Development of an Algorithm for Surveillance of Ventilator-Associated Pneumonia With Electronic Data and Comparison of Algorithm Results With Clinician Diagnoses

Abstract

Objective.Surveillance for ventilator-associated pneumonia (VAP) using standard Centers for Disease Control and Prevention (CDC) criteria is labor intensive and involves many subjective assessments. We sought to improve the efficiency and objectivity of VAP surveillance by adapting the CDC criteria to make them amenable to evaluation with electronic data.Design.Prospective comparison of the accuracy of VAP surveillance by use of an algorithm with responses to prospective queries made to intensive care physicians. CDC criteria for VAP were used as a reference standard to evaluate the algorithm and clinicians' reports.Setting.Three surgical intensive care units and 2 medical intensive care units at an academic hospital.Methods.A total of 459 consecutive patients who received mechanical ventilation for a total of 2,540 days underwent surveillance by both methods during consecutive 3-month periods. Electronic surveillance criteria were chosen to mirror the CDC definition. Quantitative thresholds were substituted for qualitative criteria. Purely subjective criteria were eliminated. Increases in ventilator-control settings were taken to indicate worsening oxygenation. Semiquantitative Gram stain of pulmonary secretion samples was used to assess whether there was sputum purulence.Results.The algorithm applied to electronic data detected 20 patients with possible VAP. All cases of VAP were confirmed in accordance with standard CDC criteria (100% positive predictive value). Prospective survey of clinicians detected 33 patients with possible VAP. Seventeen of the 33 possible cases were confirmed (52% positive predictive value). Overall, 21 cases of confirmed VAP were identified by either method. The algorithm identified 20 (95%) of 21 known cases, whereas the survey of clinicians identified 17 (81%) of 21 cases.Conclusions.Surveillance for VAP using electronic data is feasible and has high positive predictive value for cases that meet CDC criteria. Further validation of this method is warranted.