Randomized trials of “personalized,” “individualized,” and “precision” interventions are very diverse and have low transparency and high bias

BACKGROUND AND OBJECTIVES: The terms "personalized," "individualized," and "precision" are increasingly used to describe interventions, yet their operational meaning is often unclear. Despite extensive conceptual debate, no empirical work has examined how these terms are applied in randomized controlled trials (RCTs) and how transparent and reliable the results of these trials are. METHODS: We surveyed MEDLINE for RCTs published between 2020 and 2022 that used the terms "personalized," "individualized," or "precision" in their title to describe an intervention. We collected data on trial characteristics and conclusions, intervention features, individual-level tailoring, transparency indicators, and risk of bias. Interventions were classified according to the number of personalized components, modes of tailoring, and dosage variability. RESULTS: A total of 262 RCTs were included. The term "personalized" was used most frequently (n = 129, 49.2%), followed by "individualized" (n = 120, 45.8%) and "precision" (n = 13, 5.0%). Most trials compared personalized interventions versus nonpersonalized control groups (n = 225, 85.9%) and focused on therapeutic applications (n = 186, 71.0%). Behavioral, digital, and medication-based interventions predominated. Personalization was most often based on lifestyle, psychological characteristics, or disease classification, whereas genetic and omics-based features were relatively uncommon. The majority of trials implemented a single intervention (n = 177, 67.6%) tailored differently for each participant (n = 217, 82.8%), often through individualized dosage or content (n = 109, 41.6%). Of the 221 RCTs comparing personalized interventions versus nonpersonalized controls, abstract conclusions were favorable to the personalized intervention in 156 (70.6%), mixed in 36 (16.3%), and unfavorable in 29 (13.1%). Few meaningful differences were observed across studies according to the term used. Transparency indicators such as data (n = 13, 5.0%) and code sharing (n = 1, 0.4%) were rare, and most trials (n = 162, 68.6%) were judged to be at high overall risk of bias. CONCLUSION: In contemporary RCTs, the labels "personalized," "individualized," and "precision" are applied largely interchangeably to a wide range of heterogeneous interventions that are predominantly nongenomic. Most of these trials have favorable conclusions, low transparency, and high risk of bias.