179 Clinical sub-phenotypes of Staphylococcus aureus bacteraemia 8 Edinburgh cohort (n=458) ARREST cohort (n=758) SAFO cohort (n=214) P-value Skin or soft tissue infection 88 (19·2) 293 (38·7) 39 (18·2) Other 61 (13·3)6 55 (7·3)7 20 (9·3) Respiratory 28 (6·1) 29 (3·8) 4 (1·9) Urine 25 (5·5) 19 (2·5) 13 (6·1) All-cause 84-day mortality 121 (26·4) 56/388 (14·4)8 17/110 (15·5)8 <0.0001 Table 1: Characteristics of included patients. 1Not available due to participant deidentification. 2people with Child Pugh C liver cirrhosis were excluded from the SAFO trial. 3peripheral vascular disease, myocardial infarction or stroke. 4implanted cardiac devices, including pacemakers and implantable automatic cardioverter-defibrillator and Left Ventricular Assist Devices, but not including prosthetic heart valves. 5vertebral osteomyelitis, epidural abscess, native joint septic arthritis, prosthetic joint infection, deep tissue abscess. 6injection drug use and bone classified as ‘other’. 7‘other’ sources not specified. 8data shown for trial control arms. Continuous values are shown as median (interquartile range). Categorical variables are shown as count (%). Variables not available in the ARREST dataset are represented as NA. Vital signs and laboratory measurements were recorded at the time of the index blood culture in the Edinburgh and SAFO cohorts. In the ARREST trial, baseline laboratory measurements were those closest to randomisation (preceding 4 days or 1 day post randomisation) and for vital signs, the highest value within 24h of randomisation was taken. MRSA: methicillin-resistant Staphylococcus aureus Identification of sub-phenotypes using latent class analysis Eighteen class-defining variables were included in the final latent class analysis. Despite co-linearity, both creatinine and chronic kidney disease were included because creatinine provides additional information on the presence of acute kidney injury (correlation coefficient 0·66). After determination of contributing variables (Supplementary Figure 2) using the Edinburgh cohort, latent class models with one to seven classes were fitted (Table 2). For the Edinburgh and ARREST cohorts, the BIC and clinical interpretability favoured the five-class model and the size of the smallest class was acceptable (>5% of total population).
RkJQdWJsaXNoZXIy MTk4NDMw