A nomogram model for early recognition of acute necrotizing encephalopathy: a retrospective study

Introduction

Acute necrotizing encephalopathy (ANE) is a rare and devastating pediatric condition marked by rapid neurological decline, extensive symmetrical brain lesions, and poor clinical outcomes (1). Since its first described by Mizuguchi et al. in 1995 (2), ANE is a fulminant disorder with a mortality rate exceeding 30%; survivors frequently experience severe neurological sequelae such as cognitive impairment, motor dysfunction, epilepsy, and developmental delay (3). The disease typically follows viral infections, particularly influenza and other respiratory pathogens, and can progress swiftly to coma, brain herniation, and death without prompt recognition and intervention (4-6). Due to its high fatality rate and substantial long-term disability burden, ANE represents one of the most severe acute neurologic emergencies in children.

The diagnosis of pediatric ANE remains challenging in its early stages despite severe clinical consequences. Initial manifestations like fever, altered consciousness, seizures, and nonspecific laboratory abnormalities are not unique and overlap with other infectious or inflammatory encephalopathies, such as viral encephalitis, acute disseminated encephalomyelitis, or sepsis-associated encephalopathy. Characteristic neuroimaging findings are often absent at onset, and the condition typically progresses rapidly before definitive radiological evidence emerges (7). This makes early identification of high-risk children difficult in routine practice, potentially delaying intervention and worsening outcomes (8). Although considered rare, population-based data indicate ANE contributes to a substantial disease burden in children, highlighting the need for practical tools to enable timely recognition and risk stratification.

Currently, robust and easily applicable models to predict pediatric ANE based on readily available clinical and laboratory variables are lacking. To address this gap, we conducted a retrospective study of children with ANE and non-ANE encephalopathy, collecting demographic characteristics, clinical features, and laboratory indicators as candidate predictors. Using least absolute shrinkage and selection operator (LASSO) regression followed by logistic regression, we developed a nomogram to estimate the risk of ANE and evaluated its clinical utility. This model could assist clinicians in early identification of high-risk patients, thereby supporting timely clinical decisions and potentially improving outcomes. We present this article in accordance with the TRIPOD reporting checklist (available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0392/rc).

Methods

Study population and inclusion/exclusion criteria

This single-center retrospective cohort study analyzed the raw data of 370 pediatric patients admitted to the Children’s Hospital of Soochow University in Eastern China between November 2021 and February 2024. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The Medical Ethics Committee of Children’s Hospital of Soochow University approved this retrospective study (No. 2024CS103). Written informed consent was obtained from the legal guardians of all patients prior to enrollment. Following the removal of all personal identifiers, the research data were analyzed anonymously.

The inclusion criteria were as follows: (I) age between 29 days and 18 years; (II) detection of influenza virus or coronavirus disease 2019 (COVID-19) from pharyngeal swabs, sputum, or paired serum samples; (III) seizure or altered consciousness as the initial presenting symptom. Exclusion criteria comprised: (I) underlying conditions such as epilepsy, cerebral palsy, or hereditary genetic diseases; (II) recent head trauma; (III) missing clinical data exceeding 10%; (IV) onset time greater than 72 hours. We excluded 43 children, leaving 327 for the final analysis (Figure 1).

Figure 1 Flowchart of patient enrollment in this study. ANE, acute necrotizing encephalopathy; COVID-19, coronavirus disease 2019.

The diagnostic criteria for ANE followed those established by Mizuguchi et al. (9). A diagnosis required the presence of the following features: a preceding viral prodrome, such as fever, occurring 1–3 days before onset and followed by rapid neurological deterioration; cerebrospinal fluid (CSF) protein exceeding 1,000 mg/L with normal or only mildly elevated cell counts; elevated serum transaminases, with or without increased lactate dehydrogenase (LDH) and creatine kinase (CK), in the absence of hyperammonemia; characteristic brain magnetic resonance imaging (MRI) findings of symmetrical multifocal lesions, predominantly involving the bilateral thalami, brainstem tegmentum, periventricular white matter, cerebellar medulla, or internal capsules; and the exclusion of alternative conditions such as Reye’s syndrome, acute disseminated encephalomyelitis, and Wernicke’s encephalopathy. MRI findings served as the diagnostic gold standard.

To ensure that all clinical and laboratory findings were captured during the early phase of neurological deterioration and before the appearance of characteristic neuroradiological lesions, we obtained data from the first assessments performed upon emergency department (ED) arrival. The first venous blood sample was drawn immediately after ED registration, almost always within 1–2 hours of hospital presentation. All patients were brought directly to our ED by their caregivers, and the symptom onset-to-ED arrival interval was typically within a few hours.

The initial neurological examination, including the Glasgow Coma Scale (GCS) assessment, was conducted by the attending pediatrician or neurologist within the same time window, and the lowest GCS score during the first 24 hours was recorded. To minimize the potential impact of antiseizure and sedative medications on the GCS score, we cross-checked the timing of drug administration (e.g., benzodiazepines, barbiturates) with nursing and physician documentation. In addition, the variable “altered consciousness” used in subsequent analyses was not solely based on the GCS score but also incorporated the qualitative assessment of consciousness by the attending pediatric neurologist at the time of initial evaluation. Two independent investigators extracted GCS assessments from the medical records, resolving any disagreements by consensus. Due to the retrospective design, the exact interval from symptom onset to blood sampling and clinical evaluation could not be standardized; however, all laboratory and clinical data in this study were obtained before neuroimaging confirmed bilateral thalamic lesions.

Statistical analysis

Based on prior studies and clinical experience, demographic characteristics, clinical features, and laboratory examinations were collected from the included children. Categorical variables are presented as frequencies (percentages), and continuous variables with a skewed distribution as median and interquartile range (IQR). Categorical variables were analyzed using the Chi-squared (χ²) or Fisher’s exact test, while continuous variables were analyzed using the Mann-Whitney U test. Cases with missingness exceeding 10% for key variables were excluded from the analysis; remaining missing continuous values were imputed using the median. Sensitivity analyses were conducted to assess the robustness of the results.

To identify candidate predictors for the final model, logistic regression and LASSO regression were performed on the 32 candidate risk factors. Variables selected by these analyses were incorporated into a nomogram model. The single-center patient data was randomly divided into training and validation subsets at a 7:3 ratio for internal validation; the model was developed on the training set and subsequently validated on the validation set. Model discrimination was evaluated using the area under the curve (AUC) and the area under the precision-recall curve (AUPRC). Calibration was assessed with calibration curves, while clinical utility was examined via decision curve analysis (DCA) and clinical impact curves. Logistic regression was performed in SPSS version 26.0, with a two-sided P value <0.05 denoting statistical significance. LASSO regression was implemented using the “glmnet” package in R software (version 4.2.3). The nomogram and calibration curves were generated with the “rms” package, DCA and clinical impact curves were performed using the “rmda” package, and receiver operating characteristic (ROC) and precision-recall (PR) curves were constructed using the “pROC” and “PRROC” packages, respectively.

Results

Comparison of baseline characteristics between the training and validation sets

The training set comprised 31 patients with ANE and 197 patients without, while the validation set included 8 patients with ANE and 91 without. The clinical characteristics and laboratory findings of both cohorts are presented in Table S1. Except for white blood cell (WBC), LDH, and international normalized ratio (INR), which showed significant differences between cohorts (P<0.05), no significant differences were observed for other variables. The overall baseline characteristics were generally balanced between the training and validation sets.

Comparison of baseline characteristics between the ANE and non-ANE groups

Among the 327 children enrolled, 39 (11.92%) were assigned to the ANE group and 288 (88.07%) to the non-ANE group (Figure 1). The clinical diagnoses in the non-ANE group comprised febrile seizures, mild influenza-associated encephalopathy, viral encephalitis, viral meningitis, acute disseminated encephalomyelitis, immune-mediated encephalitis, and toxic encephalopathy (Figure S1).

Significant differences were observed between the ANE and non-ANE groups regarding sex, vomiting, diarrhea, Tmax, altered consciousness, and history of previous seizures (all P<0.05). In laboratory examinations, all parameters except WBC count, N%, and L% showed significant differences between groups (P<0.05; Table S2).

Identifying independent risk factors for ANE onset

In univariate logistic regression analysis, four variables (time from onset to seizure, age, WBC, and N%) were not significantly correlated with ANE (P>0.05) and were excluded from subsequent analysis. The remaining 28 candidate variables were subjected to LASSO regression with 10-fold cross-validation. Using the lambda.1se criterion, four variables with non-zero coefficients were retained: altered consciousness, aspartate aminotransferase (AST), LDH, and blood urea nitrogen (BUN). These four variables were selected for model construction (Figure 2).

Figure 2 LASSO regression analysis. (A) Selection of the optimal parameter (λ) in the LASSO model. (B) Distribution of variable coefficients in the LASSO model, with Lambda.1se having four non-zero coefficients and Lambda.min having eight non-zero coefficients. LASSO, least absolute shrinkage and selection operator.

Construction of a personalized nomogram prediction model

A nomogram was constructed based on the four selected predictors to estimate the probability of ANE onset (Figure 3). The total score was calculated from altered consciousness and the earliest laboratory measurements of AST, LDH, and BUN. The maximum total score of the nomogram was 180 points. By summing the points assigned to each variable, the probability of ANE could be estimated for individual patients.

Figure 3 Nomogram for early prediction of ANE onset risk. ANE, acute necrotizing encephalopathy; AST, aspartate aminotransferase; BUN, blood urea nitrogen; LDH, lactate dehydrogenase.

Evaluation and validation of the nomogram prediction model

The discriminative ability of the nomogram was evaluated using ROC analysis in both the training and validation cohorts (Figure 4A,4B). The AUC was 0.961 [95% confidence interval (CI): 0.929–0.992] in the training set and 0.978 (95% CI: 0.951–1.000) in the validation set.

Figure 4 ROC and PR curves of the nomogram for predicting the risk of ANE onset risk. (A) ROC curve for the training set. (B) ROC curve for the validation set. The X-axis represents 1−specificity, and the Y-axis represents sensitivity. (C) PR curve for the training set. (D) PR curve for the validation set. The X-axis represents recall (sensitivity), and the Y-axis represents precision (positive predictive value). ANE, acute necrotizing encephalopathy; AUC, area under the curve; CI, confidence interval; PR, precision-recall; ROC, receiver operating characteristic.

Given the imbalanced distribution of ANE and non-ANE cases, PR analysis was additionally performed. The AUPRC was 0.792 (95% CI: 0.608–0.932) in the training set and 0.809 (95% CI: 0.466–0.988) in the validation set (Figure 4C,4D).

The calibration curve demonstrated good agreement between the predicted and observed probabilities of ANE onset, with a mean absolute error (MAE) of 0.018 (Figure S2). DCA showed that the nomogram provided a net clinical benefit across a threshold probability range of 0% to 85% (Figure S3A). The clinical impact curve indicated good agreement between the predicted number of high-risk cases and the observed number of ANE cases (Figure S3B).

Discussion

In this retrospective study, we developed and internally validated a nomogram model to predict the risk of pediatric ANE with influenza- or COVID-19-associated neurological manifestations. By integrating clinical characteristics and routinely available laboratory parameters, the final model incorporated four variables: altered consciousness, AST, LDH, and BUN. The nomogram demonstrated favorable discrimination in both the training and validation cohorts, with high AUC and AUPRC values, and exhibited good calibration and potential clinical utility based on DCA and clinical impact curve evaluation. These findings suggest that a concise model incorporating neurological status with systemic biochemical markers could assist clinicians in early identification of children at high risk for ANE.

ANE is a fulminant and life-threatening encephalopathy that predominantly affects children and is often precipitated by viral infections, particularly influenza virus (10). Although rare, ANE carries a substantial clinical burden due to its rapid progression, high mortality, and frequent neurological sequelae in survivors. Evidence indicates that steroid pulse therapy initiated within 24 hours of ANE onset can significantly improve clinical symptoms and prognosis, highlighting the critical need for early identification (11). However, early recognition remains challenging because initial manifestations-including fever, seizures, vomiting, diarrhea, and altered mental status-are nonspecific and overlap considerably with those of other infectious or immune-mediated neurological disorders. In current practice, ANE diagnosis relies heavily on characteristic MRI findings, which typically become apparent only in advanced disease stages, potentially delaying treatment and worsening outcomes (12). Consequently, the concise nomogram model developed here may aid in the early identification of children at high risk for ANE, thereby guiding timely neuroimaging and intervention.

ANE typically progresses through three distinct clinical phases: the prodromal stage, the acute encephalopathic stage, and the recovery stage (13). The acute encephalopathic stage emerges 1–3 days after the prodrome and is characterized by rapid neurological deterioration, manifesting as altered mental status, progressive impairment of consciousness (frequently advancing to coma), seizures, and focal neurological deficits. Altered consciousness reflects acute central nervous system dysfunction and frequently indicates extensive brain parenchymal involvement or rapidly progressive cerebral edema, both core pathological features of severe ANE (14). One study reported that 100% of patients with COVID-19-associated ANE presented with altered consciousness, with an associated mortality rate of 42% (15). As a predictor with substantial weighting in the present nomogram model, altered consciousness facilitates early risk stratification and serves as a valuable indicator of disease severity. This readily assessable bedside sign may therefore help distinguish ANE from milder infection-related neurological conditions, such as febrile seizures.

ANE is not result from direct viral invasion of the central nervous system; instead, its pathogenesis is widely regarded as a post-infectious, cytokine storm-mediated process involving endothelial injury, blood-brain barrier disruption, and multi-organ stress responses (16-18). Hepatic enzyme elevation is an established early risk factor for ANE. Elevated AST levels may reflect hepatic involvement or metabolic disturbance, while LDH acts as a sensitive, though non-specific, marker of tissue hypoxia and cellular necrosis (19). A single-center study from China reported that peripheral blood tests in ANE patients demonstrated mild-to-moderate elevations in ALT and AST, along with marked increases in LDH (20). This pattern aligns with the relatively higher weighting assigned to AST and LDH in our model. Renal impairment, reflected by elevated creatinine and BUN at presentation, has also been documented in ANE patients (10,16,21). Furthermore, BUN elevation may correlate with cerebral edema (22). Our findings confirm that elevated BUN serves as an early predictor of ANE. Integrating both consciousness status and hepatorenal laboratory markers into the nomogram not only aids early diagnosis but also offers insight into disease severity. Notably, all four predictive indicators are routinely available in emergency and inpatient settings, are inexpensive, and provide rapid results. These features allow for intuitive and efficient identification of children at higher risk for ANE, significantly enhancing the model’s utility in resource-limited clinical environments. To our knowledge, this model is among the first clinical scoring systems specifically designed for the early detection of ANE.

This study addresses a distinct and clinically urgent need: the early differentiation of ANE from other neurological complications during acute viral illness, whereas prior research has focused primarily on prognostic factors or established imaging hallmarks. The resulting nomogram is parsimonious and relies on early-phase clinical data. Its development employed rigorous statistical methods, using LASSO regression to mitigate overfitting before logistic regression analysis. Given the class imbalance inherent in this rare disease cohort, model performance was robustly assessed with precision-recall curves. Calibration plots and DCA further confirm the model’s reliability and its potential net benefit across a range of clinical thresholds.

Several limitations warrant consideration. First, the overall sample size was relatively modest due to the rarity of ANE, and the number of ANE cases in the validation set was especially small; therefore, the model should be considered exploratory and requires external validation in a larger multicenter prospective cohort before considering any clinical applications. Second, baseline laboratory data from the precise time of disease onset were missing for some patients because initial blood tests were performed at outside institutions. Third, while various viral pathogens can trigger ANE, this study only included patients with confirmed influenza or COVID-19 infection to enable precise matching of controls given the available clinical sample. Finally, the current model is intended only for early ANE risk identification and does not predict long-term neurological outcomes or mortality. Future studies should aim to develop more comprehensive models that integrate dynamic data such as neuroimaging findings, inflammatory cytokine profiles, genetic susceptibility markers, and therapeutic variables.

Conclusions

In conclusion, we developed a simple, clinically applicable nomogram that incorporates altered consciousness, AST, LDH, and BUN to predict ANE risk in children with viral-associated neurological symptoms. The model demonstrated robust discrimination, calibration, and potential clinical utility in internal validation.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0392/rc

Data Sharing Statement: Available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0392/dss

Peer Review File: Available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0392/prf

Funding: This research was funded by the Soochow Municipal Science and Technology Support Program (No. GSWS2023006).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0392/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. This retrospective study received approval from the Medical Ethics Committee of Children’s Hospital of Soochow University (No. 2024CS103). Written informed consent was obtained from the legal guardians of all patients prior to enrollment.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Qin N, Wang J, Peng X, et al. Pathogenesis and Management of Acute Necrotizing Encephalopathy. Expert Rev Neurother 2023;23:641-50. [Crossref] [PubMed]
2. Mizuguchi M, Abe J, Mikkaichi K, et al. Acute necrotising encephalopathy of childhood: a new syndrome presenting with multifocal, symmetric brain lesions. J Neurol Neurosurg Psychiatry 1995;58:555-61. [Crossref] [PubMed]
3. Palazzo AF, Joseph J, Lim M, et al. Workshop on RanBP2/Nup358 and acute necrotizing encephalopathy. Nucleus 2022;13:154-69. [Crossref] [PubMed]
4. Choudhary I, Kc S, Kumar P, et al. Acute Necrotizing Encephalopathy of Childhood Associated with Human Metapneumovirus. Indian J Pediatr 2024;91:203-4. [Crossref] [PubMed]
5. Wang PY, Yang MT, Liang JS. Acute necrotizing encephalopathy caused by SARS-CoV-2 in a child. Pediatr Neonatol 2022;63:642-4. [Crossref] [PubMed]
6. Khan M, Bhattarai S, Boyce TG, et al. Acute Necrotizing Encephalopathy Associated with Coronavirus Disease 2019 in an Infant. J Pediatr 2022;247:160-2. [Crossref] [PubMed]
7. Wertheimer GSO, Brandão MB, Reis F. COVID-19-related acute necrotizing encephalopathy with new spectroscopy features. Rev Soc Bras Med Trop 2022;55:e02752022. [Crossref] [PubMed]
8. Chatur N, Yea C, Ertl-Wagner B, et al. Outcomes in influenza and RANBP2 mutation-associated acute necrotizing encephalopathy of childhood. Dev Med Child Neurol 2022;64:1008-16. [Crossref] [PubMed]
9. Mizuguchi M, Ichiyama T, Imataka G, et al. Guidelines for the diagnosis and treatment of acute encephalopathy in childhood. Brain Dev 2021;43:2-31. [Crossref] [PubMed]
10. Wu L, Peng H, Jiang Y, et al. Clinical features and imaging manifestations of acute necrotizing encephalopathy in children. Int J Dev Neurosci 2022;82:447-57. [Crossref] [PubMed]
11. Forest C, Laudisi M, Malaventura C, et al. Pediatric recurrent acute necrotizing encephalomyelitis, RANBP2 genotype and Sars-CoV-2 infection: Diagnosis, pathogenesis and targeted treatments from a case study. Eur J Paediatr Neurol 2023;42:117-21. [Crossref] [PubMed]
12. Okumura A, Mizuguchi M, Kidokoro H, et al. Outcome of acute necrotizing encephalopathy in relation to treatment with corticosteroids and gammaglobulin. Brain Dev 2009;31:221-7. [Crossref] [PubMed]
13. Lee VWM, Khoo TB, Teh CM, et al. Factors associated with outcomes of severe acute necrotizing encephalopathy: A multicentre experience in Malaysia. Dev Med Child Neurol 2023;65:1256-63. [Crossref] [PubMed]
14. Yang X, Wei M, Chu S, et al. Adult-onset Still's disease with concurrent acute necrotizing encephalopathy: a case report. BMC Neurol 2022;22:329. [Crossref] [PubMed]
15. Bridwell R, Long B, Gottlieb M. Neurologic complications of COVID-19. Am J Emerg Med 2020;38:1549.e3-7.
16. Li S, Hu D, Li P, et al. Parameters Indicating Development of Influenza-Associated Acute Necrotizing Encephalopathy: Experiences from a Single Center. Med Sci Monit 2021;27:e930688. [Crossref] [PubMed]
17. Bashiri FA, Al Johani S, Hamad MH, et al. Acute Necrotizing Encephalopathy of Childhood: A Multicenter Experience in Saudi Arabia. Front Pediatr 2020;8:526. [Crossref] [PubMed]
18. Ohashi E, Hayakawa I, Murofushi Y, et al. Recurrent acute necrotizing encephalopathy in a boy with RANBP2 mutation and thermolabile CPT2 variant: The first case of ANE1 in Japan. Brain Dev 2021;43:873-8. [Crossref] [PubMed]
19. Lee EP, Lin JJ, Chang HP, et al. Ferritin as an Effective Predictor of Neurological Outcomes in Children With Acute Necrotizing Encephalopathy. Pediatr Neurol 2024;152:162-8. [Crossref] [PubMed]
20. Fang Y, Gao Q, Jin W, et al. Clinical characteristics and prognostic analysis of acute necrotizing encephalopathy of childhood: a retrospective study at a single center in China over 3 years. Front Neurol 2023;14:1308044. [Crossref] [PubMed]
21. Gupta S, Banerjee B, Sasidharan S, et al. Clinical, laboratory, radiologic profile, and outcome in acute necrotizing encephalopathy of childhood (ANEC): a case series. J Pediatr Crit Care 2021;8:192-6.
22. Hu Y, Tian Z, Zhao B, et al. A novel variation in RANBP2 associated with infection-triggered familial acute necrotizing encephalopathy. Neurol Sci 2022;43:3973-7. [Crossref] [PubMed]