Every test should be built with a physician end-user in mind. Our team is physician-led, ensuring that development starts with the question of whether a test could be clinically actionable, and whether it fits into workflow. Our test for sepsis was built to answer a critical clinical gap: physicians usually guess about whether or not a patient needs antibiotics. Current pathogen-focused diagnostic methods fail to answer this question because they frequently miss infections that haven’t spread to the bloodstream. Thus, we focused on using the immune response to infections to answer to key clinical action items in the evaluation of an acutely ill patient: (1) does the patient need antibiotics, and (2) what level of care is required? These questions are best answered by ‘reading’ the immune response, rather than looking directly for a pathogen as a ‘needle in a haystack’.
Embrace “dirty data” in selecting and analyzing cohorts to study. Acute infections can be remarkably clinically diverse: kids and adults, inpatients and outpatients, in different settings around the world. We wanted to build a test that would work in every environment. We thus embrace this clinical heterogeneity by analyzing multiple cohorts that are diverse in their population, sample types, assays used and other factors. Our robust statistical pipeline allows us to find reproducible signal in the ‘noise’ of multiple datasets. Although this is a hard challenge, our discovery methods ensure that the diagnostics we find are generalizable to new populations.
Validate on multiple, independent and diverse cohorts. Trustworthy tests must demonstrate robust performance in independent, blinded multi-center studies.