Embracing Enrichment and Unknown Unknowns*

    loading  Checking for direct PDF access through Ovid

Excerpt

In clinical medicine and research, “enrichment” refers the selection of patients in whom an intervention is more likely to be beneficial when compared with unselected patients (1). “Prognostic” enrichment selects patients who are more likely to have a disease-related event, such as mortality. “Predictive” enrichment selects patients who are more likely to respond to a therapy based on a biological mechanism. Virtually, any type of patient-derived data, including biomarkers, have the potential to inform an enrichment strategy. By accounting for and measuring disease heterogeneity, enrichment strategies can enhance the efficiency of clinical trials and are fundamental to precision medicine.
In this issue of Critical Care Medicine, Meyer et al (2) report a post hoc, biomarker-driven analysis of a phase II trial testing the efficacy of recombinant human interleukin-1 receptor antagonist (rhIL1RA) among patients with sepsis. rhIL1RA is the endogenous antagonist of the canonical proinflammatory cytokine, interleukin-1β (IL-1β). In the original trial, rhIL1RA did not confer a survival benefit compared with placebo despite strong preclinical and clinical evidence supporting the biological plausibility of inhibiting IL-1β as a novel therapy for sepsis (3). In the current study, Meyer et al (2) report that among subjects having a pre-enrollment IL1RA concentration greater than 2,071 pg/mL, but not among those with a concentration less than that cut point, treatment with rhIL1RA was associated with a significant reduction of all-cause 28-day mortality. In essence, the investigators used pre-enrollment IL1RA data as a predictive enrichment strategy to identify a subset of patients more likely to benefit from rhIL1RA. The implications of these findings are obvious.
Also obvious are the many limitations inherent to a biomarker-based, post hoc analysis of a trial that occurred over 20 years ago. Many of these are well recognized and acknowledged by the investigators. These limitations notwithstanding, the study is of importance to the field of critical care medicine because it again reinforces the concept that clinicians and researchers in the field should strive to develop enrichment strategies as a means of managing the heterogeneity of our patients and consequently improve clinical care and research. Analogous post hoc studies recently reported enrichment strategies to identify children with septic shock who are more likely to benefit from adjunctive corticosteroids and adults with sepsis who are more likely to benefit from an antitumor necrosis factor-α strategy (4, 5).
The study by Meyer et al (2) also provides an opportunity to consider the many challenges inherent to the development and validation of effective enrichment strategies for critically ill patients. Based on existing data, the authors reasoned that perturbations of the IL-1β/IL1RA axis might identify patients with differential responses to rhIL1RA therapy. This reflects a knowledge-based approach because it is founded on existing knowledge.
Although valid because it relies on the well-accepted scientific method, a knowledge-based approach is also limited to existing knowledge and can therefore fail to consider otherwise unknown possibilities (“unknown unknowns”). As an example, elegant work by the same group of investigators previously reported that an IL1RA gene variant that leads to more efficient expression of IL1RA is associated with lower mortality from sepsis (6). It may therefore be counterintuitive that subjects with higher pre-enrollment IL1RA concentrations seemed to derive the most benefit from rhIL1RA. The investigators posit that this lack of concordance with previous data reflects the complex nuances of the IL1β/IL1RA axis. This is entirely plausible, but requires validation, and is perhaps an example of a previously “unknown unknown.”
The predictive enrichment strategy proposed by Meyer et al (2) is attractive in its relative simplicity.
    loading  Loading Related Articles