The authors reply
We believe that there is a need for generalizable (not context-specific) biomarkers in sepsis for two reasons. First, the clinical question that underlies the FAIM3-to-PLAC8 ratio or any other sepsis biomarker is largely whether a critically ill patient needs antibiotics—and here generalizability is key to a successful tool. If we use reductio ad absurdum, the context specificity that Scicluna and van der Poll (1) champion would lead to separate tests for pneumonia, skin and soft tissue infections, urinary tract infections, abdominal infections, etc—and then separate versions of each test for the emergency department, ward, and ICU. Validating and translating dozens of separate biomarkers would, of course, be unfeasible and unhelpful. Instead, what is needed is a noncontext-specific, highly generalizable diagnostic that can discriminate whether patients with systemic inflammation have bacterial infections (4, 5). Our article conclusively shows that the FAIM3-to-PLAC8 ratio is not specific to the clinical question of community-acquired pneumonia versus noninfectious respiratory distress. It is clear from the data that the FAIM3-to-PLAC8 ratio is strongly induced for all of the types of infections represented in the 39 datasets and 2,604 patients included in our study (2). Thus, we have little confidence that a patient with an extrapulmonary source of infection concomitant with respiratory systems would not be misclassified as “pneumonia” in the narrow binary context of pneumonia versus noninfectious respiratory distress. But is that not a benefit? The critical need for early antibiotics in patients with sepsis from any source (6) suggests that a generalizable biomarker for bacterial infection would fit best into clinical workflows.
Second, it is of course true that a targeted study of gene expression is better than a microarray study; however, high-fidelity gene expression studies done for validation of a diagnostic in a population not significantly different from the discovery population (i.e., same inclusion/exclusion criteria, same enrolling site) are not a guarantee of generalizability. Performing multiple trials across a broad range of patient types from different centers is the only way to ensure the generalizability of a diagnostic test. Leveraging the existing clinical microarray data is an excellent way to rapidly test a diagnostic across multiple patient groups. There may even be an ethical argument that not to do so would risk harm to new trial participants in “reinventing the wheel” by not making full use of the contributions of existing trial participants (7). In any case, we find it hard to understand any argument against widespread testing of proposed interventions for a problem as difficult and urgent as improving sepsis care.