Chapter 3 Survival Prediction Index
A variety of systems that place a numeric score on the severity of disease are used in human critical care,1-4 and similar systems are being developed for use in the small animal intensive care unit (ICU).5-12 Equations have also been developed for large animals but are beyond the scope of this chapter.13-16 Some systems globally classify the degree of physiologic derangement regardless of the diagnosis, and some are designed to be applied to specific situations, for example peritonitis or trauma. These systems have the common goal of attempting to objectively classify the severity of disease; in other words, to predict the outcome within a given patient population.
All scoring systems are developed for the same purpose: to categorize patients into groups with similar severity of disease. This is particularly important for risk stratification of groups of patients for clinical research studies. Clinical research trials are necessary to advance our understanding of the pathophysiology and management of disease. However, small animal patients with naturally occurring disease are usually a mixture of ages and breeds, may be brought for treatment at different stages of the problem, may have received a variety of treatments, and may have multiple concurrent diseases in addition to the problem being studied.
Although research animals with experimentally induced disorders are important models of naturally occurring disease in both humans and animals and provide a much more homogeneous population, the use of these animal models has certain disadvantages. In particular, experimentally induced models of disease may be expensive and time consuming to produce, and the disorders may not be identical to those that occur in clinical patients. Thus, results from experimental studies may not apply directly to clinical cases.
Our ability to draw conclusions from clinical trials of management techniques or new therapies is therefore often hampered by our inability to define homogeneous patientpopulations that can be compared. For example, if we were testing a new drug for treatment of dogs with autoimmune hemolytic anemia, the results might be difficult to interpret unless the test group has been proven to have the same type and severity of hemolysis, the same degree of systemic illness, and therefore the same risk of mortality as the standard treatment group. This problem can be addressed partially by developing an index for scoring the severity of disease, which attempts to place a numeric value on the degree of illness in the patients included in the clinical trial. Patients can then be categorized into groups with a similar severity of disease, which then allows comparison of groups for clinical trials.
Similarly, these scoring systems can also be used if a specific test result is being studied to determine its relationship to the severity of disease. They may also have some utility for triage of patients to objectively and prospectively allocate resources of staffing and equipment. Objective characterization of the severity of disease also allows quality control and comparison of actual outcomes between institutions and within institutions over time.
Ideally, such indices should be independent of the diagnosis and therefore applicable to any patient and any disease. In addition, the ideal prediction index should use readily available information that can be collected early during hospitalization, before beginning the clinical trial. If it can be shown that the severity of illness is initially statistically similar in two groups that are then treated differently, the results of the clinical trial carry more weight.
The statistical accuracy of severity-of-disease scoring systems can be assessed by evaluation of the area under the receiver operating characteristic curve, sensitivity and specificity, odds ratios, or positive and negative predictive indices. The predictive accuracy of logistic regression equations is usually estimated using receiver operating characteristic (ROC) curves, which demonstrate the tradeoff between the true-positive rate (sensitivity) and the false-positive rate (1 minus specificity) at varying predictive cut-points. The area under the ROC curve (AUC) represents the probability that a randomly selected “survivor” has a larger predicted probability of survival than a randomly selected “nonsurvivor,” and is therefore a measure of the predictive value of the equation. The higher the AUC value (closer to 1), the more accurate is the equation. In human outcome prediction equations, AUC values commonly are obtained as high as 0.85 to 0.90.
The survival prediction index (SPI) was developed as a method of scoring the severity of disease of critically ill dogs in the ICU.5-7 The system comprises parameters that are independent of the diagnosis, and are part of the routine monitoring and evaluation of ICU patients. Data for this calculation can therefore be collected early during hospitalization and before interventions being tested in clinical trials.
A pilot study was conducted initially to develop an SPI using data from 200 dogs admitted to the ICU at the Veterinary Hospital of the University of Pennsylvania.5 The SPI was calculated by logistic regression analysis, using clinical parameters collected within the first 24 hours after admission to the ICU. For the pilot study, all of the data were collected by one person. The parameters were chosen to reflect the function of vital organ systems, the severity of underlying physiologic derangement, and the extent of physiologic reserve (Box 3-1).
|Age (years)||Age (years)|
|Body weight (kg)||Respiration rate (breaths/min)|
|Rectal temperature (°F)||Mean arterial pressure (mm Hg)|
|Heart rate (beats/min)||Service of entry (surgical or medical)|
|Respiration rate (breaths/min)||Packed cell volume (%)|
|Mean arterial pressure (mm Hg)||Creatinine (mg/dl)|
|Oxygen saturation SaO2 (%)||Albumin (g/dl)|
|Neurologic disease (Y/N)|
|Service of entry (surgical or medical)|
|Chronic disease (Y/N)|
|Packed cell volume (%)|
|Total solids (g/dl)|
|White blood cell count (cells/mm3)|