• Acute renal failure;
  • Model;
  • Risk factor;
  • Score;
  • Survival


  1. Top of page
  2. Abstract
  3. Materials and Methods
  4. Results
  5. Discussion
  6. Acknowledgment
  7. References


Information regarding acute kidney injury (AKI) in cats is limited, and there are no reliable tools to objectively assess disease severity and predict outcome.


To assess clinical signs, clinicopathologic abnormalities, etiology, and outcome of cats with AKI, and to develop models using clinical metrics and empirically derived scores to predict outcome.


One hundred and thirty-two client-owned cats.


Retrospective study. Bivariate logistic regression analyses were performed to identify variables predictive of 30-day survival. Continuous variables outside the reference range were divided into quartiles to yield quartile-specific odds ratios (OR) for survival. Models were developed incorporating weighting factors assigned to each quartile based on the OR. A predictive score for each model was calculated for each cat by summing all weighting factors. A second, multivariable logistic regression model was created from actual values of the same variables. Receiver operating characteristic curve analyses were performed to determine the models' performance. Models were further tested using a subset of cases not used in initial assessment.


Fifty five of 132 cats (42%) remained dialysis-independent for at least 30 days after discharge, and the remaining 77 cats either died (n = 37, 28%) or were euthanized (n = 40, 30%). The most common etiology was ureteral obstruction (n = 46, 35%). Higher scores were associated with decreased probability of survival (P < .001). Models correctly classified outcomes in 75–77% of the cases and 84–89% of the cases in the subsequent evaluation.

Conclusions and Clinical Importance

Models can provide objective guidance in assessing AKI prognosis and severity, but should be validated in other cohorts of cats.


acute kidney injury


alanine aminotransferase


acute uremia


area under curve


blood urea nitrogen


complete blood count


chronic kidney disease


odds ratio


red blood cells


receiver operating characteristic


Veterinary Medical Teaching Hospital

Acute kidney injury (AKI) represents a spectrum of disease associated with a sudden onset of renal parenchymal injury most typically characterized by generalized failure of the kidneys to meet the excretory, metabolic, and endocrine demands of the body, resulting in acute uremia (AU).[1] At advanced stages of AKI, retained uremic solutes accumulate with dysregulation of acid-base homeostasis, metabolic processes, and fluid and electrolyte balance.[1] AU results from volume and hemodynamic abnormalities, intrinsic injury to the kidney, and postrenal causes. Prerenal azotemia or volume-responsive AKI manifests from decreased renal perfusion, most typically from dehydration or hypovolemia; however, it generally is limited in severity, resolves rapidly with timely correction of the underlying cause, and does not typically predispose to AU.[1]

Acute kidney injury develops from multiple etiologies including prolonged ischemia, nephrotoxins, intrinsic kidney diseases, infectious causes, and obstruction or rupture of urinary outflow.[1, 2] Of those, ureteral obstruction is a leading cause of kidney injury in cats in North America with variable clinical presentations resembling AKI and manifest as AU or chronic kidney disease (CKD).[2, 3],1

The management of AU is aimed at eliminating the initiating cause of kidney injury and controlling clinical consequences of uremia until recovery of kidney function. In the severest stages, medical management often is ineffective, and animals might die within days from the consequences of uremia before recovery of renal function can occur. Renal replacement therapies, including intermittent hemodialysis, can restore effectively electrolyte, acid-base, and fluid balance, and eliminate retained uremic solutes (uremia toxins), excessive water loads, and endogenous and exogenous toxins, thereby prolonging survival and improving the potential for recovery.[4] Despite the extended survival afforded by intermittent hemodialysis compared with conventional medical treatment, renal injury might be too severe for recovery. The principal factors that determine renal recovery, and thus short- and long-term prognosis, are the inciting cause, the extent of kidney damage, and the presence of comorbid disorders.[3] Despite advances in both human and veterinary medicine, the case fatality for AKI remains unacceptably high,[5-11] and there is limited information in the veterinary literature forecasting the prognosis of cats experiencing AU and their outcome before recommending expensive and invasive therapies.[2, 7, 10, 12-14]

Staging and scoring systems are based on clinical and laboratory findings, commonly used in human medicine, and might help define disease severity and predict outcome. Many scoring systems have become available in human medicine over the years, predominantly for critically ill patients.[15-17] In veterinary medicine, the use of scoring systems has been limited to the assessment of trauma, critical illness, and surgery.[6, 18, 19] Recently, a prognostic index has been proposed for cats with AKI,[11] but to date, only a single scoring system has been developed to help predict outcome for dogs with AKI managed with intermittent hemodialysis.[8] Developed using a retrospective clinical setting, the scoring strategy was simple yet accurate in predicting therapeutic outcome.[8] An objective scoring system for cats with AU might likewise help assess disease severity and prognosis, and facilitate decision-making of the initiation of costly renal replacement treatment.

The objective of this study was to assess clinical severity of AKI in cats presenting with AU and managed with intermittent hemodialysis. A companion objective was to more precisely characterize the clinical signs, clinicopathologic abnormalities, common etiologies, and outcome of these same cats to develop clinically applicable predictive models that could aid the clinician in objectively assessing disease severity and prognosis for recovery.

Materials and Methods

  1. Top of page
  2. Abstract
  3. Materials and Methods
  4. Results
  5. Discussion
  6. Acknowledgment
  7. References

Animals and Data Collection

The medical records of all cats presented with AU to the University of California, William R. Prichard Veterinary Medical Teaching Hospital (VMTH) between January 1993 and February 2007 and managed with intermitted hemodialysis were reviewed. Cats with urinary system rupture, with urethral obstruction, or that underwent renal transplantation were excluded.

Acute uremia was defined by the following: (1) acute onset of clinical signs; (2) history and physical examination consistent with AKI (anuria, oliguria, vomiting, and inappetence) or ureteral obstruction; and (3) azotemia (serum creatinine >3 mg/dL and urine specific gravity <1.020). Data retrieved from the electronic medical records included signalment, history, physical examination findings, blood pressure (measured by indirect Doppler or oscillometry), complete blood count (CBC), serum chemistry, urinalysis, venous blood gas analysis, number and identity of extrarenal organs involved, etiology, and outcome. Surviving cats were defined as those that remained dialysis independent for at least 30 days after discharge. Nonsurviving cats were defined as those that either died or were euthanized because of poor prognosis. Cats that were euthanized at their owner's request, before being treated for at least 14 days with hemodialysis, were excluded.

Laboratory Findings

Blood and urine specimens for CBC, biochemistry profile, blood gas analysis, and urinalysis were collected at initial presentation, and analyses were performed by routine methods at the diagnostic laboratories of the VMTH.


Etiology was classified when known as lily intoxication, ethylene glycol intoxication, ureteral obstruction, pyelonephritis, renal lymphoma, multicentric lymphoma, and hemodynamic dyscrasia. Ethylene glycol intoxication was documented on the basis of known exposure to antifreeze, detection of serum ethylene glycol or glycolic acid, or histologic evidence of tubular necrosis and calcium oxalate crystals within the renal tubules. Hemodynamic dyscrasia was defined as a hypotensive episode documented before the development of AU and suspected as the cause of kidney injury. Lily toxicosis was diagnosed based on a history of exposure, clinical signs, and characteristic histopathologic lesions. Imaging surveys (radiography, ultrasonography, nephropyelogram, and computed tomography) or surgery (when imaging was inconclusive) was used to document ureteral obstruction.

Organ System Involvement

Extrarenal organ involvement was classified based on clinical signs, laboratory abnormalities, or radiographic or ultrasonographic evidence of organ damage. Respiratory system involvement was considered when either increased respiratory effort was observed at the presenting physical examination or significant radiographic changes such as alveolar pattern were evident on thoracic radiographs. Neurological system involvement was diagnosed when seizures were part of the cat's current history or when significant neurological abnormalities such as ataxia were noted at presentation. Hepatic involvement was defined when serum alanine aminotransferase (ALT) activity or bilirubin concentrations were above 200 U/L (RR: 27–101) and 1.0 mg/dL (RR: 0–0.2), respectively. Evidence of cardiovascular involvement was based on echocardiography, the presence of heart murmurs, with or without arrhythmias.

Statistical Analysis

The statistically derived scoring system described for cats was patterned after that described previously and characterized to predict outcome in dogs with AKI undergoing hemodialysis.[8] Because of the long study span, an exact Cochran–Armitage test of the successive (over time) binomial proportions was performed to assess changes in case fatality over the study period.

Stage I Analysis—Variable Selection

A series of logistic regression analyses were performed initially to screen a series of variables present at initial presentation for potential association with 30 days of survival (dialysis-independent) after discharge; continuous variables were entered as linear terms regardless of the nature of their functional relationships (eg, linear, quadratic, etc.). Among these, but not exclusively, were temperature, heart rate, respiratory rate, CBC parameters, serum biochemistry parameters, venous blood gas analysis measures, and extrarenal organ involvement. Variables with P ≤ .10 were considered for Stage II analysis.

Stage II Analysis—Generation of Weighting Factors for Models Using Scores

Continuous variables identified in Stage I (eg, red blood cells [RBC] count) were partitioned into normal or abnormal ranges as either increased or decreased from the reference range. For variables whose values above the reference range were detrimental to survival, any outlying value below the reference range was included with reference range values. Conversely, for variables whose values below the reference range were considered detrimental to outcome, any outlying value above the reference range also was included with reference values. The abnormal range for each variable was further partitioned into quartiles, and the quartiles were subsequently treated as categorical variables. Logistic regression models were then constructed using the reference range as the reference category, and quartile-specific weighting factors were calculated based on the odds ratios (OR) for survival. For variables for which there was no reference range (eg, body weight), the first quartile was used as the reference category.

The weighting factors for each quartile were defined as the OR where the relationship with survival was direct, or, as the reciprocal of the OR when the relationship with survival was inverse. Quartile-specific categories with similar OR were combined to reduce the number of estimated categories and to add precision without adversely affecting validity. Similarly, when no statistically significant difference (P > .10) was observed between the quartile-specific categories and the reference range category, they were combined.

Stage III Analysis—Model Development

Predictive models were generated contingent upon the previous analyses using logistic regression. In Model A, the OR for each quartile or combined quartiles of a variable (from Stage II analysis above) was rounded to its integer value and assigned as the weighting factors for that quartile. For Model B, the weighting factor for each quartile was assigned as the exact OR determined from Stage II analysis. Models C and D were constructed in a similar manner except an etiology, weighted with the rounded or exact OR, respectively, was included additionally when known. Model E was a multivariable model that included the same variables as Models C and D, but because of model nonconvergence caused by sparse data, (1) the number of known etiologies was reduced to three, with all other known and unknown etiologies combined into a reference category; and (2) ataxia was removed.

Stage III Analysis—Model Assessment

In Models A through D, the sum of the weighting factors of each variable produced a final predictive score for each cat. In all 4 models, the value 1 was assigned as the weighting factor to the reference category. Models were assessed initially on cats with complete data (Models A though D, 94 cats; Model E, 101 cats). Receiver operating characteristic (ROC) curve analysis was performed to determine sensitivities and specificities for outcome prediction at different cutoff points. The optimal cutoff point was chosen as the value associated with the fewest misclassifications (ie, maximizing Youden's index).[20] Area under the ROC curve (AUC) was calculated as an additional assessment of its respective performance. To further evaluate the models, cats with missing data, which were not assessed initially, were used (38 cats for Models A though D and 31 cat for Model E). Missing data were replaced with the average value of the variable based on the outcome group (survivor versus nonsurvivor). Evaluation was performed using the established cutoff points based on the initial assessment.

Statistical analyses were performed by the SPSS statistical software.2 For all tests, unless specified otherwise, P < .05 was considered statistically significant.


  1. Top of page
  2. Abstract
  3. Materials and Methods
  4. Results
  5. Discussion
  6. Acknowledgment
  7. References

A total of 132 cats fulfilled the inclusion criteria. Of these, five (3.8%) were intact males, three (2.3%) were intact females, 76 (57.6%) were castrated males, and 48 (36.3%) were spayed females. The median age was 83 months (range, 4–271). Median body weight was 5 kg (range, 2.4–11.5).

The most common breeds were mixed breed (115 cats, 87%), Siamese (11 cats, 8%), Abyssinian, Himalayan, Maine coon, American shorthair, Persian, Bombay, and Burmese (1 each, 0.8%).

Clinical Signs and Blood Pressure

Median body temperature was 37.7°C (range, 34.2–39.7); median heart rate was 160 beats/min (range, 66–240); and median respiratory rate was 40 breaths/min (range, 20–80). The most common abnormalities in vital signs included hypothermia (75 cats, 64%), bradycardia (<160 beats/min, 42 cats, 35%), and increased respiratory rate (>30 breaths/min, 75 cats, 70%) (Table 1). Notably, 73 (55%) cats were anuric at presentation, 66 (50%) showed signs indicative of abdominal pain, 62 (47%) had a recent history of vomiting, and 61 (46%) suffered from fluid overload.

Table 1. History and physical examination findings in 132 cats with acute uremia at presentation
Clinical Signn (%)
Lethargy117 (89)
Inappetence105 (80)
Anuria73 (55)
Abdominal pain66 (50)
Vomiting62 (47)
Fluid overload61 (46)
Dehydration25 (19)
Distended urinary bladder15 (11)
Ataxia11 (8)
Polyuria/polydipsia8 (6)
Diarrhea3 (2)

Median systolic blood pressure was 144 mmHg (range, 70–250). A systolic blood pressure >150 mmHg was documented in 38 cats (38%). Only 1 cat presented with concurrent hypothermia, bradycardia, and hypotension.

Extrarenal Organ Involvement

Sixty-five cats (49%) had gastrointestinal tract involvement, 26 (20%) had respiratory involvement, 22 (17%) liver involvement, 20 (15%) nervous system involvement, 14 (11%) cardiovascular involvement, 4 (3%) endocrine system involvement, and 2 (1.5%) had pancreatic involvement.

Clinicopathologic Data

Most common hematologic abnormalities included anemia and leukocytosis (Table 2). Serum biochemistry abnormalities included azotemia, hyperkalemia, hypochloremia, and hyperphosphatemia. Median ALT activity, aspartate aminotransferase activity, and total serum bilirubin concentration were above their reference ranges in 26, 35, and 64% of the cats, respectively (Table 3).

Table 2. Complete blood count parameters in 132 cats with acute uremia at presentation
ParameternMedian (range)n (%)Ref Rng
Below Ref RngAbove Ref Rng
  1. Ref Rng, reference range; MCV, mean corpuscular volume; MCH, mean corpuscular hemoglobin; MCHC, mean corpuscular hemoglobin concentration.

Red blood cells (×106/μL)1126.1 (1.5–11.5)80 (71.4)1 (0.9)7–10.5
Hemoglobin (g/dL)1129.1 (2.2–17.1)68 (60.7)1 (0.9)10–16
Hematocrit (%)11326.8 (8.2–51.1)68 (60.2)1 (0.9)30–50
MCV (fL)11245.5 (35.3–60.7)15 (13.4)8 (7.1)42–53
MCH (pg/cell)11215.2 (12.4–19)2 (1.8)9 (8)13–17
MCHC (g/dL)11233.3 (26.8–38.2)3 (2.7)48 (42.9)30–33.5
White blood cells (×103/μL)11212.4 (3.8–47.1)1 (0.9)44 (39.3)4.5–14
Bands (×103/μL)1120 (0–4.3)0 (0)18 (16)0–0.2
Neutrophils (×103/μL)11210.2 (2.7–40.4)0 (0)72 (64.3)2–9
Lymphocytes (×103/μL)1120.9 (0–5.7)68 (60.7)0 (0)1–7
Monocytes (×103/μL)1120.24 (0–1.84)7 (6.3)18 (16.1)0.05–0.6
Eosinophils (×103/μL)1120.059 (0–6)72 (64.3)1 (0.9)0.15–1.1
Basophils (×103/μL)1120 (0–0.36)0 (0)1 (0.9)0–0.2
Platelets (×103/μL)110233 (34–600)28 (25.5)1 (0.9)180–500
Table 3. Serum biochemistry parameters in 132 cats with acute uremia at presentation
ParameternMedian (range)n (%)Ref Rng
Below Ref RngAbove Ref Rng
  1. Ref Rng, reference range; BUN, blood urea nitrogen; ALT, alanine aminotransferase; AST, aspartate aminotransferase; ALP, alkaline phosphatase.

Albumin (g/dL)1182.4 (1.5–3.5)32 (27.1)0 (0)2.2–4.6
Albumin/globulin ratio1050.7 (0.4–1.2)7 (6.7)0 (0)0.5–1.7
ALP (U/L)10623 (1–71)19 (17.9)0 (0)14–71
ALT (U/L)10564 (2–840)9 (8.6)27 (25.7)27–101
Anion gap (mEq/L)11736 (19–47)0 (0)106 (90.6)13–27
AST (U/L)10547 (5–1098)6 (5.7)37 (35.2)17–58
Bilirubin (mg/dL)1240.5 (0–16.1)0 (0)79 (63.7)0–0.2
BUN (mg/dL)131228 (68–456)0 (0)131 (100)18–33
BUN/creatinine ratio13013 (6.8–33.7)16 (12.3)40 (30.8)10–15
Calcium (mg/dL)1229.4 (3.9–15)45 (36.9)12 (9.8)9–10.9
Chloride (mmol/L)118108 (88–131)106 (89.8)1 (0.8)117–126
Cholesterol (mg/dL)105134 (65–314)10 (9.5)1 (1)89–258
CO2 (mmol/L)12212 (4–31)85 (69.7)8 (6.6)15–21
Creatinine (mg/dL)13117.2 (5–44.2)0 (0)131 (100)1.1–2.2
Globulin (g/dL)1053.7 (2–5.7)11 (10.5)3 (2.9)2.8–5.4
Glucose (mg/dL)106117 (38–996)4 (3.8)50 (47.2)63–118
Phosphorus (mg/dL)12615.2 (1.1–27.7)12 (9.5)108 (85.7)6.6–8.4
Potassium (mmol/L)1236 (3.1–10.9)5 (4.1)93 (75.6)3.6–4.9
Sodium (mmol/L)120150 (122–169)65 (54.2)9 (7.5)151–158
Total protein (g/dL)1086.1 (3.5–8.9)70 (64.8)2 (1.9)6.6–8.4

Etiology and Outcome

Forty-six cats (38.4%) were diagnosed with ureteral obstruction, and 28 (21.2%) underwent surgery (eg, ureterotomy, neoureterocystostomy) during hospitalization to correct the obstruction (Table 4).

Table 4. Etiology and outcome of 132 cats with acute uremia
Etiologyn% of TotalSurvival (%)
Ureteral obstruction4634.867.4
Ethylene glycol intoxication129.18.3
Lily intoxication64.516.7
Renal lymphoma32.32.3
Polycystic kidney disease10.80.0
Transplant rejection10.80.0
Multi-centric lymphoma10.80.0

Fifty-five cats (42%) remained dialysis independent for at least 30 days post discharge, and the remaining 77 cases either died (n = 37, 28%) or were euthanized (n = 40, 30%). There was no significant difference (P = .73) in case fatality over the study period. Excluding hemodynamic causes (an etiology documented in only 2 cats), ureteral obstruction was the etiology with the most favorable outcome, with a 67% survival rate (Table 4). Pyelonephritis also had a favorable outcome (57% survival), whereas lily and ethylene glycol intoxications carried grave prognoses (17% and 8% survival, respectively).

Odds ratios and their 95% confidence intervals were calculated to compare the likelihood of survival between each etiology and ureteral obstruction. Animals with ethylene glycol and lily intoxications had odds of not surviving 23 times (P = .004, 95% CI 2.7–200) and 10 times (P = .040, 95% CI 1.1–100) greater than animals with ureteral obstruction, respectively. Cats whose etiology remained unknown had 6 times lower odds of survival compared with cats with ureteral obstruction (P = .001, 95% CI 2.4–13.7). No significant differences in survival were identified between ureteral obstruction and the remaining etiologies.

Scoring System

Of all the variables (~70) screened initially for a possible relationship with survival in Stage I analysis, 18 had a P ≤ .10. Ten of these were continuous variables, 5 categorical variables reflected extrarenal system involvement, and the remaining 3 were underlying etiologies. Neither sex nor breed was associated with survival. Two of the continuous variables deemed detrimental to survival in Stage I (osmolality, colloid oncotic pressure) could not be used in subsequent analyses or formal modeling because their data were missing from a large number of patient records. To avoid collinearity, only RBC was used as a hematologic parameter despite the additional Stage I significance of hematocrit and hemoglobin. Similarly, instead of using both albumin and total protein, only the latter was employed in further analyses. Neither overall survival nor survival within the various etiologies was associated with either plasma creatinine or BUN concentrations.

Vomiting and ureteral obstruction were positively associated with survival. Of the 62 cases in which vomiting had been documented, nearly 50% had ureteral obstruction, followed by 34% with unknown etiology, 8% with EG intoxication, 6.4% with lily intoxication, and only 3% with pyelonephritis. When cases with ureteral obstruction were excluded, vomiting was still positively associated with survival (P = .003).

The following variables were included for subsequent analyses: body weight, total protein, RBC, body temperature, ataxia, fluid overload, respiratory involvement, vomiting, ethylene glycol intoxication, lily intoxication, and ureteral obstruction. Ninety-four animals' records had complete data for all the variables identified in Stage I analysis and were used to assess performance of each model.

Some variables (eg, RBC) showed a threshold phenomenon, in which the 1st 3 quartiles yielded the same OR as the reference group, and only the 4th quartile was detrimental for survival. Other continuous variables demonstrated a more progressive association with survival, with greater deviations from the reference range yielding lower OR for survival and therefore higher weighting factors.

Initially, 4 models were generated (Tables 5, 6). Outcome scores ranged from 4 to 28 for Model A (average, 11.6; SD, 5.3), 4.5 to 28 for Model B (average, 12; SD, 5.1), 1 to 39 for Model C (average, 13.1; SD, 7.3), and 1.9 to 39.1 for Model D (average, 13.7; SD, 7). In all models, a higher score was associated with lower probability of survival.

Table 5. Models A and C: Outcome prediction for 95 cats with acute uremia managed with hemodialysis, based on the integer value of the OR, with (Model C) or without including etiology (Model A)
Model A
Body weight (kg)>5 ≤5
Weighting factor1 3
Total protein (g/dL)>5 ≤5
Weighting factor1 5
RBC (×106/μL)>4.86 ≤4.86
Weighting factor1 7
Body temperature (°F)>99.3598 < X ≤ 99.35≤98
Weighting factor137
Weighting factor81 
Fluid overloadYesNo 
Weighting factor21 
Respiratory involvementYesNo 
Weighting factor31 
Weighting factor(−) 31 
Model C: when the etiology is known, an additional weighting factor is added
EG intoxicationYesNo 
Weighting factor91 
Lily intoxicationYesNo 
Weighting factor41 
Ureteral obstructionYesNo 
Weighting factor(−) 51 
Table 6. Models B and D: Outcome prediction for 95 cats with acute uremia managed with hemodialysis, based on exact odds ratios, with (Model D) or without including etiology (Model B)
Model B
Body weight (kg)>5 ≤5
Weighting factor1 3.36
Total protein (g/dL)>5 ≤5
Weighting factor1 4.6
RBC (×106/μL)>4.86 ≤4.86
Weighting factor1 6.7
Body temperature (°F)>99.3598 < X ≤ 99.35≤98
Weighting factor13.26.6
Weighting factor8.11 
Fluid overloadYesNo 
Weighting factor2.31 
Respiratory involvementYesNo 
Weighting factor2.91 
Weighting factor(−) 2.51 
Model D: when an etiology is known, an additional weighting factor is added
EG intoxicationYesNo 
Weighting factor9.11 
Lily intoxicationYesNo 
Weighting factor3.71 
Ureteral obstructionYesNo 
Weighting factor(−) 4.61 

Models A, B, C, and D yielded sensitivities/specificities of 74%/75%, 74%/75%, 62%/88%, and 66%/83%, respectively (Tables 7, 8). Optimal cutoff scores (ie, those that minimized overall misclassification) with their respective sensitivities, specificities, number of correctly classified cases, and AUC values for each model are presented in Table 7 and Figure 1. All models performed similarly with regard to classification (75–77% correctly classified). When etiology was known, specificity increased at the expense of sensitivity (Models C and D). When applying the models to cats not used for initial assessment (38 cats with missing data) using the same cutoffs, Models A, B, C, and D yielded sensitivities/specificities of 82%/85%, 89%/83%, 90%/86%, and 71%/88%, respectively, with correct classification of 84–89% of the cases.


Figure 1. Receiver operating characteristics (ROC) analyses for Models A though D. Models A and C were generated based on the integer value of the odds ratio for each variable. Models B and D were generated based on the exact odds ratio. In Models C and D, etiology was included. AUC, area under the curve.

Download figure to PowerPoint

Table 7. Cutoff scores and their respective sensitivities and specificities for Models A–D. Models A and C were generated based on the integer-equivalent value of the odds ratio for each variable. Models B and D were generated based on the exact odds ratio. In Models C and D, etiology was included
Model AModel BModel CModel D
  1. Sen, sensitivity value; Spe, specificity value.

  2. Bolded font represents optimal cutoff points with the fewest misclassifications.

10 74 75 10.4 74 75 10 62 88 12 66 83
Table 8. Performance characteristics of 5 predictive models. Models A and C utilized scores generated from the rounded integer value of the odds ratio for each variable. Models B and D were generated from the exact odds ratio value. Etiology was included in Models C and D. Model E was a multivariable logistic regression model. Ninety-four and 101 cats had complete data and were used for initial model assessment of Models A–D and Model E, respectively. Numbers in parentheses are results for 38 and 31 cats with missing data that were used for further model assessment of Models A–D and Model E, respectively
ModelOptimal Cutoff PointSensitivitySpecificityCorrectly Classified (%)AUC
  1. AUC, area under the ROC curve.

  2. a

    Probability cutoff point.

A1074 (82)75 (85)75 (87)0.81
B10.474 (89)75 (83)75 (86)0.81
C1062 (90)88 (86)77 (89)0.85
D1266 (71)83 (88)76 (84)0.86
E0.41a82 (80)77 (91)77 (87)0.86

Variables selected from the Stage I analysis were also used to create the following multivariable logistic regression model (Model E): logit (p) = 0.31 × (body weight) + 0.59 × (temp) − 0.7 × (fluid overload) + 0.46 × (RBC) − 2.11 × (Lily intoxication) − 2.11 × (ethylene glycol intoxication) + 2.3 × (ureteral obstruction) − 63.95. When applying the model to the 101 cats that had data on all variables in the model, the optimal probability cutoff point was 0.41, corresponding to sensitivity and specificity of 82% and 77%, respectively, and 80/101 (79%) were classified correctly. When applying the model to cats after missing data analysis (31 cats), using the same optimal probability cutoff point, sensitivity and specificity were 80% and 91%, respectively, and 27/31 (87%) were classified correctly.

The relationship between prior probability of survival and positive predictive value after applying Models C and E is presented in Figure 2.


Figure 2. Relationship between prior probability of survival and positive predictive value for Models C and E.

Download figure to PowerPoint


  1. Top of page
  2. Abstract
  3. Materials and Methods
  4. Results
  5. Discussion
  6. Acknowledgment
  7. References

This study describes the clinical signs, clinicopathologic features, and the outcomes of a large population of cats with severe AKI requiring hemodialysis. In addition, different models (eg, a multivariable regression model and a clinically applicable scoring system) that can aid in the objective assessment of disease severity and prognosis prediction of cats with AU undergoing hemodialysis treatment are proposed.

The clinical presentation of the cats in this study was consistent with AU. The majority of cases were lethargic and inappetent. Hypothermia, a consistent finding in AU, was common and most likely resulted from retained uremia solutes.[21]

The prevalence of increased blood pressure in this population of uremic cats was generally lower than reported for dogs with AKI and is in accordance with previously published reports of blood pressure in cats with CKD.[22-24] Fifty-five percent of cases were anuric. Oliguria could not be adequately assessed as urine production, a prognostic factor in other studies,[5, 25] was not measured because urinary catheters were not implanted routinely.

Anemia was the most common abnormality revealed on the CBC and was characterized as nonregenerative in most cats based on the absence of characteristics of regeneration (ie, reticulocytes, polychromasia). This finding is more characteristic of CKD, which might have coexisted in some cats. The appearance of nonregenerative anemia also might have resulted from inflammation, acute blood loss (eg, gastrointestinal tract), and aggressive fluid treatment with subsequent overhydration. Similar results were observed in a review of 136 cats with ureteral urolithiasis.[26] However, in a different review of acute intrinsic renal failure in cats, the mean (±SD) hematocrit at initial diagnosis was 37 ± 7.5%.[10] Many cats with nephrolithiasis and ureteral obstruction that present with a seemingly AU have acute-on-CKD with anemia proportionate to the severity and duration of the underlying CKD component.

A few clinicopathologic findings at the time of presentation were not consistent with AKI. Surprisingly, a large proportion of cats demonstrated increased ALT activity as well as increased serum bilirubin concentrations. Increased ALT activity also has been reported in a previous study in a high percentage of cats with CKD.[27] The increase in liver enzyme activity and serum bilirubin concentration in this study might reflect liver damage secondary to ischemic or hypoxia injury, toxic insults, or pancreatitis.

The most commonly identified etiology of AKI was ureteral obstruction (35%), with ethylene glycol intoxication the next most frequent etiology representing only 9% of cases. This observation represents a shift in the most prevalent cause of AKI in cats requiring hemodialysis since the mid-1990s. In the 5-year period between 1993 and 1998, only 10% of cats were presented with ureteral obstruction as the cause of AKI, and 35% of cats had toxic etiologies (primarily ethylene glycol). In the subsequent 5-year period from 1999 to 2003, the prevalence of ureteral obstruction increased to 51%, and toxic etiologies decreased to 11%.1 A steady increase in the incidence of ureteral calculi from the late 1990s was previously described, and paralleled with the change documented in the incidence of calcium oxalate and struvite urolithiasis in the last 3 decades.[26, 28-30]

The case fatality was 58.3%, with the remaining 55 cases (41.7%) surviving independent of dialysis for longer than 30 days. This agrees with other studies of uremic cats treated with hemodialysis.[8, 10, 11, 31] It should be noted, however, that definitions of survival as well as duration of follow-up vary among the different studies, as might severity of kidney injury and extent of pre-existing CKD. These limitations further underscore the need for objective staging and scoring systems that can be used to index disease severity and allow more valid comparisons between studies performed in different clinical settings and at different times.

Etiology had a significant impact on the survival in cats managed with hemodialysis for AU as it does for dogs.[8] Ureteral obstruction and pyelonephritis were associated with favorable outcomes (67% and 57% survival, respectively), whereas lily and ethylene glycol intoxication had exceptionally poor prognoses (17% and 8% survival, respectively). In animals managed by hemodialysis, the reversibility of the injury, rather than its severity, is considered a more important indicator of prognosis, because the clinical and clinicopathologic consequences of uremia that influence mortality with conventional treatment can be managed with extracorporeal renal replacement treatment. The relatively positive outcome for ureteral obstruction was likely associated with surgical correction of the inciting cause. Nevertheless, prognosis for a favorable outcome for AKI after ureteral obstruction is highly dependent on available and experienced surgical expertise, opportunity for ureteral stenting, and degree of pre-existing (chronic) kidney injury.

For several etiologies, the number of cases was conspicuously low, and the reported survival rates should be interpreted cautiously. The severity of ethylene glycol and lily toxicoses, the delayed presentation for treatment, and the irreparable damage to the kidneys probably account for their grave prognoses.[7, 10, 32] Ethylene glycol negatively affects survival in dogs with AKI in a similar evaluation.[8] In a different study, 3 of 9 cats with ethylene glycol-induced AKI survived after hemodialysis treatment for at least 4 months.[31] However, these uncharacteristically high survival rates in cats with ethylene glycol intoxication were associated with prolonged courses of hemodialysis, and incomplete recovery of renal function upon cessation of treatment. In this study, 5 of 17 cats with ethylene glycol intoxication underwent renal transplantation and were excluded from analyses. The timing of diagnosis and initiation of hemodialysis treatment in cats with ethylene glycol intoxication was also critical for survival and might explain the differences in outcomes between studies.[33, 34]

Scoring System

The scoring systems were developed to facilitate prediction of 30-day postdischarge survival of cats with AU managed with hemodialysis. Cats with severe AKI are subject to prolonged hospitalization and intensive treatment, including hemodialysis, associated with high treatment costs. Consequently, the objective assessment of survival outcome at the time of presentation could play a pivotal role in the decision to pursue such a treatment or opt for euthanasia. An AKI scoring systems also has potential application to facilitate quantitative comparison of the severity of AKI in cats from different studies. Many such scoring systems have become available in human medicine over the years to predict prognosis in critically ill patients. In veterinary medicine, however, the use of scoring systems is limited, and only 1 scoring system for dogs with AKI has been developed.[8, 19, 35]

Of the variables considered to potentially influence survival screened in Stage I, only 18 had a P ≤ .10. Hyperkalemia, a historical cause of mortality in AKI, was not one of these, in contrast to a previous report[10] and in agreement with a recent one.[11] This counterintuitive finding relates to the efficacy of hemodialysis to control potassium homeostasis, so that hyperkalemia no longer compromises the animal. Similarly, the degree of azotemia was not significantly associated with survival, in agreement with previous studies.[8, 10, 11, 31] These observations in cats and comparable observations in dogs undergoing hemodialysis contravene the hypothesis that severity of azotemia is a risk factor for AKI in dogs managed without hemodialysis.[5]

Vomiting was positively associated with survival. Of the 62 cases in which vomiting had been documented, nearly 50% had ureteral obstruction as an etiology. The positive predictive effect of vomiting might be partly ascribed to its association with ureteral obstruction, which had a favorable outcome in this study. Similarly, in the scoring system for dogs with AKI, increased ALT activity was positively associated with survival likely attributable to its association with leptospirosis as an etiology, which typically had a good outcome for survival.[8] Nevertheless, when cases with ureteral obstruction were excluded from the analysis, vomiting was still positively associated with survival (P = .003), suggesting that an association between vomiting and ureteral obstruction cannot explain entirely the positive association of vomiting with survival.

Four statistics-based scoring systems and a multivariate model were developed to test performance at predicting survival outcome in cats presenting with AKI. The multivariate approach provided the most accurate predictive fit; however, the performance of all models was comparable. Nonetheless, we specifically sought to create an integer-based scoring system that could easily be implemented in an ICU context, as has been done in human scoring systems. When using the AUC of the ROC analysis for assessing model performances, models were considered good (ie, >0.80) at discriminating survivors from nonsurvivors. Whenever the etiology was known, the specificity increased at the expense of sensitivity: fewer animals were classified incorrectly as survivors, but model performance only slightly improved. The association between specific etiologies and specific clinical signs or clinicopathological parameters (eg, ureteral obstruction and vomiting, hypocalcemia, and EG intoxication) might have obscured differences between the models as these parameters served as proxies for etiologies.

Models A or B, which were independent of etiology, might be more pragmatic as often etiological cause is unknown at case presentation, and decisions to initiate hemodialysis must be made early in the disease course. Etiology might remain unknown throughout hospitalization and hence could not facilitate prognostic projections.

The overall classification performance of all models was very good, but there were differences between models' predictions of survivors and nonsurvivors, and the use of scoring systems should be used cautiously when predicting the outcome in individual cats. Models should be used in conjunction with other laboratory tests, clinical assessment, and clinician experience, rather than serve as the sole means by which decisions are made. Discretionary selection of outcome classification cutoff points can maximize sensitivities and specificities according to clinician's (and owner's) preferences. The higher the cutoff point, the higher the sensitivity (fewer false negatives) and the lower the specificity (greater false positives).

A similar scoring system, developed for dogs with AKI, showed better results with an AUC of 0.88–0.91.[8] These differences might be explained by the differences in etiologies of AKI between dogs and cats. Perhaps, the most important factor that accounts for species differences between scoring systems is the observation that AU in dogs represents bona fide acute intrinsic kidney injury, whereas many cats presenting for hemodialysis had considerable degrees of pre-existing CKD before the development of AU, ie, so-called “acute-on-chronic” kidney disease. Therefore, cats with AU in this study probably represented a more heterogeneous group than dogs with intrinsic AKI with no chronic component.

There are several drawbacks to this study. First, medical records were reviewed retrospectively, and data for several key variables were missing from a substantial number of cases that could not be included in the initial statistical analyses. For example, urine production, a reportedly important prognostic factor, could not be included in the models. Second, hemodialysis is only available in a small number of referral centers, and most cats typically received treatment before arrival. This might have resulted in changes in some clinicopathologic findings at presentation. Third, in the development of the scoring system, both surgical cases and cases managed solely by medical means were included. Surgical cases (eg, ureteral obstruction) might have had a better prognosis as the inciting cause could potentially be resolved. Nevertheless, ureteral obstruction constituted a substantial cause of AU in cats, and surgical cases were therefore included in this study. Fourth, this study considered euthanized cats as a nonsurvivor group. To minimize this effect, euthanized cats with no meaningful attempt at management were excluded from the study, and euthanasia was performed because of poor prognosis for recovery.


  1. Top of page
  2. Abstract
  3. Materials and Methods
  4. Results
  5. Discussion
  6. Acknowledgment
  7. References

Conflict of Interest Declaration: Authors disclose no conflict of interest.

  1. 1

    Pantaleo V, Francey T, Fischer JR, Cowgill LD. 2004. Application of hemodialysis for the management of acute uremia in cats: 119 cases (1993–2003). ACVIM Forum 2004, Minneapolis, MN (abstract)

  2. 2

    SPSS 15.0 for Windows; SPSS Inc, Chicago, IL


  1. Top of page
  2. Abstract
  3. Materials and Methods
  4. Results
  5. Discussion
  6. Acknowledgment
  7. References