This study was approved by the Ethics Committee of Guangzhou Women and Children’s Medical Center (2017121406), and the trial was registered prior to patient enrollment at the Chinese Clinical Trial Registry (http://chictr.org.cn,ChiCTR1800015340, principal investigator: Bi Lian Li, date of registration: 2018/03/24). Written informed consent was obtained from all patients’ legal guardians or parents before the study.

Children undergoing interventional therapy, aged between 3 and 72 months, requiring general anesthesia were enrolled in this clinical trial. The inclusion criteria included adequate hematological, hepatic, and renal functions (National Cancer Institute, 2017); no consumption of DEX or any other sedative within a week before surgery; and classification as class II and III on the American Society of Anesthesiologists (ASA) scale. The exclusion criteria were as follows: 1) patients with neurological disorders; 2) organ dysfunction; 3) significant developmental delays or behavior problems; 4) allergy or hypersensitive reaction to DEX; 5) current treatment with α-adrenergic, β-adrenergic agonists, antagonists, or enzyme inducers or enzyme inhibitors of DEX; 6) the presence of active respiratory symptoms or rhinorrhea that might influence nasal drug absorption.

DEX at 3 mcg·kg⁻¹ (Ai Bei Ning; Jiang Su Heng Rui Medicine Co. Ltd.) was administered via atomization using the Mucosal Atomization Device (Teleflex MAD Nasal; Research Triangle Park) 30–40 min before propofol induction. The target UMSS levels (University of Michigan Sedation Scale, UMSS, Supplementary Table S1) (Malviya et al., 2002) were 2–4 after DEX initiation before propofol induction. The time interval between dexmedetomidine administration and propofol induction was recorded. Subsequently, all patients received 2 mg·kg⁻¹ propofol, 0.3 mcg·kg⁻¹ sufentanil, and 0.2 mg·kg⁻¹ cisatracurium besylate and had laryngeal mask airway placement or tracheal intubation. General anesthesia was maintained with sevoflurane inhalation (0.5%–3%). Medications administered before and during the study period were recorded. Significant signs (changeover 20% from the baseline values) were recorded every 15 min after the completion of the operation until the patient was awake and ready to be discharged.

We hypothesized that the individual variations in DEX-induced hemodynamic instability could be partially explained by clinical characteristics, DEX pharmacokinetics, and pharmacogenetics. The primary outcome was hemodynamic instability, defined as a composite of hypotension and/or bradycardia, with hypotension being a 20% reduction from age-related baseline values in systolic or diastolic blood pressure and bradycardia being a 20% reduction from age-related baseline values in heart rate (Yuen et al., 2008). In cases of hypotension or bradycardia during the procedure, intravenous epinephrine (40–100 μg) or dopamine (<3 μg kg·min⁻¹) was administered. Successful sedation was defined as achieving a UMSS score of 2–4 within 30 min, as assessed by an experienced pediatric anesthesiologist. Hypoxia was defined as SpO₂ ≤ 93% or ≥5% decrease from baseline. These identified events were used to assess the association between hemodynamic instability and the collected factors.

Two milliliters of blood were collected at baseline and 60 min after DEX administration from each patient. Plasma DEX was measured by a validated high-performance liquid chromatography/tandem mass spectrometry (LC–MS/MS) method (Guan et al., 2019). In brief, sample preparation was performed by liquid–liquid extraction. Plasma aliquots of 100 μL were mixed with a stable isotope-labeled internal standard (dexmedetomidine-D₃) and extracted with 500 μL ethyl acetate. After 3 min of vortex mixing and 10 min of centrifugation (4°C, 15,000 rpm), the supernatant was evaporated to dryness and the residues were reconstituted in the mobile phase. Isocratic HPLC separation was performed with an ACQUITY BEH C18 column (2.1 mm × 50 mm, 1.7 μm particle size; Waters, United States) and a gradient mobile phase consisting of acetonitrile and 1‰ formic acid in water (flow rate 0.3 mL·min⁻¹) at 40°C. Mass spectrometric detection was carried out with a TSQ Quantum triple-quadrupole instrument using a positive selected reaction mode. The precursor–fragment ion pairs detected were m/z 201.3 → 95.1 for DEX and m/z 204.2 →98.0 for the internal standard. Mass parameters were 4,000 V ion spray voltage, 20 psi sheath gas pressure, 280°C vaporizer temperature, 5 psi auxiliary gas pressure, and 350°C capillary temperature. The lower limit of reliable quantitation of the method was 0.05 ng·mL⁻¹, and the linear range was 0.05–10 ng·mL⁻¹ with a correlation coefficient ≥0.99. The within- and between-run accuracy and precision of the bioassay (coefficient of variation) was within ±15% in the relevant concentration range. No significant carry-over and matrix effects were observed.

The pharmacokinetic parameters for each individual were estimated using a Bayesian approach and the post hoc analysis. The pharmacokinetic parameters and residual variability were assessed based on a previous pharmacokinetic study of intranasal DEX (Li et al., 2022). The population pharmacokinetic model served as a priori information for Bayesian forecasting, utilizing a limited sampling strategy. The apparent clearance (CL/F, Q/F) and volume of distribution (V/F, V₂/F) for the population pharmacokinetic model are described in Equations 1–4: C L / F = 53.0   ∗   W e i g h t 70 0.75 ∗ P M A 3.0 P M A 3.0 + 44.7 3.0 , (1) V / F = 57.5 ∗ W e i g h t 70 , (2) Q / F = 243.8 ∗ W e i g h t 70 0.75 , (3) V 2 / F = 71.8 ∗ W e i g h t 70 , (4) where CL/F is apparent clearance of the central compartment, V/F is the central volume of distribution, Q/F is apparent clearance of the peripheral compartment, V₂/F is the peripheral volume of distribution, weight is the patient’s body weight, and PMA is the postmenstrual age.

The individual plasma concentrations–time profiles were estimated to calculate pharmacokinetic parameters using the population pharmacokinetic model, incorporating allometrically scaled parameters and a limited number of plasma concentrations (Guan et al., 2019; Li et al., 2022).

Genomic DNA was extracted from peripheral blood cells by the phenol–chloroform extraction method, as described by Blin and Stafford (1976). The quality of the extracted DNA was assessed by measuring the absorbance ratio at A₂₆₀/A₂₈₀, with values ranging from 1.8 to 2.0 being considered acceptable. The concentrations of extracted DNA were detected using a NanoDrop 2000 Spectrophotometer (Thermo Fisher Scientific, Rockford, IL, United States).

The primary outcomes of interest were SNPs in the genomic DNA, which were measured using DNAs extracted from blood cells in the Agena Sequenom MassARRAY Analyzer 4 system (MALDI-TOF platform). In brief, the regions of the genome containing each SNP in extracted DNA samples (20 ng·mL-1) were amplified by a PCR amplifier (Mastercycler^® nexus, Eppendorf, Hamburg, Germany), and then, an extension PCR reaction was performed in a 384-well plate (Thermo Fisher Scientific, Rockford, IL, United States). After that, a single terminator nucleotide base extends the DNA fragment. The terminator bases were desalted and dispensed into a SpectroCHIP^® Array (Sequenom, San Diego, CA, United States). The detection of alleles was conducted using a MassARRAY mass spectrometer (Sequenom, San Diego, CA, United States). MassArray Typer 4.0 software (Sequenom, San Diego, CA, United States) was used to analyze the data. Inspection of the clusters was performed to ensure a clear cluster separation with a satisfactory signal-to-noise cut-off. SpectroChip data with a more than 10% call rate in the blank check or with a more than 25% call rate in the blank control were considered invalid and were repeated.

All SNPs associated with dexmedetomidine in this study were selected based on one or more of the following criteria: (1) minor allele frequency (MAF) greater than 0.05 for the Chinese population in the GRCH37 database (http://grch37.ensembl.org/index.html), (2) previously reported associations with DEX, and (3) previously reported functional effects. The candidate SNPs were genotyped using the Agena Sequenom MassARRAY platform (San Diego, CA, United States). The primers of each SNP are presented in Supplementary Table S2.

Continuous variables were presented as medians with interquartile range (IQR), while categorical variables were expressed as frequencies (%). The Hardy–Weinberg equilibrium for candidate SNPs was assessed using a χ² test, with a P-value < 0.05 indicating a deviation from equilibrium.

The cumulative incidence of hemodynamic instability was evaluated using the Kaplan–Meier method. Univariate Cox models were fit for those variables for the outcomes, and hazard ratios (HRs), 95% CI, and P values were computed for each variable using the cox.zph function of the survival package in R. Time at zero was recognized as the initiation of DEX administration, and the last follow-up time was defined as the completion of the operation.

Least absolute shrinkage and selection operator (LASSO) regression with 100-repeated fivefold cross-validation was applied to identify the most predictive factors for DEX-associated hemodynamic instability using the R package glmnet. A total of 270 patients were randomly split into the training (60%) and test cohorts (40%). Initially, the lambda sequence was computed, followed by model fitting for each fold omission. The optimal lambda value was determined based on the standard deviation and average error across folds. The elasticnet mixed parameters (0≤α ≤ 1) were used, where α = 1 for the LASSO penalty and α = 0 for the Ridge penalty. After 10,000 iterations of the LASSO model, the frequency of variables was calculated. Variables were ranked based on the frequency, and these were progressively added to the test cohort. The process continued until the area under the receiver operating characteristic curve (AuROC) in the test cohort showed no further significant improvement.

A score-based nomogram model was developed based on multivariate Cox regression analysis. Nomogram is a statistical model that provides individualized risk assessment. The calibration and discrimination performances of the nomogram model were evaluated alongside accuracy, specificity, and sensitivity using the receiver operating characteristic (ROC) curve. The optimal threshold prediction value of the multivariate regression model was determined using the maximum Youden’s index.

Statistical analyses and data management were conducted using R version 4.2.2 (R Foundation for Statistical Computing, Vienna, Austria). Pharmacokinetic estimation was conducted using Phoenix NLME (Version 7.0, Certara L.P. Pharsight), and the non-compartment model was calculated in the WinNonlin mode. Graphs were plotted using GraphPad Prism 8 (GraphPad Software Incorporated, San Diego, CA, United States).

In total, 285 potentially eligible pediatric patients received DEX pre-operatively during the study timeframe, and 270 patients met the inclusion criteria. There were 15 dropouts because of previous exposure to DEX (n = 2) or chloral hydrate (n = 3), previous enrollment (n = 2), and nasal symptom (n = 2). The patients’ flow diagram is shown in Figure 1. Data from 270 pediatric participants (median (IQR) aged 22 (8–42) months (146 female and 124 male individuals) are presented in Table 1.

In this study, hypotension and bradycardia were commonly observed but were not life-threatening side effects, and only four patients required interventions to treat hypotension/bradycardia. One patient exhibited oxygen saturation below 93% at 45 min post-induction of anesthesia. Using the Kaplan–Meier analysis, the cumulative incidence of hemodynamic instability within 90 min after DEX administration was estimated to be 16.3% at 90 min (Figure 2). Among the enrolled patients, at least one episode of hemodynamic instability occurred in 52 patients (209 events). After achieving the target UMSS at sedation onset time [median (IQR): 16 min (12–22 min)] after intranasal DEX administration, they experienced induction anesthesia and received fixed doses of propofol (2 mg·kg⁻¹), sufentanil (0.3 mcg·kg⁻¹), and cisatracurium besylate (0.2 mg·kg⁻¹). The first hemodynamic instability commonly occurred at the median time of 20 min (15–30 min) after DEX administration, and approximately 14.4% (39 out of 270) of children experienced it after the initiation of anesthesia induction.

Plasma samples from 270 pediatric patients were collected at 60 min after DEX administration. The median concentration level at 60 min after DEX initiation was 0.429 ng·mL⁻¹ (IQR: 0.332–0.523 ng·mL⁻¹). Median plasma concentrations at different time points (0–6 h) along with fitted PK profiles were simulated using a validated population pharmacokinetic model (shown in Supplementary Figure S1). All the model-derived pharmacokinetic parameters are shown in Table 2.

Subgroup analysis on patients with hypotension or bradycardia comparing patients without events showed a significant difference in DEX clearance after 35-min DEX initiation [hypotension vs. normotension: 0.76 vs. 0.98 (P = 0.029); bradycardia vs. normal: 0.74 vs. 0.95 (P = 0.0078)] (Table 2).

All allelic distributions for the candidate SNPs conformed to the Hardy–Weinberg equilibrium (p > 0.05). Among the candidate SNPs, univariate analysis showed that 21 of the remaining 134 tag SNPs could serve as significant predictors for DEX-induced hemodynamic instability (FDR <0.05) and were used for the next multivariable Cox analysis. Of the clinical variables tested, sex (p < 0.001), age (p < 0.001), weight (p < 0.001), height (p < 0.001), premature birth (p = 0.006), creatinine (p < 0.001), ALT (p < 0.001), cisatracurium besylate (p = 0.04), and sevoflurane (p = 0.002) had significant correlations with DEX-induced hemodynamic instability. All the concomitant medications were recorded and analyzed in the univariate analysis, and only concomitant propofol (p < 0.001) had a significant impact on DEX-associated hemodynamic instability. In the further subgroup analysis, patients with concomitant sevoflurane were divided into three dosage groups (<1.0, 1.0–2.0, and >2.0%), and the risk of hemodynamic instability increased with the dosage of sevoflurane (p = 0.004). Of the pharmacokinetic variables tested, peak time (p < 0.001) and DEX clearance (p < 0.001) demonstrated significant associations with outcomes. The results of univariate analysis are shown in Figure 3 and Table 3.

The complete dataset was randomly split into training and validation datasets. The training dataset comprised 60% of the total patients, and the remainder constituted the test cohort. LASSO regression was used to evaluate the frequencies of variables from all the candidate factors. The changes in areas under the curve (AUCs) after adding the variables and the average AuROC values for the training and test datasets over 100 replications are shown in Figure 4.

Multivariable Cox models were analyzed, and the following variables were retained in the model: weight, dexmedetomidine clearance, concomitant propofol at induction anesthesia, UGT2B10 rs1841042, CYP2A6 rs8192733, ADRA2B rs3813662, CACNA2D2 rs2236957, NR1I2 rs3814057, and CACNB2 rs10764319. In the training cohort, the AUC values for the training and the test cohorts were 0.881 and 0.762, respectively (Figures 4C, D). Figure 5 shows the nomogram based on the above predictors containing 270 pediatric patients. The calibration curves demonstrated an optimal consistency between the nomogram-predicted and the actual observed risk probability (Supplementary Figure S2).