Inverse probability weighting leads to more accurate incidence estimates for healthcare associated infections in intensive care units, results from two national surveillance systems.

Abstract

Background: Two main approaches are employed to monitor healthcare associated infections (HAIs): longitudinal surveillance, which allows to measure incidence rates, and point prevalence surveys (PPS). PPS are less time-consuming; however, they are affected by length-biased sampling, which can be corrected through inverse probability weighting. We assessed the accuracy of this method by analysing data from two Italian national surveillance systems.

Methods: Ventilator associated pneumonia (VAP) and central-line associated bloodstream infection (CLABSI) incidence measured through a prospective surveillance system (GiViTI) was compared to incidence estimates obtained through conversion of crude and inverse probability weighted prevalence of the same HAIs in intensive care units (ICUs) measured through a PPS. Weighted prevalence rates were obtained after weighting all patients inversely proportional to their time-at-risk. Prevalence rates were converted into incidence per 100 admissions using an adapted version of the Rhame and Sudderth formula.

Results: Overall, 30988 patients monitored through GiViTI, and 1435 patients monitored through the PPS were included. A significant difference was found between incidence rates estimated based on crude VAP and CLABSI prevalence and measured through GiViTI (relative risk, RR 2.5 and 3.36; 95% confidence interval, CI 1.42 - 4.39 and 1.33 - 8.53, p = 0.006 and 0.05 respectively). Conversely, no significant difference was found between incidence rates estimated based on weighted VAP and CLABSI prevalence and measured through GiViTI (p = 0.927 and 0.503 respectively).

Conclusion: When prospective surveillance is not feasible, our simple method could be useful to obtain more accurate incidence rates from PPS data.