185 Clinical sub-phenotypes of Staphylococcus aureus bacteraemia 8 Hendriks et al definition of low-risk SAB30, with the majority predicted to belong to sub-phenotype B. Sub-phenotype B could therefore represent a rational target for expanded investigation of earlier oral switch in SAB, providing a data-driven definition of ‘low-risk’ SAB. Furthermore, the risks of adjunctive agents might outweigh the limited potential to improve on already good outcomes, as exemplified by our finding that adjunctive rifampicin potentially caused increased mortality in this subphenotype. Inclusion of this sub-phenotype in trials of combination therapy should be done cautiously. Sub-phenotype E (IDU SAB) was associated with complicated disease but despite this, low mortality. Sub-phenotype C (community-acquired metastatic SAB) had complicated disease and worse microbiologic outcomes, but without clear predisposing factors. Patients in this sub-phenotype had a lower Charlson Comorbidity Index and generally lacked an obvious source for bacteraemia. The possible benefit of adjunctive rifampicin warrants further investigation in this subphenotype, in addition to alternative adjunctive agents including antimicrobials (e.g. clindamycin, currently being evaluated in the adjunctive treatment domain of the SNAP trial31) and anti-staphylococcal lysins (Exebacase)32. These sub-phenotypes also provide a framework for investigation of immunobiology in SAB, and could facilitate identification of treatable traits, for example defective phagocyte responses that could be therapeutically re-calibrated11. Our study has several strengths. The sub-phenotypes were replicated in analysis of an observational cohort and a large trial cohort with permissive inclusion criteria. Four of the sub-phenotypes could also be identified in a smaller trial with more restrictive inclusion criteria. Trial populations of SAB differ from real-life cohorts, including patient characteristics and mortality rates9,29. It is therefore re-assuring that despite the differences between the cohorts (Table 1), the core features of the identified subphenotypes were reproducible, suggesting the findings are generalisable. Outcomes differed between sub-phenotypes but were not included as class-defining variables, and overall the association between sub-phenotype and outcome was consistent across the cohorts. To allow prospective sub-phenotype prediction of individual patients, future work will aim to identify a sub-set of variables that can be used as predictive markers of sub-phenotype membership. Our study has important limitations. First, despite using model parameters such as BIC there is a degree of subjectivity with the class selection based on clinical interpretability. Second, the class-defining variables included were restricted to routinely available clinical data. Inclusion of inflammation biomarkers could provide biological insights. Third, the included cohorts were from countries with a low prevalence of MRSA. The USA300 MRSA clone is prevalent in the USA and independently associated with metastatic disease3. Replication in a cohort with higher MRSA prevalence will be required. Fourth, the cohorts differed in inclusion criteria and the variables available for analysis. Fifth, the definitions of microbiologic outcomes used in the cohorts were different, preventing direct comparison. Sixth, the Edinburgh cohort was a retrospective observational study, without structured prospective monitoring of microbiologic outcomes. Detection of persistent or recurrent SAB was opportunistic, relying on healthcare attendance and blood cultures
RkJQdWJsaXNoZXIy MTk4NDMw