Abstract
We aimed to explore the predictive role of radiomic features of body compositions in the occurrence of delayed awakening after flexible ureteroscopic lithotripsy, and further develop a predictive model to identify patients at higher risk. We analyzed the data from 314 admitted patients undergoing total intravenous general anesthesia at two institutions. Binary logistic regression was used to assess univariable and multivariable associations of clinical factors with delayed awakening. Radiomic features of body compositions were deeply extracted from preoperative computed tomography (CT) scans using ResNet50 model. We developed a combined model by integrating significant clinical variables with radiomic score. SHapley Additive explanation (SHAP) method was adopted for model interpretation and helping clinicians understand the predictive results. At multivariable analysis, CT-derived visceral adipose tissue area and skeletal muscle index were independent predictors of postoperative delayed awakening. The combined model had area under curves (AUCs) of 0.85(95%CI: 0.82, 0.89), 0.83(95%CI: 0.80, 0.87) and 0.80(95%CI: 0.76, 0.84) in the derivation, internal and external test datasets, respectively. 8 radiomic features were selected using Spearman rank correlation test and the least absolute shrinkage and selection operator logistic regression analysis. This combined model showed better discrimination and calibration than clinical model alone. Furthermore, SHAP plots quantified the impact of key variables on prediction at the global level. The combined model may allow estimation of the risk of postoperative delayed awakening after flexible ureteroscopic lithotripsy.
Supplementary Information
The online version contains supplementary material available at 10.1038/s41598-026-50706-9.
Keywords: delayed awakening, body composition analysis, deep learning, flexible ureteroscopic lithotripsy, clinical decision
Subject terms: Diseases, Medical research, Risk factors, Urology
Introduction
Delayed awakening is failure to restore consciousness following general anesthesia within 30–60 min and its incidence is about 9%1,2. It is a common complication following general anesthesia and represents a challenge for anesthesiologists. Postoperative delayed awakening was significantly associated with a prolonged Post-Anesthesia Care Unit stay3. About 10% patients with delayed awakening and agitation experienced unplanned Intensive Care Unit admission after general anesthesia. It may significantly increase the risks associated with prolonged intubation such as ventilator-associated pneumonia and airway trauma, particularly in patients with underlying respiratory disease4. Patients with delayed awakening presented with an altered mental state featuring lack of adequate response to stimuli. This abnormal awakening trajectory could lead to emergence agitation5–7. It showed close association with respiratory complications such as an increased risk of aspiration due to the lack of protective airway reflexes. Accurate prediction of delayed awakening before operation is important for clinicians to evaluate the reasonableness of the procedure and improve treatment options.
Postoperative delayed awakening is one of representative neurological adverse events after surgery8. To the best of our knowledge, recovery time from anesthesia was related to the doses of medications9–11. A patient with an increased body fat has to receive relatively higher doses to attain the same peak plasma concentration12. It was attributed to the redistribution of the lipid-soluble anesthetic agents in body fat. This indicated that body composition may play a role in determining the speed of recovery after general anesthesia13. In recent years, radiomics, an emerging research methodology, offers the potential for discovering novel imaging diagnostic and predictive biomarkers14. However, the predictive role of radiomic features of body compositions in the occurrence of delayed awakening remained unclear.
This study was conducted to investigate the relationships between body compositions and delayed awakening. The secondary endpoint was to develop an easy-to-use tool for identifying susceptible patients at high risk of developing delayed awakening in advance. This would help clinicians adjust anesthetic regimens accordingly. Notedly, flexible ureteroscopy is characterized by relatively fixed operative time, minimal surgical trauma and stable hemodynamics. The substantial homogeneity effectively minimized confounding influences from anesthesia, surgical trauma and procedural differences on postoperative delayed awakening, thus providing a key prerequisite for exploring the core predictive role of radiomic features of body compositions in our study. Additionally, most patients who underwent flexible ureteroscopy were elderly, with a high prevalence of metabolic abnormalities (e.g., visceral obesity, sarcopenia) and comorbidities. Reportedly, these patients are prone to postoperative delayed awakening in the operation of urinary surgery13,15,16. However, few tools were developed for predicting postoperative delayed awakening in these patients.
Results
Baseline characteristics
In this retrospective multicenter study, we initially enrolled 340 patients who underwent general anesthesia between January 2022 and December 2024. 12 patients were excluded (with the history of taking opioid drugs, n = 2; low quality CT images, n = 10) for the derivation and internal test datasets, and 14 patients were excluded (with the history of taking opioid drugs, n = 1; low quality CT images, n = 13) for the external test dataset. Finally, a total of 314 eligible patients were included in this study. 160 patients were randomly assigned into derivation dataset for model construction. 49 patients were assigned into internal test dataset and 105 patients for external dataset. The flowchart of study design is displayed in Fig. 1.
Fig. 1.
Flow diagram of study population.
The baseline features of study population are shown in Table 1. The incidences of postoperative delayed awakening were 13.1% (n = 21) in the derivation cohort, 14.2% (n = 7) in the internal test cohort, and 15.2% (n = 16) in the external test cohort, respectively. In the derivation dataset, the median age and body-mass index (BMI) were 62 ((interquartile range (IQR), 15) years and 22.3 (IQR, 3.2), respectively. In addition, we compared anesthetic drug dosage and anesthesia duration between patients with and without delayed awakening. We found no significant difference in drug doses and anesthesia duration between the two groups (Supplementary Table 1).
Table 1.
Clinical characteristics of study population.
| Characteristics | Derivation dataset n = 160 |
Test datasets | |
|---|---|---|---|
| internal n = 49 |
external n = 105 |
||
| Age, year a | 62 (15) | 60 (12) | 65(11) |
|
Gender b male |
74(46.3) | 23(46.9) | 41(39.1) |
| female | 86(53.7) | 26(53.1) | 64(60.9) |
| BMI, kg/m2 a | 22.3(3.2) | 23.1(3.5) | 23.4(3.6) |
| ASA grade b | |||
| I/II | 139(86.9) | 40(81.6) | 87(82.9) |
| III | 21(13.1) | 9(18.4) | 18(17.1) |
| aCCI score a | 2(2) | 2(2) | 3(2) |
| Surgery duration≥60 min b | 64(40) | 18(36.7) | 23(21.9) |
| Intraoperative strong opioids given b | 14(8.7) | 4(8.2) | 5(4.8) |
| Intraoperative hypothermia+b | 39(24.3) | 10(20.4) | 24(22.9) |
| Intraoperative Hypotension++b | 31(19.3) | 7(14.2) | 15(14.3) |
| delayed awakening b | 21(13.1) | 7(14.2) | 16(15.2) |
| Body composition a | |||
| VAT, cm2 | 82.3(30.2) | 88.2(32.1) | 85.9(42.5) |
| SAT, cm2 | 53.6(28.6) | 57.1(30.3) | 59.2(35.7) |
| SMI, cm2/m2 | 27.6(9.5) | 26.9(8.9) | 30.1(10.8) |
Note: a. Quantitative values are Median (IQR); b. Categorical variables are n (%).
+Hypothermia is defined as a core body temperature below 36°C. ++Intraoperative Hypotension is defined as adecrease in mean arterial pressure by 20%.
BMI, body mass index; ASA, American Society of Anesthesiologists score; aCCI, age-adjusted Charlson Comorbidity Index; VAT, visceral adipose tissue area; SAT, subcutaneous adipose tissue area; SMI, skeletal muscle index; IQR, interquartile range.
Quantitative CT imaging analysis
We analyzed body composition parameters of all subjects using Slice-O-Matic software. An axial CT slice was selected at the level of the lumbar vertebra 1 (L1). Quantitative CT imaging analysis revealed that the median visceral adipose tissue area (VAT), subcutaneous adipose tissue area (SAT) and skeletal muscle index (SMI) were 82.3(IQR, 30.2), 53.6(IQR, 28.6) and 27.6(IQR, 9.5), respectively. In our study, 60 patients’ images from derivation dataset were selected at random for inter-class correlation coefficients (ICCs) calculation to ensure the reproducibility. Results revealed that the ICC values of VAT, SAT and SM were 0.96, 0.95 and 0.93 respectively.
Identification of independent predictors
Binary logistic regression analyses were performed to assess univariable and multivariable associations of clinical factors with postoperative delayed awakening. Analysis results were summarized in Fig. 2. In multivariable regression analysis, age (OR, 1.41(95%CI, 1.21–1.65)), age-adjusted Charlson Comorbidity Index (aCCI) (OR, 1.54(95%CI, 1.35–1.76)), intraoperative hypotension (OR, 1.26(95%CI, 1.02–1.36)), VAT (OR, 1.95(95%CI, 1.76–2.52)) and SMI (OR, 0.68(95%CI, 0.50–0.78)) were identified as independent predictors of delayed awakening.
Fig. 2.
Forest plots for logistic regression analysis. The associations between each variable and delayed awakening were evaluated using univariable (A) and multivariable (B) regression analysis. The vertical line represents the value of no effect. Data are presented as the OR value with 95%CI. Abbreviation: BMI, body mass index; ASA, American Society of Anesthesiologists; aCCI, age-adjusted Charlson Comorbidity Index; VAT, visceral adipose tissue area; SAT, subcutaneous adipose tissue area; SMI, skeletal muscle index; OR, odds ratio; CI, confidence interval.
Radiomics score calculation
Based on the CT image data, VAT, SAT and skeletal muscle (SM) areas were segmented using 3D slicer software. A representative CT image for the segmentations of body compositions is displayed in Fig. 3. Dice Similarity Coefficient (DSC) can reflect the spatial consistency of segmentation among different observers. In our study, 50 patients’ images from derivation dataset were selected at random for DSC calculation to evaluate the consistency of the manual segmentation. The average DSC values for VAT, SAT and SM areas were 0.93 ± 0.11, 0.94 ± 0.17 and 0.90 ± 0.10. Furthermore, a total of 3409 CT image features of three segmented tissue types were extracted from each region of interest (ROI) using the pre-trained classification model ResNet50, in which the final fully connected layer was removed. Of these deep learning (DL) radiomic features, 1877 DL radiomic features were correlated with delayed awakening by using the Spearman rank correlation test. By using least absolute shrinkage and selection operator (LASSO) logistic regression model, 8 most valuable radiomic features were selected to develop the DL score and the formula was generated as follows:
Fig. 3.
A representative CT image for the segmentation of body compositions. Red area represents the area of SAT. Green area represents the area of skeletal muscle. Blue area represents the area of VAT. Abbreviation: VAT, visceral adipose tissue area; SAT, subcutaneous adipose tissue area; SM, skeletal muscle area.
DL score = 1.195 + 0.357*feature_627(VAT) -0.204*feature_892(SAT) + 0.408* feature_395(SM) + 0.628* feature_1150(SM) -0.417* feature_1041(VAT) -0.605* feature_923(SM) + 0.352* feature_458(VAT) -0.811* feature_503(SAT).
Model performance
The eXtreme Gradient Boosting classification (XGBC) model was built by combing DL score and independent clinical variables. It was optimized by grid search and the optimal parameter combinations of this model are as follows: number of trees, 7; max_depth, 6; learning rate, 0.01; subsample, 0.25; colsample_bytree = 0.2. This model had area under curves (AUCs) of 0.85(95%CI: 0.82, 0.89) for derivation dataset, 0.83(95%CI: 0.80, 0.87) for internal dataset and 0.80(95%CI: 0.76, 0.84) for external dataset, respectively (Table 2). The predictive performance of clinical model is summarized in Supplementary Table 2. Delong test revealed that the combined model significantly outperformed clinical model (P < 0.05). Furthermore, Decision curve analysis (DCA) analysis revealed that the XGBC model had a larger net benefit in the whole cohort (Fig. 4A). The calibration plot for the models for predicting postoperative delayed awakening is presented in Fig. 4B and C. This model was calibrated in the derivation and test datasets.
Table 2.
Performance of the XGBC model in predicting the delayed awakening from anesthesia.
| Performance | Derivation dataset | Internal dataset | External dataset |
|---|---|---|---|
| AUC (95%CI) | 0.85(0.82, 0.89) | 0.83(0.80, 0.87) | 0.80(0.76, 0.84) |
| Sensitivity (%) | 82.5(79.6, 85.4) | 80(78.2, 84.2) | 77.3(73.2, 79.2) |
| Specificity (%) | 81.3(79.3, 84.6) | 78.8(75.4, 81.4) | 76(72.1, 78.9) |
| Accuracy (%) | 81.9 | 79.4 | 76.6 |
| PPV (%) | 79.6 | 77.1 | 74.5 |
| NPV (%) | 83.9 | 81.5 | 78.6 |
AUC, area under of ROC curve; CI, confidence interval; PPV, positive predictive value; NPV, negative predictive value.
Fig. 4.
Assessment and interpretability of the model. (A) decision curve analysis for the XGBC and clinical models in the whole dataset. (B) calibration curves of XGBC model in both datasets. (C) calibration curves of clinical model in both datasets. (D) features importance ranking of XGBC model. (E) SHAP summary plot of the XGBC model illustrates the detailed contribution of variables to prediction at the global level. The colors represent the magnitude of the feature values.
Sensitivity analyses
We conducted sensitivity analysis by redefining delayed awakening as 45 min and 60 min after anesthesia termination. First, CT-derived VAT (OR = 1.87, 95%CI (1.22–2.86) for 45 min; OR = 1.78, 95%CI (1.15–2.76) for 60 min) and SMI (OR = 0.69, 95%CI (0.55–0.94) for 45 min; OR = 0.71, 95%CI (0.55–0.94) for 60 min) remained independent predictors for delayed awakening. Second, the XGBC model still showed good discrimination and calibration with the thresholds of 45-minute and 60-minute. The AUC values for 45 and 60 min were 0.83 (95%CI (0.79–0.87) and 0.81 (95%CI (0.73–0.83)) for the derivation dataset, 0.79 (95%CI (0.75–0.83) and 0.75 (95%CI (0.70–0.79)) for the internal test dataset, and 0.76 (95%CI (0.71–0.79) and 0.74 (95%CI (0.70–0.77)) for the external test dataset respectively. Although the AUC values slightly decreased, they still indicated good predictive performance. It was consistent with the clinical expectation that a longer threshold would reduce the model’s ability to identify early delayed awakening. Third, DCA analysis showed that the XGBC model still had a significant net clinical benefit for the 45-minute and 60-minute thresholds in the whole cohort, Calibration curves presented good consistency between the predicted and actual risks (Supplementary Figs. 1 and 2).
Explanation of the XGBC model at the global level
Feature importance was ranked for the XGBC model, and the SHapley Additive exPlanation (SHAP) method was used to quantify variable contributions to global predictions. The feature importance rankings were calculated in Fig. 4D. At the global level, the SHAP summary plot revealed that DL score, SMI and VAT were the three most important variables in predicting postoperative delayed awakening (Fig. 4E).
Discussion
Delayed return of consciousness after general anesthesia was a potentially serious event. It presents a challenge for anesthetists to accurately evaluate the risk of delayed awakening. In this retrospective study, we found that CT-derived VAT and SMI were significantly associated with postoperative delayed awakening in patients undergoing flexible ureteroscopic lithotripsy. A novel XGBC model was generated for risk estimation based on independent clinical variables and DL score. This model showed good discrimination, calibration and clinical utility. SHAP analysis was adopted to help clinicians better understand the predictive results.
In the current study, quantitative CT measurements of body compositions were conducted. Notedly, previous studies revealed that body compositions were associated with neurodegenerative disease and neurological complications after surgery5,17,18. As the main findings of our study, body composition parameters (VAT and SMI) showed close association with delayed awakening from general anesthesia. To our knowledge, obesity is recognized as a risk factor of postoperative delayed awakening19. It could be attributed to the volume of distribution of fat-soluble anesthetics increased in obese patients and drug elimination slowed down. At the termination of anesthesia, these anesthetics stored in adipose tissue diffuse into circulation, causing delayed awakening. Visceral obesity was defined as the excessive accumulation of visceral fat in the abdominal cavity20–22. In this study, we observed that patients with high VAT had a higher risk of postoperative delayed awakening. Compared to the conventional indicator called BMI, VAT may be a more accurate predictor of delayed awakening. Additionally, loss of skeletal muscle mass was associated with adverse events after surgery23,24. In general, elderly patients with low muscle radiodensity have less reserve of multiple physiological systems. This may be the reason that they are prone to surgical complications including delayed awakening.
Notably, the radiomic features of VAT and SMI in our study are quantitative parameters extracted via radiomic analysis of preoperative CT scans, which can more accurately reflect the intrinsic characteristics of body composition compared to conventional clinical indicators such as BMI. BMI only reflects the relationship between weight and height while radiomic-derived features directly quantify visceral fat accumulation. SMI normalizes skeletal muscle mass by height, providing more precise and adjustable biomarkers. The radiomics-based model may capture pathophysiological information that cannot be obtained from routine clinical data. This is the main reason why it outperformed the clinical model.
Therefore, a deep analysis of the quality of body composition may provide novel parameters related to delayed awakening. The present study is the first report about the predictive role of radiomic features of body compositions in predicting postoperative delayed awakening. As shown in the presented study, radiomics features of body compositions can be used to identify patients at high risk of developing delayed awakening, indicating the validity of DL-based radiomic features.
In addition, consistent with previous studies, variables such as age, aCCI score, and intraoperative hypotension had impacts on postoperative delayed awakening9,13,25,26. Although not all risk factors were modifiable, it was considered in our study. Surgical patients with advanced age or high aCCI scores should require more attention and receive additional monitoring.
This XGBC model offered guidance to clinicians on the development of risk-reduction strategies. In clinical setting, clinicians could be aware of the risk of postoperative delayed awakening with the help of this combined model. When a patient has a high risk of delayed awakening predicted by this model, further strategies should be considered such as optimizing the use of anesthetic drugs. On the contrary, only regular surveillance is recommended for the low-risk patients. Such a strategy can effectively help reallocate societal healthcare resources.
Our study has some limitations. First, this was a retrospective study that may be subject to inherent sampling biases. Second, the lack of fine-tuning of ResNet50 may lead to the extraction of partially irrelevant deep radiomic features. Third, more confounding factors should be considered such as use of reversal agents. The use of reversal agents in our study followed strict and unified clinical indications, and there was no arbitrary use in clinical practice. Besides, we screened candidate predictors using univariable analysis prior to multivariable modeling. Although this is a widely adopted practice, it may carry potential instability in variable selection. Finally, our model was developed and validated based on the homogeneous cohort of this specific surgery with unified anesthesia management, so this model was limited to patients undergoing flexible ureteroscopic lithotripsy. However, the core predictive factors identified in this study (VAT and SMI) were universal body composition parameters associated with anesthetic metabolism. This provided a theoretical basis for the generalization of the model to other surgeries.
Methods
Study design and participants
This study was conducted in strict compliance with the 1975 Declaration of Helsinki and its later amendments. The study protocol was approved by the Institutional Review Board of Wujiang Fifth People’s Hospital (KWK2024-032, March 2024). Written informed consents were waived for all patients due to its retrospective design and no additional invasive interventions involved in the study.
The inclusion criteria were as follows: (1) age ≥ 18 years old (2) receiving flexible ureteroscopic lithotripsy under total intravenous anesthesia (3) complete clinical data and CT images. Patients with the history of taking opioid drugs or low-quality CT images were excluded from this retrospective study. We collected demographics, laboratory tests and postoperative adverse events from the electronic record system. In addition, an external test dataset was independently collected from the Affiliated Changshu Hospital of Nantong University, with independent patient enrollment, data curation, CT scanning and anesthesia practice to guarantee fully independent external validation. More information on total intravenous anesthetic protocol can be obtained in Supplementary material S1.
Definitions of delayed awakening from anesthesia
Delayed awakening is defined as not having awakened 30 min after the termination of general anesthesia. Termination of general anesthesia is defined as the time point of complete cessation of all intravenous anesthetic drug infusions in this study.
Images acquisition
CT images were obtained using a 64-slice spiral CT. Within 30 days before surgery, all patients underwent CT scans at 120 kVp tube voltage and 220–280 mA tube current. All CT data was obtained from picture archiving and communication system and saved in the format of DICOM. To mitigate variability across different CT scanners, all images were resampled to an isotropic voxel size of 1 × 1 × 1 mm3. This standardization was achieved using cubic spline interpolation to ensure spatial consistency across the dataset. The intensity values of CT images were standardized by the method of Z-score.
Quantitative CT analysis
According to predefined HU threshold, VAT (− 150 to − 50 HU), SAT (− 190 to − 30 HU) and SM (− 29 to 150 HU) were calculated respectively27. Considering the close correlation between total muscle mass and height, the SMI was then calculated by normalizing SM by square height (m2)28–30. Two clinicians independently measured body composition parameters when blind to clinical information using Slice-O-Matic (TomoVision, Magog, Canada, version 5.0) software. Furthermore, the ICCs were adopted for evaluating the interobserver reproducibility of body composition calculation.
ROI segmentation
We manually outlined these areas with the threshold of the same specified values as quantitative analysis using 3D slicer software (Version 5.8.1). Two experts reviewed and manually labelled CT images when blind to clinical information. The dispute on the ROI delineation would be settled after discussion. All observers received standardized training on CT body composition segmentation before the start of study. A standardized operating manual was formulated to minimize subjective variation from manual delineation, thereby improving the reproducibility of segmentation.
Feature extraction and dimension reduction
The largest cross-sectional slice and its two nearest slices on the z-axis were selected as the input data for the ResNet50. We extracted the DL radiomic features from the ROIs. More information on feature extraction was provided in Supplementary material S2.
We adopted the Spearman rank correlation test and the LASSO logistic regression analysis to remove unimportant DL features. Before the LASSO analysis, the Spearman correlation analysis only serves as a crude pre-filter to remove features that are not related to the clinical outcomes without multiple testing correction. The DL score was then calculated based on their corresponding weights. More information on parameter selection can be obtained in Supplementary material S3.
Model development and evaluation
In the derivation dataset, univariable logistic regression analysis was used to screen candidate variables. Next, those with P < 0.05 were then entered into multivariable logistic regression analysis to identify independent risk factors. A clinical model was built based on multivariable logistic regression analysis. The XGBC model was developed by combining significant clinical variables and DL score31. The Python software was used for model construction and parameter optimization. The best parameters and eligible variables were obtained by the method of grid-search32. More details are described in Supplementary material S4.
For visualizing the contributions to the sample’s positive and negative predictions, SHAP plot was used to explain the impact of selected features on predictions at the global level33,34.
Furthermore, to improve the rigor and reproducibility of this study, we performed sensitivity analyses in a subset of patients.
Statistical analysis
Continuous data was shown as the median (IQR) and compared using the Mann–Whitney U test or Student’s t test as appropriate. Categorical data was compared using the chi-square test. We calculated the AUC values to assess the predictive performance of this model. DCA and calibration curve were conducted to assess the clinical utility and stability of models. We compared the statistical significance between ROC curves using Delong test. Sample size was estimated using pmsampsize package (R2cs = 0.28, parameters = 7; R software, version 4.3.2)35. This indicates that at least 145 individuals are required. IBM SPSS version 26.0 and Python software (version 3.6.2) were used for data analysis. P < 0.05 was considered statistically significant in all analyses.
Conclusions
In the present study, VAT and SMI were identified as independent predictors of delayed awakening. The combined model was developed specifically for estimating the risk of postoperative delayed awakening among patients underwent flexible ureteroscopic lithotripsy.
Abbreviations: XGBC, eXtreme Gradient Boosting classification; SHAP, SHapley Additive explanation; aCCI, age-adjusted Charlson Comorbidity Index; VAT, visceral adipose tissue area; SMI, skeletal muscle index; DL deep learning.
Electronic Supplementary Material
Below is the link to the electronic supplementary material.
Abbreviations
- CT
Computed tomography
- VAT
Visceral adipose tissue area
- SAT
Subcutaneous adipose tissue area
- SM
Skeletal muscle area
- SMI
Skeletal muscle index
- ROI
Region of interest
- DL
Deep learning
- aCCI
Age-adjusted Charlson Comorbidity Index
- LASSO
Least absolute shrinkage and selection operator
- XGBC
EXtreme Gradient Boosting classification
- AUC
Area under the curve
- ROC
Receiver operating characteristic
- DCA
Decision curve analysis
- DSC
Dice similarity coefficient
- IQR
Iinterquartile range
- BMI
Body-mass index
- ASA
American society of anesthesiologists
- SHAP
SHapley Additive exPlanation
Author contributions
1.Conceptualization：CWR, LY, GL, WXQ2. Methodology: CWR, LY, GL, WXQ, LJ3. Data collection and analysis: LY, GL, LY, CJH, LJ4. Original Draft, review and editing: all authors.
Data availability
The data used during the current study are available from the corresponding author on reasonable request. A data dictionary covering the analysis codes has been deposited at https://github.com/xiaobai00789/XGBoost.git (DOI: https://doi.org/10.5281/zenodo.18786838).
Declarations
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Yao Lu and Xiaqiu Wu contributed equally to this work.
Contributor Information
Weirong Cai, Email: 39305486@qq.com.
Li Geng, Email: lvlvli1007@163.com.
References
- 1.Kalra, S. & Wadhwa, R. Role of amino acid infusion in delayed recovery from neuromuscular blockers. Indian J. Anaesth.54, 166–168. 10.4103/0019-5049.63649 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Bu, F. D., Si, S. K., Zhang, D. B. & Chi, Y. L. Effect of invasive acupuncture on awakening quality after general anesthesia: systematic review and meta-analysis. Front. Med.11, 1502619. 10.3389/fmed.2024.1502619 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Alghamdi, L., Filfilan, R., Alghamdi, A., Alharbi, R. & Kayal, H. Factors Associated With Prolonged-Stay Patients Within the Post-anesthesia Care Unit: A Cohort Retrospective Study. Cureus16, e60092. 10.7759/cureus.60092 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Miskovic, A. & Lumb, A. B. Postoperative pulmonary complications. Br. J. Anaesth.118, 317–334. 10.1093/bja/aex002 (2017). [DOI] [PubMed] [Google Scholar]
- 5.Shirasaka, T. et al. New indicator of postoperative delayed awakening after total aortic arch replacement. Eur. J. Cardiothorac. Surg.47, 101–105. 10.1093/ejcts/ezu141 (2015). [DOI] [PubMed] [Google Scholar]
- 6.Barak, M., Greenberg, Z. & Danino, J. Delayed awakening following inadvertent high-dose remifentanil infusion in a 13 year old patient. J. Clin. Anesth.23, 322–324. 10.1016/j.jclinane.2010.07.005 (2011). [DOI] [PubMed] [Google Scholar]
- 7.Venincasa, M. J. et al. Elevated preoptic brain activity in zebrafish glial glycine transporter mutants is linked to lethargy-like behaviors and delayed emergence from anesthesia. Sci. Rep.11, 3148. 10.1038/s41598-021-82342-w (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Kong, H., Xu, L. M. & Wang, D. X. Perioperative neurocognitive disorders: A narrative review focusing on diagnosis, prevention, and treatment. CNS Neurosci. Ther.28, 1147–1167. 10.1111/cns.13873 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Thomas, E., Martin, F. & Pollard, B. Delayed recovery of consciousness after general anaesthesia. BJA Educ.20, 173–179. 10.1016/j.bjae.2020.01.007 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Tzabazis, A., Miller, C., Dobrow, M. F., Zheng, K. & Brock-Utne, J. G. Delayed emergence after anesthesia. J. Clin. Anesth.27, 353–360. 10.1016/j.jclinane.2015.03.023 (2015). [DOI] [PubMed] [Google Scholar]
- 11.Dong, D. et al. Morbid Obesity Alters Both Pharmacokinetics and Pharmacodynamics of Propofol: Dosing Recommendation for Anesthesia Induction. Drug Metab. Dispos.44, 1579–1583. 10.1124/dmd.116.071605 (2016). [DOI] [PubMed] [Google Scholar]
- 12.Hug, C. C. Jr. & Murphy, M. R. Tissue redistribution of fentanyl and termination of its effects in rats. Anesthesiology55, 369–375. 10.1097/00000542-198110000-00006 (1981). [DOI] [PubMed] [Google Scholar]
- 13.Cascella, M., Bimonte, S. & Di Napoli, R. Delayed Emergence from Anesthesia: What We Know and How We Act. Local. Reg. Anesth.13, 195–206. 10.2147/lra.S230728 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Hu, C. et al. Deep learning radio-clinical signatures for predicting neoadjuvant chemotherapy response and prognosis from pretreatment CT images of locally advanced gastric cancer patients. Int. J. Surg. (London England). 109, 1980–1992. 10.1097/js9.0000000000000432 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Dong, T., Lei, S. & Ding, G. Comparative Effectiveness of Propofol-Sufentanil vs Propofol-Fentanyl in Elderly Patients Undergoing Transurethral Resection of the Prostate. Ther. Clin. Risk Manag.21, 1773–1782. 10.2147/tcrm.S549323 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Zhang, Q. et al. Risk factors for delayed recovery in postanesthesia care unit after surgery: a large and retrospective cohort study. Int. J. Surg. (London England). 109, 1281–1290. 10.1097/js9.0000000000000364 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Winter, Y., Pieper, L., Klotsche, J., Riedel, O. & Wittchen, H. U. Obesity and Abdominal Fat Markers in Patients with a History of Stroke and Transient Ischemic Attacks. J. stroke Cerebrovasc. diseases: official J. Natl. Stroke Association. 25, 1141–1147. 10.1016/j.jstrokecerebrovasdis.2015.12.026 (2016). [DOI] [PubMed] [Google Scholar]
- 18.Xu, S. et al. Association Between Body Composition Patterns, Cardiovascular Disease, and Risk of Neurodegenerative Disease in the UK Biobank. Neurology103, e209659. 10.1212/wnl.0000000000209659 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Weber, J. et al. Context-sensitive decrement times for inhaled anesthetics in obese patients explored with Gas Man®. J. Clin. Monit. Comput.35, 343–354. 10.1007/s10877-020-00477-z (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Stranahan, A. M. Visceral adiposity, inflammation, and hippocampal function in obesity. Neuropharmacology205, 108920. 10.1016/j.neuropharm.2021.108920 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Silveira, E. A., Kliemann, N., Noll, M., Sarrafzadegan, N. & de Oliveira, C. Visceral obesity and incident cancer and cardiovascular disease: An integrative review of the epidemiological evidence. Obes. reviews: official J. Int. Association Study Obes.22, e13088. 10.1111/obr.13088 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Deng, R. et al. Association Between Visceral Obesity Index and Diabetes: A Systematic Review and Meta-analysis. J. Clin. Endocrinol. Metab.109, 2692–2707. 10.1210/clinem/dgae303 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Li, H. et al. Combination of albumin-globulin score and skeletal muscle index predicts long-term outcomes of intrahepatic cholangiocarcinoma patients after curative resection. Clin. Nutr.40, 3891–3900. 10.1016/j.clnu.2021.04.038 (2021). [DOI] [PubMed] [Google Scholar]
- 24.Li, Z. et al. Association of Perioperative Skeletal Muscle Index Change With Outcome in Colorectal Cancer Patients. J. Cachexia Sarcopenia Muscle. 15, 2519–2535. 10.1002/jcsm.13594 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Bayable, S. D., Amberbir, W. D. & Fetene, M. B. Delayed awakening and its associated factor following general anesthesia service, 2022: a cross-sectional study. Annals of medicine and surgery () 85, 4321–4328,) 85, 4321–4328, (2012). 10.1097/ms9.0000000000001103 (2023). [DOI] [PMC free article] [PubMed]
- 26.Lin, H. T. et al. Predictors for delayed awakening in adult glioma patients receiving awake craniotomy under monitored anesthesia care. Journal of neuro-oncology 165, 361–372, (2023). 10.1007/s11060-023-04494-1 [DOI] [PMC free article] [PubMed]
- 27.Bian, L. et al. Associations of radiological features of adipose tissues with postoperative complications and overall survival of gastric cancer patients. Eur. Radiol.32, 8569–8578. 10.1007/s00330-022-08918-w (2022). [DOI] [PubMed] [Google Scholar]
- 28.van Dijk, D. P. J. et al. Low skeletal muscle radiation attenuation and visceral adiposity are associated with overall survival and surgical site infections in patients with pancreatic cancer. J. Cachexia Sarcopenia Muscle. 8, 317–326. 10.1002/jcsm.12155 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Rizzo, S. et al. CT-Derived Body Composition Values and Complications After Pneumonectomy in Lung Cancer Patients: Time for a Sex-Related Analysis? Front. Oncol.12, 826058. 10.3389/fonc.2022.826058 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Heus, C. et al. Visceral obesity and muscle mass determined by CT scan and surgical outcome in patients with advanced ovarian cancer. A retrospective cohort study. Gynecol. Oncol.160, 187–192. 10.1016/j.ygyno.2020.10.015 (2021). [DOI] [PubMed] [Google Scholar]
- 31.Liang, D. et al. Perspective: Global Burden of Iodine Deficiency: Insights and Projections to 2050 Using XGBoost and SHAP. Adv. Nutr. (Bethesda Md). 16, 100384. 10.1016/j.advnut.2025.100384 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Ji, G. W. et al. Integrating Machine Learning and Tumor Immune Signature to Predict Oncologic Outcomes in Resected Biliary Tract Cancer. Ann. Surg. Oncol.28, 4018–4029. 10.1245/s10434-020-09374-w (2021). [DOI] [PubMed] [Google Scholar]
- 33.Rodríguez-Pérez, R. & Bajorath, J. Interpretation of Compound Activity Predictions from Complex Machine Learning Models Using Local Approximations and Shapley Values. J. Med. Chem.63, 8761–8777. 10.1021/acs.jmedchem.9b01101 (2020). [DOI] [PubMed] [Google Scholar]
- 34.Zhou, S. N. et al. Feasibility of machine learning-based modeling and prediction using multiple centers data to assess intrahepatic cholangiocarcinoma outcomes. Ann. Med.55, 215–223. 10.1080/07853890.2022.2160008 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Riley, R. D. et al. Calculating the sample size required for developing a clinical prediction model. BMJ368, m441. 10.1136/bmj.m441 (2020). [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The data used during the current study are available from the corresponding author on reasonable request. A data dictionary covering the analysis codes has been deposited at https://github.com/xiaobai00789/XGBoost.git (DOI: https://doi.org/10.5281/zenodo.18786838).