Optimal dose of intra‐operative dexmedetomidine for postoperative delirium prevention: a reply

We thank Guo and Liu for their comments [1] regarding our study on the dose-dependent association between dexmedetomidine and postoperative delirium after non-cardiac surgery [2]. We appreciate the opportunity to discuss important methodological considerations related to confounding, exposure and outcome specification and to highlight the strengths of our study. Guo and Liu suggest there are imbalances between exposure groups regarding surgical duration, complexity and anaesthesia-related factors. They argue these could have biased the association between dexmedetomidine dose and delirium and raise concerns regarding confounding by indication [1]. We agree that an administration rate of 4% is consistent with a more specific as opposed to broad intra-operative administration of dexmedetomidine, reflecting real-world clinical practice [3]. To ensure these factors do not bias results, we included these variables in our multivariable model and, in addition, normalised dexmedetomidine dose to surgical duration in a confirmatory sensitivity analysis [2]. To account for potential changes in practice over the 16-year study period of our analysis, we included the year of the procedure as a primary confounder in the model. Guo and Liu further raise the possibility of bias related to provider preference. To address this, we conducted an additional mixed-effects regression where we incorporated the individual anaesthesia provider as a random effect, which yielded robust results. This was an adjusted OR 0.66 (95%CI 0.47–0.93) (p = 0.017) for lower vs. adjusted OR 1.08 (95%CI 0.85–1.37) (p = 0.550) for higher doses of dexmedetomidine, each compared with no administration. While we acknowledge that residual confounding cannot be ruled out in a retrospective analysis such as ours, it is unlikely to explain fully the non-linear association observed across multiple modelling approaches and sensitivity analyses. Guo and Liu raise the concern that the cumulative dose used to derive the primary exposure variable is inherently entangled with procedure duration and administration route. We addressed this issue by ascertaining the exposure using duration-normalised dosing (μg.kg-1.min-1), subgroups of administration modalities and rate-based thresholds derived from randomised trials, which all yielded robust findings [2]. It is true that the boundaries of the proposed range depend on the statistical model. However, the consistent identification of an optimum range across multiple exposure definitions and across fractional polynomial models, restricted cubic splines and threshold regression, heavily supports a non-linear dose–response relationship. We acknowledge fully that the pharmacokinetic and pharmacodynamic heterogeneity across different dosing strategies and regimens cannot be addressed fully in a single study and future prospective trials are now warranted to refine the optimal dosing range further to inform clinical decision-making and to establish potential context-dependency. Guo and Liu note that delirium identification leveraging a combination of chart-based ascertainment, Confusion Assessment Method assessments and diagnostic codes [2] may under-detect hypoactive delirium. While it is possible that a subset of patients suffering from (mainly hypoactive) delirium might not have been identified, the incidence of delirium has been shown to vary substantially even among randomised controlled trials [4, 5]. A sensitivity analysis restricted to patients exposed to Confusion Assessment Method assessments in the ICU, mimicking these conditions with a delirium incidence of 15%, confirmed our findings. Lastly, any delirium misclassification is unlikely to differ by exposure groups and might therefore bias results toward the null rather than generating a spurious association.