| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Guest Access | Sign In via User Name/Password
|
|
|
|||||||
Guest Access | Sign In via User Name/Password |
|||||||||
Dr. Glance is Associate Professor of Anesthesiology, and Dr. Szalados is Associate Professor of Anesthesiology and Vice-Chairman, Department of Anesthesiology, University of Rochester School of Medicine and Dentistry.
Correspondence to: James E. Szalados, MD, FCCP, Department of Anesthesiology, University of Rochester Medical Center, 601 Elmwood Ave, Rochester, NY 14642
In this issue of CHEST (see page 539), Sirio et al compare critical care outcomes in the United States and Japan after accounting for differences in severity of disease and case mix using APACHE (acute physiology and chronic health evaluation) III. One of the objectives of international comparisons of critical care delivery systems is to determine if differences in resource use translate into differences in health-care outcomes. This study found that the observed and expected mortality rates in the Japanese cohort were identical. This finding is notable given the fact that hospital length of stays in Japan are significantly longer than they are in the United States due to differences in discharge policies. The authors point out that longer hospital length of stays are usually associated with higher hospital mortality rates, inferring that the aggregate outcomes in the Japanese data set may be better than expected. However, this study does not include information on the processes of care that may have led to improved outcomes relative to the American cohort. This study reminds us that one of the primary goals of outcomes information is to identify high-performance hospitals or health-care delivery systems so that we can uncover the "best practices" responsible for their superior outcomes and then implement them in other settings.1 Although there is little evidence to suggest that we are making significant progress toward accomplishing this goal, the work of O’Connor et al2 in the Northern New England Cardiovascular Disease Study Group demonstrates the potential for improving mortality rates through quality-improvement initiatives grounded in outcomes information. Furthermore, following the publication of the Institute of Medicine report on patient safety, there now exists a national mandate to search for processes of care that will minimize medical errors and improve patient outcomes.3
The question then arises: How do we formulate an agenda that will help translate research on ICU outcomes into improvements in the quality of care for ICU patients? First, we need a national contemporary population-based ICU outcomes database; Project IMPACT,4 a multicenter outcomes database sponsored by the Society of Critical Care Medicine, may be an excellent starting point. Second, we need to examine the comparative performance characteristics of existing ICU prediction models within this database to determine which model is best, or alternatively, to develop a better model in order to have a "single yardstick" against which to measure ICU performance. Third, the use of a fixed end point such as 30-day or 90-day mortality should be substituted for hospital mortality. Fourth, we need to look at other outcomes besides mortality. Fifth, once we have identified high-performance ICUs, we must use these centers of excellence as data laboratories in order to identify best practices. Finally, and most importantly, we must insure that these best practices are widely disseminated and adopted in other centers.
A national database is necessary in order to better understand the connection between medical interventions, processes of care, costs, and health-care outcomes in the ICU.5 Increasingly, third-party payers and purchasers are pressing health-care systems to collect data on health-care outcomes. Clinical performance data will be used as one of the criteria for accreditation by the Joint Commission, the National Committee on Quality Assurance, and the Health Care Financing Administration.6 The decision by the critical-care community to collect and utilize outcomes data on a national scale can either originate and be directed by physicians, or can be mandated and directed by government agencies, payers, and accreditation bodies. We propose that all ICUs participate in the formation of a national ICU-outcomes database. Universal participation will eliminate the possibility of selection bias, in contrast to voluntary participation that may result in a nonrepresentative group of ICUs. Adequate funding is necessary for the success of such a national data set. One potential source of seed money is the Agency for Healthcare Research and Quality. Further funding might be obtained by charging third-party payers for the cost of data collection and analysis. The creation of a national data repository, as opposed to numerous regional databases, is critical to the formation of a broadly representative data set so that research findings based on this database can be generalized to other ICUs in this country.
At present, there is no consensus on which ICU prediction model—APACHE II, APACHE III, mortality probability model (MPM) II, or simplified acute physiology score (SAPS) II—is the best. Unfortunately, there can never be a "gold standard" against which severity-adjustment models will be judged in order to achieve criterion validity. Furthermore, different prediction models will sometimes disagree on the identity of quality outliers,7 prompting the need to identify or develop a single "best" model. Perhaps one of the greatest barriers to quality improvement is that our measures of quality are imperfect and that the science of risk adjustment in critical care, despite 20 years of research, is still a work in progress.8 The fundamental problem is that most ICU prediction models fail to accurately predict outcomes when they are transported to data sets independent of the original database used to develop the models.9 However, poor model fit in an external data set does not necessarily mean that the model is faulty. It may instead be caused by differences in the quality of care between the reference data set used to develop the model, and the external data set used to validate the model.10 One possible solution to this problem is to initially include all the data elements from APACHE II, APACHE III, MPM II, and SAPS II in the national database. Then, once the database is of sufficient size, the models could be customized to the database and their comparative performance assessed. Alternatively, a new model could be derived using this database, taking advantage of recently introduced powerful modeling techniques, such as fractional polynomials.11 A single best model could then be selected as the basis for benchmarking performance. Customization of existing scoring systems would allow ICUs to compare their performance to their peer institutions, as opposed to comparing themselves to a benchmark based on outcomes data collected > 10 years ago. Other authors8 12 also support the use of customized models based on contemporary databases. Regardless of which model is adopted, it is critical that the model coefficients be periodically updated so that institutions can benchmark against a contemporary reference point.
It may also be necessary to revise current models so that 30-day or 90-day mortality replaces in-hospital mortality as the outcome of interest. Using in-hospital mortality rates can make it difficult to interpret outcomes differences across institutions if the discharge policies of these institutions are not the same. For example, transfer of long-term ICU patients to separate subacute facilities, or earlier hospital discharges, may cause a decrease in the in-hospital mortality rate. By redefining the end point as vital status at 90 days, Teres and Lemeshow13 have proposed an outcome end point that is temporally closer to the end of an episode of acute critical illness. Vital status at 90 days is a more clinically relevant end point than in-hospital mortality, and is more likely to reflect true differences in quality of care.
Although mortality is an important dimension of quality of care, quantitating the impact of medical interventions on health outcomes should not be limited to measuring survival alone.14 ICU outcome information should be expanded to incorporate quality-of-life measures. Functional outcomes could be measured using a modification of the functional independence measure used in trauma registries. This measure evaluates patient function across three domains—feeding, locomotion, and expression—and grades patient function along a spectrum ranging from complete independence to complete dependence.15
In the end, we should not lose sight of the purpose of collecting outcomes information. The goal is to take advantage of the variation in health-care processes and structures to identify best practices associated with superior outcomes. The challenge is to construct a national database, along with appropriate risk-adjustment tools, which allows us to identify ICUs that deliver high-quality care. We can then use these hospital outliers as "data laboratories" to uncover the best practices that contribute to their improved outcomes. Finally and most importantly, we must then disseminate this information to the international critical care community.
References
This article has been cited by other articles:
|
|
 |
|
  P. L. Graham and D. A. Cook Prediction of Risk of Death Using 30-Day Outcome: A Practical End Point for Quality Auditing in Intensive Care Chest, April 1, 2004; 125(4): 1458 - 1466. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
