Risk factors for invasive mechanical ventilation, extracorporeal membrane oxygenation, and mortality in children with severe adenovirus infection in the pediatric intensive care unit: a retrospective study

Abstract

Background

Adenovirus infection causes considerable morbidity and mortality in pediatric patients, primarily those affected by severe respiratory system involvement. Although prevalent, it often presents vague indications, making accurate diagnosis and management challenging. This study aims to set some risk factors for invasive mechanical ventilation, ECMO, and mortality in children with severe adenovirus infection admitted to PICU.

Methods

We evaluated 66 children with severe adenovirus infection admitted to the PICU of Children’s Hospital, Zhejiang University School of Medicine, from 2018 to 2019. Data on general conditions, clinical manifestations, laboratory findings, pathogenetic and radiological discoveries, treatments, therapeutic efficacy, and outcomes were collected. Machine learning models were used to predict the need for invasive mechanical ventilation, ECMO, and mortality.

Results

Of the 66 patients, 5 died, and 61 survived. Significant factors related to mortality included heart failure (p = 0.005), pericardial effusion (p = 0.032), septic shock (p = 0.009), hemoglobin levels (p = 0.013), lactate dehydrogenase (p = 0.022), albumin (p = 0.035), normal creatinine levels (p = 0.037), and pneumothorax (p = 0.002). Additional risk factors for invasive mechanical ventilation included acute respiratory distress syndrome and encephalopathy. Low breath sounds were identified as a risk factor for ECMO. For predicting poor outcomes, including invasive mechanical ventilation, ECMO, or mortality, the random forest model using these factors demonstrated high accuracy, with an area under the curve of 0.968.

Conclusions

The study indicates poor prognosis in children with severe adenovirus infection is significantly related to comorbidities and clinical symptoms. Machine learning models can accurately predict adverse outcomes, providing valuable insights for management and treatment. Identifying high-risk patients using these models can improve clinical outcomes by guiding timely and appropriate interventions.

Trial registration

The article is a retrospective study without a clinical trial number, so it is not applicable.

Introduction

Human adenoviruses comprise a group of non-enveloped viruses with double-stranded linear DNA [1]. Adenovirus infection is at high risk in pediatrics, accounting for 2–7% of respiratory diseases and 4–20% of pneumonia [2–4]. While most infections are self-limiting or present with minor respiratory symptoms, in some cases, such as immunocompromised children, adenoviruses can manifest as a severe, life-threatening disease requiring intensive care and characterized by a high mortality rate [5, 6]. The incidence of respiratory failure and acute respiratory distress syndrome (ARDS) is higher among infants and children admitted to the PICU with adenovirus respiratory infections [7, 8]. The reported mortality rate for pediatric patients with lower respiratory tract infections due to adenoviruses requiring PICU admission ranges from 7 to 22%. The incidence of adenovirus infection is significantly higher in children with compromised immune functions, and the mortality rate can reach 50% [9].

Mechanical ventilation (MV) is commonly used in managing critically ill children suffering severe adenovirus infections. MV is not a direct cause of complications. Still, rather, it is an intervention that is indispensable in patients with severe respiratory failure, in some cases associated with multiple organ failure [10, 11]. Although MV may consequently lead to a range of secondary complications, for example, ventilator-associated pneumonia, these complications arise because of the severe illness of the patient and not quite due to the MV [12]. Mechanical ventilation is an absolute requisite for these patients because of an ongoing respiratory crisis appearing to be independently determinative of mortality in severe adenovirus pneumonia [13]. Mechanical ventilation is a therapeutic measure that provides supportive respiratory function in critically ill patients rather than causing other infections that complicate the clinical picture.

This study aims to investigate the demographic and clinical characteristics of pediatric patients with severe adenovirus respiratory infection pneumonia in our hospital and to analyze the risk factors for MV, ECMO, and mortality. The study will provide research evidence for the clinical identification of high-risk populations with poor prognosis.

Methods

Setting

This retrospective study reviewed 66 children admitted to the PICU of Children’s Hospital, Zhejiang University School of Medicine, and National Clinical Research Center for Child Health diagnosed with severe adenovirus pneumonia between January 2018 and December 2019. The diagnosis was made based on positive adenovirus tests on respiratory specimens, and the hospital’s medical record management system created the list of patients. All the children were subjected to chest X-ray or computed tomography scan to evaluate the severity of pneumonia; during the acute stage of the infection, respiratory secretions compellingly mentioned the application of indirect immunofluorescence virus detection for identifying the type causing the infection. Blood and/or bronchoalveolar lavage (BAL) cultures were performed on patients with suspected bacterial, mycoplasma, or fungal infections. Symptomatic adenovirus-infected children were diagnosed clinically to be either moderate or severe pneumonia. Study participants with incomplete data were excluded. The study was approved by the Children’s Hospital, Zhejiang University School of Medicine, and National Clinical Research Center for Child Health (Approval No.2024-IRB-0222-P-01), and concurrence from parents or guardians was obtained in accordance with hospital policy; ethical committee waiver consent was given for this retrospective study.

Cohort description

The cohort consisted of young patients ranging from neonates to adolescents. A majority of the patients were in good health before the study, although some suffered from underlying comorbidities such as immunodeficiency disorders. The cohort encompassed healthy children with no significant pre-existing chronic conditions alongside children with immunocompromised conditions arising due to prior medical treatment or congenital disease.

Data collection

Information was recorded for each patient about general condition, clinical presentation, laboratory findings, etiologic and radiological results, treatment provided, and outcome on the first day of admission to the PICU, arterial PCO2, PaO2, and hypoxemia measured as P/F were recorded. The Pediatric Acute Lung Injury Consensus Conference criteria were adopted to define ARDS [14]. Septic shock was defined according to the International Pediatric Sepsis Consensus Conference criteria [15]. Preceding clinical history revealed symptoms of heart failure in combination with echocardiographic findings that included identification of systolic dysfunction, decreased ejection fraction, and the presence of pulmonary edema or pleural effusion. Encephalopathy (neurological complications including seizures, encephalitis, encephalopathy, Reye’s syndrome, and all other infection-related neurological diseases) was defined according to the standards espoused by the Centers for Disease Control and Prevention [16].

Inclusion and exclusion criteria

Eligibility criteria were children under 18 years of age who tested positive for adenovirus and were clinically diagnosed with moderate to severe pneumonia. Exclusion criteria were children with incomplete clinical information or other non-adenoviral respiratory infections.

Follow-up of clinical parameters

Clinical parameters such as PCO₂, FiO₂, CRP, Hb, and LDH were recorded primarily on the first day of PICU admission to assess the severity of the illness. These parameters were not monitored systematically during hospitalization since the primary objective was identifying early predictors of poor outcomes. However, those parameters were monitored clinically as per standard of care protocols.

Data analyses

SPSS version 24.0 and R version 4.1.1 were used for all the statistical calculations. The children were divided into two groups according to their survival status and whether MV/ECMO was required. Parameter data were presented as means and standard deviations or medians and interquartiles. They were compared with t-tests or the Kruskal-Wallis rank sum test, respectively, if normally distributed (Shapiro-Wilk normality test). Categorical variables were presented as the number and percentage of occurrences within each category, and comparisons between groups were made using a chi-square or Fisher’s exact test.

Machine learning model fitting took place with MV/ECMO or death being designated as poor outcomes. Feature variables were selected with a cutoff contribution score to the area under the curve (AUC) ≥ 0.2. A 5-fold cross-validation of the prediction performance of random forest models, logistic regression models, decision tree models, and weighted K-nearest neighbor models concerning poor outcomes was done. Receiver operating characteristic curves (ROC) and AUC determined the fitted accuracy. A two-tailed p-value less than 0.05 was considered to be significant.

Results

Demographic information and patient outcomes

Features for clinical data were collected from 66 children. Five died, while 61 were alive. The cohort comprised males and females with a median age of 23 months (2 months to 15 years). Primary outcomes in this study were obtention of mechanical ventilation, ECMO, and mortality.

Factors associated with patient outcomes

The most significant clinical/demographic variables associated with specific outcomes were heart failure (p = 0.005), the presence of pericardial effusion (p = 0.032), the onset of septic shock (p = 0.009), the levels of HB (p = 0.013), levels of lactate dehydrogenase (p = 0.022), levels of albumin (p = 0.035), and normal creatinine (p = 0.037). Also, during treatment, length of hospital stay (p = 0.008), number of antibiotics used (p = 0.002), use of antifungal medication (0.011), MV (p = 0.006), and ECMO (p = 0.016) showed a significant correlation with patient outcomes (Table 1).

Table 1.

Comparative results of different death outcomes in children with severe adenovirus infection admitted to PICU

Variable		Survival group	Death group	Statistic	p value
		Median (IQR)/n	Median (IQR)/n
N		61	5
Hospitalization duration		17(11)	45 (32)	7.056	0.008
Sex	Boy	35	3
	Girl	26	2	0.013	1.000
Age (Month)		23 (31)	17 (24)	0.494	0.482
Fever higher than 39℃	Yes	55	5
	No	6	0	0.541	1.000
Fever days		13 (12)	17 (11)	-0.252	0.811
Fever more than 5 days	Yes	54	5
	No	7	0	0.642	0.649
Fever more than 15 days	Yes	25	3
	No	36	2	0.684	0.667
Cough more than 15 days	Yes	57	5
	No	4	0	0.349	1.000
Vomiting/diarrhea	Yes	7	0
	No	54	5	0.642	0.641
Cyanosis	Yes	5	0
	No	56	5	0.443	1.000
Convulsions	Yes	3	1
	No	58	4	1.846	0.248
Breath count more than 60 time/minutes	Yes	29	4
	No	32	1	1.948	0.343
Heart rate more than 180 times/minutes	Yes	26	3
	No	35	2	0.566	0.655
Moist rales	Yes	60	5
	No	1	0	0.083	1.000
Wheeze	Yes	54	4
	No	7	1	0.315	1.000
Low breath sounds	Yes	16	3
	No	45	2	2.571	0.145
Respiratory failure	Yes	60	5
	No	1	0	0.083	1.000
ARDS	Yes	8	2
	No	53	3	2.598	0.173
Heart failure	Yes	8	4
	No	53	1	13.897	0.005
Pericardial effusion	Yes	18	4
	No	43	1	5.302	0.032
Gastrointestinal bleeding	Yes	0	1
	No	61	4	12.388	0.071
Encephalopathy	Yes	11	3
	No	50	2	4.870	0.061
Septic shock	Yes	12	4
	No	49	1	9.158	0.009
Acute renal failure	Yes	2	1
	No	59	4	2.978	0.212
PCO2		40.9 (19.6)	54.7 (29.7)	3.712	0.054
WBC		7.28 (10.34)	26.44 (27.5)	0.324	0.569
CRP		14.4 (37.48)	56 (43.03)	1.626	0.202
HB*		102.23 ± 18.17	81.2 ± 11.99	3.599	0.013
PLT		254 (193)	59 (177)	2.676	0.102
Normal ALT/AST ratio	Yes	35	1
	No	26	4	2.604	0.160
LDH		670 (632.5)	1596 (658)	5.242	0.022
Normal CK-MB	Yes	15	2
	No	46	3	0.574	0.595
Albumin		31 (6.9)	26.1 (2.2)	4.448	0.035
Normal creatinine	Yes	58	3
	No	3	2	8.123	0.037
Pneumonia grade	Medium	13	0
	Sever	48	5	1.327	0.594
Pleural effusion	Yes	32	5
	No	29	0	4.240	0.060
Pneumothorax	Yes	8	4
	No	53	1	13.897	0.002
Bronchoscopy	Yes	27	4
	No	34	1	2.370	0.184
Plastic bronchitis	Yes	4	0
	No	57	5	0.349	1.000
Concurrent infection
Bacterial	Yes	33	5
	No	28	0	3.986	0.070
Virus	Yes	18	2
	No	43	3	0.241	1.000
Fungi	Yes	3	0
	No	58	5	0.258	1.000
Antimicrobial drugs
Anti-virus	Yes	22	3
	No	39	2	1.125	0.366
Antibiotic type	1	5	0
	2	23	0
	3	16	0
	4	12	1
	5	2	1
	6	1	2
	7	2	1	24.253	0.002
Anti-fungi	Yes	12	4
	No	49	1	9.158	0.011
Duration of steroid use (day)		11 (7)	15 (9)	1.417	0.234
γ globulin	Yes	58	5
	No	3	0	0.258	1.000
MV	Yes	11	4
	No	50	1	10.104	0.006
ECMO	Yes	6	3
	No	55	2	9.874	0.016

Abbreviations: ALT: alanine transaminase; ARDS: acute respiratory distress syndrome; AST: aspartate aminotransferase; CK-MB: creatine kinase-MB; CRP: C-reaction protein; ECMO: extracorporeal membrane oxygenation; HB: hemoglobin; IQR: interquartile range; LDH: lactate dehydrogenase; MV: Mechanical ventilation; PCO₂: partial pressure of carbon dioxide; PICU: pediatric intensive care unit; PLT: platelet; WBC: white blood cell

*: Mean (standard deviation)

Note: Bold fonts represent statistically significant differences.

Mechanical ventilation (MV) outcomes

A total of 15 patients needed MV. Notably associated with MV in the present study were diminished breath sounds (p = 0.001), ARDS (p < 0.001), heart failure (p < 0.001), pericardial effusion (p = 0.004), encephalopathy (p < 0.001), septic shock (p = 0.006), PCO₂ levels (p = 0.01), LDH levels (p = 0.031), albumin levels (p = 0.009), pleural effusion (p = 0.043), and pneumothorax (p < 0.001). During the treatment period, disease severity was significantly correlated with duration of hospitalization (p < 0.001), number of antibiotic choices (p = 0.001), antifungal application (p = 0.001), and duration of steroid use (p = 0.004; Table 2).

Table 2.

Comparative results of whether implemented invasive MV in children with severe adenovirus infection admitted to PICU

Variable		No-MV application	MV application	Statistic	p value
		Median (IQR)	Median (IQR)
N		51	15
Hospitalization duration		15 (7.5)	45 (25.5)	29.817	< 0.001
Sex	Boy	31	7
	Girl	20	8	0.946	0.385
Age (Month)		24 (30)	19 (28.5)	0.312	0.576
Fever higher than 39℃	Yes	48	12
	No	3	3	2.795	0.130
Fever days		12 (11.5)	17 (10.5)	-1.197	0.246
Fever more than 5 days	Yes	45	14
	No	6	1	0.318	0.687
Fever more than 15 days	Yes	19	9
	No	32	6	2.455	0.155
Cough more than 15 days	Yes	49	13
	No	2	2	1.803	0.227
Vomiting/diarrhea	Yes	4	3
	No	47	12	1.807	0.328
Cyanosis	Yes	3	2
	No	48	13	0.919	0.582
Convulsions	Yes	3	1
	No	48	14	0.013	1.000
Breath count more than 60 time/minutes	Yes	22	11
	No	29	4	4.227	0.072
Heart rate more than 180 times/minutes	Yes	20	9
	No	31	6	2.033	0.239
Moist rales	Yes	50	15
	No	1	0	0.299	1.000
Wheeze	Yes	46	12
	No	5	3	1.131	0.348
Low breath sounds	Yes	8	11
	No	43	4	18.789	0.001
Respiratory failure	Yes	50	15
	No	1	0	0.299	1.000
ARDS	Yes	0	10
	No	51	5	40.071	0.001
Heart failure	Yes	2	10
	No	49	5	30.675	0.001
Pericardial effusion	Yes	12	10
	No	39	5	9.706	0.004
Gastrointestinal bleeding	Yes	0	1
	No	51	14	3.452	0.209
Encephalopathy	Yes	5	9
	No	46	6	17.475	0.001
Septic shock	Yes	8	8
	No	43	7	8.945	0.006
Acute renal failure	Yes	2	1
	No	49	14	0.201	1.000
PCO 2		39.6 (16.25)	54.7 (20.5)	6.610	0.010
WBC		7.28 (11.13)	9.69 (14.29)	0.064	0.801
CRP		14 (30.55)	42.87 (57.53)	1.450	0.229
HB*		99.26 ± 17.71	105.33 ± 21.25	-1.010	0.325
PLT		227 (190.5)	254 (265)	0.247	0.619
Normal ALT/AST ratio	Yes	30	6
	No	21	9	1.656	0.248
LDH		613 (601)	1101 (1095)	4.648	0.031
Normal CK-MB	Yes	13	4
	No	38	11	0.008	1.000
Albumin		31 (7.15)	26.7 (6.9)	6.731	0.009
Normal creatinine	Yes	49	12
	No	2	3	4.279	0.069
Pneumonia grade	Medium	12	1
	Severe	39	14	2.084	0.266
Pleural effusion	Yes	25	12
	No	26	3	4.516	0.043
Pneumothorax	Yes	3	9
	No	48	6	22.820	0.001
Bronchoscopy	Yes	23	8
	No	28	7	0.316	0.764
Plastic bronchitis	Yes	3	1
	No	48	14	0.013	1.000
Concurrent infection
Bacterial	Yes	27	11
	No	24	4	1.973	0.245
Virus	Yes	15	5
	No	36	10	0.084	1.000
Fungi	Yes	2	1
	No	49	14	0.201	1.000
Antimicrobial drugs
Anti-virus	Yes	16	9
	No	35	6	4.037	0.063
Antibiotic type	1	5	0
	2	22	1
	3	13	3
	4	9	4
	5	1	2
	6	1	2
	7	0	3	23.314	0.001
Anti-fungi	Yes	7	9
	No	44	6	13.514	0.001
Duration of steroid use (day)		10 (7)	15 (13)	8.358	0.004
γ globulin	Yes	48	15
	No	3	0	0.924	0.576

Abbreviations: ALT: alanine transaminase; ARDS: acute respiratory distress syndrome; AST: aspartate aminotransferase; CK-MB: creatine kinase-MB; CRP: C-reaction protein; ECMO: extracorporeal membrane oxygenation; HB: hemoglobin; IQR: interquartile range; LDH: lactate dehydrogenase; MV: Mechanical ventilation; PCO2: partial pressure of carbon dioxide; PICU: pediatric intensive care unit; PLT: platelet; WBC: white blood cell

*: Mean (standard deviation)

Note: Bold fonts represent statistically significant differences.

Extracorporeal membrane oxygenation (ECMO) outcomes

ECMO was started in nine children. In addition to displaying markedly diminished breath sounds (p = 0.003), ARDS (p < 0.001), heart failure (p < 0.001), pericardial effusion (p < 0.001), encephalopathy (p = 0.001), and septic shock (p = 0.001), factors significantly associated with ECMO therapy also included creating barriers to leukocyte function as C-reactive protein (CRP)(p = 0.011), its partner lactate dehydrogenase (LDH)(p = 0.002), albumin, with the computer signal representation of albumin level showing an effective measure (p < 0.001), normal creatinine (p = 0.018), pleural effusion (p = 0.010), and pneumothorax (p = 0.048). During the treatment, the quantity of significant correlation will show how the patients developed the length of stay (p < 0.001), number of applied antibiotics (p = 0.001), finally applied antifungal (p = 0.004), length of steroid treatment (p = 0.002), and undergone mechanical ventilation (MV) (p < 0.001) (Table 3).

Table 3.

Comparative results of whether implemented invasive ECMO in children with severe adenovirus infection admitted to PICU

Variable			No-ECMO application	ECMO application	Statistic	p value
			Median (IQR)	Median (IQR)
N			57	9
Hospitalization duration			17 (8)	45 (22)	17.242	< 0.001
Sex	Boy		34	4
	Girl		23	5	0.736	0.466
Age (Month)			22 (30)	31 (31)	0.518	0.472
Fever higher than 39℃	Yes		51	9
	No		6	0	1.042	0.615
Fever days			12 (12)	18 (4)	3.824	0.051
Fever more than 5 days	Yes		50	9
	No		7	0	1.236	0.579
Fever more than 15 days	Yes		21	7
	No		36	2	5.332	0.034
Cough more than 15 days	Yes		53	9
	No		4	0	0.672	0.649
Vomiting/diarrhea	Yes		5	2
	No		52	7	1.483	0.241
Cyanosis	Yes		4	1
	No		53	8	0.186	1.000
Convulsions	Yes		3	1
	No		54	8	0.467	1.000
Breath count more than 60 time/minutes	Yes		26	7
	No		31	2	3.216	0.136
Heart rate more than 180 times/minutes	Yes		23	6
	No		34	3	2.185	0.153
Moist rales	Yes		56	9
	No		1	0	0.160	1.000
Wheeze	Yes		51	7
	No		6	2	0.998	0.591
Low breath sounds	Yes		12	7
	No		45	2	12.200	0.003
Respiratory failure	Yes		56	9
	No		1	0	0.160	1.000
ARDS	Yes		2	8
	No		55	1	44.074	0.001
Heart failure	Yes		3	9
	No		54	0	46.895	0.001
Pericardial effusion	Yes		14	8
	No		43	1	14.474	0.001
Gastrointestinal bleeding	Yes		0	1
	No		57	8	6.431	0.131
Encephalopathy	Yes		8	6
	No		49	3	12.883	0.001
Septic shock	Yes		9	7
	No		48	2	16.263	0.001
Acute renal failure	Yes		2	1
	No		55	8	1.035	0.353
PCO2			40.9 (19.6)	50.9 (22.5)	1.257	0.262
WBC			7.88 (10.87)	3.51 (18.38)	2.151	0.142
CRP			13.8 (31.33)	56 (48.82)	6.440	0.011
HB*			98 ± 29	95 ± 17	0.490	0.633
PLT			254 (196)	90 (186)	3.597	0.058
Normal ALT/AST ratio	Yes		34	2
	No		23	7	4.391	0.072
LDH			613 (628)	1596 (1355)	9.552	0.002
Normal CK-MB	Yes		15	2
	No		42	7	0.068	1.000
Albumin			31 (6.7)	24.5 (2.6)	12.283	< 0.001
Normal creatinine	Yes		55	6
	No		2	3	9.874	0.018
Pneumonia grade	Medium		13	0
	Severe		44	9	2.556	0.194
Pleural effusion	Yes		28	9
	No		29	0	8.168	0.010
Pneumothorax	Yes		8	4
	No		49	5	4.832	0.048
Bronchoscopy	Yes		24	7
	No		33	2	3.971	0.071
Plastic bronchitis	Yes		3	1
	No		54	8	0.467	1.000
Concurrent infection
Bacterial	Yes		32	6
	No		25	3	0.353	0.705
Virus	Yes		18	2
	No		39	7	0.322	0.712
Fungi	Yes		3	0
	No		54	9	0.496	1.000
Antimicrobial drugs
Anti-virus		Yes	19	6
		No	38	3	3.670	0.072
Antibiotic type		1	5	0
		2	23	0
		3	15	1
		4	10	3
		5	2	1
		6	2	1
		7	0	3	27.123	0.001
Anti-fungi		Yes	10	6
		No	47	3	10.213	0.004
Duration of steroid use (day)			10 (8)	21 (15)	9.783	0.002
γ globulin		Yes	54	9
		No	3	0	0.496	1.000
MV		Yes	6	9
		No	51	0	35.432	0.001

Abbreviations: ALT: alanine transaminase; ARDS: acute respiratory distress syndrome; AST: aspartate aminotransferase; CK-MB: creatine kinase-MB; CRP: C-reaction protein; ECMO: extracorporeal membrane oxygenation; HB: hemoglobin; IQR: interquartile range; LDH: lactate dehydrogenase; MV: Mechanical ventilation; PCO₂: partial pressure of carbon dioxide; PICU: pediatric intensive care unit; PLT: platelet; WBC: white blood cell

*: Mean (standard deviation)

Note: Bold fonts represent statistically significant differences.

Clarification of MV and ECMO as therapeutic interventions

This study describes MV and ECMO as treatment interventions rather than poor outcomes. Poor outcomes were defined as MV, ECMO, or death.

Machine Learning Model for Predicting Poor outcomes

Machine Learning Models for Outcome Prediction were employed to predict the poor outcome. The feature selection was based on AUC, concluding with the following eight features: heart failure, ARDS, encephalopathy, diminished breath sounds, pneumothorax, and septic shock. Random forest (AUC: 0.968) and K-Nearest Neighbor (KNN) (AUC: 0.966) outperform the logistic regression (AUC: 0.946) while the decision tree model was comparatively poorer (AUC: 0.783) (Fig. 1).

Fig. 1.

Internal cross-validation AUC result distribution (A) and ROC curve (B) fitted by four machine learning models for poor outcomes in children with severe adenovirus infection

Discussion

This retrospective study examined the characteristics of 66 children admitted to the PICU following severe adenovirus infection. Results indicate complications and comorbidities associated with heart failure, pneumothorax, pericardial effusion, and septic shock pose a significant mortality risk in these children. Factors associated with mechanical ventilation (MV) and extracorporeal membrane oxygenation (ECMO) included decreased breath sound, ARDS, and encephalopathy. The study also utilized machine learning models to predict poor outcomes with excellent diagnostic accuracy, indicating that recognizing key clinical features may enhance clinical decision-making.

Adenovirus infection causes diseases in children throughout the year and has wide tissue tropism, affecting the respiratory, gastrointestinal, immune, and neurological systems. This makes diagnosis and treatment very complex for these children, especially those with immature immune systems [17]. Unlike other types of community-acquired pneumonia (CAP), such as those caused by Mycoplasma pneumoniae (Mycoplasma pneumonia) or Staphylococcus aureus, severe adenovirus pneumonia usually presents itself with non-specific clinical symptoms, making diagnosis difficult [18, 19]. This presents a challenge in itself, along with the rapid progression to more severe stages, causing a high mortality rate and long-term sequelae; hence, identifying the risk factors of poor prognosis is essential to reduce the mortality and occurrence of severe sequelae.

Among them, the overall mortality rate of our study was 9.1%, which is higher than the mortality rate reported in other studies, such as Hao et al. (6.88%) [20] and Wang et al. (3.3%) [13]. A higher mortality rate in our cohort might be due to the severe illness among the children analyzed or a mixture of more virulent adenovirus F serotypes. Our findings, nevertheless, indicate that severe complications, for example, heart failure, ARDS, and septic shock, are key predictors of poor outcomes [21].

Mechanical ventilation, or MV, is one of the linchpins of therapy for patients with severe hypoxemia or respiratory failure. It is a life-preserving agent with risks, including ventilator-associated pneumonia, barotrauma, and hemodynamic instability [22]. MV should not in any way be implied as the one responsible for complications; actually, complications are a manifestation of the underlying severity of illness with multiple organ failure, which propels toward MV. MV is normally applied as a mode of intervention and seldom as a “poor outcome.” In this study, MV was utilized in the treatment of severe respiratory distress and some residual or overtly acute problems, including ARDS, septic shock, or encephalopathy, all of which attest to significant severity of critical illness.

The ECMO intervention has been used since the 1970s as one kind of cardiopulmonary support for critically ill patients who failed on conventional MV. ECMO has gained attention and has saved innumerable lives during the COVID-19 pandemic [23]; however, it comes with significant complications: bleeding, infection, pneumothorax, as well as higher mortality during both short- and long-term use [24, 25–27]. Even though MV and ECMO are intended to save lives, they are frequently implemented in cases already cataclysmic. This study considers the need for MV, ECMO, or death as poor outcomes that exhibit correspondence with high mortality and morbidity.

An unexpected finding in this study was the association between normal creatinine levels and the need for MV and ECMO. Creatinine is regarded as a renal biomarker, whereas elevated creatinine levels are generally correlated with kidney injury. However, in our study, normal creatinine levels were found to correlate with the recourse to MV and ECMO, hinting at a possible role for renal reserve or baseline kidney function in managing critically ill children. One explanation for this association is that normal creatinine functions more as a marker for better baseline renal function, implying better baseline organ reserve. Children with higher renal reserve might survive multi-organ failure for a more extended period. Another possibility is that creatinine is only one part of a more intricate interplay of clinical factors, and its levels may not, in all cases, reliably predict the severity of issues linked to cardiopulmonary dysfunction and organ failure, all within the context of a multi-system disease like adenovirus infection. Additional investigations are needed to analyze this association and its pertinent implications.

This study also emphasized using machine learning models to predict poor outcomes in children with severe adenovirus infection. Models were developed using clinical complications, comorbidities, and symptoms as features, achieving excellent accuracy (AUC = 0.968 for the random forest model). The current study findings build on previous studies demonstrating an application of machine learning for predicting mortality and complications in critically ill children. For example, Sánchez et al. used EEG to predict outcomes in pediatric ICU patients, with an AUC of 0.82 [28], while Kwizera A. et al. applied random forest models to predict in-hospital mortality in children with acute infections, achieving an AUC of 0.8 [29].

This is the first study to utilize machine learning models to predict poor outcomes for children with severe adenovirus infection specifically. The models developed in this study may assist clinicians in identifying high-risk patients earlier, enabling timely intervention and more personalized treatment plans. However, it must be noted that this model still needs to be externally validated before it can be considered for broader clinical application. External validation is essential to provide insight into the model’s generalizability and performance over clinical situations.

The authors acknowledge several limitations of this study. The small sample size might impede the generalization of the findings. The study’s event rate for poor outcomes, including mortality and the need for mechanical ventilation or ECMO, was low, which, if anything, could affect power. The data for this study were gathered from January 2018 to December 2019, and additional years of data were not collected owing to the study’s retrospective nature and the relatively rare occurrence of severe adenovirus infections in children during the study period. These limitations can be addressed with future multi-centered studies with larger sample sizes and extended durations of data collection to confirm and expand these findings.

In these multi-center cohorts, future studies should aim to prospectively validate these findings to ensure that the identified risk factors and machine learning models can be generalized. Further use of creatinine levels, renal reserve, and other biomarkers of critical illness or organ failure with adenoviral infections in children needs to be studied.

Conclusions

The study shows highly useful results that will give insight into the concrete results and risk factors of seriousness in children with severe adenoviral infections. However, there are limitations to this research. First, it is observed that severe adenoviral infections have a comparatively low incidence, and the clinical signs and symptoms can vary from person to person owing to seasonal or geographical differences. Hence, multi-center large sample studies should be performed to establish these findings with validation. Second, while machine learning models used in this study have presented predictions with reasonable accuracy, external validity is yet to be confirmed by the limited number of poor outcome cases in the cohort studied. Third, the clinical factors and the treatment preferences of the family of the child influenced the decision between MV and ECMO, thus rendering the classification of MV, ECMO, and death as poor outcomes somewhat subjective.

Future studies should focus on the prospective validation of these findings in larger cohorts and external datasets. In parallel, a more comprehensive framework must be implemented for decision-making regarding MV and ECMO, considering care’s clinical, ethical, and financial aspects.

Abbreviations

PICU

Pediatric Intensive Care Unit

ECMO

Extracorporeal membrane oxygenation

BAL

Bronchoalveolar lavage

PCO₂

Pressure of carbon dioxide

PaO₂

Partial pressure of oxygen

SD

Standard deviation

IQR

Interquartile range

AUC

Area under the curve

ROC

Receiver operating characteristic curves

HB

Hemoglobin

KNN

K-Nearest Neighbors

CAP

Community-acquired pneumonia

Author contributions

All authors read and approved the final manuscript.

Funding

The authors have not received any funding support.

Data availability

All data generated or analysed during this study are included in this. Further enquiries can be directed to the corresponding author.

Declarations

Ethics approval and consent to participate

The study was approved by the Children’s Hospital, Zhejiang University School of Medicine, and the National Clinical Research Center for Child Health (No.2024-IRB-0222-P-01). Written informed consent was obtained from all individuals included in this study, and consent to participate was obtained from parents or guardians.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Conflict of interest

The authors declare no conflict of interest.