Methods: We used data from the Michigan Hospital Medicine Safety (HMS) Consortium, a 51-hospital Blue Cross Blue Shield-of Michigan funded collaborative to create a CLABSI prediction tool. CLABSI was determined by National Healthcare Safety Network (NHSN) criteria. A multivariable model that used a bootstrap resampling method with stepwise variable selection (including demographic data, comorbidities, admission diagnosis, CLABSI history, and device characteristics), was fit to create a risk prediction tool. Candidate predictor variables with p-values <0.25 in the univariate analyses for association with CLABSI were considered for inclusion in 1,000 bootstrap samples. Candidate variables were then used in a logistic generalized estimating equation (GEE) model, controlling for hospital-level clustering. Regression coefficients were used to assign points to each predictor to calculate a CLABSI risk score. Model fit and predictive value of candidate models were assessed using standard approaches.
Results: Of the 19,080 patients with PICCs, 287 (1.5%) had an NHSN-confirmed CLABSI. Predictive factors from the multivariate model included: CLABSI history (OR=4.24, 95%CI 2.34, 7.70), chemotherapy via PICC (OR=4.02, 95%CI 2.76, 5.84), TPN via PICC (OR=2.46, 95%CI 1.78, 3.41), presence of another CVC at the time of PICC placement (OR=2.10, 95%CI 1.57, 2.80), multiple PICC lumens (OR=1.81, 95%CI 1.31, 2.51), dwell time of 14-days or more (OR=1.63, 95%CI 1.29, 2.06), hemoglobin of 11gm/dL or less (OR=1.49 (1.13, 1.97), and age 65 or older (OR=1.41, 95%CI 1.10, 1.82). The AUC for the “optimism” corrected model was 0.74 (95% CI: 0.71, 0.77).
Conclusions: We derived and internally validated a CLABSI-risk prediction tool using a large cohort of hospitalized medical patients. Future studies that focus on external validation of the tool and methods to implement the score to inform risk and potentially avoid CLABSI would be welcomed.