Table 3 Designing clinical trials to address heterogeneity within ARDS
From: Promises and challenges of personalized medicine to guide ARDS therapy
Strategy | Type of heterogeneity | Specifics | Pros | Cons | Examples |
|---|---|---|---|---|---|
Subgroup analysis | Any | Pre-specify subgroups for analysis on completion of traditional RCT | Acknowledges uncertainty about best matching of treatment to subgroup/phenotype | Inefficient; too many subgroups may result in false positives | Liu et al., activated protein C in ARDS [103] |
Prognostic enrichment | Severity | Restrict enrollment to patients with more severe ARDS (lower PaO2/FiO2 ratio) | Likely enhances ability to detect treatment response, as relative risk reduction translates into higher absolute risk reduction if mortality is high | Reduces generalizability; may miss benefit in milder ARDS | PROSEVA trial [104] |
Predictive enrichment | Biologic, physiologic, radiographic | Restrict enrollment to patients with specific abnormalities targeted by chosen therapy (e.g., inflammation of a certain level, for an anti-inflammatory therapy) | May identify treatment-responsive subset by better matching therapy with phenotype | Reduces generalizability; requires either understanding of or assumptions about best way to personalize treatment; no proof of “non-response” in excluded patients | RECOVERY tocilizumab trial [45] |
Explicit comparison of personalized versus non-personalized therapy | Any | Randomize patients to personalized arm (with specific therapies based on subgroup/phenotype) vs standard-of-care arm | Explicit test of whether personalized strategy improves outcomes; tests effectiveness as well as efficacy to some degree | Complexity of design; misclassification may bias toward null; requires either understanding of or assumptions about best way to personalize treatment | LIVE trial [23] |
Adaptive design | Any | Pre-specify subgroups and stratify randomization; adjust target population or randomization based on interim analyses of subgroup-specific results | Acknowledges uncertainty about best matching of treatment to subgroup/phenotype but with greater efficiency than standard RCT; allows “learning as you go” | Complexity of design; more sophisticated analytic approaches may be needed | Bhatt and Mehta (review) [105] |