| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Department of Cardiothoracic Surgery, Erasmus Medical Center, P O Box 2040, Room BD 569, Rotterdam, 3000 CA, the Netherlands
(Email: a.kappetein{at}erasmusmc.nl).
The demand for intensive care unit (ICU) resources are increasing, and similarly, so are the costs incurred with the postoperative care of patients needing an intensive care bed. The frequent shortages of intensive care beds have led to renewed efforts to define criteria for admission and discharge. Identifying patients at high risk for postoperative complications and length of ICU stay would facilitate decisions to allocate resources and to plan schedules for operations. High-risk patients could be electively scheduled for surgery in a series rather than parallel. While scheduling the operative program, a case mix of patients that includes patients needing ICU admission and those who are likely to have uncomplicated recovery could potentially prevent the blocking of beds in the intensive care unit.
Brunelli and colleagues [1] aimed to develop and validate a scoring system to predict intensive care unit (ICU) admission for complications after major lung resection and performed a multicenter study.
These kind of computer-based multivariable regression models are now widely available and are powerful tools that are used frequently to assess quality of care and to facilitate medical decision making. The medical literature overflows with articles offering to help clinicians make better predictions. Different prediction models aim to help triage patients in the emergency room, to decide between medical or surgical therapy, to predict ICU and hospital stay, and to predict hospital mortality. Uncritical application of modeling techniques, however, can result in models that poorly fit the data set at hand, or do not always perform as well for other patients as those from whom the data models were derived. The validity of a model needs to be assessed in new groups of patients. The predicted probabilities can be calculated with the model and compared with the actually observed outcomes. The different aspects of the validity and usefulness of a prediction model are: (1) calibration, which is the agreement between predicted probabilities and observed probabilities; (2) discrimination, which is the ability of the model to distinguish subjects with different outcomes; (3) the clinical usefulness of the ability of the model to improve the decision-making process (ie, the will of the added information from the new model translated into important changes in patient management; and (4) whether the new model will actually be used by practitioners at least as often (and hopefully more often) than the older model.
A well-known limitation of a model is that it is based on data from one institution, and therefore, subject to the efforts of local practices and case mix. Baseline risk between groups might be subtly different and might not be detected by a scoring system. Often clinical judgment also seems to be quite reliable and is critical in guiding assessment of patients who have a high operative risk, need intensive care admission, and subsequently have a slower postoperative recovery. Subjective clinical judgment is difficult to capture in a database, and it is subsequently difficult to use as a variable in a model.
Assuming that valid statistical models can be created that are better than most physicians at predicting outcomes, good calibration and discriminative ability are not sufficient for a model to be clinically useful. It is difficult to implement them in daily practice because the physician should be familiar with the model and know when to use it; despite score cards, pocket calculators, and personal digital assistants, calculation might be time consuming, and not all patient information is available at the same moment.
Practitioners will need to learn the strengths and limitations of these new tools, and need to discover settings where they can be most appropriately used. It is important to keep in mind that it is difficult to develop a model that is well-calibrated across diverse settings, because a model always consists of a limited number of predictors, leaving some variation unexplained between patients.
The computational complexity of the model described in the current article [1] is limited, but the information that needs to be entered in the model might not always be at hand. The positive characteristics of the current study are that it is a multicenter study and that the data were validated by bootstrapping simulation. However, to be adopted into clinical practice, the model should be easy to use and demonstrate an acceptable validity in different patient populations. Hopefully other clinical researchers will start using the current model and will report on the validity in the near future. Clinicians may only then accept the prediction model to aid in resource management within their ICU.
 |
References |
|---|
 Top References |
|---|
Related Article
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| ANN THORAC SURG | ASIAN CARDIOVASC THORAC ANN | EUR J CARDIOTHORAC SURG |
| J THORAC CARDIOVASC SURG | ICVTS | ALL CTSNet JOURNALS |
