Background context: There is very little evidence to guide treatment of patients with spinal surgical site infection (SSI) who require irrigation and debridement (I&D) in deciding need for single or multiple I&Ds or more complex wound management such as vacuum-assisted closure dressing or soft-tissue flaps.
Purpose: The purpose of this study was to build a predictive model that stratifies patients with spinal SSI, allowing us to determine which patients will need single versus multiple I&D. The model will be validated and will serve as evidence to support a scoring system to guide treatment.
Study design: A consecutive series of 128 patients from a tertiary spine center (collected from 1999 to 2005) who required I&D for spinal SSI were studied based on data from a prospectively collected outcomes database.
Methods: More than 30 variables were identified by extensive literature review as possible risk factors for SSI and tested as possible predictors of risk for multiple I&D. Logistic regression was conducted to assess each variable's predictability by a "bootstrap" statistical method. A prediction model was built in which single or multiple I&D was treated as the "response" and risk factors as "predictors." Next, a second series of 34 different patients meeting the same criteria as the first population were studied. External validation of the predictive model was performed by applying the model to the second data set, and predicted probabilities were generated for each patient. Receiver operating characteristic curves were constructed, and the area under the curve (AUC) was calculated.
Results: Twenty-four of one hundred twenty-eight patients with spinal SSI required multiple I&D. Six predictors: anatomical location, medical comorbidities, specific microbiology of the SSI, the presence of distant site infection (ie, urinary tract infection or bacteremia), the presence of instrumentation, and the bone graft type proved to be the most reliable predictors of need for multiple I&D. Internal validation of the predictive model yielded an AUC of 0.84. External validation analysis yielded AUC of 0.70 and 95% confidence interval of 0.51 to 0.89. By setting a probability cutoff of .24, the negative predictive value (NPV) for multiple I&D was 0.77 and positive predictive value (PPV) was 0.57. A probability cutoff of .53 yielded a PPV of 0.85 and NPV of 0.46.
Conclusions: Patients with positive methicillin-resistant Staphylococcus aureus culture or those with distant site infection such as bacteremia were strong predictors of need for multiple I&D. Presence of instrumentation, location of surgery in the posterior lumbar spine, and use of nonautograft bone graft material predicted multiple I&D. Diabetes also proved to be the most significant medical comorbidity for multiple I&D. The validation of this predictive model revealed excellent PPV and good NPV with appropriately chosen probability cutoff points. This study forms the basis for an evidence-based classification system, the Postoperative Infection Treatment Score for the Spine that stratifies patients who require surgery for SSI, based on specific spine, patient, infection, and surgical factors to assess a low, indeterminate, and high risk for the need for multiple I&D.
Copyright © 2012 Elsevier Inc. All rights reserved.