This study retrospectively collected clinical data from 752 hospitalized pediatric patients through the hospital’s electronic medical record system. The extracted data encompassed multidimensional information including demographic characteristics, clinical details, laboratory test results, imaging features, extrapulmonary involvement, and co-infections. Severe or critical human metapneumovirus (hMPV) infection is defined as meeting diagnostic criteria for hMPV infection and presenting with at least one of the following clinical manifestations: 1) Respiratory failure necessitating mechanical ventilation; 2) Occurrence of shock; 3) Involvement of other organ failure requiring intensive care unit (ICU) admission [14][15].

Inclusion criteria: 1) Age ≤ 14 years; 2) Confirmed hMPV infection by nucleic acid testing of respiratory specimens.

Exclusion criteria: 1) Hospital-acquired infection; 2) Incomplete case records.

The decision to transfer pediatric patients to the PICU is determined by the attending physician based on the severity of their condition and their need for advanced life support.

Respiratory secretion specimens were collected via throat swabs and used for PCR detection of respiratory pathogens, including hMPV, adenovirus, RSV, parainfluenza virus, bocavirus, influenza virus A/B, rhinovirus, enterovirus, Mycoplasma pneumoniae, and Chlamydia pneumoniae. Additionally, deep sputum or bronchoalveolar lavage fluid was collected for external testing using targeted next-generation sequencing (tNGS) to detect multiple respiratory pathogens, including 49 bacterial species (e.g., Haemophilus influenzae, Streptococcus pneumoniae), 35 viral species (e.g., hMPV, adenovirus), and 16 fungal species (e.g., Mycoplasma, Chlamydia, Rickettsia, Candida albicans).

This study incorporated demographic data as covariates into the analysis to control for potential confounding effects. Demographic data encompass basic characteristics and pre-existing conditions. Basic characteristics include age and gender. Pre-existing conditions include prior comorbidities and multiple diagnosis status. Prior comorbidities encompass congenital cardiovascular disease, congenital airway anomalies, neurological disorders, metabolic disorders, and immunodeficiency. Multiple diagnosis status is defined as the simultaneous presence of two or more confirmed diagnoses.

We collected various clinical, laboratory, and imaging indicators associated with PICU admission risk. Extrapulmonary involvement is a binary variable encompassing gastrointestinal, neurological, and circulatory system involvement, all diagnosed based on clinical symptoms and relevant examinations. Infection-related characteristics included viral co-infection status, pneumonia severity, fever duration, and respiratory distress status.

All data in this study were analyzed using R-4.5.1 statistical software. Prior to formal statistical analysis, systematic quality control was conducted on the raw data. Through logical verification and missing value imputation, the integrity and accuracy of the dataset included in the analysis were ensured. For different types of research data, corresponding descriptive statistical methods were employed: Quantitative data (e.g., age, hemoglobin levels, inflammatory markers, and other laboratory indicators) were expressed as mean ± standard deviation ( x ¯ ± s) after confirming normal distribution via the Shapiro-Wilk test. Independent samples t-tests were used for intergroup comparisons.

To avoid the interference of multicollinearity on regression analysis results, we first employed the Variance Inflation Factor (VIF) for collinearity diagnosis. VIF reflecting the strength of multicollinearity by quantifying the degree to which one independent variable is linearly explained by other independent variables [16][17]. A VIF value exceeding 5 was used as the criterion for identifying significant multicollinearity, and variables meeting this criterion were subsequently excluded. Variables screened for multicollinearity were incorporated into a logistic regression model. The final results of the regression analysis are presented as odds ratios (OR) and 95% confidence intervals (95% CI).

This study conducted pre-specified subgroup analyses based on clinical characteristics. The proposed subgroups include sex subgroup (females, males), disease severity subgroups (mild-to-moderate pneumonia, severe pneumonia), and underlying disease subgroups (presence or absence of congenital disorders). By comparing the differences in OR values and 95% CIs across these subgroups, we will assess the consistency of risk factor effects across different populations, thereby providing a basis for individualized clinical risk assessment. The statistical significance level for this study is defined as a P-value < 0.05.

Among 752 hospitalized children (Table 1), 37 (4.9%) required PICU admission while 715 (95.1%) were managed in the general ward. The mean age was similar between groups (2.40 ± 1.65 vs. 2.66 ± 2.57 years, P = 0.547), and sex distribution did not differ (males: 60.28% vs. 62.16%, P = 0.819). Compared with ward patients, PICU patients showed markedly lower hemoglobin levels (101.78 ± 14.38 vs. 116.08 ± 11.77 g/L, P < 0.001), higher neutrophil percentages (63.00 ± 15.04 vs. 48.90% ± 18.34%, P < 0.001), lower lymphocyte percentages (29.24 ± 12.88 vs. 40.33 ± 19.36%, P < 0.001), higher CRP (63.15 ± 89.61 vs. 29.73 ± 42.62 mg/L, P = 0.03), and higher PCT (5.78 ± 9.26 vs. 1.72 ± 3.28, P = 0.012). Co-infection was more common in the PICU (48.65% vs. 18.18%, P < 0.001), as was bronchoscopy use (24.32% vs. 2.80%, P < 0.001). Severe pneumonia predominated among PICU patients (89.19% vs. 13.01%), whereas mild-to-moderate pneumonia was rare (5.41% vs. 79.58%; both P < 0.001). Multiple diagnoses were substantially more frequent in the PICU (72.97% vs. 8.53%, P < 0.001). Five pre-existing comorbidities including congenital cardiovascular disease (8.11% vs. 1.96%, P = 0.046), congenital airway anomalies (13.51% vs. 3.08%, P = 0.004), neurological disorders (16.22% vs. 2.24%, P < 0.001), metabolic disorders (8.11% vs. 0.14%, P < 0.001), and immunodeficiency (5.41% vs. 0.28%, P = 0.013) were significantly enriched in the PICU group. PICU children had longer fever duration (8.59 ± 5.39 vs. 4.87 ± 1.91 days, P < 0.001) and markedly prolonged hospital stay (19.43 ± 16.14 vs. 6.05 ± 3.05 days, P < 0.001), as well as a higher rate of respiratory distress (83.78% vs. 12.31%, P < 0.001). Extrapulmonary involvement was also more prevalent in the PICU, including gastrointestinal (37.84% vs. 6.57%), neurological (48.65% vs. 7.27%), and circulatory involvement (8.11% vs. 0.70%; all P < 0.001). Radiologically, PICU patients more frequently exhibited patchy opacities/infiltrates (64.86% vs. 52.45%), pleural effusion (5.41% vs. 0.14%), and pneumonic/inflammatory changes (21.62% vs. 10.07%; P < 0.001). On CT, pneumonia/inflammation (40.54% vs. 5.87%) and airway disease (5.41% vs. 1.96%) were notably higher in the PICU group, whereas “normal/none” findings were much less frequent (43.24% vs. 87.27%; P < 0.001).

Table 1. Baseline characteristics of participants (N = 752).

In the initial VIF assessment (Table 2), Seventeen predictors demonstrated substantial multicollinearity, with HGB showing the highest VIF value (97.05), followed by cough (56.26), neutrophil percentage (49.80), SII (25.75), CT category (23.57), lymphocyte percentage (23.48), NLR (22.38), PLT (19.97), fever (18.98), LDH (10.42), PLR (8.11), duration of fever (8.06), WBC (7.33), X-ray category (6.42), AST (6.29), albumin (5.22), and hospital stay (5.11); all of these variables were removed based on the VIF > 5 threshold. After filtering, the remaining variables displayed substantially lower VIF values. For example, age decreased from 4.57 to 2.03 (−55.6%), ALT from 3.70 to 1.53 (−58.5%), PNR from 3.10 to 1.48 (−52.3%), sex from 2.69 to 2.11 (−21.5%), PICU status from 2.37 to 1.90 (−20.0%), and CRP from 2.88 to 2.50 (−13.2%). Smaller reductions were observed in respiratory distress (3.28 to 3.04), multiple diagnoses (1.67 to 1.60), pathogen detection (1.37 to 1.30), congenital airway anomalies (1.30 to 1.20), and gastrointestinal involvement (1.30 to 1.27). Three predictors had VIF values close to 1 both before and after filtering, including hematologic involvement (1.04 to 1.02), renal involvement (1.03 to 1.02), and renal disease (1.01 to 1.00).

Table 2. VIF assessment before and after variable filtering.

Figure 1 presents the correlation heatmaps of all study variables before and after VIF-based filtering. Panel (a) shows the full correlation matrix, in which many variables exhibit moderate to strong positive correlations, reflected by concentrated red blocks along the diagonal and multiple off-diagonal clusters, indicating substantial multicollinearity across hematologic markers, inflammatory indicators, clinical features, and radiological categories. Panel (b) shows the correlation structure after removing variables with VIF > 5. The reduced matrix displays markedly fewer high-correlation clusters, and the remaining variables demonstrate weaker and more sparse correlation patterns, indicating a substantial reduction in multicollinearity.

(a)

(b)

Figure 1. Comparison of heatmaps before and after VIF filtering. (a) Heatmap of all variables prior to VIF assessment; (b) Heatmap after exclusion of variables with VIF > 5.

Table 3 shows the associations between extrapulmonary involvement and PICU admission based on univariate and multivariate logistic regression analyses. In univariate models, higher CRP (OR = 1.01; 95% CI: 1.01 - 1.01; P < 0.001), higher PCT (OR = 1.11; 95% CI: 1.06 - 1.17; P < 0.001), higher ALT (OR = 1.01; 95% CI: 1.01 - 1.01; P = 0.028), and lower PNR (OR = 0.84; 95% CI: 0.74 - 0.96; P = 0.010) were associated with PICU admission, although none of these remained significant after adjustment. Compared with mild-to-moderate pneumonia, severe pneumonia showed a strong univariate association with PICU admission (OR = 100.95; 95% CI: 23.82 - 427.79; P < 0.001), which decreased substantially in the multivariate model (OR = 11.12; 95% CI: 0.99 - 125.49; P = 0.051). The presence of multiple diagnoses was consistently associated with higher odds of PICU admission both before (OR = 28.95; 95% CI: 13.38 - 62.62; P < 0.001) and after adjustment (OR = 36.44; 95% CI: 8.76 - 151.53; P < 0.001). Co-infection also showed significant associations in both analyses (univariate OR = 4.26; adjusted OR = 6.18; both P ≤ 0.004). Among comorbidities, congenital airway anomalies remained significant after adjustment (OR = 7.86; 95% CI: 1.23 - 50.04; P = 0.029), while metabolic disorders also retained significance (OR = 51.35; 95% CI: 1.53 - 1724.87; P = 0.028). Respiratory distress demonstrated a strong association with PICU admission (univariate OR = 36.81; 95% CI: 14.93 - 90.74; P < 0.001), and remained significant in the multivariate model (OR = 9.96; 95% CI: 1.66 - 59.95; P = 0.012). For extrapulmonary involvement, gastrointestinal involvement (adjusted OR = 5.52; 95% CI: 1.33 - 22.93; P = 0.019) maintained a significant association with PICU admission, whereas neurological and circulatory involvement lost significance after adjustment.

Table 3. Multivariable logistic regression analysis of the association between extrapulmonary involvement and PICU admission.

In the overall cohort of 752 children (Figure 2), gastrointestinal involvement was associated with higher odds of PICU admission (OR = 5.52; 95% CI: 1.33 - 22.93; P = 0.019). Across sex groups, the association was not statistically significant for either females (OR = 17.58; 95% CI: 0.60 - 516.15; P = 0.096) or males (OR = 4.18; 95% CI: 0.55 - 31.95; P = 0.168), with no evidence of interaction (P_interaction = 0.476). Among diagnostic categories, the association remained significant only in severe pneumonia (OR = 11.59; 95% CI: 1.63 - 82.55; P = 0.014), while P_interaction was 0.409. Similar associations were observed in children with bronchoscopy not performed (OR = 9.56; 95% CI: 1.48 - 61.76; P = 0.018). In the subgroup without multiple diagnoses, gastrointestinal involvement remained associated with PICU admission (OR = 47.08; 95% CI: 1.09 - 2040.30; P = 0.045), whereas the interaction test was not significant (P_interaction = 0.600). Significant associations were also observed in children without congenital cardiovascular disease (OR = 5.84; P = 0.016), without congenital airway anomalies (OR = 4.54; P = 0.050), without neurological disorders (OR = 4.69; P = 0.116, nonsignificant), without metabolic disorders (OR = 8.56; P = 0.007), without hematologic disorders (OR = 5.55; P = 0.018), without immunodeficiency (OR = 5.43; P = 0.020), and without respiratory distress (OR = 0.00; P = 1.000), while the interaction P-values remained above 0.05 across all comorbidity subgroups. For clinical symptoms, wheezing (P_interaction = 0.519) and hoarseness (P_interaction = 0.216) did not modify the association. Respiratory distress showed a statistically significant interaction (P_interaction = 0.038), with strong associations in children with respiratory distress (OR = 38.63; 95% CI: 3.72 - 400.92; P = 0.002). Across extrapulmonary involvement subgroups—including renal, neurological, circulatory, and hematologic involvement—no significant effect modification was detected, with interaction P-values all above 0.20.

Figure 2. Subgroup analysis of the association between extrapulmonary involvement and PICU admission.