Electronic health record-derived care-exposure burden and central line-associated bloodstream infection among critically ill adults: a multi-source retrospective cohort study.

Background: Central line-associated bloodstream infection (CLABSI) remains a clinically important healthcare-associated infection in intensive care units. Routinely collected electronic health record (EHR) data may capture patient-level care-process complexity not reflected by conventional catheter-related variables alone. Objective: To examine whether an EHR-derived care-exposure burden score was associated with adjudicated CLABSI or CLABSI-like bloodstream infection among critically ill adults with central venous catheter exposure. Methods: This multi-source retrospective cohort study used harmonised data from MIMIC-IV, the eICU Collaborative Research Database, and a Chinese real-world critical care cohort. Adult patients with documented central venous catheter exposure and an ascertainable post-catheterisation observation window were included. The primary care-exposure burden score summarised cumulative care-process complexity using charting frequency, unique caregivers, concurrent vascular lines, catheter duration, pre-insertion mechanical ventilation duration, antibiotic-type count, and 24-h fluid input and output after catheterisation. Components were log-transformed, robustly standardised within each data source, averaged into a composite score, and categorised into database-specific quartiles. Multivariable logistic regression estimated adjusted odds ratios (aORs) and 95% confidence intervals (CIs). A 48-h landmark-compatible sensitivity analysis recalculated the score after excluding total catheter duration. Results: The final cohort included 12,693 patients: 10,868 from MIMIC-IV, 1,489 from eICU, and 336 from the Chinese cohort. Overall, 910 patients developed adjudicated CLABSI or CLABSI-like bloodstream infection (7.2%). Event rates were 4.7, 19.1, and 33.6%, respectively. Patients with events had higher care-exposure burden scores than those without events [median 0.27 (IQR, 0.00-0.46) vs. -0.02 (IQR, -0.21 to 0.20); Conclusion: Higher EHR-derived care-exposure burden was associated with higher odds of adjudicated CLABSI or CLABSI-like bloodstream infection. The score should be interpreted as a pragmatic patient-level marker of cumulative care complexity and infection-prevention prioritisation, rather than as a causal measure of care quality or nursing performance. Prospective validation is required before clinical implementation.