Bacteremia Detection in Second or Subsequent Blood Cultures Among Hospitalized Patients in a Tertiary Care Hospital

This cohort study assesses the likelihood of detecting microbiological positivity or bacteremia in second or subsequent blood cultures among hospitalized patients while the first culture is still incubating after 24 hours.


Definitions
We defined the first BC set as the first two BC bottles collected in the ED (BC collected at t0), whereas the additional BC bottles were collected at least 24 hours later. Each bottle was associated with date and time information for the following events: patient admission; BC collection; incubation start/end as reported by the automated blood culture system; and diffusion to the patient's electronic health record (EHR). The following values were derived: pre-analytical time (PAT) corresponding to the time elapsed between the sample collection and the start of incubation as it can be affected by laboratory operating hours, and microbiological time-to-positivity (mTTP) corresponding to the time elapsed between the start and the end of the incubation for a positive BC (eFigure). During the study period, our laboratory opening hours were 6:00am -7:00pm. Outside this timeframe, BC were incubated © 2022 Zanella MC et al. JAMA Network Open.
at room temperature in the general accessioning area until the next morning when they were loaded on the automated BC system.

Conditional Probabilities for Independent Events and Primary Outcome
Conditional probabilities for independent events were used to estimate the primary outcome.
Where P(A) is the probability that the first BC set is not positive after 24 hours of incubation; P(B) is the probability that an additional BC becomes positive within 5 days of incubation. P(B|A) is the probability that event B happens when event A has occurred.

Data Extraction and Processing
Data were extracted from the EHR to create a specific study database and then processed to filter out non-relevant data.

EHR Review and Screening Process
We screened our final table for potential BC contaminants whenever culture reported one of the following microorganisms: Corynebacterium spp, Cutibacterium acnes and coagulasenegative staphylococci. Each care episode reporting at least one of these microorganisms prompted an EHR chart review. The contaminant was then confirmed and flagged whenever it was clearly documented in the EHR, either by the infectious diseases' consultant or the physician in charge of the patient.

Statistical Analysis
We used the Kruskal-Wallis one-way analysis of variance for numeric variables and the Chisquared test for categorical variables; p-values are shown with a maximum of 3 decimals and the level for statistical significance was set at P≤.05 (two-sided), if not specified otherwise.
Due to the non-negligible probability to get a type I error caused by the multiple comparison in subgroup analyses for the performance assay, we used P≤.005 as the cut-off for significance by applying the Bonferroni correction. McNemar's test was used to assess the statistical significance in sensitivity and specificity changes as depicted by Fagan's nomogram. All data and statistical analyses were performed using R software version 4.1.0 (2021-05-18) and the following extension packages: dplyr, table1, bdpv, DTComPair, caret, ggplot2, pROC, epiDisplay and MASS in their latest available version. the accuracy of our model, we performed a cross-validation by training it using 80% of the data and kept the remaining 20% to assess its performance. The best fitting model (AIC=391.5) included: mTTP; source of sampling (i.e., intravenous/intra-arterial line or peripheral venous punctures); gender; transplantation status; direct examination under the microscope for microorganisms (e.g., morphology, Gram staining or identification as a yeast);

Multivariate Logistic Regression, Adjusted Odds Ratios and ROC Curve
and culture type (aerobic, anaerobic, Lytic MycoF). Adjusted odds ratios (aOR) were calculated with the corresponding 95% CI.