Search

Background: Patients with mechanical ventilators (MV) and central lines (CL) have increased rates of infection during hospitalization compared with other patients. Our objectives were to 1) model changing infection rates varied over device duration adjusting for changes in risk factor distribution throughout that time and 2) to determine whether changing rate models were a more appropriate analysis than constant rate models. Methods: Electronic healthcare records of inpatients at UNC Hospitals between January 1, 2002 and December 31, 2007 who used central line(s) or mechanical ventilator were collected. We modeled changing rates across duration of device placement via discrete-time hazards regression, adjusting for age, gender, race/ethnicity, comorbidities, patient location, and hospital service. We compared changing rate to constant rate models via differences in error and F-tests. Results: MV-associated infection peaked around 7 days; CL-associated infection peaked around 14 days. Rates varied between 2 and 10 infections per 1000 device-days. MV infection rates were higher among patients with trauma, central venous catheters, or blood-stream infection. Infection rates were lower on the floor compared to ICU. After 11 days, medical and pediatric patients had decreased MV infection rates compared with surgical patients. With CL, rate of infection was higher in Black individuals, and patients in intensive care units (ICUs) or on a surgical service. In the first week, males had increased infection risk compared to females. Changing rate models fit the observed rates better than constant rate models for surgical and ICU patients, and the UNC Hospitals overall population. Various definitions of device exposure indicated that changing rate models were better for at least some patient groups. Conclusion: Variation in rates across duration of time at risk remained after adjusting for the changing population at risk. Awareness of patients at increased risk, monitoring throughout hospitalization and timing interventions and surveillance appropriately could reduce the incidence of MV and CL associated infections. Investigators should be aware that when comparing patient groups or different hospitals without accounting for the changing rate of infection across device duration estimates may not be meaningful.