Fig. 1

Flow chart of antimicrobial prescription filtering and linking to diagnostic microbiology. Of 840 prescriptions collected by pharmacists, samples were excluded if they were prophylactic (n = 11; breakdown: 7 peri-operative or otherwise surgical and 4 medical), topical (n = 21), or lacking unique patient hospital identification numbers, IDs (n = 426). The 382 remaining prescriptions were grouped into 276 “Prescription Episodes” (PEs) of antimicrobials given for the same patient illness (see main text). Contemporaneous diagnostic microbiology from blood, urine, sputum or CSF were linked to prescriptions with matching hospital IDs. Contaminants (as initially flagged up in the microbiologist comment on the result, then confirmed on case review by authors WLH and SP) were transferred from “positive” to “negative” and a single CSF sample was moved from “negative” to “positive” due to elevated white cell count without any microorganism identified. All PEs with positive pathogenic microbiology (n = 38) and a randomised sample of the 124 PEs with negative microbiology (n = 44) were reviewed, with 9 PEs removed at this stage due to the linked microbiology sample being deemed irrelevant to the antimicrobial prescription. This left a final set for detailed case review of 33 PEs with positive microbiology and 40 with negative microbiology