing will have no effect unless the organization is willing and able to align its behavior with the evidence, once presented.
Leading Practices
Several leading practices come to the surface in the course of examining organizations that practice clinical intelligence:
- Information Leads to Action: Data gathering must be clinically relevant and lead to changes in practice within the organization. Changes in practice which result from identifying areas of improvement in clinical processes should result in measurably positive outcomes.
- Ongoing Measurement: The process of measurement should be circular, closing the loop and driving further improvement.
- Enterprise-wide Data: The collection of data should encompass the widest possible scope within the organization if it is to be most useful. Oftentimes seemingly unrelated systems, when data is combined and analyzed, will reveal relationships key to the understanding of process gaps and inefficiencies. In addition, as our ability to draw from a wider pool of data increases, the closer we get to being able to measure real outcomes, rather than surrogates.
- Clinician Governance: The process owners must decide how data is collected and interpreted. No effort at standardization or reporting can be successful unless the data means something to those who will use it. IT professionals must yield their traditional ownership of all things data to an unprecedented level of clinical oversight to ensure that the data becomes actionable information.
- Self-Service: Ideally users should be able to access current information on their own performance and on those who report to them through standard views, such as dashboards, but also through adhoc or custom views of the data, where appropriate. Physicians have long recognized and valued the evidence-based approach to providing care, but have had limited access to specific feedback about their own performance.
Where It Works
One example of an institution which demonstrates many of these key practices is Intermountain Health Care (IHC).[1] IHC has an extremely robust quality infrastructure, with areas of focus across the entire 19-facility system, examining processes both in inpatient and outpatient realms. Their data infrastructure includes a clinical data repository that connects their facilities and allows for access to clinical data transactions (patient visit documentation, medications, problems lists, etc.), across the enterprise (See Figure 1). In addition, this repository feeds a data warehouse, designed for reporting capability, research and knowledge discovery.
Among their primary care initiatives are measures programs in diabetes, asthma, depression, pneumonia and congestive heart failure. Within the Diabetes Care Management System, they extract a registry of more than 25,000 diabetes patients from the data warehouse and follow diabetes longitudinal care measures including cholesterol, hemoglobin A1c (HgbA1c) and urine microalbumin screening. Every primary care provider in the organization gets a quarterly report on how well their specific patients are doing in these measures over time, as well as a comparison to their colleagues performance. This has been powerful medicine. In the four years between 1998 and 2002, they saw improvements of 25 percent in some of these measures, with plans to expand the program yearly and increase clinicians access to real-time, clinically focused data.
Another excellent example of the use of clinical intelligence at IHC involved the study of induced labor.[2] IHC extracted data on 85,000 live births from its data warehouse and looked at the timing of induced labor for women from 37 to 39 weeks of gestation and later, comparing Bishop scores (a measure of cervical readiness for labor), and outcomes for neonates and the likelihood of emergency cesarean section. Conventional wisdom among physicians practicing obstetrics was that there was no negative impact from pharmacological induction after 37 weeks, provided the cervix appeared to be receptive. Their ability to extract data on neonatal admissions to the intensive care unit and rates of C-section demonstrated otherwise, however. They found a 4.1 percent increase in NICU admissions among children delivered by induction at 37 weeks, as well as a higher rate of these inductions going to C-section for failure to progress and other complications. They have since instituted tighter controls on labor inductions across the system, raising the threshold and asking for clinical indications to early induction.
Challenges in Implementing Clinical Intelligence
As illustrated in the IHC example, physicians behavior has been shown in multiple studies to be highly responsive to data, both generalizable from clinical studies as well as specific feedback on their own performance. The caveat, however, is that there needs to be trust in the quality of the information. Access to data is essential, but not sufficient. There must be a perception that the right elements are being measured in the right ways, and that apples and oranges are not being confused with each other. This also relates to issues of standardization and data integrity which invariably come into question when clinicians are not principally involved in managing the process of data collection.
In order to build the infrastructure for clinical intelligence, disparate systems across the provider enterprise need to move toward a standardized set of vocabularies, data elements, policies and procedures so that integration of information can be achieved in a meaningful way. This is often a daunting task, and is a process of evolution, rather than revolution. Some information is easier to move to a structured data warehouse environment than other types, and will be warehoused first, but may represent too limited a set of data to be clinically useful until a critical mass of different systems are tapped. Among the barriers to this are cost, poor knowledge management and documentation in IT departments, inflexible legacy systems, and the failure to adopt standard terms and data elements so that the data is internally consistent.
Perhaps most critical to the success of any effort to implement clinical intelligence is a strong governance structure, from the executive suite down to the provider level.Without clear enterprise- level guidance on how to facilitate the process of moving clinicians and IT professionals toward agreement on standards and procedures across the institution, very little will be accomplished, at very great expense. In addition, an ongoing joint working group of IT and clinician leaders is necessary for the maintenance and integration of constantly evolving dictionaries, processes and standards. A centralized knowledge management system for use by this group in designing modifications, tracking history and avoiding redundancy is indispensable. And finally there needs to be a group of clinicians and administrators with the power and will to redesign workflow in response to their examination of the resulting data. Often this will involve changes to existing clinical information systems, such as alerts and design changes, as well as work process revisions, such as timing the administration of antibiotics or moving patients through the facility.
Conclusion
Though the challenges are significant, the pressures on healthcare organizations of all kinds to measure their own quality of care are only accelerating. In the competitive environment, key differentiators will be based on publicly available data. Some private payers have already begun to demand proof of quality in what they are paying for, and Medicare will soon be doing so. As large institutions move away from paper as an information vehicle, they are confronted with the opportunity to leverage the new electronic medium to meet these demands. There is just no other way to do so in many cases. But clinical systems designed for realtime care provision will not be able to fill this need. A robust clinical intelligence architecture that is designed to extract aggregate data and use it effectively will be essential, as medical care emerges from the dark ages of information management.
Endnotes
- Journal of Managed Care Pharmacy, Vol. 9, No. 6, 553-8.
- Desert Morning News (Salt Lake City), Oct. 20, 2003.
