Outcome Definition and Contaminant Identification

We defined a blood culture “set” as a single culture acquisition, typically an inoculation of two blood culture bottles. We grouped all sets into blood culture “instances,” defined as a sequence of sets from a single patient where each set was within 1 hour of the last. We categorized each instance as positive, negative, or contaminated. To do this, we identified all the unique organisms that were isolated and labeled them as either potential contaminants or true pathogens. We labeled all instances that grew true pathogens as positive. Organisms deemed potential contaminants included coagulase-negative staphylococci, Propionibacterium species, Bacillus species, Corynebacterium species, and viridans group streptococci. For these, we labeled the organism as a contaminant if it was present in only one set of a culture instance (15) or in only one of the two bottles in an instance comprised of a single set. We adjudicated instances with potential contaminants that did not meet these criteria (i.e., multiple sets grew a potential contaminant) by an independent review of the medical record by two separate clinicians who were blinded to each other’s decisions (A.N.Z. and C.C.M.). A third clinician (T.D.B.) adjudicated discordant decisions, also in a blinded fashion. We labeled instances as contaminated if a contaminant was the only growth. We labeled instances with both true pathogens and contaminants as positive. We labeled instances with growth of multiple true pathogens as a single positive instance. For individual isolate modeling, these polymicrobial culture instances were labeled as positive for all true pathogen isolate classes that were grown (i.e., an instance with growth of Staphylococcus aureus and Escherichia coli was labeled as both GP and GN).