The Systemic Inflammatory Response Syndrome and Predictors of Infection and Mortality in 1068 Critically Ill Newborn Foals

Abstract

Background: Sepsis has been defined in humans as the concurrent proven or suspected presence of microbial infection and the systemic inflammatory response syndrome (SIRS). Sepsis is the leading cause of morbidity and mortality in neonatal foals. The clinical utility of using SIRS or its individual components to predict infection and mortality in critically ill foals is currently unknown.

Objectives: Assess the ability of history and signalment, clinical findings, laboratory results, and SIRS-related indices to predict infection and mortality in critically ill foals.

Animals: Retrospective, multi-center, cross-sectional study using a convenience sample of 1068 critically ill foals < 3 days of age admitted to 16 veterinary referral hospitals in 4 countries.

Methods: Data were retrieved from medical records. Infection was defined as the presence of bacteremia (positive blood culture) or clinical identification of an infected focus on admission. Univariate non-parametric and categorical methods, multivariate logistic regression, and classification tree methods were used for statistical analysis.

Results: Foal age at admission and presence of toxic neutrophils were independent predictors of infection, whereas SIRS-related indices were not predictive of infection. In-hospital mortality was 24%. Independent predictors for mortality were hypokinetic pulses, cold extremities, presence of seizures, blood L-lactate concentration > 6.0 mmol/L, and increased serum potassium and total bilirubin concentrations.

Conclusions and clinical importance: The presence of infection in critically ill newborn foals was not predicted by SIRS indices. Cardiovascular dysfunction was strongly associated with mortality, suggesting that maintaining adequate perfusion and pulse pressure should be important treatment goals.

Keywords: bacteremia; hyperlactatemia; sepsis; shock.

FIGURE 1

Receiver operating characteristic (ROC) curve for predicting infection in 455 critically ill newborn foals. Infection was defined as the presence of bacteremia (positive blood culture) or clinical identification of an infected focus at a non‐blood site on admission. The left panel is the ROC curve for the logistic regression model obtained using history, signalment and physical examination data when available for > 50% of foals (n = 406; ROC = 0.62; red diagonal line indicates a test of no predictive value). At the optimal cut point (foal age > 55 h), Se = 0.52, Sp = 0.68, and accuracy = 0.62. The probability of infection increased almost linearly with age over the first 3 days (right panel; the blue shaded area represents the 95% confidence interval for prediction; open circles indicate the age of the foal on admission).

FIGURE 2

Receiver operating characteristic (ROC) curves for predicting infection in 455 critically ill newborn foals. Infection was defined as the presence of bacteremia (positive blood culture) or clinical identification of an infected focus at a non‐blood site on admission. The left panel is the ROC curve for the logistic regression model obtained using laboratory data when available for > 50% of foals (n = 270; ROC = 0.64; red diagonal line indicates a test of no predictive value). At the optimal cut point (toxic neutrophils present), Se = 0.62, Sp = 0.66, and accuracy = 0.64. The right panel is the ROC curve for the logistic regression model obtained using all the data when available for > 50% of foals (n = 270; ROC = 0.69). At the optimal cutpoint (toxic neutrophils present and foal age > 55 h), Se = 0.67, Sp = 0.61, and accuracy = 0.63.

FIGURE 3

Receiver operating characteristic (ROC) curves for predicting in‐hospital mortality of critically ill newborn foals. The left panel is the ROC curve for the logistic regression model obtained using history, signalment and physical examination data when available for > 50% of foals (n = 420; ROC = 0.77; predictors are hypokinetic pulses and cold extremities). The right panel is the ROC curve for the logistic regression model obtained using laboratory data when available for > 50% of foals (n = 384; ROC = 0.76; predictors are increased blood L‐lactate concentration and potassium concentration). The bottom panel is the ROC curve for the logistic regression model obtained using all data when available for > 50% of foals (n = 264; ROC = 0.81; predictors are hypokinetic pulses and increased total bilirubin concentration).

FIGURE 4

Classification tree‐derived decision tree for in‐hospital mortality in 1 068 critically ill newborn foals. Each node includes the total number of foals and the number of foals with in‐hospital mortality, expressed as a percentage. Branches to the left in a split reflect a lower probability of mortality, whereas branches to the right (gray shading) reflect a higher probability of mortality. Rectangles represent subgroups that are not able to be further subdivided. The presence of hypokinetic pulses was the strongest predictor of in‐hospital mortality, with increased mortality associated with hyperlactatemia (blood L‐lactate concentration > 6.0 mmol/L). In contrast, in foals with normal arterial pulse pressures as assessed by palpation, the presence of seizures was also predictive of in‐hospital mortality.