Randomization in clinical trials of titrated therapies: Unintended consequences of using fixed treatment protocols*

Abstract

Clinical trial designs that randomize patients to fixed treatment regimens may disrupt preexisting relationships between illness severity and level of therapy. The practice misalignments created by such designs may have unintended effects on trial results and safety. To illustrate this problem, the Transfusion Requirements in Critical Care (TRICC) trial and the Acute Respiratory Distress Syndrome Network low tidal volume (ARMA) trial were analyzed. Publications before TRICC indicated that clinicians used higher transfusion thresholds in patients with ischemic heart disease compared with younger, healthier patients (p = .001). The trial, however, randomized patients (n = 838) to liberal (10 g/dL hemoglobin) or restrictive (7 g/dL) transfusion thresholds. Thirty-day mortality was different and opposite in the liberal compared with the restrictive arm depending on presence (21 vs. 26%) or absence (25 vs. 16%) of ischemic heart disease (p = .03). At baseline in ARMA, consistent with prior publications, physicians set ventilator volumes lower in patients with high airway pressures and poor compliance (8.4-10.6 mL/kg interquartile range) than patients with less severe abnormalities (9.6-12 mL/kg) (p = .0001). In the trial, however, patients (n = 861) were randomized to low (6 mL/kg) or high (12 mL/kg) tidal volumes. In patients with low compliance (<0.6 mL/kg), 28-day mortality was higher when tidal volumes were raised rather than lowered (42 vs. 29%), but this effect was reversed in patients with higher compliance (21 vs. 37%; p = .003). In TRICC and ARMA, randomization to fixed treatment regimens disrupted preexisting relationships between illness severity and therapy level. This created noncomparable subgroups in both study arms that received care different and opposite from titrated care, that is, practice misalignments. These subgroups reduced the interpretability and safety of each trial. Characterizing current practice, incorporating current practice controls, and using alternative trial designs to minimize practice misalignments should improve trial safety and interpretability.