Table 4 Main features of the fully automated algorithm for CRBSI/CLABSI detection according to PRAISE

Type of system

Fully automated

HAI targeted

Intravascular catheter infections (definitions adapted from ECDC)

Date sources

Electronic Health Record

Validation method

Comparison with prospective manual surveillance of BSI conducting according to ECDC definitions

Comparator

Manual surveillance of bloodstream infections over a 6-year period (2016 to 2021)

Data type included

Administrative data, microbiology lab data, individual intravascular catheters data extracted from data EHR

Patient population

All adult patients admitted to the intensive care unit

Indicators

CRBSI, CLABSI, ICU-onset BSI

Denominators

Catheter-days and patient-days

Sensitivity

83% (95%CI: 43.7–96.9)

Specificity

100% (95%CI: 99.5–100)

Lessons learned

- Several cut-offs for different parameters in the algorithm were set arbitrarily and would require further in-depth sensitivity analyses (e.g., delay between two blood cultures with the same common contaminant to consider the episode; delay to consider positive specimens with the same bacteria as in the blood cultures to exclude a CRBSI).

- Some ECDC rules were not transposable in a fully automated algorithm because of the lack of availability or accuracy of the data in the IT system: (e.g., quantitative blood culture ratio CVC blood sample/peripheral blood sample; differential time of blood culture positivity)

- Some data are difficult to capture in a fully automated algorithm because of the lack of standardisation (e.g., culture from pus from the insertion site of the catheters are frequently mislabelled and difficult to identify in the microbiology database).