A novel clinical scoring system for outcome prediction in dogs with acute kidney injury managed by hemodialysis

Abstract

Background: No reliable tool to predict outcome of acute kidney injury (AKI) exists.

Hypothesis: A statistically derived scoring system can accurately predict outcome in dogs with AKI managed with hemodialysis.

Animals: One hundred and eighty-two client-owned dogs with AKI.

Methods: Logistic regression analyses were performed initially on clinical variables available on the 1st day of hospitalization for relevance to outcome. Variables with P< or = .1 were considered for further analyses. Continuous variables outside the reference range were divided into quartiles to yield quartile-specific odds ratios (ORs) for survival. Models were developed by incorporating weighting factors assigned to each quartile based on the OR, using either the integer value of the OR (Model A) or the exact OR (Models B or C, when the etiology was known). A predictive score for each model was calculated for each dog by summing all weighting factors. In Model D, actual values for continuous variables were used in a logistic regression model. Receiver-operating curve analyses were performed to assess sensitivities, specificities, and optimal cutoff points for all models.

Results: Higher scores were associated with decreased probability of survival (P < .001). Models A, B, C, and D correctly classified outcomes in 81, 83, 87, and 76% of cases, respectively, and optimal sensitivities/specificities were 77/85, 81/85, 83/90 and 92/61%, respectively.

Conclusions and clinical relevance: The models allowed outcome prediction that corresponded with actual outcome in our cohort. However, each model should be validated further in independent cohorts. The models may also be useful to assess AKI severity.