A simple, rapid, and cost-effective model for predicting critical influenza a infection in children: a multicentre, retrospective cohort study

Abstract

Background

Owing to the absence of straightforward and scientifically validated screening and evaluation tools for the timely identification, diagnosis, and treatment of critical influenza A infection in children, this study aimed to construct an effective model for the early identification of patients at high risk of progressing to critical influenza A infection.

Methods

The prediction model was developed using the registration data of children diagnosed with influenza A who were admitted to Wuxi Children’s Hospital, the Children’s Hospital Affiliated to Soochow University and the Children’s Hospital Affiliated to Fudan University. Patients were randomly divided into a training group and a validation group at a 7:3 ratio. A logistic regression model was established based on the least absolute shrinkage and selection operator to construct the nomogram. The performance of the nomogram was evaluated by the area under the characteristic curve (AUC), calibration ability, decision curve analysis (DCA) and clinical impact curve analysis (CICA).

Results

A total of 170 hospitalized children with influenza A infection were identified, including 92 severe patients and 78 critical patients. The model was composed of the following five predictors: loss of appetite, seizure ≥ 2 times, altered neutrophil-to-lymphocyte ratios, haemoglobin levels, and total number of complications. The AUC of the model in the training set was 0.905, and the specificity and sensitivity were 91.1% and 77.8%, respectively. The score has been translated into an online risk calculator that is freely available to the public (https://iavchildren.shinyapps.io/DynNomapp/).

Conclusion

This predictive model, which is based on clinical history and commonly used laboratory test values, is valuable for predicting the risk of critical influenza A infection in hospitalized children.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12887-025-06085-7.

Introduction

Influenza A virus has the most frequent variation, and there are epidemics of different extents every year. Every 2 ~ 3 years, important antigenic variants that lead to outbreaks are identified in epidemiology. 45% of hospitalized patients were younger than 18 years old [1]. In some epidemic seasons, the annual infection rate of influenza in children was approximately 50% [2, 3]. A study on influenza infection rates in children and adults in the 2018–19 epidemic season in northern China revealed that the overall infection rate of seasonal influenza in children was 31%, and the infection rate of subtype A (H3N2) was 17%, both of which were significantly higher than those in adults (21% and 10%, respectively) [4]. In November 2022, the influenza-associated hospitalization rate among children < 5 years of age was 12.6 per 100 000, which is comparable to the peak levels typically observed in high-severity seasons [5]. Higher rates of influenza-related hospitalization also lead to a more serious economic burden, with costs 10 times greater than those of outpatient departments [6–8]. Although influenza A virus causes simple febrile respiratory tract infection, it can cause disorders affecting multiple organs, such as the lungs, heart, brain, liver, and muscle, which can lead to life-threatening viral or secondary bacterial pneumonia [9]. The related critical illness and death present health hazards and demand medical resources such as beds, ventilators, and intensive care units (ICUs).

During the 2022–2023 season, which began earlier than usual in several regions, more than 97.6% of the influenza viruses identified were influenza A subtypes, predominantly H3N2 (71.9%). This season exhibited an unusually high disease burden among children, marked by elevated hospitalization rates [10]. The rate of admission to the paediatric intensive unit (PICU) after influenza infection was between 10% and 26%. The mortality rate of patients in the PICU was reported to be 10.5%−47% [11]. Critically ill children often need intensive medical resources. Therefore, early identification of high-risk children who may progress to critical influenza A infection will help provide the resources needed for appropriate care for those at high risk and reduce their modality while not exhausting medical resources for those at low risk.

Given these challenges, our study introduced a novel predictive model designed to identify children at high risk of progressing to critical influenza A infection at an early stage. This model is notable because of its incorporation of dynamic clinical markers such as decreased appetite, frequent seizures, altered neutrophil-to-lymphocyte ratios (NLRs), and haemoglobin levels, in addition to a tally of complications. By accurately forecasting the likelihood of severe outcomes, this model not only aims to guide more tailored and effective interventions but also has the potential to alleviate the economic and logistical strains on healthcare systems by reducing unnecessary or inappropriate care. Our findings, which are based on multicentre clinical data and reliable statistical analysis, have been validated across different cohorts. These discoveries are expected to advance the field of paediatric influenza management and open new pathways for improving patient treatment outcomes and optimizing resource allocation in healthcare settings.

Materials and methods

Study design and participants

This study was performed in accordance with the Declaration of Helsinki, with approvals of the Medical Ethics Board of Wuxi Children’s Hospital (WXCH2022-12-105), Children’s Hospital of Soochow University (2023CS034), Children’s Hospital of Fudan University in Shanghai [(2021) 431]. This study was registered in the National Medical Research Registration (registration number: MR-32-24-023405). All the methods used in this study were conducted in accordance with the Declaration of Helsinki. The parents or legal guardians of all the children enrolled in the study provided informed consent for their child.

This retrospective study was conducted in three tertiary hospitals, Wuxi Children’s Hospital, Children’s Hospital of Soochow University and Children’s Hospital of Fudan University in Shanghai. The study population consisted of children younger than 18 years of age who were admitted to hospitals between February 2023 and February 2024 with clinically and laboratory-confirmed influenza A virus infections; outpatients, emergency hospitalizations for less than 2 h, and patients with incomplete documentation of clinical outcomes were excluded. The main outcome was severe/critical influenza A virus infection. The patients were divided into two groups based on the severity of influenza A virus infection (severe and critical), defined according to the “Influenza Diagnosis and Treatment Program (2020 edition) [12]”. The following criteria are used for inclusion in the severe group, with any one criterion sufficient: 1) Persistent high fever (> 3 days) with severe cough, purulent or blood-tinged sputum, or chest pain. 2) Rapid breathing, difficulty breathing, and cyanosis of lips or mucous membranes. 3) Altered mental status (e.g., lethargy, agitation, or seizures). 4) Severe vomiting, diarrhea, and dehydration. 5)Concurrent pneumonia. 6) Significant worsening of pre-existing conditions. 7) Other clinical conditions requiring hospitalization.For inclusion in the critical group, meeting any one of the following criteria is sufficient: 1) Respiratory failure with acute necrotizing encephalopathy. 2) Acute necrotizing encephalopathy. 3) Shock. 4) Multiple organ dysfunction. 5) Other severe conditions requiring intensive monitoring and treatment. 6) Other clinical conditions requiring hospitalization. A professional clinical team reviewed, extracted, and crosschecked the data, and two clinicians independently checked the recorded results. This study was conducted in accordance with the guidelines of the Declaration of Helsinki and the principles of good clinical practice. The study was approved by the medical ethics committees of the participating hospitals.

Data collection and definition

The data were extracted from electronic medical records and microbiological laboratory reports and included demographic characteristics, prehospital clinical manifestations, underlying diseases, prehospital laboratory test results (white blood cell count, percentage of neutrophils, absolute value of neutrophils, percentage of lymphocytes, C-reactive protein [CRP], procalcitonin [PCT]), complications within 24 h after admission, time from onset to treatment, and anti-influenza treatment performed before admission. Children with underlying diseases were defined as those with (but not limited to) one or more of the following underlying medical conditions: autoimmune disease; immunosuppressant use; connective tissue disease; tumours; haematological, renal, or cardiovascular diseases (except hypertension); endocrine and metabolic diseases; neurologic diseases (including moderate to severe developmental delay); bronchial foreign body; congenital bronchopulmonary dysplasia; bronchomalacia; asthma; and tuberculosis. The complications included respiratory complications, cardiac complications, gastrointestinal complications, musculoskeletal complications, neurological complications, renal complications, sepsis, and septic shock [9].

Variable selection and model construction

The baseline characteristics are described first. The normality of the distribution was tested with the Kolmogorov–Smirnov test. Normally distributed data are presented as the mean ± standard deviation (SD), and nonnormally distributed data are expressed as the median and interquartile range (IQR). The qualitative data are expressed as percentages. Differences were compared between the training and validation sets. The mean values of the two sets were compared by independent-sample t tests for quantitative data with a normal distribution and the nonparametric independent-sample Mann–Whitney test for quantitative data without a normal distribution. The chi-square test was used to compare rates between two groups. For a small amount of missing clinical data, the missing values were imputed using the expectation‒maximization method in SPSS statistical software, version 26 (IBM SPSS, Chicago, Illinois).

To improve the reliability of our model and prevent data overinterpretation, we randomly divided the 170 subjects into a training set and a validation set at a ratio of 7:3 without replacement. The training set was used for baseline variable selection and nomogram development. Univariate logistic regression was used to identify the factors associated with critical influenza A infection in the training set. In addition, to minimize the potential collinearity between variables and model overfitting, least absolute shrinkage and selection operator (LASSO) regression and 10-fold cross-validation were used to select predictors. A logistic regression model was established by using LASSO regression analysis to identify variables.

The model constructed is not unique. Models obtained using only statistical methods often ignore important clinical indicators. Therefore, on the basis of examples from the literature, we can subjectively add a specific indicator to obtain a new model; calculate the Akaike information criterion (AIC), log likelihood, and C index; and compare them with clinical significance to determine the final predictors [10]. We reported adjusted β coefficients for each predictor with standard errors (SEs) and corresponding odds ratios (ORs) and 95% confidence intervals (CIs). Finally, the nomogram was drawn. A nomogram sets different scores for each index included in the model according to the weight ratio and numerical size of the model, and the prediction probability is calculated after combining the scores. The final nomogram is the early prediction scale.

Model validation

The ability of the nomogram and risk factors to predict critical illness in children with influenza A were quantified using receiver operating characteristic (ROC) curve analysis in the training set and verified in the validation set. The calibration of the nomogram was evaluated by a calibration curve, and the Hosmer-Lemeshow test was performed to evaluate the goodness of fit. Decision curve analysis (DCA) can be used to estimate the net benefit of a model on the basis of the difference between true-positive and false-positive results, weighted by the probability of the selected threshold risk probability. If the predictive model’s net profit curve is higher than that of “full treatment” or “no treatment” within a reasonable risk threshold, then the model has clinical utility. On this basis, we performed clinical impact curve analysis (CICA) on the model. CICA will determine the estimated number of high-risk people for each risk threshold and visually display the proportion of cases (true positives). Statistical analysis was carried out using SPSS 26.0, and R software (version 4.4.1) was used to construct all the graphics using the following R packages: “Hmisc”, “rms”, “ggplot2”, “pROC”, “car”, ”openxlsx”, “gtsummary”, “tidyverse”, “Table”, “glmnet” and other packages. A two-sided α less than 0.05 was considered statistically significant.

Results

Baseline characteristics of patients

According to the inclusion and exclusion criteria, a total of 170 children with influenza A virus infections were identified, including 92 severely ill children and 78 critically ill children. The median age of those in the severe group was 4.31 years (range, 0.17–15 years). The median age in the critical group was 5.53 years (range, 0.58-15 years), as shown in Table 1. There were no significant differences in sex between the severe group and the critical group. Compared with severely ill children, critically ill children had more complications and a shorter time from symptom onset to medical treatment (median days, 1.66 vs. 2.63, p < 0.001). There were also significant differences in clinical symptoms; inspiratory triconcave signs; pathological reflexes; capillary refilling times; NLRs; and haemoglobin, platelet, CRP, and PCT levels (p < 0.05) (Table 1). The original data were divided into a training set and a validation set using 7:3 nonrepetitive random sampling (Fig. 1). There were no significant differences in age, sex, or disease type between the training and validation cohorts (Table 2). The clinical characteristics and laboratory findings of the training cohort are summarized in Table 3.

Table 1.

Baseline of children

	Severe (N = 92)	Critical (N = 78)	P value1
Demographics
Age, y, Median (IQR)	4.31 (2.00,6.00)	5.53 (3.25,7.75)	0.004
Age < 2	17/92 (18%)	7/78 (9.0%)	0.076
Age < 5	52/92 (57%)	30/78 (38%)	0.019
Gender			0.51
Female	40/92(43%)	30/78 (38%)
Male	52/92 (57%)	48/78 (62%)
Basic disease	10/92 (11%)	12/78 (15%)	0.38
Premature birth	1/92 (1.1%)	0/78 (0%)	> 0.99
Chronic lung disease	2/92 (2.2%)	0/78 (0%)	0.5
cardiovascular diseases	0/92 (0%)	2/78 (2.6%)	0.21
Digestive system disease	0/92 (0%)	2/78 (2.6%)	0.21
Neuro/neuromuscular diseases	5/92 (5.4%)	5/78 (6.4%)	> 0.99
Hereditary metabolic diseases	0/92 (0%)	3/78 (3.8%)	0.095
Developmental delay/precocity	0/92 (0%)	5/78 (6.4%)	0.019
Low immune function (tumor,	0/92 (0%)	3/78 (3.8%)	0.095
blood tumor disease, transplantation,
HIV, etc.)
Othera	2/92 (2.2%)	4/78 (5.1%)	0.41
Signs and symptoms
Fever	86/92 (93%)	77/78 (99%)	0.13
Highest temperature,◦C, Median (IQR)	39.2 (39.0,40.0)	39.7 (39.0,40.2)	0.009
Fever duration, d, Median (IQR)	2.92 (1.00,4.00)	2.39 (1.00,3.00)	0.067
Cough	79/92 (86%)	52/78 (67%)	0.003
Myalgia or fatigue	15/92 (16%)	1/78 (1.3%)	< 0.001
Vomiting	29/92 (32%)	27/78 (35%)	0.67
bellyache	9/92 (9.8%)	6/78 (7.7%)	0.63
diarrhoea	8/92 (8.7%)	12/78 (15%)	0.18
Loss of appetite	18/92 (20%)	42/78 (54%)	< 0.001
dizzy	7/92 (7.6%)	6/78 (7.7%)	0.98
headache	9/92 (9.8%)	6/78 (7.7%)	0.63
Blurred vision	0/92 (0%)	0/78 (0%)
seizure ≥ 2 times	4/92 (4.3%)	28/78 (36%)	< 0.001
Oliguria	1/92 (1.1%)	3/78 (3.8%)	0.33
Alimentary tract hemorrhage	0/92 (0%)	2/78 (2.6%)	0.21
Tachypneab	31/92 (34%)	45/78 (58%)	0.002
Increased heart rateb	42/92 (46%)	62/78 (79%)	< 0.001
Hypotension	0/92 (0%)	2/78 (2.6%)	0.21
Diminished breath sound	1/92 (1.1%)	4/78 (5.1%)	0.18
Inspiratory tri-concave sign	4/92 (4.3%)	10/78 (13%)	0.045
Flaring of nares	0/92 (0%)	4/78 (5.1%)	0.042
	Severe(N = 92)	Critical(N = 78)	P value1
Signs and symptoms
Nodding breathing	0/92 (0%)	0/78 (0%)
Meningeal irritation sign	2/92 (2.2%)	6/78 (7.7%)	0.14
Pathological reflex	0/92 (0%)	12/78 (15%)	< 0.001
CRT, s			< 0.001
2	92/92 (100%)	67/78 (86%)
3	0/92 (0%)	8/78 (10%)
4	0/92 (0%)	2/78 (2.6%)
5	0/92 (0%)	1/78 (1.3%)
Laboratory test
WBC (× 109/L), Median (IQR)	7.5 (4.7,9.1)	8.3 (5.2,11.3)	0.24
N%,Median (IQR)	59 (47,73)	70 (61,83)	< 0.001
L%,Median (IQR)	31 (17,41)	21 (11,29)	< 0.001
NLR, Median (IQR)	3.3 (1.2,4.4)	5.8 (2.1,7.2)	< 0.001
Hb(g/L), Median (IQR)	127 (121,133)	123 (116,129)	0.016
PLT (× 109/L), Median (IQR)	235 (166,264)	208 (143,265)	0.11
CRP(mg/L), Median (IQR)	12 (1,10)	23 (3,21)	0.003
PCT(ng/ml), Median (IQR)	0.3 (0.0,0.5)	7.3 (0.5,10.2)	< 0.001
Complication
Pneumonia	30/92 (33%)	46/78 (59%)	< 0.001
ARDS	0/92 (0%)	2/78 (2.6%)	0.21
Myositis and rhabdomyolysis	16/92 (17%)	2/78 (2.6%)	0.002
Neuropsychiatric events	11/92 (12%)	43/78 (55%)	< 0.001
Sepsis	2/92 (2.2%)	7/78 (9.0%)	0.082
Sepsis shock	0/92 (0%)	2/78 (2.6%)	0.21
Other complicationsc			< 0.001
0	91/92 (99%)	61/78 (78%)
1	1/92 (1.1%)	11/78 (14%)
2	0/92 (0%)	4/78 (5.1%)
3	0/92 (0%)	1/78 (1.3%)
5	0/92 (0%)	1/78 (1.3%)
Total number of complications			< 0.001
0	34/92 (37%)	13/78 (17%)
1	56/92 (61%)	23/78 (29%)
2	2/92 (2.2%)	27/78 (35%)
3	0/92 (0%)	5/78 (6.4%)
4	0/92 (0%)	4/78 (5.1%)
5	0/92 (0%)	5/78 (6.4%)
7	0/92 (0%)	1/78 (1.3%)
Interval between symptom onset to medical service, days, d, Median (IQR)	2.63 (1.00,4.00)	1.66 (1.00,2.00)	0.034
Antiviral agent (within 48 h)	39/92 (42%)	34/78 (44%)	0.88

1 Pearson’s Chi-squared test; Mann-Whitney U test; Fisher’s exact test

Abbreviations CRT capillary refill time, WBC white blood cell, NLR neutrophil-lymphocyte ratio, Hb Haemoglobin, PLT Platelets, CRP C-reactive protein, PCT procalcitonin, ARDS Acute respiratory distresssyndrome

a Others include mental illness, systemic lupus erythematosus

b Tachypnea adjusted for age; Increased heart rate adjusted for age

c Gastrointestinal complications,eurological complications, renal complications, etc [9] 

Fig. 1.

Flow chart of study participants in the training and validation groups

Table 2.

Baseline Characteristics of the Study Cohort

Characteristic	Training Cohort	Validation Cohort	P Value
	(n = 119)	(n = 51)
Age, mean ± SD, y	4.82 ± 3.19	4.97 ± 2.56	0.54
Gender			0.86
Female	50 (42%)	20 (39.2%)
Male	69 (58%)	31 (60.8%)
Basic disease			0.96
No	103 (86.6%)	50 (98%)
Yes a	16 (13.4%)	6 (11.8%)
Disease type			0.76
Severe	63 (52.9%)	29 (56.9%)
Critical	56 (47.1%)	22 (43.1%)

^aPatients with 1 of the following: Chronic lung disease, cardiovascular diseases, Digestive system disease, Neuro/neuromuscular diseases, Hereditary metabolic diseases, Developmental delay/precocity, Low immune function (tumor, blood tumor disease, transplantation, HIV, etc.), mental illness, systemic lupus erythematosus, etc

Table 3.

Univariate logistic regression for predicting critical IAV-associated patients in the training cohort

Variables	Severe(N = 63)	Critical(N = 56)	Univariate analysis OR (95% CI)	P value
Demographics
Age, y, Median (IQR)	4.29 (2.00,6.00)	5.43 (3.75,8.00)	1.12(1-1.27)	0.054
Age < 2	14/63 (22%)	7/56 (13%)	0.5(0.18–1.31)	0.17
Age < 5	37/63 (59%)	24/56 (43%)	0.53(0.25–1.09)	0.085
Gender			1.08(0.52–2.24)	0.844
Female	27/63(43%)	23/56(41%)
Male	36/63 (57%)	33/56 (59%)
Basic disease	7/63 (11%)	9/56 (16%)	1.53(0.53–4.59)	0.431
Signs and symptoms
Fever	59/63 (94%)	55/56 (98%)	3.73(0.53–74.15)	0.246
Highest temperature,◦C, Median (IQR)	39.2(39.0,39.8)	39.7 (39.0,40.0)	1.54(1.04–2.38)	0.04
Cough	54/63 (86%)	33/56 (59%)	0.24(0.09–0.56)	0.002
Myalgia or fatigue	10/63 (16%)	1/56 (1.8%)	0.1(0.01–0.53)	0.028
Loss of appetite	13/63 (21%)	31/56 (55%)	4.77(2.17–10.97)	0
seizure ≥ 2 times	2/63 (3.2%)	20/56 (36%)	16.94(4.59–110.1)	0
Tachypnea	23/63 (37%)	31/56 (55%)	2.16(1.04–4.55)	0.041
Increased heart rate	31/63 (49%)	47/56 (84%)	5.39(2.34–13.45)	0
Laboratory test
WBC (× 109/L), Median (IQR)	7.5 (4.3,9.2)	8.5 (5.2,11.9)	1.04(0.97–1.12)	0.286
N%,Median (IQR)	58 (45,74)	70 (61,84)	1.04(1.02–1.06)	0.001
L%,Median (IQR)	32 (16,44)	22 (11,29)	0.96(0.94–0.99)	0.003
NLR, Median (IQR)	3.2 (1.1,4.7)	5.8 (2.0,7.3)	1.17(1.06–1.31)	0.004
Hb(g/L), Median (IQR)	127 (122,133)	121 (116,129)	0.95(0.92–0.99)	0.009
PLT (× 109/L), Median (IQR)	247 (170,277)	210 (145,262)	1.00(0.99-1.00)	0.082
CRP(mg/L), Median (IQR)	14 (1,13)	20 (3,17)	1.01(0.99–1.02)	0.242
PCT(ng/ml), Median (IQR)	0.3 (0.0,0.5)	5.7 (0.2,9.9)	3.1(1.75–7.84)	0.003
Complication
Pneumonia	20/63 (32%)	29/56 (52%)	2.31(1.1–4.93)	0.028
Myositis and rhabdomyolysis	10/63 (16%)	2/56 (3.6%)	0.2(0.03–0.79)	0.041
Neuropsychiatric events	6/63 (9.5%)	29/56 (52%)	10.2(4.01–29.92)	0
Total number of complications			3.46(2.09–6.31)	0
0	26/63 (41%)	12/56 (21%)
1	35/63 (56%)	13/56 (23%)
2	2/63 (3.2%)	19/56 (34%)
3	0/63 (0%)	5/56 (8.9%)
4	0/63 (0%)	2/56 (3.6%)
5	0/63 (0%)	4/56 (7.1%)
7	0/63 (0%)	1/56 (1.8%)
Interval between symptom onset to	2.59 (1.0,4.0)	1.70 (1.0,2.0)	0.75(0.57–0.94)	0.02
medical service, days, d,Median (IQR)

Abbreviations:WBC white blood cell, HbHaemoglobin, PLT Platelets, CRP C-reactive protein, PCT procalcitonin

Risk factor screening and model construction

A total of 59 features were collected from each patient in the training cohort, and 37 insignificant factors were eliminated through univariate logistic regression analysis (Table 3). The remaining 22 independent variables were used in the LASSO regression analysis; 14 variables with the smallest mean square error values were selected (Fig. 2 A and B). Owing to the large number of variables and the relatively small number of cases, according to Harrell’s quasi-test, we finally selected five independent variables for multivariate logistic regression analysis and modelling. The optimal model was selected based on the AIC, log-likelihood, and C-index and combined with clinical significance; the final model consisted of the following five predictors: loss of appetite, ≥ 2 seizures, NLR, haemoglobin, and total number of complications (Table 4 and Table S1). The predictive nomogram for the risk of becoming critically ill with influenza A infection generated using the five features selected from the training cohort is shown in Fig. 2C.

Fig. 2.

Construction of prediction nomogram in children with critical influenza a. The nomogram is composed of loss of appetite, seizure ≥ 2 times, neutrophil-lymphocyte ratio(NLR), Haemoglobin (Hb), and total number of complications. A Least absolute shrinkage and selection operator (LASSO) coefficient profiles (y-axis) of the 22 features. Variables selection: LASSO multiple logistic regression model; coefficient distribution map construction: the log (λ) sequence. The top x-axis shows the average number of predictors. B Identification of the optimal penalization coefficient (λ) in the LASSO model was performed via 10-fold cross-validation based on minimum criteria. C Nomogram predicting the risk of critically ill children with influenza A. To use this nomogram in clinical management, an individual patient’s value is located on each variable axis, and a line is plotted upward to calculate the number of points received for each variable value. The sum of these scores is located on the total points axis and draws a line straight down to find the probability of critical IAV. D Decision curve compares the net clinical benefits of 3 scenarios in predicting the probability of critical IAV: a perfect prediction model (gray line), screen none (horizontal solid black line), and screen based on the nomogram (red line). E The clinical impact curve of the nomogram plotted the number of children with influenza A classified as high-risk at each high-risk threshold, and the number of children classified as high-risk and critically ill with influenza A

Table 4.

Final multivariable model predicting severe/critical IAV

Variables	β(SE)	Wald Z	OR(95%CI)	P Value
Loss of appetite	0.5658	2.71	4.641(1.531–14.069)	0.0067
seizure ≥ 2 times	0.8926	3.43	21.359(3.713–122.860)	0.0006
NLR	0.0645	1.41	1.095(0.965–1.243)	0.1588
Hb(g/L)	0.0249	−2.45	0.941(0.896–0.988)	0.0141
Total number of complications	0.3542	3.65	3.645(1.820–7.297)	0.0003

SE standard error, OR odds ratio, CI confidence interval

An online calculator based on our predictive model was developed in which clinicians enter the values of the 5 variables required for the risk score, and an automatic calculation of the likelihood (with 95% CIs) that a hospitalizedchild with IAV will develop critical illness is performed (https://iavchildren.shinyapps.io/DynNomapp/) (Fig. 3 A and B).

Fig. 3.

Calibration curve and receiver operating characteristic curve for performance to distinguish between critically ill and severe influenza A children in the training cohort A and C and validation cohort B and D respectively. Abbreviation: AUC, area under the receiver operating characteristic curve

Model verification

To evaluate the clinical applicability of our risk prediction nomogram, DCA and CICA were performed. The DCA and CICA results visually revealed that the nomogram had a superior overall net benefit within wide and practical ranges of threshold probabilities and impacted patient outcomes (Fig. 2D and E). The calibration curve of the nomogram for the risk prediction of critically ill patients with influenza A infection revealed that there was good agreement between the predictions and actual observations in the training cohort and the validation cohort. The Hosmer-Lemeshow test revealed that there was no significant difference between the training set and the validation set (training set, p = 0.938; validation set, p = 0.946), which indicates that the diagnostic accuracy of the nomogram is high (Fig. 4).

Fig. 4.

The Online Web-Based Calculatora for Predicting Critical Illness Among Children With IAV

In the training cohort, the nomogram had a high AUC of 0.905 (95% CI, 0.849–0.960) in distinguishing children with severe influenza A infection, with a sensitivity of 91.1% and a specificity of 77.8%. Consistent with the training cohort, the AUC of the nomogram in the validation cohort was 0.868, and the specificity and sensitivity were 77.3% and 86.2%, respectively (Figs. 3C and D).

Discussion

Early identification of children with influenza A infection at risk of progression to critical disease is crucial for improving the management and optimal use of medical resources. In this study, we found that a decreased appetite, ≥ 2 seizures, increased NLR, decreased haemoglobin, and increased number of complications were associated with a greater risk of critical influenza A infection. Moreover, we developed a highly effective risk prediction nomogram incorporating these five features, which demonstrated significantly high sensitivity and specificity in distinguishing individuals with critical influenza A infection from those with severe infection. DCA and CICA confirmed that our nomogram provides significant net clinical benefits, making it a valuable tool for the individualized and precise assessment of critical risk from influenza A in children.

Previously, it was reported that most very serious influenza cases, that is, cases that may result in admission to the ICU or death, occur mainly in preschool-aged and school-aged children [11, 13]. Our findings are similar to those reported in the literature. In our training set, the median age of children with severe influenza A infection was 4.29 years, and the median age of children with critical influenza A infection was 5.43 years. Fever is the primary manifestation of the body’s response to pathogen invasion, and the peak temperature can reflect the severity of infection in children. In this study, the peak temperature before admission in the critical group was greater than that in the severe group (39.7 °C > 39.2 °C, P = 0.04). Additionally, our research revealed that the proportion of patients with cough, myalgia or fatigue symptoms was significantly lower in the critically ill group than in the severe group, suggesting that these symptoms are not indicators of disease severity. Studies have shown that the respiratory rate is more reliable than oxygen saturation in predicting the severity of influenza in children [14]. Unfortunately, we did not include the prehospital oxygen saturation index in this study. We observed that the presence of shortness of breath and an increased heart rate were more prevalent in critically ill children than in those with severe illness, and these factors were identified as risk factors for critical illness.

Interestingly, the presence of non-respiratory symptoms was associated with the severity of influenza A infection, potentially indicating systemic involvement. For example, gastrointestinal symptoms are an early risk factor for severe disease [13]. Our study revealed that children who experienced a decreased appetite were at increased risk of developing critical influenza A infection. Diet is a very important part of our daily life, and loss of appetite is often a warning of physical condition. Although flu-related neurological symptoms are less common than respiratory symptoms in children, they have been increasingly reported and are now recognized as the third leading cause of hospitalization after respiratory and systemic symptoms [14]. Studies have reported that neurological complications occur in 7–11% of hospitalized patients [15–17]. Although most neurological complications are temporary, permanent sequelae and death are not rare [18–20]. Other studies have suggested that PICU admissions and mortality are related to neurologic events [21]. All children who are admitted with neurological findings, especially during the influenza season, should be evaluated for influenza-associated neurological complications even if their respiratory complaints are mild or nonexistent [22].

The presence of multiple complications significantly increases the risk of critical influenza A infection. The frequency of complications in the respiratory, cardiovascular, gastrointestinal, or nervous systems or in kidneys, as well as septic shock, was notably greater in the critically ill group. Our study revealed that pneumonia and neuropsychiatric events were the main complications in critically ill children with influenza A infection, with incidence rates exceeding 50%. Additionally, while most children in the severe group presented with 0–1 complications, those in the critically ill group had at least 2 complications. Moreover, several critically ill children exhibited multiple organ dysfunction, leading to a poor prognosis.

Whole blood cell counts are part of a routine laboratory examination; the haemoglobin (Hb) level is an important value that is associated with malnutrition and immune response inhibition. A low haemoglobin level is a significant indicator of potential inflammatory processes. While studies are limited, some evidence suggests that higher haemoglobin and platelet levels are protective factors [23]. Conversely, low haemoglobin levels or anaemia are linked to severe outcomes, including respiratory failure, ICU admission, mechanical ventilation, and increased risk of mortality [24]. This haemoglobin effect may reflect the severity of the disease, underlying complications or malnutrition [24]. Our study corroborates this finding, revealing that haemoglobin levels in the critically ill group were lower than those in the severe group.

Blood count results allow us to calculate various ratios that indicate ongoing inflammatory processes in children, one of which is the NLR. Although the NLR did not reach statistical significance in our multivariate logistic regression analysis, it was significant in both the univariate regression and LASSO regression analyses. Furthermore, clinical studies have consistently demonstrated the sensitivity of the NLR in the diagnosis and stratification of systemic infections, sepsis, and bacteraemia, as well as its robust predictive and prognostic value [25–27]. Therefore, we included this metric in our model.

Dynamic changes in the NLR occur several hours before the clinical state, serving as an early warning for clinicians of ongoing pathological processes [28]. In patients with severe symptoms, the number of neutrophils increases [29], whereas the number of lymphocytes decreases [30]. The NLR has been reported to be a very rapid and valid immunological marker of ongoing inflammation, infection, tissue injury, stress response to a disease, and organ dysfunction and reflects disease severity [28]. Our retrospective study revealed that the NLR was greater in the critical group than in the severe group (5.2 vs. 3.8, P = 0.004). The proportion of children with an NLR above 7 was significantly greater in the critical group than in the severe group (29% vs. 11%). Previous studies have also revealed that severe influenza A virus infection increases the number of circulating neutrophils. Neutrophils, one of the earliest types of immune cells to infiltrate infection sites, mediate virus clearance by phagocytosing dead or dying virus-infected cells, releasing granular media, and producing neutrophil extracellular traps to immobilize virus particles [31]. Zhu et al. reported that high levels of neutrophil extracellular traps were associated with poor prognosis in patients with severe influenza A infection [32]. Studies have also shown that lymphocytes undergo apoptosis through Fas-Fas ligand signalling after exposure to influenza A virus [33]. B Liu et al. reported that infection with the influenza A (H1N1) virus pdm09 triggered thymocyte apoptosis and thymic atrophy [34]. During the 2009 pandemic, studies showed that the level of specific CD8 + T cells was negatively correlated with the severity of disease caused by infection with the H1N1 pdm09 virus [35]. These results suggest that the progression of influenza A virus infection is related to an imbalance in neutrophil and lymphocyte levels and that the NLR may be a potential biomarker.

CRP levels have been consistently shown to be significantly greater in H1N1 influenza patients who develop severe disease. Denitsa Vasileva et al. demonstrated that serum CRP levels can be employed—in combination with other biomarkers—to predict the complications of H1N1 influenza [36]. According to another report, ferritin levels ≥ 625 ng/mL and PCT values ≥ 0.275 ng/mL were also effective in determining severity and mortality, with negative predictive values of 93.7% and 98.2%, respectively [37]. At present, there are few clear strategies to determine which H1N1 influenza patients are likely to have severe outcomes. Although some studies on the development of predictive models for severe influenza infection in children have been conducted, most are single-centre retrospective studies, and potential selection bias cannot be avoided. There is a lack of screening tools for the early identification of critical influenza A infection in children. To our knowledge, this study is the first to develop a nomogram that uses simple laboratory indicators combined with risk factors for clinical symptoms to establish a clinical prediction model that can be quickly applied in clinical practice.

Compared with previous studies, this study has several advantages. First, our risk model incorporates variables that are readily available in outpatient and emergency departments, providing a practical and cost-effective quantitative prediction tool that relies on only five features that are relatively easy to obtain from inexpensive routine blood tests. Second, to increase the robustness of our findings, we included sample size data from three centres and performed internal verification. The performance of our nomogram has been shown to be effective for clinical application. This nomogram combines early clinical and haematological indicators to facilitate the prediction of critical influenza A infection occurrence in children, providing a visual and intuitive tool to aid in the early identification and screening of critical cases.

There are several limitations in this study. First, the number of cases is not large, which in principle may render our calculation results, including the regression model, insufficiently robust. Although our sample size is small, the correlation P values of many comparisons are small, which offsets this deficiency. The small sample size could also obscure other important associations due to the possibility of Type II errors. Second, despite incorporating samples from three centres, the data from the training and validation sets were derived from retrospective studies, which may lead to additional bias. Consequently, further research based on an independent prospective cohort is warranted. In the practical application of the model, it is necessary to be aware of overfitting. Third, regional differences in admission thresholds and hospitalization management may influence the results. We also acknowledge that rapid mutations in the influenza A virus could affect the performance of the model. Therefore, we recommend constantly updating these models for better performance.

In conclusion, the newly established predictive model, which utilizes clinical history and routine laboratory test values, has demonstrated significant potential in predicting the risk of severe or critical influenza A infection in hospitalized children. Given its promising performance, this model warrants validation across diverse healthcare settings to ascertain its broader applicability. If effective, it could serve as a crucial tool for health facility managers, potentially alleviating medical resource shortages and reducing mortality rates among this vulnerable population.

Abbreviations

IAV

Influenza A virus

CRT

Capillary refill time

WBC

White blood cell

NLR

Neutrophil-lymphocyte ratio

Hb

Haemoglobin

PLT

Platelets

CRP

C-reactive protein

PCT

Procalcitonin

ARDS

Acute respiratory distress syndrome

ICUs

Intensive Care Units

PICU

Paediatric Intensive Unit

NLRs

Neutrophil-to-Lymphocyte Ratios

SEs

Standard Errors

ORs

Odds Ratios

CIs

Confidence Intervals

ROC

Receiver Operating Characteristic

SD

Standard deviation

IQR

Interquartile range

CI

Confidence interval

LASSO

Least absolute shrinkage and selection operator

AUC

Area under the receiver operating characteristic curve

DCA

Decision curve analysis

CIC

Clinical impact curve analysis

Authors’ contributions

PZ, JQ and LL designed the study. SX, LJ, YW, WC, and ZB collected the epidemiological and clinical data. SX summarized all the data. YG, and SX analyzed and drafted the manuscript. All authors revised the final manuscript. The authors read and approved the final manuscript.

Funding

This paper was supported by the Wuxi Taihu Lake Talent Plan (Grant No.DJTD202304), the Natural Science Foundation of Jiangsu Province (Grants No BK20230189), Wuxi Municipal Bureau on Science and Technology (grant number: K20221033), the Top Talent Support Program for young and middle-aged people of Wuxi Health Committee (grant number: BJ2023089).

Data availability

The data that support the findings of this study are available from the corresponding authors upon reasonable request.

Declarations

Ethics approval and consent to participate

This study was approved by the Medical Ethics Committee of Wuxi Children’s Hospital (no. WXCH2022-12-105) and registered in the Chinese Clinical Trial Registry (registration number: MR-32-24-023405). All the methods used in this study were conducted following the Declaration of Helsinki. All children enrolled in the study provided informed consent from their parents or legal guardians.

Consent for publication

No individual participant data is reported that would require consent to publish from the participant.

Competing interests

The authors declare no competing interests.