Medical progress is reflected in the advance from broad clinical syndromes to mechanistically coherent diagnoses. By this metric, research in sepsis is far behind other areas of medicine-the word itself conflates multiple different disease mechanisms, whilst excluding noninfectious syndromes (e.g., trauma, pancreatitis) with similar pathogenesis. New technologies, both for deep phenotyping and data analysis, offer the capability to define biological states with extreme depth. Progress is limited by a fundamental problem: observed groupings of patients lacking shared causal mechanisms are very poor predictors of response to treatment. Here, we discuss concrete steps to identify groups of patients reflecting archetypes of disease with shared underlying mechanisms of pathogenesis. Recent evidence demonstrates the role of causal inference from host genetics and randomised clinical trials to inform stratification analyses. Genetic studies can directly illuminate drug targets, but in addition they create a reservoir of statistical power that can be divided many times among potential patient subgroups to test for mechanistic coherence, accelerating discovery of modifiable mechanisms for testing in trials. Novel approaches, such as subgroup identification in-flight in clinical trials, will improve efficiency. Within the next decade, we expect ongoing large-scale collaborative projects to discover and test therapeutically relevant sepsis archetypes.
Keywords: Causal inference; Genetics; Machine learning; Sepsis; Stratification; Trials.
© 2024. Springer-Verlag GmbH Germany, part of Springer Nature.