Prediction of postoperative desaturation and bradycardia in neonates undergoing laparoscopic inguinal hernia repair: a retrospective cohort study

Article information

Anesth Pain Med. 2026;.apm.25395

Publication date (electronic) : 2026 May 13

doi : https://doi.org/10.17085/apm.25395

Sujung Park¹, Dongwoo Chae², Hyo-Jin Byon³, Min-Soo Kim^,3, Jeongrim Lee^,4

¹Department of Anesthesiology and Pain Medicine, Asan Medical Center, University of Ulsan College of Medicine, Seoul, Korea

²Department of Pharmacology, Yonsei University College of Medicine, Seoul, Korea

³Department of Anesthesiology and Pain Medicine, Anesthesia and Pain Research Institute, Yonsei University College of Medicine, Seoul, Korea

⁴Department of Anesthesia, University of Iowa Carver College of Medicine, Iowa City, IA, USA

Corresponding authors Min-Soo Kim, M.D., Ph.D. Department of Anesthesiology and Pain Medicine, Anesthesia and Pain Research Institute, Yonsei University College of Medicine, 50 Yonsei-ro, Seodaemun-gu, Seoul 03722, Korea Tel: 82-2-2228-5784 Fax: 82-2-2227-7897 E-mail: kmsviola@yuhs.ac.kr

Jeongrim Lee, M.D., Ph.D. Department of Anesthesia, University of Iowa Carver College of Medicine, 200 Hawkins Drive, Iowa City, IA 52242, USA Tel: 1-319-800-5166 E-mail: leejeongrim@gmail.com

Sujung Park and Dongwoo Chae contributed equally to this study.This study was presented at the 29th Annual Meeting of the Korean Society of Pediatric Anesthesiology, Yonsei University, Seoul, Korea, on June 12, 2021

Received 2025 September 11; Revised 2026 January 14; Accepted 2026 March 11.

Abstract

INTRODUCTION

Apnea of prematurity (AOP) is a well-documented perioperative complication in pediatric anesthesia [1,2]. Known risk factors for AOP include a history of apnea, lower gestational age (GA), low birth weight, need for mechanical ventilation after birth, need for supplemental oxygen after birth, and comorbidities such as patent ductus arteriosus, intraventricular hemorrhage, and bronchopulmonary dysplasia (BPD) [3].

Despite advancements in anesthetic techniques, relatively few studies have explored updated predictive approaches for perioperative apnea and related complications. Moreover, there are limited effective tools for predicting the risk of apnea. In light of constrained intensive care resources and the increasing prevalence of outpatient surgery, the need for accurate risk stratification is more critical than ever.

Recent studies have introduced modern computational approaches to enhance clinical prediction models [4]. While machine learning techniques can improve risk classification and handle complex interactions among clinical variables, traditional statistical models such as logistic regression remain widely used because of their interpretability and clinical relevance [5]. Integrating these approaches—using computational methods to support feature selection and model robustness while maintaining a regression-based predictive framework—may improve accuracy without compromising interpretability [6].

In this study, we aimed to develop a machine learning–based model to predict postoperative desaturation and bradycardia in preterm and term infants who underwent laparoscopic inguinal hernia repair (IHR) under general anesthesia with sevoflurane or desflurane. Using a retrospective cohort design, we identified key clinical risk factors and incorporated them into the machine learning model intended to assist anesthesiologists in the early identification of high-risk patients.

MATERIALS AND METHODS

We conducted a retrospective cohort study at Severance Hospital, a tertiary university hospital in South Korea. The Institutional Review Board (IRB) of Severance Hospital (IRB #4-2019-1293) waived the requirement for informed consent due to the nature of the study. This study adhered to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines [7].

We included all neonates and infants with a postconceptual age (PCA) < 60 weeks who underwent IHR under general anesthesia between February 2013 and September 2019.

The standard surgical procedure for all primary IHRs at our institute is laparoscopy. Preterm and term births were defined as GA < 37 and ≥ 37 weeks, respectively. Eligible patients were identified through electronic medical records and included those who underwent laparoscopic surgery for unilateral, bilateral, or recurrent inguinal hernia. The standard anesthesia procedure for IHR in our hospital is general anesthesia, with sevoflurane or desflurane as maintenance anesthesia. Patients who were intubated and receiving ventilator support or those requiring vasopressors due to hemodynamic instability during the preoperative period were excluded from the study.

We collected data on the following demographic and clinical variables: sex, GA, postnatal age and PCA at surgery, birth weight, body weight at surgery, medical history and comorbidities (e.g., BPD or a history of apnea in preterm infants), urgency of surgery (emergency or elective), preoperative hemoglobin level, anesthetic agents used, operation time, anesthesia time, intraoperative opioid dose, length of hospital stay, operation to discharge time, and postoperative ward (e.g., general ward or neonatal/pediatric intensive care unit). We also examined postoperative complications within 48 h after surgery, including desaturation and bradycardia. A desaturation event was defined as peripheral oxygen saturation (SpO₂) < 85% or a need for mask-assisted ventilation, as documented in the electronic medical records. A bradycardia event was defined as heart rate < 80 bpm.

Continuous variables are presented as the mean ± standard deviation (range), and categorical variables are expressed as the number of patients (proportion).

The original dataset was randomly split into training and validation datasets in a ratio of 7:3. To reduce bias, this process was repeated 100 times, generating 100 development-validation dataset groups.

Features identified in prior univariate analyses were evaluated by linear discriminant analysis to determine their relative importance values within each of the 100 development-validation datasets. The features were ranked by averaging the importance values across all dataset groups. The features were selected in three stages. First, univariate logistic regression was performed for each feature. Features with P < 0.05 were chosen. Second, Pearson correlation analysis was performed to identify highly correlated features. Among correlated features, those with lower P values were retained. Third, multivariate logistic regression was applied for model development in the training dataset. The developed model was verified using the receiver operating characteristic (ROC) curve in the validation dataset. Features were sequentially added to the model based on descending mean importance values. After each addition, the area under the ROC curve (AUC) was evaluated. The optimal number of features was determined by identifying the configuration that achieved the highest mean AUC across the 100 development-validation dataset groups.

Multivariate logistic regression was conducted on the original dataset, incorporating the features identified during the selection process. Model performance was evaluated by calculating the AUC. A P value < 0.05 was considered statistically significant. All statistical analyses, including machine learning modeling, were performed using Python 3.6 (Python Software Foundation) and R version 4.0.5 (R Foundation).

RESULTS

Of the 664 infants who underwent laparoscopic IHR during the study period, 401 of them met the eligibility criteria and were included in the analysis (Fig. 1). The characteristics of the patients are summarized in Table 1. The mean PCA of the patients was 45.7 weeks, and the mean body weight was 4948.9 g. The patients were administered inhalation-based anesthesia (sevoflurane [46.9%] or desflurane [53.1%]). Postoperative desaturation and bradycardia occurred in 9.2% (37 of 401) and 4.5% (18 of 401) of the infants, respectively.

Fig. 1.

Flow chart of patient selection for the cohort study.

Table 1.

Demographic and Clinical Characteristics (n = 401)

Following univariate logistic regression analyses of all available features of postoperative complications, only those with P < 0.05 were selected for further analysis. Pearson correlation analysis was then performed among the selected features. If a correlation coefficient of > 0.6 was observed between two features, the feature with the higher P value was excluded from the analysis.

For desaturation and bradycardia, 13 and 11 features, respectively, were ranked according to their relative importance through linear discriminant analysis (Fig. 2). The change in the AUC as a function of the number of features is illustrated in Fig. 3. The maximum predictive performance was achieved with three and two features for desaturation and bradycardia, respectively. The final multivariate logistic regression models included the following features: desaturation (cardiac comorbidity, GA, and body weight at surgery); bradycardia (GA and cardiac comorbidity) (Table 2). Cardiac comorbidity emerged as the most important risk factor for both complications. The odds ratios (95% confidence intervals) were 5.112 (1.881–13.888) for desaturation and 26.597 (3.190–221.850) for bradycardia, respectively.

Fig. 2.

Feature importance for predicting (A) desaturation and (B) bradycardia.

Fig. 3.

Model performance with increasing number of features for (A) desaturation and (B) bradycardia. The graphs display changes in the AUC as features were sequentially added to the model based on their mean importance. The vertical line indicates the optimal number of features. AUC: area under the receiver operating characteristic curve, CI: confidence interval.

Table 2.

Multivariate Logistic Regression Analysis of Risk Factors for Postoperative Complications

Nomograms were constructed to visually represent the predictive model for each postoperative complication (Fig. 4). For practical application, clinicians can estimate individual risk by locating each patient’s value on each predictor axis, summing the corresponding points, and projecting the total score to the predicted probability scale. This allows rapid bedside estimation of postoperative risk without the need for electronic calculators.

Fig. 4.

Nomograms for predicting the risk of (A) desaturation and (B) bradycardia after laparoscopic hernia repair. For each predictor, read the points assigned on the 0–100 scale at the top and sum these points. Find the number on the ‘total points’ scale to estimate the likelihood of each complication. Intrauterine period is measured in weeks and body weight at operation in kilograms (kg). The risk scale represents predicted probability (e.g., 0.2 indicates a 20% estimated risk).

For example, in an infant born at 34 weeks of gestation with a ventricular septal defect (cardiac comorbidity) and a body weight of 3.0 kg at the time of surgery, the estimated risk can be calculated using the nomograms.

In the desaturation nomogram, cardiologic history (YES) corresponds to 33 points, gestational age of 34 weeks corresponds to 30 points, and body weight at surgery of 3 kg corresponds to 85 points, yielding a total of 148 points, which corresponds to an estimated desaturation risk of approximately 35%.

In the bradycardia nomogram, gestational age of 34 weeks corresponds to 45 points and the presence of cardiac comorbidity corresponds to 93 points, yielding a total of 138 points, which corresponds to an estimated bradycardia risk of approximately 12%.

DISCUSSION

In this study, we identified key risk factors for desaturation and bradycardia following general anesthesia for laparoscopic IHR in infants. Among infants with a PCA < 60 weeks, the presence of cardiac comorbidities, lower GA, and lower body weight at surgery significantly predicted these adverse postoperative events. In a hypothetical case of an infant weighing 4 kg, born at 28 weeks’ GA with underlying cardiac disease, the estimated risks of desaturation and bradycardia would be approximately 40% and 30%, respectively. Such individualized risk estimates may assist clinicians in determining postoperative monitoring strategies, including the need for prolonged observation, cardiorespiratory monitoring, or intensive care unit admission, rather than relying solely on PCA–based criteria.

Current clinical guidelines recommend that preterm infants with a PCA < 46 weeks should be admitted for postoperative monitoring after general anesthesia [8]. However, the study by Coté et al. [9], which influenced these recommendations, was limited in its evaluation of patient history. It considered factors such as necrotizing enterocolitis, respiratory distress syndrome, BPD, administration of opioids and relaxants, birth weight, apnea history, anemia, GA, and PCA but did not include important intraoperative variables such as detailed comorbidity profiles or weight at surgery. In contrast, our study incorporated a broader set of clinical variables, including detailed medical histories and weight at the time of surgery. Among highly correlated features such as GA, PCA, and postnatal age, those with lower P values were selected. The results revealed the greater importance of GA than PCA for predicting postoperative complications after IHR.

Previous studies [8,10-12] have indicated that infants with chronic lung disease might be predisposed to postoperative apnea. However, our analysis found a moderate negative correlation (r = –0.62) between GA and coexisting pulmonary disease. Therefore, coexisting pulmonary disease, which had a higher p-value, was excluded in favor of GA. The results suggest that pulmonary comorbidities may reflect underlying immaturity (i.e., lower GA), rather than independently increasing risk.

Our findings are consistent with those of Jiang et al. [13], who demonstrated that low body weight was a risk factor for postoperative bradycardia and desaturation in preterm infants. Similarly, Bawazir [14] reported that infants undergoing early surgery, with a lower body weight and PCA, were more likely to experience postoperative apnea. These results likely reflect the role of body weight as a surrogate marker of overall health status [13].

Anemia can contribute to AOP, and a possible treatment modality for AOP is the transfusion of red blood cells to increase oxygen-carrying ability [1,2,15]. Postoperative complications such as desaturation and bradycardia, which can be caused by anemia, may improve after the transfusion of red blood cells [16]. However, this association may only be relevant in cases of severe anemia [17]. In our cohort, the mean hemoglobin level was approximately 11.0 g/dl, which is within a normal range for infants. Accordingly, hemoglobin level did not emerge as a significant predictor of postoperative complications. This finding is consistent with that of Westkamp et al. [15], who demonstrated the minimal benefit of blood transfusion on the incidence of AOP in infants with moderate anemia.

Postoperative respiratory complications are more common in preterm and ex-preterm infants undergoing IHR compared with full-term infants [18]. In our study, the complication rates among full-term infants were 1.25% for bradycardia and 2.57% for desaturation. However, even among term neonates, the presence of multiple comorbidities significantly increased the risk of postoperative complications.

There are some limitations in this study. First, its retrospective design introduces the potential for reporting and documentation bias. Postoperative complications such as desaturation or bradycardia events were recorded by the nursing staff and documented in electronic medical records, which may vary in accuracy. Second, our study focused on observable endpoints (bradycardia and desaturation), rather than apnea per se, as the definition of apnea—cessation of breathing for > 15 s or associated symptoms such as bradycardia, hypotonia, or cyanosis—was not consistently documented in the available charts. Third, we were unable to account for the severity of disease. Patients with severe BPD or mild pneumonia were grouped under the same “pulmonary history” category to allow for simplified scoring. Lastly, we did not investigate the duration and severity of complications. However, as severe complications usually occur in the early postoperative period, their occurrence remains clinically useful in determining the need for postoperative monitoring or admission.

In conclusion, our study presents a clinically applicable risk scoring system to predict postoperative desaturation and bradycardia in infants undergoing laparoscopic IHR. In this cohort of infants younger than 60 weeks PCA, the presence of a cardiac comorbidity, lower GA, and reduced body weight at surgery were key risk factors. These findings support the use of individualized risk stratification to guide postoperative monitoring decisions, potentially reducing unnecessary admissions while identifying high-risk infants who require closer observation.

Article information Continued