By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Wiley Online Library will be unavailable on Saturday 7th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 08.00 EDT / 13.00 BST / 17:30 IST / 20.00 SGT and Sunday 8th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 06.00 EDT / 11.00 BST / 15:30 IST / 18.00 SGT for essential maintenance. Apologies for the inconvenience.
The work was performed at University of Tennessee and Colorado State University. Presented in part as an abstract at the Veterinary Emergency and Critical Care Society Meeting in San Antonio, TX, September 2010.
Corresponding author: Jacqueline C. Whittemore, DVM, DACVIM, PhD, Department of Small Animal Clinical Sciences, the College of Veterinary Medicine, University of Tennessee, 2407 River Drive, Knoxville, TN 37996-4550; e-mail: email@example.com.
Background: Microalbuminuria and C-reactive protein (CRP) are predictors of morbidity and survival in critically ill human patients.
Hypothesis/Objectives: To evaluate results of microalbuminuria assays (untimed single-sample urine albumin concentration [U-ALB] and the urine albumin : creatinine ratio [UACR]), serum CRP, and survival predictor index (SPI2) scores as predictors of survival in critically ill dogs.
Animals: Seventy-eight dogs admitted to intensive care units at University of Tennessee (UT) and Colorado State University (CSU).
Methods: Prospective observational study. Critically ill dogs were eligible for enrollment, unless euthanized because of financial constraints. Samples were collected within 3 hours of admission. Spearman's rank-correlation coefficients were determined for U-ALB, UACR, CRP, and SPI2. U-ALB, UACR, CRP, and SPI2 were assessed for associations with 7- and 30-day survival by Mann-Whitney U-tests and receiver operating characteristic (ROC) curves. P-values < .0125 were considered significant.
Results: UT (n = 49) and CSU (n = 29) patients did not differ significantly. Forty percent (31/78) of dogs died. SPI2 was inversely correlated with U-ALB (rs=−0.39, P < .001) and UACR (rs=−0.41, P < .001). CRP was not correlated with SPI2 (P= .019), U-ALB (P > .1), or UACR (P > .1). U-ALB and UACR had very high correlation (rs= 0.95, P < .001). SPI2, U-ALB, and UACR differed significantly for survivors and nonsurvivors. SPI2, U-ALB, and UACR had areas under the ROC curve (AUC) from 0.68 to 0.74 for survival prediction.
Conclusions and Clinical Importance: Albuminuria and SPI2, but not CRP, are associated with survival in critically ill dogs. Suboptimal AUCs limit the value of microalbuminuria testing for clinical risk assessment. Additional studies are necessary to determine the usefulness of microalbuminuria testing in patient risk stratification for prospective research.
area under receiver operating characteristic curve
Colorado State University
receiver operating characteristic
sequential organ failure assessment
survival predictor index
untimed single-sample urine albumin concentration
urine albumin : creatinine ratio
University of Tennessee
An objective illness scoring system has been devised for dogs, the survival predictor index (SPI2), based on the acute physiology and chronic health evaluation (APACHE) scoring system used in human medicine.1 The SPI2 score weights readily available clinical data points to objectively determine the patient's probability of survival, independent of the underlying disease process or processes, as a percentage. Data points included in the SPI2 are mean arterial pressure, respiratory rate, creatinine, PCV, albumin, age, and medical versus surgical status. The higher the SPI2 score, the greater the likelihood is that the patient will survive. Receiver operating characteristic (ROC) curve analysis was used to determine the accuracy of the SPI2 in categorizing patient survival during validation studies. The area under the ROC curve (AUC) was judged sufficiently robust to support use of the SPI2 in categorizing patient severity of illness for research purposes, but it was not considered large enough for use as a prognostic tool in individual patients.1
Another tool that has been used to evaluate severity of disease in dogs with a variety of disorders is C-reactive protein (CRP).2–5 Increases in CRP concentration occur and resolve in direct proportion to the degree of tissue inflammation but are not specific for individual disease processes.2 One recent study found no difference in CRP concentrations for critically ill dogs admitted to an intensive care unit that survived compared with nonsurvivors.3
Microalbuminuria is defined as amounts of albumin in the urine that are greater than normal but below the limits of detection of standard urine dipstick assays.6–8 Increasing age is associated with increased urinary albumin concentration in dogs7,8; this association was shown to persist independent of disease status in 1 study.7 The age-dependent increase in urinary albumin concentration complicates establishment of a quantitative definition for microalbuminuria in dogs; currently, a range of 1–30 mg/dL is used.7,8 In critically ill human beings, increased untimed single-sample urine albumin concentration (U-ALB) or an increased urine albumin : creatinine ratio (UACR) at time of admission and increasing UACR over the course of hospitalization have been associated with increased morbidity and mortality.6,9–14 The UACR determined within 15 minutes of admission to the intensive care unit has been found to be as good a predictor of morbidity and mortality (AUC 0.61–0.74) as APACHE2 (AUC 0.72–0.80), simplified acute physiology scores (AUC = 0.79), and sequential organ failure assessment (SOFA) scores (AUC 0.72) at 24 hours.6,10 In 1 recently published study of dogs and cats admitted to an intensive care unit, albuminuria was associated with an increased risk of mortality.8
Identification of objective biomarkers of disease severity with results that can be reported quickly may allow for better allocation of client and hospital resources and limit the use of expensive and invasive therapeutics to those patients at greatest risk of mortality, potentially leading to improved survival in the critical care setting. The primary objective of our study was to evaluate for associations among SPI2 score, CRP concentration, U-ALB, and UACR, and survival in critically ill dogs. Our secondary aim was to evaluate for associations among SPI2 score, CRP concentration, U-ALB, and UACR. Our null hypothesis was that there would be no association between any of the dependent and independent variables evaluated.
Materials and Methods
Dogs admitted to the University of Tennessee (UT) Veterinary Teaching Hospital Intensive Care Unit or Colorado State University (CSU) Veterinary Teaching Hospital Critical Care Unit from June 1, 2009 to December 1, 2009 were eligible for prospective enrollment in the study. Informed consent was obtained as part of study enrollment. The experimental design was approved by the UT Institutional Animal Care and Use Committee and the CSU Institutional Animal Care and Use Committee. All cases were managed by their attending clinicians. To be eligible for inclusion in the study, dogs had to (1) be considered ill enough to medically warrant hospitalization in an intensive care unit by both the attending clinician and the enrolling investigator, (2) have a CBC, serum biochemical profile, and urinalysis collected within 3 hours of admission, (3) have serum and urine collected within 3 hours of admission for performance of CRP and microalbuminuria assays, respectively, and (4) have a mean arterial blood pressure measurement determined within 24 hours of hospitalization. Unowned dogs, dogs anuric upon admission to the intensive care unit, and dogs euthanized for financial reasons were excluded from the study. All patients from each institution were enrolled by 1 individual (B.A.M. or M.V.C.) to ensure the inclusion and exclusion requirements were met, particularly with regard to severity of illness and reasons for euthanasia. Written guidelines defining intensive care hospitalization criteria are not utilized at the study institutions. At UT, licensed veterinary technicians and veterinary assistants are available to assist with treatments, observation orders, and fluid therapy in the wards; therefore, none of these necessitate intensive care hospitalization as a matter of course. At CSU, a separate step-down unit is available for hospitalization of noncritical patients that require more intensive management. Because admission to either unit is, however, at the final discretion of the attending clinician, occasionally noncritical patients may be admitted. Therefore, in cases where there was uncertainty about whether a particular dog was truly ill enough to medically warrant hospitalization in the intensive care unit and thus the study, additional investigators at the participating institution (J.C.W., D.I.M., T.B.H.) were consulted to determine eligibility. Enrolling investigators (B.A.M. and M.V.C.) reviewed case files, consulted with attending clinicians, and consulted with additional investigators as necessary (J.C.W., D.I.M., and T.B.H.) to confirm whether the decision to perform euthanasia was made based on cost of treatment.
Cost of stay and length of hospital stay were determined from the medical record. Information on short- and longer-term survival was obtained by follow-up phone calls to owners and referring veterinarians. Short-term survival was defined as dogs surviving 7 days from admission and longer-term survival was defined as dogs surviving 30 days from admission.
Sample Collection Protocol
Blood and urine were collected within 3 hours of admission for all dogs. Serum was separated from clotted blood within 30 minutes by centrifugation at 1,800 ×g for 3 minutes. Serum and urine samples were stored at −80°C pending completion of study enrollment. All samples were assayed together after study enrollment was completed to remove the potential for interassay variability to affect results.
Disease Severity Indices
The SPI2 score was calculated for each patient as described previously1 by an individual blinded to the CRP, U-ALB, and UACR results using the following equations:
The SPI2 score calculation results in a value between 0.00 and 1.00, reflecting the percentage likelihood of survival, with 1.00 equaling 100% chance of survival.
Serum CRP concentrationsa were determined according to the manufacturer's instructions by an individual blinded to the results of other assays and indices. CRP concentrations >60 μg/mL were recorded as 60.5 μg/mL for purposes of statistical analysis. U-ALB and UACRs were objectively determined as described previously7 by an individual blinded to the CRP results and SPI2 scores.
Descriptive statistics were generated for each variable. Samples were analyzed for normality by the Kolmogorov-Smirnov test. Outliers noted on box-and-whisker plots were double-checked to rule out data entry errors. Because of non-normal distribution of some measures, nonparametric testing was adopted for all analytes. Nonparametric Mann-Whitney U-tests were performed to determine whether the UT and CSU enrollment populations differed significantly for age, weight, cost of stay, time of stay, SPI2 score, CRP concentration, UACR, or U-ALB. Based on lack of significant differences between the 2 populations (see “Results”), they were combined for analyses.
Spearman's rank-correlation coefficients (rs) among U-ALB, UACR, CRP concentration, and SPI2 scores were determined. Spearman's rank-correlation coefficients for U-ALB or UACR and CRP concentration also were obtained and for U-ALB and UACR. Mann-Whitney U-tests were performed to determine whether survivors and nonsurvivors differed significantly for SPI2 score, U-ALB, UACR, or CRP concentration. Variables significantly associated with survival by Mann-Whitney U-test were assessed by ROC curve analysis for accuracy in prediction of 7- and 30-day survival. Ninety-five percent confidence intervals were calculated for ROC curves because the statistical program utilized cannot generate 99% confidence intervals, which would be more appropriate given the multiplicity of analyses. Threshold of significance for all P-values (2-sided) was adjusted to account for multiple analyses by a Bonferroni-like correction within each set of tests. Statistical analyses were performed and figures created by a commercially available statistical software package.b
There were 78 dogs enrolled in the study (UT 49 dogs, CSU 29 dogs). There were no significant differences between UT and CSU study populations by a Bonferroni-corrected P-value of .006. All P-values exceeded .4 except for CRP concentration (P-value = .04) (data available upon request from the corresponding author). Patient populations therefore were combined for further analysis.
Diagnoses for the 78 dogs are summarized by 30-day survival status in Table 1. The total number of diagnoses is greater than the total number of cases because several dogs had multiple diagnoses. No dogs were lost to follow-up. Short-term survival was 69% (54/78), and 60% (47/78) of dogs survived 30 days from time of admission. Six of the 31 (19%) nonsurvivors died; 1 dog (3%) died, was resuscitated, and then was euthanized because of persistent nonresponsive mental status; and 24 dogs (77%) were euthanized.
Table 1. Diagnoses by organ system for 78 dogs hospitalized in the intensive care units at the University of Tennessee and Colorado State University, categorized by survival status at 30 days.
Survivors (n = 46)
Nonsurvivors (n = 32)
The total number of diagnoses is greater than the total number of cases because several dogs had multiple diagnoses.
8: Uncharacterized mass (1), toxicity (3), trauma (4)
4: Uncharacterized mass (4)
Spearman's Rank-Correlation Coefficients
SPI2 score had significant but low correlations with U-ALB (P < .001, rs=−0.39) and UACR (P < .001, rs=−0.41). U-ALB and UACR had almost perfect correlation (P < .001, rs= 0.95). CRP concentration was not correlated with U-ALB (P > .1), UACR (P > .1), or SPI2 score (P= .019, rs=−0.28), based on a Bonferroni-adjusted P-value of .010.
Results of Mann-Whitney U-tests for survivors and nonsurvivors are presented in Table 2. Variables significantly associated with 7-day survival were U-ALB and SPI2 score, based on a Bonferroni-adjusted P-value of .0125. U-ALB, UACR, and SPI2 score were significantly associated with 30-day survival.
Table 2. Mann-Whitney U-test results for survivors and nonsurvivors in a population of dogs hospitalized in the intensive care units at the University of Tennessee and Colorado State University.
ROC curves, AUCs, and 95% confidence intervals for variables associated with survival by Mann-Whitney U-test are presented in Figures 1 and 2 and Table 3.
Table 3. Receiver operating characteristic (ROC) curve results in a population of dogs hospitalized in the intensive care units at the University of Tennessee and Colorado State University.
AUC, area under the ROC curve.
P-value is for the null hypothesis that the AUC = 0.5. See Table 2 for other abbreviations.
*Significant using a Bonferroni-adjusted P-value threshold for significance of .0125.
Clinical scoring systems are widely used in human medicine to quantify disease severity. Commonly used human clinical scoring systems include the APACHE and SOFA scoring systems.6,9,10 Veterinary clinical scoring systems are not as widely utilized, but their use continues to increase.15 By definition, increasing SPI2 scores should be associated with increased survival. Not surprisingly, increasing SPI2 score was associated with 7- and 30-day survival in this population of critically ill dogs. CRP concentration was not associated with SPI2 score or microalbuminuria measurements. Consistent with previous studies,3–5 CRP concentration also was not associated with short- or longer-term survival. Significant, but biologically weak, inverse correlations were documented between SPI2 score and microalbuminuria measurements (U-ALB and UACR). In keeping with studies on critically ill human beings,6,9–14 U-ALB and UACR were both associated with survival. AUCs for UACR (0.74) and U-ALB (0.68–0.74) for survival were similar to those documented for human critically ill patients (AUC 0.61–0.74).6,10
CRP is a sensitive acute phase reactant, often increasing within 4 hours of tissue injury.2 A number of veterinary studies have identified associations between CRP concentrations and disease severity.3–5 No association has been identified, however, between CRP concentrations and survival in previous studies of critically ill dogs, immune-mediated hemolytic anemia, and pancreatitis.3–5 Consistent with these reports, no association was seen between CRP concentration and survival in this study. In contrast to previous results in human beings,10 CRP concentration was not associated with microalbuminuria. It is possible that lack of associations with CRP concentration reflects the heterogeneity of diseases afflicting critically ill dogs. Significant variability has been identified in CRP concentrations for different disease etiologies. 16,17 Time to peak CRP concentration post-insult has been reported to vary by disease etiology, from 1 to 10 days for trauma versus infection.2 Significant interdog variability also has been identified for dogs with the same disease processes16,17 and severity.17 Additionally, not all diseases that may increase CRP concentration would be anticipated to have similar responses to therapy.
In critically ill humans, presence of microalbuminuria at admission and increasing UACR over the course of hospitalization have been associated with increased morbidity and mortality.6,9–14 Several studies have shown the UACR determined within 15 minutes of admission to the intensive care unit to be as good a predictor of morbidity and mortality in critically ill people as APACHE2 and SOFA scores at 24 hours.6,9,10 Unfortunately, published studies in humans were not standardized with regard to the way microalbuminuria was quantified, reference intervals or cut-offs, time of sampling, or inclusion and exclusion criteria. For these reasons, meta-analysis of the data was not possible.18 Currently, no clear recommendation exists for the incorporation of microalbuminuria screening into the management of critical care cases.18
Microalbuminuria can be measured by multiple techniques. The UACR has been determined to be clinically and statistically superior to other measurement techniques in human medicine,9 where it has gained prominence as a biomarker of disease. Interpretation of UACR results in humans can be complicated by differences in urine creatinine excretion between the sexes and among races.19,20 Because creatinine excretion is not affected by sex in dogs,21 it is possible that the UACR may be more sensitive than U-ALB in dogs. In this study, equivalent results were found, probably because urine specific gravity is adjusted to 1.010 before U-ALB assay performance.
Increasing age has been associated with an increased prevalence of microalbuminuria in 3,041 apparently healthy dogs.c Nonstandardized follow-up evaluation of 572 dogs led to identification of previously undiagnosed disease in 322 cases, suggesting that occult disease may be a confounding factor.22 Results of 2 other studies in dogs, however, documented independent associations with microalbuminuria for age and presence of systemic disease.7–8 In 1 study,8 albuminuria was present in 72% (23/32) and 55% (40/73) of dogs admitted to an intensive care unit for stabilization or postoperative recovery, respectively. Presence of albuminuria was a significant risk factor for mortality in dogs in that study.8
ROC curve analysis is increasingly being utilized to evaluate the diagnostic accuracy of tests. For an ROC curve, the entire spectrum of possible sensitivities is plotted against the entire range of specificities, expressed as the false positive rate (1-specificity), based on alterations in the diagnostic cut-off for a positive versus negative result.23 Because ROC curve analysis allows assessment of a test's diagnostic accuracy over the entire continuum of sensitivity and specificity thresholds, it allows evaluation of different cut-offs to maximize sensitivity, specificity, and overall accuracy. Because a perfect test would have 100% sensitivity and specificity, the false positive rate would be 0 at any sensitivity and the AUC would equal 1.00. In contrast, a test with no discriminating power would have equal likelihood of being correct or incorrect, resulting in an AUC of 0.5. In our study, the AUCs for UACR (0.74) and U-ALB (0.68–0.74) were almost identical to results for critically ill human beings (AUC 0.61–0.74).6,10 U-ALB and the UACR were significantly lower for dogs surviving 30 days versus nonsurvivors. Although these associations are statistically quite strong and may warrant further exploration, they appear to have limited clinical utility for prognostication. For example, the U-ALB has a 46% sensitivity for 30-day survival with a specificity of 94% if a threshold of <1 mg/dL (the published cut-off for albuminuria in normal dogs) is used. Using this cut-off, 94% of dogs that died were positive for U-ALB but so were 54% of survivors. Assessment of patient risk using this cut-off might result in unnecessary interventions in, or worse, euthanasia of over half of survivors in a population with similar mortality rates. Alternately, if a threshold of <30 mg/dL is used (the cut-off for overt albuminuria), the sensitivity for survival reaches 91% but with a concomitant decrease in specificity to 32%. In this setting, only 9% of survivors in a population with a similar mortality rate would have a positive U-ALB result, but 68% of nonsurvivors would be miscategorized as being at lower risk for mortality. Further exploration may be warranted given the impact of small sample size on the precision of our ROC curves, as evidenced by wide 95% confidence intervals. ROC curves ideally should be performed with a minimum of 100 subjects, preferably 100 per group.23 Because ROC curve analysis was performed in this study using 78 cases, the curves generated are imprecise. This results in wide confidence intervals and limits application of these results to populations that differ from the enrolled population with regard to disease prevalences, relative morbidity, or survival rates. It also limits comparison of the accuracy of the different analytes with one another. Based on overlap in the 95% confidence intervals, there is no apparent or significant difference in the accuracy of the microalbuminuria assays versus the SPI2 score for predicting survival status. Based on ROC curve and Spearman's rank-correlation coefficient results, enrollment of a minimum of 770 subjects would have been necessary to detect a statistically significant difference between the ROC curves.
There were a number of additional limitations to our study. The biggest limitation to the study is the large proportion of nonsurvivors that were euthanized. Few dogs enrolled in our study died from their diseases. Instead, most nonsurvivors were euthanized within the first few days after hospitalization. Attending clinicians and case files for all euthanized patients were consulted to determine the reason for euthanasia. In all cases, euthanasia was performed because of substantial worsening of disease and prognosis, not because of financial constraints. However, it is reasonable to assume that financial considerations influenced the decision-making process for at least some of the dogs, particularly when combined with worsening clinical condition, and that some patients euthanized might have survived if unlimited treatment had been pursued. Because euthanasia is unlikely to be eliminated from veterinary practice, the inclusion of euthanized cases primarily limits our biological understanding of potential associations between critical illness and the studied biomarkers, not the application of our results to veterinary medicine.
The enrollment window of 3 hours was longer than that utilized in most studies of microalbuminuria in critically ill human patients, although 1 study noted a strong association between UACR 6 hours post-admission and outcome.9 Some patients may have received supportive therapy, including IV fluids and medications, during this time. The impact of stabilization performed before sample collection on study results is unknown. For example, interventions associated with increased glomerular permselectivity or albuminuria may have increased albuminuria concentrations. Given the rapidity of CRP increases in response to injury, the delay in obtaining samples also may have resulted in increased CRP concentrations in animals that sustained injury peracutely. Alternately, by decreasing the instability of the patient overall, it is possible that stabilization of some patients before collection of samples may have improved abnormalities in analyte concentrations. Finally, changes in biomarkers before and after stabilization may be more meaningful in risk determination for an individual patient than single time point evaluation in comparison with a larger population with a continuum of diseases and times of presentation relative to individual insults.
Finally, extrapolation of results to dissimilar patient populations should be performed with care. Because the study was performed using patients admitted to tertiary care facilities, results may not be directly applicable to primary care populations. Additionally, because inclusion in the study required disease severe enough to medically warrant hospitalization in a critical or intensive care unit, results may not be applicable to all intensive care populations at tertiary care facilities. Institutions that utilize their intensive care units for patients requiring merely fluid therapy, observation, or postoperative recovery may have very different survival rates.
In conclusion, CRP concentration was not associated with survival or any of the other analytes evaluated. Significant inverse correlations were noted between SPI2 and microalbuminuria but these correlations were of low biological relevance. Microalbuminuria and SPI2 score differed significantly for survivors versus nonsurvivors, and AUCs on ROC curve analysis for microalbuminuria were similar to results obtained in studies of critically ill people. Additional studies involving larger sample sizes and a broader spectrum of disease severity will be necessary to determine the reproducibility and clinical usefulness of microalbuminuria assays in patient risk assessment and stratification.
b MedCalc, version 11.3.3, MedCalc Software, Mariakerke, Belgium
c Radecki S, Donnelly R, Jensen WA et al. Effect of age and breed on the prevalence of microalbuminuria in dogs. J Vet Intern Med 2003;17:110 (abstract)
The authors thank Ann Reed for assistance and consultation in statistical analysis of the data.
This project was supported by a grant from Heska Corporation, 3760 Rocky Mountain Avenue, Loveland, CO 80538. The funders had no involvement in the design or performance of the study, writing of the manuscript, or the decision to submit the manuscript for publication.