Modeling sepsis induced by methicillin-resistant Staphylococcus aureus infection: a human/ovine approach

Introduction

Methicillin-resistant Staphylococcus aureus (MRSA) is an invasive pathogen in critically ill patients, and is commonly the cause of nosocomial pneumonia. Infection with MRSA can lead to bacteremia, septic shock, and multisystem dysfunction. Previous work has demonstrated that a MRSA pneumonia-sepsis model in sheep mimics the vascular state in human sepsis. In the present study, we sought to combine sheep and clinical data from humans to determine whether common parameters existed across sepsis with regard to coagulopathy. Secondly, we wanted to model and provide estimates of MRSA bacterial load in both species.

Methods

Nineteen sheep with acute lung injury and 14 human patients were incorporated into this sepsis model. In sheep, pneumonia was induced by inoculating the airway with ~2.5 × 10¹¹ colony-forming units (CFU) MRSA. Thirteen of the sheep had smoke injury induced through inhalation of cotton smoke. All human patients were retrospectively studied and were bacteremic with MRSA from varying primary infection sites. Initial bacterial load in humans was modeled using clinical and microbiologic data available at the start of sepsis, while the initial load in sheep was the inoculating amount of bacteria. Load continues throughout the study period and is modified by vital signs and antibiotic coverage. The bacterial load as well as clinical and laboratory parameters are inputs, with the output parameter being prothrombin time (PT). In order to minimize overparameterization of the population, the model was allowed to estimate PT using only three parameters. Data were modeled for 24 to 48 hours.

Results

Bacterial load was estimated to range from between 10⁸ and 10¹¹ CFU, with the high end of the range being similar to the inoculum used to induce pneumonia in the sheep. The highest-ranking parameters in estimating PT were calcium, potassium, and bacterial load. When using calcium alone, the model estimate agreement with measured PT was r² = 0.25. Combining calcium and potassium improved agreement (r² = 0.34), while using all three parameters further improved the estimate (r² = 0.37).

Conclusions

Through progression modeling we were able to provide prediction of coagulopathy and bacterial load across two different species of animals infected with the same organism.