Pediatric septic shock estimation using deep learning and electronic medical records.

BACKGROUND: Diagnosing pediatric septic shock is difficult due to the complex and often impractical traditional criteria, such as systemic inflammatory response syndrome (SIRS), which result in delays and higher risks. This study aims to develop a deep learning-based model using SIRS data for early diagnosis in pediatric septic shock cases. METHODS: The study analyzed data from pediatric patients (<18 years old) admitted to a tertiary hospital from January 2010 to July 2023. Vital signs, lab tests, and clinical information were collected. Septic shock cases were identified using SIRS criteria and inotrope use. A deep learning model was trained and evaluated using the area under the receiver operating characteristics curve (AUROC) and area under the precision-recall curve (AUPRC). Variable contributions were analyzed using the Shapley additive explanation value. RESULTS: The analysis, involving 9,616,115 measurements, identified 34,696 septic shock cases (0.4%). Oxygen supply was crucial for 41.5% of the control group and 20.8% of the septic shock group. The final model showed strong performance, with an AUROC of 0.927 and AUPRC of 0.879. Key influencers were age, oxygen supply, sex, and partial pressure of carbon dioxide, while body temperature had minimal impact on estimation. CONCLUSIONS: The proposed deep learning model simplifies early septic shock diagnosis in pediatric patients, reducing the diagnostic workload. Its high accuracy allows timely treatment, but external validation through prospective studies is needed.

Streamlining pediatric vital sign assessment: innovations and insights.

Accurate assessment of pediatric vital signs is critical for detecting abnormalities and guiding medical interventions, but interpretation is challenging due to age-dependent physiological variations. Therefore, this study aimed to develop age-specific centile curves for blood pressure, heart rate, and respiratory rate in pediatric patients and create a user-friendly web-based application for easy access to these data. We conducted a retrospective cross-sectional observational study analyzing 3,779,482 records from the National Emergency Department Information System of Korea, focusing on patients under 15 years old admitted between January 2016 and December 2017. After applying exclusion criteria to minimize the impact of patients' symptoms on vital signs, 1,369,608 records were used for final analysis. The box-cox power exponential distribution and Lambda-Mu-Sigma (LMS) method were used to generate blood pressure centile charts, while heart rate and respiratory rate values were drawn from previously collected LMS values. We developed comprehensive age-specific centile curves for systolic, diastolic, and mean blood pressure, heart rate, and respiratory rate. These were integrated into a web-based application ( http://centile.research.or.kr ), allowing users to input patient data and promptly obtain centile and z-score information for vital signs. Our study provides an accessible system for pediatric vital sign evaluation, addressing previous limitations and offering a practical solution for clinical assessment. Future research should validate these centile curves in diverse populations.

Congenital central hypoventilation syndrome in korea: 20 years of clinical observation and evaluation of the ventilation strategy in a single center.

Congenital central hypoventilation syndrome (CCHS) is a rare genetic disorder characterized by hypoventilation due to impaired breathing control by the central nervous system and other symptoms of autonomic dysfunction. Mutations in paired-like homeobox 2 B (PHOX2B) are responsible for most cases of CCHS. Patients with CCHS have various phenotypes and severities, making the diagnosis difficult. This study aimed to present a comprehensive single-center experience of patients with CCHS, including key clinical features, treatment strategies, and outcomes. A retrospective chart review was performed for patients diagnosed with CCHS between January 2001 and July 2023 at Seoul National University Children's Hospital. Finally, we selected 24 patients and collected their demographic data, genotypes, ventilation methods, and clinical features related to autonomic dysfunction. The relationship between the clinical manifestations and genotypes was also examined. All patients used home ventilators, and tracheostomy was performed in 87.5% of patients. Fifteen (62.5%) patients had constipation and nine (37.5%) were diagnosed with Hirschsprung disease. Arrhythmia, endocrine dysfunction, and subclinical hypothyroidism were present in nine (37.5%), six patients (25.0%), and two patients (16.7%), respectively. A significant number of patients exhibited neurodevelopmental delays (19 patients, 79.2%). There was a correlation between the phenotype and genotype of PHOX2B in patients with CCHS. (r = 0.71, p < 0.001).   Conclusion: There was a positive correlation between paired-like homeobox 2 B mutations (especially the number of GCN repeats in the polyalanine repeat mutations sequence) and clinical manifestations. This study also demonstrated how initial treatment for hypoventilation affects neurodevelopmental outcomes in patients with CCHS. What is Known: • Congenital central hypoventilation syndrome is a rare genetic disorder characterized by hypoventilation and dysfunction of autonomic nervous system. • The disease-defining gene of CCHS is PHOX2B gene - most of the cases have heterozygous PARMs and the number of GCN triplets varies among the patients(20/24 - 20/33). What is New: • We have noted in the Korean patients with CCHS that there is a correlation between genotype (number of GCN repeats) and severity of phenotype. • National support for rare diseases allowed for a prompter diagnosis of patients with CCHS in Korean population.

Development of a deep learning model for predicting critical events in a pediatric intensive care unit.

BACKGROUND: Identifying critically ill patients at risk of cardiac arrest is important because it offers the opportunity for early intervention and increased survival. The aim of this study was to develop a deep learning model to predict critical events, such as cardiopulmonary resuscitation or mortality. METHODS: This retrospective observational study was conducted at a tertiary university hospital. All patients younger than 18 years who were admitted to the pediatric intensive care unit from January 2010 to May 2023 were included. The main outcome was prediction performance of the deep learning model at forecasting critical events. Long short-term memory was used as a deep learning algorithm. The five-fold cross validation method was employed for model learning and testing. RESULTS: Among the vital sign measurements collected during the study period, 11,660 measurements were used to develop the model after preprocessing; 1,060 of these data points were measurements that corresponded to critical events. The prediction performance of the model was the area under the receiver operating characteristic curve (95% confidence interval) of 0.988 (0.9751.000), and the area under the precision-recall curve was 0.862 (0.700-1.000). CONCLUSIONS: The performance of the developed model at predicting critical events was excellent. However, follow-up research is needed for external validation.

Healthcare service use and medical outcomes of tracheostomy-dependent children: a nationwide study.

BACKGROUND: Despite the rising trend of tracheostomies in children, there is a lack of comprehensive resources for families to navigate the challenges of living with a tracheostomy, emphasising the need for evidence-based support in understanding postoperative care and long-term adjustments. This study aimed to examine the pattern of using healthcare services and nationwide medical outcomes in children who underwent a tracheotomy before the age of 2 years. METHODS: This retrospective study used the National Health Insurance System database from 2008 to 2016 and included all children codified with tracheotomy procedure codes before their second birthday. Healthcare utilisation, such as medical costs, number of hospital visits, home healthcare nursing and medical diagnoses on readmission, in the first 2 years after tracheotomy was evaluated. Multivariable logistic regression analysis was used to determine the factors affecting mortality. RESULTS: In total, 813 patients were included in this study. Their use of healthcare services and the accompanying expenses were higher than the national medians for similar age groups; however, both metrics decreased in the second year. The major causes of admission within 2 years of surgery were respiratory and neurological diseases. The mortality rate within 2 years was 37.8%. Higher risks of mortality were associated with having two or more complex chronic conditions. Use of home healthcare nursing services was associated with a lower mortality risk. CONCLUSION: Paediatric patients with more complex chronic conditions tended to have higher mortality rates within 2 years after surgery. However, receiving home healthcare nursing was significantly associated with a reduced risk of death. Many causes of hospitalisation may be preventable with education and supportive care. Therefore, further research for establishing an integrated care system for these patients and their caregivers is required.

Individualized estimation of arterial carbon dioxide partial pressure using machine learning in children receiving mechanical ventilation.

BACKGROUND: Measuring arterial partial pressure of carbon dioxide (PaCO2) is crucial for proper mechanical ventilation, but the current sampling method is invasive. End-tidal carbon dioxide (EtCO2) has been used as a surrogate, which can be measured non-invasively, but its limited accuracy is due to ventilation-perfusion mismatch. This study aimed to develop a non-invasive PaCO2 estimation model using machine learning. METHODS: This retrospective observational study included pediatric patients (< 18 years) admitted to the pediatric intensive care unit of a tertiary children's hospital and received mechanical ventilation between January 2021 and June 2022. Clinical information, including mechanical ventilation parameters and laboratory test results, was used for machine learning. Linear regression, multilayer perceptron, and extreme gradient boosting were implemented. The dataset was divided into 7:3 ratios for training and testing. Model performance was assessed using the R2 value. RESULTS: We analyzed total 2,427 measurements from 32 patients. The median (interquartile range) age was 16 (12-19.5) months, and 74.1% were female. The PaCO2 and EtCO2 were 63 (50-83) mmHg and 43 (35-54) mmHg, respectively. A significant discrepancy of 19 (12-31) mmHg existed between EtCO2 and the measured PaCO2. The R2 coefficient of determination for the developed models was 0.799 for the linear regression model, 0.851 for the multilayer perceptron model, and 0.877 for the extreme gradient boosting model. The correlations with PaCO2 were higher in all three models compared to EtCO2. CONCLUSIONS: We developed machine learning models to non-invasively estimate PaCO2 in pediatric patients receiving mechanical ventilation, demonstrating acceptable performance. Further research is needed to improve reliability and external validation.

Development of a deep learning model that predicts critical events of pediatric patients admitted to general wards.

Early detection of deteriorating patients is important to prevent life-threatening events and improve clinical outcomes. Efforts have been made to detect or prevent major events such as cardiopulmonary resuscitation, but previously developed tools are often complicated and time-consuming, rendering them impractical. To overcome this problem, we designed this study to create a deep learning prediction model that predicts critical events with simplified variables. This retrospective observational study included patients under the age of 18 who were admitted to the general ward of a tertiary children's hospital between 2020 and 2022. A critical event was defined as cardiopulmonary resuscitation, unplanned transfer to the intensive care unit, or mortality. The vital signs measured during hospitalization, their measurement intervals, sex, and age were used to train a critical event prediction model. Age-specific z-scores were used to normalize the variability of the normal range by age. The entire dataset was classified into a training dataset and a test dataset at an 8:2 ratio, and model learning and testing were performed on each dataset. The predictive performance of the developed model showed excellent results, with an area under the receiver operating characteristics curve of 0.986 and an area under the precision-recall curve of 0.896. We developed a deep learning model with outstanding predictive power using simplified variables to effectively predict critical events while reducing the workload of medical staff. Nevertheless, because this was a single-center trial, no external validation was carried out, prompting further investigation.

Development of a model to predict vancomycin serum concentration during continuous infusion of vancomycin in critically ill pediatric patients.

Vancomycin is a frequently used antibiotic in intensive care units, and the patient's renal clearance affects the pharmacokinetic characteristics of vancomycin. Several advantages have been reported for vancomycin continuous intravenous infusion, but studies on continuous dosing regimens based on patients' renal clearance are insufficient. The aim of this study was to develop a vancomycin serum concentration prediction model by factoring in a patient's renal clearance. Children admitted to our institution between July 1, 2021, and July 31, 2022 with records of continuous infusion of vancomycin were included in the study. Sex, age, height, weight, vancomycin dose by weight, interval from the start of vancomycin administration to the time of therapeutic drug monitoring sampling, and vancomycin serum concentrations were analyzed with the linear regression analysis of the mixed effect model. Univariable regression analysis was performed using the vancomycin serum concentration as a dependent variable. It showed that vancomycin dose (p < 0.001) and serum creatinine (p = 0.007) were factors that had the most impact on vancomycin serum concentration. Vancomycin serum concentration was affected by vancomycin dose (p < 0.001) and serum creatinine (p = 0.001) with statistical significance, and a multivariable regression model was obtained as follows: Vancomycin serum concentration (mg/l) = -1.296 + 0.281 × vancomycin dose (mg/kg) + 20.458 × serum creatinine (mg/dl) (adjusted coefficient of determination, R2 = 0.66). This prediction model is expected to contribute to establishing an optimal continuous infusion regimen for vancomycin.

Outcomes of extracorporeal membrane oxygenation support in pediatric hemato-oncology patients.

BACKGROUND: In this study, we reviewed the outcomes of pediatric patients with malignancies who underwent hematopoietic stem cell transplantation (HSCT) and extracorporeal membrane oxygenation (ECMO). METHODS: We retrospectively analyzed the records of pediatric hemato-oncology patients treated with chemotherapy or HSCT and who received ECMO in the pediatric intensive care unit (PICU) at Seoul National University Children's Hospital from January 2012 to December 2020. RESULTS: Over a 9-year period, 21 patients (14 males and 7 females) received ECMO at a single pediatric institute; 10 patients (48%) received veno-arterial (VA) ECMO for septic shock (n=5), acute respiratory distress syndrome (ARDS) (n=3), stress-induced myopathy (n=1), or hepatopulmonary syndrome (n=1); and 11 patients (52%) received veno-venous (VV) ECMO for ARDS due to pneumocystis pneumonia (n=1), air leak (n=3), influenza (n=1), pulmonary hemorrhage (n=1), or unknown etiology (n=5). All patients received chemotherapy; 9 received anthracycline drugs and 14 (67%) underwent HSCT. Thirteen patients (62%) were diagnosed with malignancies and 8 (38%) were diagnosed with non-malignant disease. Among the 21 patients, 6 (29%) survived ECMO in the PICU and 5 (24%) survived to hospital discharge. Among patients treated for septic shock, 3 of 5 patients (60%) who underwent ECMO and 5 of 10 patients (50%) who underwent VA ECMO survived. However, all the patients who underwent VA ECMO or VV ECMO for ARDS died. CONCLUSIONS: ECMO is a feasible treatment option for respiratory or heart failure in pediatric patients receiving chemotherapy or undergoing HSCT.

Pediatric Extracorporeal Membrane Oxygenation in Korea: A Multicenter Retrospective Study on Utilization and Outcomes Spanning Over a Decade.

BACKGROUND: Over the last decade, extracorporeal membrane oxygenation (ECMO) use in critically ill children has increased and is associated with favorable outcomes. Our study aims to evaluate the current status of pediatric ECMO in Korea, with a specific focus on its volume and changes in survival rates based on diagnostic indications. METHODS: This multicenter study retrospectively analyzed the indications and outcomes of pediatric ECMO over 10 years in patients at 14 hospitals in Korea from January 2012 to December 2021. Four diagnostic categories (neonatal respiratory, pediatric respiratory, post-cardiotomy, and cardiac-medical) and trends were compared between periods 1 (2012-2016) and 2 (2017-2021). RESULTS: Overall, 1065 ECMO runs were performed on 1032 patients, with the annual number of cases remaining unchanged over the 10 years. ECMO was most frequently used for post-cardiotomy (42.4%), cardiac-medical (31.8%), pediatric respiratory (17.5%), and neonatal respiratory (8.2%) cases. A 3.7% increase and 6.1% decrease in pediatric respiratory and post-cardiotomy cases, respectively, were noted between periods 1 and 2. Among the four groups, the cardiac-medical group had the highest survival rate (51.2%), followed by the pediatric respiratory (46.4%), post-cardiotomy (36.5%), and neonatal respiratory (29.4%) groups. A consistent improvement was noted in patient survival over the 10 years, with a significant increase between the two periods from 38.2% to 47.1% (P = 0.004). Improvement in survival was evident in post-cardiotomy cases (30-45%, P = 0.002). Significant associations with mortality were observed in neonates, patients requiring dialysis, and those treated with extracorporeal cardiopulmonary resuscitation (P < 0.001). In pediatric respiratory ECMO, immunocompromised patients also showed a significant correlation with mortality (P < 0.001). CONCLUSION: Pediatric ECMO demonstrated a steady increase in overall survival in Korea; however, further efforts are needed since the outcomes remain suboptimal compared with global outcomes.

Eleven years of experience in operating a pediatric rapid response system at a children's hospital in South Korea.

BACKGROUND: Various rapid response systems have been developed to detect clinical deterioration in patients. Few studies have evaluated single-parameter systems in children compared to scoring systems. Therefore, in this study we evaluated a single-parameter system called the acute response system (ARS). METHODS: This retrospective study was performed at a tertiary children's hospital. Patients under 18 years old admitted from January 2012 to August 2023 were enrolled. ARS parameters such as systolic blood pressure, heart rate, respiratory rate, oxygen saturation, and whether the ARS was activated were collected. We divided patients into two groups according to activation status and then compared the occurrence of critical events (cardiopulmonary resuscitation or unexpected intensive care unit admission). We evaluated the ability of ARS to predict critical events and calculated compliance. We also analyzed the correlation between each parameter that activates ARS and critical events. RESULTS: The critical events prediction performance of ARS has a specificity of 98.5%, a sensitivity of 24.0%, a negative predictive value of 99.6%, and a positive predictive value of 8.1%. The compliance rate was 15.6%. Statistically significant increases in the risk of critical events were observed for all abnormal criteria except low heart rate. There was no significant difference in the incidence of critical events. CONCLUSIONS: ARS, a single parameter system, had good specificity and negative predictive value for predicting critical events; however, sensitivity and positive predictive value were not good, and medical staff compliance was poor.

Comparison of paediatric basic life support guidelines endorsed by member councils of Resuscitation Council of Asia.

Background: Paediatric cardiac arrest outcomes, especially for infants, remain poor. Due to different training, resource differences, and historical reasons, paediatric cardiac arrest algorithms for various Asia countries vary. While there has been a common basic life support algorithm for adults by the Resuscitation Council of Asia (RCA), there is no common RCA algorithm for paediatric life support.We aimed to review published paediatric life support guidelines from different Asian resuscitation councils. Methods: Pubmed and Google Scholar search were performed for published paediatric basic and advanced life support guidelines from January 2015 to June 2023. Paediatric representatives from the Resuscitation Council of Asia were sought and contacted to provide input from September 2022 till June 2023. Results: While most of the components of published paediatric life support algorithms of Asian countries are similar, there are notable variations in terms of age criteria for recommended use of adult basic life support algorithms in the paediatric population less than 18 years old, recommended paediatric chest compression depth targets, ventilation rates post-advanced airway intra-arrest, and first defibrillation dose for shockable rhythms in paediatric cardiac arrest. Conclusion: This was an overview and mapping of published Asian paediatric resuscitation algorithms. It highlights similarities across paediatric life support guidelines in Asian countries. There were some differences in components of paediatric life support which highlight important knowledge gaps in paediatric resuscitation science. The minor differences in the paediatric life support guidelines endorsed by the member councils may provide a framework for prioritising resuscitation research and highlight knowledge gaps in paediatric resuscitation.

Change in Severity and Clinical Manifestation of MIS-C Over SARS-CoV-2 Variant Outbreaks in Korea.

BACKGROUND: There is difference in the incidence of multi-system inflammatory syndrome in children (MIS-C) in patients with different variants of severe acute respiratory syndrome coronavirus 2, however, little is known about the epidemiology in Asian countries. We investigated and compared the epidemiology of the MIS-C during omicron-dominant period with that of previous periods in South Korea. METHODS: We obtained clinical, epidemiological and laboratory data on MIS-C cases from national MIS-C surveillance in South Korea. We defined pre-delta period as January 2020-May 2021; delta period as June 2021-December 2021; and omicron period as January 2022-April 2022. We describe the clinical characteristics and outcomes of MIS-C patients by period. RESULTS: A total of 91 cases were assessed to be MIS-C cases. Number of MIS-C cases have increased from six cases during pre-delta period to 66 cases during omicron period, while the incidence rate (the number of MIS-C cases per 100,000 cases of reported coronavirus disease 2019) has decreased from 38.5 cases per 100,000 (95% confidence interval [CI], 14.1-83.9) during pre-delta period to 1.6 cases per 100,000 (95% CI, 1.2-2.0) during omicron periods. During pre-delta period, 66.7% and 100% had hypotension and gastrointestinal involvement, respectively; while during omicron period, 12.1% and 6.1% had such clinical manifestations. Fifty percent of pre-delta MIS-C patients were taken intensive care unit (ICU) cares, while 10.6% of patients during omicron periods were in ICUs. CONCLUSION: Omicron period were associated with less severe clinical manifestation compared to pre-delta and delta periods. Although incidence rate of MIS-C was lower for the omicron period than pre-delta and delta periods, number of patients reported with MIS-C may pose a substantial clinical burden.

Evaluation of off-label medication use and drug safety in a pediatric intensive care unit.

Safety and efficacy are essential in the process of disease treatment. However, off-label medication use is inevitable because various medications do not contain regulatory labels for pediatric use. We aimed to examine off-label medication use and analyze the risk factors correlated with adverse drug reactions (ADRs). This study was performed retrospectively using electronic medical data from a pediatric intensive care unit (PICU) of a tertiary hospital in Korea from July 2019 to June 2020. A total 6,183 prescribed medications from 502 PICU patients were examined in the present study. A total of 80% were infants or children, and 96.0% of them were treated with off-label medications. It was discovered that 4,778 off-label cases (77.2%) of the top 100 drugs had prescriptions with dosage (67.8%). Drugs prescribed to patients admitted to the cardiothoracic department (odds ratio [OR], 3.248; p = 0.019), total number of medications (OR, 1.116; p = 0.001), and length of PICU stay of ≥ 7 days (OR, 4.981; p = 0.008) were significantly associated with ADRs. ADRs were noted to be more severe in off-label use (p = 0.0426). For appropriate medication use, evidence regarding the safety of off-label medications is required and ultimately reflected in the official regulation.

Effect of Diurnal Variation of Heart Rate and Respiratory Rate on Activation of Rapid Response System and Clinical Outcome in Hospitalized Children.

Heart rate and respiratory rate display circadian variation. Pediatric single-parameter rapid response system is activated when heart rate or respiratory rate deviate from age-specific criteria, though activation criteria do not differentiate between daytime and nighttime, and unnecessary activation has been reported due to nighttime bradycardia. We evaluated the relationship between rapid response system activation and the patient's clinical outcome by separately applying the criteria to daytime and nighttime in patients < 18. The observation period was divided into daytime and nighttime (8:00−20:00, and 20:00 to 8:00), according to which measured heart rate and respiratory rate were divided and rapid response system activation criteria were applied. We classified lower nighttime than daytime values into the 'decreased group', and the higher ones into the 'increased group', to analyze their effect on cardiopulmonary resuscitation occurrence or intensive care unit transfer. Nighttime heart rate and respiratory rate were lower than the daytime ones in both groups (both p values < 0.001), with no significant association with cardiopulmonary resuscitation occurrence or intensive care unit transfer in either group. Heart rate and respiratory rate tend to be lower at nighttime; however, their effect on the patient's clinical outcome is not significant.

Left atrial thrombosis with invasive pulmonary aspergillosis in children with immunodeficiency.

Invasive aspergillosis is a major cause of infectious disease in immunocompromised patients; however, cardiac involvement in pulmonary aspergillosis is not well-known. Two paediatric patients undergoing chemotherapy were diagnosed with cardiac aspergilloma, accompanied by pulmonary aspergillosis. In both patients, antibiotic and antifungal treatments were initiated immediately after the pneumonia was diagnosed; however, both died of multiple cerebral thromboembolisms.

YH29407 with anti-PD-1 ameliorates anti-tumor effects via increased T cell functionality and antigen presenting machinery in the tumor microenvironment.

Among cancer cells, indoleamine 2, 3-dioxygenase1 (IDO1) activity has been implicated in improving the proliferation and growth of cancer cells and suppressing immune cell activity. IDO1 is also responsible for the catabolism of tryptophan to kynurenine. Depletion of tryptophan and an increase in kynurenine exert important immunosuppressive functions by activating regulatory T cells and suppressing CD8+ T and natural killer (NK) cells. In this study, we compared the anti-tumor effects of YH29407, the best-in-class IDO1 inhibitor with improved pharmacodynamics and pharmacokinetics, with first and second-generation IDO1 inhibitors (epacadostat and BMS-986205, respectively). YH29407 treatment alone and anti-PD-1 (aPD-1) combination treatment induced significant tumor suppression compared with competing drugs. In particular, combination treatment showed the best anti-tumor effects, with most tumors reduced and complete responses. Our observations suggest that improved anti-tumor effects were caused by an increase in T cell infiltration and activity after YH29407 treatment. Notably, an immune depletion assay confirmed that YH29407 is closely related to CD8+ T cells. RNA-seq results showed that treatment with YH29407 increased the expression of genes involved in T cell function and antigen presentation in tumors expressing ZAP70, LCK, NFATC2, B2M, and MYD88 genes. Our results suggest that an IDO1 inhibitor, YH29407, has enhanced PK/PD compared to previous IDO1 inhibitors by causing a change in the population of CD8+ T cells including infiltrating T cells into the tumor. Ultimately, YH29407 overcame the limitations of the competing drugs and displayed potential as an immunotherapy strategy in combination with aPD-1.

S/F and ROX indices in predicting failure of high-flow nasal cannula in children.

BACKGROUND: A high-flow nasal cannula (HFNC) is a useful respiratory support for children with respiratory distress; however, it elevates the risk of belated intubation. Recently, indices based on percutaneous oxygen saturation, a fraction of inspired oxygen, and respiratory rate (RR) have been suggested for the prediction of HFNC failure. We aimed to evaluate various indices predicting HFNC failure in children who started receiving HFNC at a tertiary center for 27 months. METHODS: Cases of HFNC failure were classified as hypoxic respiratory failure (HRF) or non-HRF (NHRF) depending on the cause of intubation. The ratio of percutaneous oxygen saturation to the fraction of inspired oxygen (S/F), the ratio of S/F to RR (ROX), the ratio of S/F to RR/median RR (ROX-M), and the ratio of S/F to z-score of RR (ROX-Z) were calculated and compared between groups. RESULTS: Of the 152 cases, 45 (29.6%) failed to wean from the HFNC support, of which 21 (46.7%) were HRFs and 24 (53.3%) were NHRFs. The S/F and ROX-M at 6 h and 3 h, respectively, predicted HRF well with a high area under the curve. Initial hypercapnia and low weight were good predictors for NHRF. CONCLUSIONS: For the management of children with HFNC, these risk factors and indicators should be monitored to make an early decision regarding intubation.