On the Robustness of Preoperative Intranasal Dexmedetomidine for Postoperative Delirium Prevention: Methodological Considerations [Letter] Authors: Hongyu Gai, Minggan Bao, Bo Chen Published: 2026 Dear editor We read with great interest the randomized trial by Liu et al1 investigating preoperative intranasal dexmedetomidine for postoperative delirium (POD) prevention in older adults undergoing joint arthroplasty. The authors should be commended for addressing a clinically important question in a vulnerable population. However, several methodological issues warrant closer examination, as they may affect the interpretation of the findings. Sample Size Calculation and Statistical Fragility The trial was powered based on an assumed 14% absolute risk reduction (23% vs 9%). The actual results showed a 10.4% reduction (20.7% vs 10.3%)—a smaller effect size than anticipated. With only 36 delirium events in total, the study may be underpowered for the observed effect magnitude. Recent literature has highlighted that many positive perioperative randomized trials are statistically fragile; we wonder whether the authors have calculated the fragility index for their primary outcome, which would help readers gauge the robustness of the P value of 0.030. Mediation Analysis Assumptions The authors report that 64% of the treatment effect on POD was mediated by prolonged preoperative sleep duration. However, the methods section explicitly states that the mediation model “did not adjust for potential mediator-outcome confounders or treatment-mediator interactions”. This is a critical limitation. Valid causal mediation analysis requires the sequential ignorability assumption—essentially, that there are no unmeasured confounders of the mediator-outcome relationship.2 In this perioperative context, numerous factors such as intraoperative hypotension, anesthesia duration, postoperative pain, and inflammatory response could confound the association between sleep and delirium. None of these were adjusted for in the mediation model. Furthermore, the proportion mediated had a confidence interval extending from 11.2% to 110.8%—a range that crosses 100%, indicating an unstable estimate. We would appreciate the authors’ perspective on whether a more comprehensive mediation model, adjusting for key perioperative confounders, would yield similar estimates. Safety Monitoring Duration The authors monitored patients for only 2 hours after intranasal administration, then relied on spontaneous reporting. In elderly patients, dexmedetomidine can cause profound and sustained hypotension. Given that the drug’s peak effect and the physiological nocturnal blood pressure dip both occur during the early morning hours (2:00–4:00 AM), we are concerned that clinically significant hypotension or bradycardia may have been missed. Large cohort studies have consistently shown that even brief episodes of intraoperative hypotension (MAP < 65 mmHg) are associated with myocardial injury and acute kidney injury.3,4 A longer monitoring period—or at least intermittent nocturnal checks—would have provided more reassurance regarding the safety profile in this age group. Additionally, the authors did not report whether any patients required unscheduled blood pressure measurements or experienced falls during the night, which are relevant safety endpoints. Sleep Measurement Precision The use of Fitbit Charge 2 to quantify sleep architecture in minute-level detail raises validity concerns. Consumer wristband devices have shown only modest agreement with polysomnography for sleep staging, with discrepancies often exceeding 15–20 minutes for REM and deep sleep.5 Reporting differences of 10–11 minutes in REM sleep as statistically significant may therefore reflect measurement error rather than true physiological differences. We acknowledge that wearable devices are practical for large trials, but the authors should discuss this limitation more explicitly and perhaps present sleep data as categorical or using minimal clinically important differences. Inconsistent Exclusion Criteria and Baseline Data The study excluded patients with “severe pre-existing sleep disorders or a history of routine use of hypnotic medications,” yet the baseline PSQI scores (median 11–12) indicate significant sleep disturbances in both groups. This suggests that many enrolled patients actually had chronic poor sleep, which contradicts the exclusion criterion. It would be helpful if the authors clarified how “severe” was defined and whether any enrolled patients met formal insomnia criteria. This inconsistency affects the generalizability of the sleep‑mediated mechanism. We fully acknowledge the value of this well-conducted trial and do not question the overall positive finding. Nonetheless, we believe that addressing these points—particularly the fragility of the primary outcome, the limitations of the mediation analysis, and the safety monitoring gap—would strengthen the manuscript and help readers interpret the results with appropriate caution.