Incremental value of body composition indices in discriminating poorly controlled asthma in children: a cross-sectional study

Introduction

Asthma is the most common chronic respiratory disease in children worldwide and affects approximately 300 million people globally, imposing a substantial burden on affected families and healthcare systems (1). However, despite advances in modern treatment regimens, a considerable proportion of children with asthma have poorly controlled asthma, facing risks of frequent exacerbations, reduced quality of life, and impaired lung function (2). Therefore, early identification of children at high risk of poorly controlled asthma is of paramount importance for optimizing treatment strategies and improving long-term outcomes.

Currently, asthma control assessment relies primarily on clinical symptom scores and lung function parameters. However, systematic reviews indicate that assessment models based on traditional clinical indicators have limited discriminatory ability, with area under the curve (AUC) typically ranging from 0.6 to 0.8 and poor generalizability during external validation (3-5). Consequently, this moderate predictive performance suggests that relying solely on clinical symptom scores and lung function parameters may not adequately capture the underlying pathophysiological mechanisms of poorly controlled asthma. This highlights an urgent need to explore novel biomarkers that can capture systemic metabolic and inflammatory changes.

Bioelectrical impedance analysis (BIA) is a noninvasive and rapid body composition assessment method that has been widely applied in pediatrics (6). It not only evaluates fat and muscle mass, but also provides parameters such as phase angle (PhA) and extracellular water-to-total body water ratio (ECW/TBW) that reflect cellular health and fluid distribution (7). Extensive research has shown that decreased PhA and elevated ECW/TBW are closely associated with chronic inflammation and malnutrition (8,9). Specifically, in chronic obstructive pulmonary disease (COPD), PhA has been confirmed as an effective indicator for discriminating disease severity and prognosis (10). However, the application value of these body composition indices—which can reveal systemic physiological disturbances—in assessing asthma control in children has not been systematically explored.

Previous studies have primarily focused on the association between obesity and asthma (11), but have lacked in-depth investigation into the independent discriminatory roles of more refined body composition indices, such as ECW/TBW and PhA, in asthma control. The obesity-related asthma phenotype is often accompanied by systemic low-grade inflammation. Adipokines and proinflammatory cytokines secreted by visceral fat may contribute to the pathological process of poorly controlled asthma by affecting fluid balance and cell integrity (12). Therefore, incorporating body composition indices into asthma control assessment models may help identify systemic metabolic and inflammatory abnormalities that cannot be captured by traditional indicators.

Based on these research gaps, this study aimed to systematically evaluate the value of BIA-derived body composition indices (ECW/TBW, PhA, and visceral fat area) in discriminating poorly controlled asthma in children and to construct an enhanced discriminatory model integrating body composition parameters with traditional clinical indicators. We hypothesized that body composition indices could reflect underlying systemic inflammation and metabolic abnormalities in children with poorly controlled asthma, thereby significantly improving discriminatory ability beyond traditional assessment and providing a novel assessment tool for precision management of childhood asthma. We present this article in accordance with the STROBE and TRIPOD reporting checklists (available at https://tp.amegroups.com/article/view/10.21037/tp-2026-1-0102/rc).

Methods

Study design and participants

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of The First Affiliated Hospital of Xinjiang Medical University (No. K202307-28) and informed consent was obtained from legal guardians of all individual participants. This single-center cross-sectional study consecutively enrolled children with asthma who met the inclusion criteria at the Department of Pediatrics, The First Affiliated Hospital of Xinjiang Medical University. The enrollment period was from January 1, 2025, to September 1, 2025. All children, in a fasting state, completed anthropometric measurements, body composition analyses, pulmonary function testing, and hematological examinations to assess their body composition, lung function status, and inflammatory marker levels.

Inclusion criteria: (I) aged 6–14 years; (II) met the diagnostic criteria for childhood asthma according to the Global Initiative for Asthma (GINA) 2024 guidelines (13); (III) had received standardized asthma treatment for ≥6 months; (IV) no severe acute exacerbations within the past month; (V) written informed consent signed by guardians.

Exclusion criteria: (I) comorbid other chronic respiratory diseases; (II) comorbid severe cardiovascular, liver or kidney diseases, or endocrine disorders; (III) use of systemic glucocorticoids within the past month; (IV) implanted metal devices or pacemakers.

This study initially screened 311 children with asthma. According to the inclusion and exclusion criteria, after excluding children with incomplete clinical data (n=112), those who used systemic glucocorticoids within the past month (n=36), those with severe underlying diseases (n=5), and those with acute asthma exacerbations (n=55), a total of 103 children were finally enrolled for analysis using age- and sex-matched methods. Figure 1 illustrates the study flow.

Figure 1 Study flow diagram. AUC, area under the curve; ECW/TBW, extracellular water-to-total body water ratio; FEV₁, forced expiratory volume in 1 second; FVC, forced vital capacity; PhA, phase angle; VFA, visceral fat area.

Assessment of asthma control status

The Childhood Asthma Control Test (C-ACT) was used to assess asthma control status. The C-ACT is designed for children aged 12 years or younger and has a total score ranging from 0 to 27 points. The Asthma Control Test (ACT) is intended for children older than 12 years, with a total score ranging from 0 to 25 points. Based on the appropriate test scores for each age group, children were divided into a well-controlled group (C-ACT or ACT score ≥20, n=62) and a poorly controlled group (C-ACT or ACT score <20, n=41). All assessments were completed by research personnel who received standardized training to ensure consistency and accuracy of the evaluations.

Inhaled corticosteroid (ICS) treatment information

ICS treatment data were retrospectively collected from medical records for all enrolled children. The types and daily doses of ICS were recorded and converted to budesonide equivalents (µg/day) to enable standardized comparison across different ICS formulations. ICS doses were classified as low, medium, or high according to the age-specific daily dose categories defined in the GINA report (13). Specifically, for children aged 6–11 years, low dose was defined as ≤200 µg/day, medium dose as >200–400 µg/day, and high dose as >400 µg/day of budesonide equivalent; for children aged ≥12 years, low dose was defined as ≤400 µg/day, medium dose as >400–800 µg/day, and high dose as >800 µg/day.

Body composition measurement

Body composition was measured using the InBody 720 multifrequency segmental bioelectrical impedance body composition analyzer (InBody Co., Ltd., Seoul, South Korea). Subjects were required to fast for at least 4 hours; empty their bladder; remove shoes and socks; wear light clothing; stand on the measurement platform; hold the hand electrodes firmly; and remain still for approximately 2 minutes to complete the measurement. Measured parameters included VFA, PhA, ECW/TBW, skeletal muscle mass (SMM), and its percentage of body weight (SMM/weight).

Pulmonary function testing

Pulmonary function testing was performed using the MasterScreen Pulmonary Function Testing System (JAEGER, Hoechberg, Germany). Before testing, the procedures and precautions were explained in detail to the subjects and their guardians, and the tests were conducted by professionally trained technicians according to standard operating procedures. Testing parameters included forced expiratory volume in 1 second (FEV₁), forced vital capacity (FVC), FEV₁/FVC ratio, peak expiratory flow (PEF), maximal mid-expiratory flow (MMEF), and maximal expiratory flows at 75%, 50%, and 25% of FVC (MEF75, MEF50, MEF25). The main pulmonary function parameters collected in this study included FEV₁ % predicted, FVC % predicted, FEV₁/FVC ratio, and PEF % predicted.

Inflammatory marker measurement

Serum C-reactive protein (CRP) and interleukin-6 (IL-6) levels were measured using enzyme-linked immunosorbent assay (ELISA); peripheral blood eosinophil counts were measured using chemiluminescence; fractional exhaled nitric oxide (FeNO) levels were measured using a FeNO analyzer. All test results were obtained through the hospital’s electronic medical record system.

Statistical analysis

Baseline characteristics comparison

Statistical analyses were performed using SPSS 29.0 and R 4.5.2 software. The Shapiro-Wilk test was first used to assess the normality of continuous variables. Continuous variables with normal distribution were expressed as mean ± standard deviation (SD), and independent samples t-tests were used for between-group comparisons; continuous variables with skewed distribution were expressed as median [interquartile range (IQR)], and Mann-Whitney U tests were used for between-group comparisons. Categorical variables were expressed as frequencies, and Chi-squared tests were used for between-group comparisons. P<0.05 was considered statistically significant, with P values reported to three decimal places, and reported as P<0.001 when P<0.001.

Univariable logistic regression analysis

Univariable logistic regression analysis was used to evaluate the discriminatory value of each candidate variable for poorly controlled asthma. Asthma control status (poorly controlled =1, well-controlled =0) was used as the dependent variable. Candidate variables were selected based on the following criteria: (I) variables with P<0.20 in the baseline characteristics comparison of this study; (II) indicators commonly used in clinical practice to assess body composition and asthma control status. According to these criteria, 12 candidate variables were selected, including (I) body composition indices: ECW/TBW, PhA, VFA, and SMM; (II) pulmonary function parameters: FEV₁/FVC, FVC % predicted, and FEV₁ % predicted; (III) inflammatory markers: IL-6 and CRP; and (IV) demographic characteristics: age, sex, and body mass index (BMI), included as potential confounders. For each variable, the odds ratio (OR) and its 95% confidence interval (CI) were calculated, and the AUC was calculated via receiver operating characteristic (ROC) curves to evaluate the discriminatory ability of each variable. P<0.05 was considered statistically significant.

Multivariable logistic regression model construction

Two multivariable logistic regression models were constructed to discriminate poorly controlled asthma. The base model included traditional clinical variables: age, sex, FEV₁ % predicted, and FEV₁/FVC ratio. The enhanced model incorporated body composition indices VFA, PhA, and ECW/TBW in addition to the base model variables. All variables were entered simultaneously using the Enter method in both models. Prior to modeling, multicollinearity was assessed using variance inflation factor (VIF) analysis among discriminator variables. Five-fold cross-validation was employed to evaluate model stability. Poorly controlled asthma (1 = poorly controlled, 0 = well-controlled) was used as the dependent variable. Continuous variables were standardized using Z-score transformation (subtracting the sample mean and dividing by the sample SD) before entering the logistic regression models, so that ORs represent the change in odds per one SD increase in each predictor. Adjusted ORs, 95% CIs, and P values were reported, with P<0.05 considered statistically significant. To evaluate the potential confounding effect of ICS therapy on the association between body composition indices and asthma control status, a sensitivity analysis was conducted by adding the standardized ICS dose (budesonide equivalent, µg/day) as an additional covariate to the enhanced model. The remaining model specifications were identical to the primary enhanced model analysis.

Internal validation and model robustness

To address concerns regarding potential overfitting due to the low events-per-variable (EPV) ratio in the enhanced model, several additional validation analyses were performed. First, bootstrap internal validation with 1,000 resamples was conducted to estimate the optimism-corrected AUC and the calibration slope (shrinkage factor) for both the enhanced model and a parsimonious model. The optimism was calculated as the mean difference between the AUC obtained on each bootstrap sample (training performance) and the AUC of the bootstrap-derived model applied to the original sample (test performance). The optimism-corrected AUC was obtained by subtracting the mean optimism from the apparent AUC. Second, a parsimonious model was constructed including only the three strongest discriminators identified in univariable and multivariable analyses (ECW/TBW, PhA, and FEV₁/FVC), yielding an improved EPV of 13.7. This model was evaluated using the same bootstrap validation procedure. Third, least absolute shrinkage and selection operator (LASSO) penalized logistic regression was performed as a sensitivity analysis, with the regularization parameter selected via 5-fold cross-validation. This approach simultaneously performs variable selection and coefficient shrinkage, providing an independent assessment of which discriminators remain informative under penalization.

Model performance evaluation

Model discriminatory ability was evaluated by the AUC and its 95% CI, with DeLong’s test used to compare AUC differences between the two models. Sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV) were calculated. The selection of optimal cutoff values was guided by clinical priorities, such as balancing sensitivity and specificity. To reduce false positives and avoid excessive medical interventions, we adopted a specificity-priority principle, selecting cutoff values with specificity constrained to be at least 75%. Additionally, calibration was assessed visually using calibration curves, and the Hosmer-Lemeshow goodness-of-fit test was performed.

Incremental value assessment

Multiple metrics were used to evaluate the incremental value of body composition indices. The AUC increment (ΔAUC) was calculated to assess improvement in discriminatory ability, with a ΔAUC greater than 0.10 considered clinically meaningful. The net reclassification index (NRI) was used to reflect the proportion of patients correctly reclassified, and the integrated discrimination improvement (IDI) was used to reflect improved concordance between predicted probabilities and actual outcomes. Additionally, decision curve analysis (DCA) was plotted to evaluate the clinical net benefit across different predicted risk thresholds, complementing the other metrics. A reclassification table was constructed to show the number of patients reclassified between the base model and the enhanced model.

Feature importance analysis

The Random Forest algorithm from the Python scikit-learn package was used to evaluate the relative importance of each variable in the enhanced model. The parameters were set as n_estimators =1,000, max_depth =10, and random_state =42. Feature importance was calculated based on the mean decrease in Gini impurity, with higher values indicating a greater contribution of that variable to model discriminatory.

Results

Baseline characteristics of study participants

This study enrolled 103 children with asthma, including 41 (39.8%) in the poorly controlled asthma group, and 62 (60.2%) in the well-controlled asthma group; their baseline characteristics are shown in Table 1.

Demographic characteristics and general data

There were no significant differences between the two groups in age, sex, BMI, body fat percentage, weight, or height (P>0.05), indicating good comparability between groups.

Treatment information

Children with poorly controlled asthma received significantly higher ICS doses compared with the well-controlled group [median 600 (IQR, 200, 800) vs. 300 (IQR, 200, 400) µg/day, P<0.001]. The proportion of children on high-dose ICS was markedly greater in the poorly controlled group (29.3% vs. 1.6%), whereas the majority of children in the well-controlled group were on low-dose ICS (77.4% vs. 36.6%; χ²=23.32, P<0.001) (Table 1).

Pulmonary function parameters

The poorly controlled asthma group had a significantly lower FEV₁/FVC ratio compared with the well-controlled group (0.73±0.08 vs. 0.79±0.08, P<0.001), indicating more severe airway obstruction. Notably, percent predicted FVC (82.4%±6.8% vs. 79.4%±6.9%, P=0.03) and percent predicted PEF (79.3%±10.0% vs. 74.0%±10.7%, P=0.01) were higher in the poorly controlled group, a finding that requires cautious interpretation.

Body composition indices

Body composition analysis showed that the poorly controlled group had significantly decreased PhA (4.6°±0.8° vs. 5.2°±0.7°, P<0.001) and significantly elevated ECW/TBW [0.40 (IQR, 0.39, 0.41) vs. 0.38 (IQR, 0.37, 0.39), P<0.001] compared with the well-controlled group, indicating impaired cell membrane integrity and dysregulated fluid balance. SMM significantly decreased (16.3±4.8 vs. 18.4±5.1 kg, P=0.03), while the SMM/weight ratio significantly increased (35.2%±5.6% vs. 31.8%±5.6%, P=0.003). There was no significant difference in VFA between the two groups [150.0 (IQR, 110.3, 181.9) vs. 133.7 (IQR, 82.2, 176.4) cm², P=0.19]. Box plots demonstrated significant differences in ECW/TBW and PhA between the two groups (P<0.001), while the difference in VFA was not statistically significant (P=0.19) (Figure 2A).

Figure 2 Body composition indices and univariable analysis for discriminating poorly controlled asthma. (A) Box plots comparing body composition indices between well-controlled and poorly controlled groups. Each box represents the IQR, with the horizontal line indicating the median value. Whiskers extend to 1.5× IQR. Individual data points are displayed as scatter dots. Significant differences between groups are observed for ECW/TBW (P<0.001) and PhA (P<0.001), while VFA shows no significant difference (P=0.19). (B) Forest plot illustrating the ORs and 95% CIs of body composition indices from univariate logistic regression analysis. ECW/TBW demonstrates the highest discriminatory value (OR =4.20, 95% CI: 2.33–7.56, P<0.001), while PhA exhibits a strong protective effect (OR =0.34, 95% CI: 0.20–0.58, P<0.001). VFA shows a non-significant trend (OR =1.34, 95% CI: 0.89–2.00, P=0.15). The vertical dashed line represents OR =1.0 (no effect). CI, confidence interval; ECW/TBW, extracellular water-to-total body water ratio; IQR, interquartile range; OR, odds ratio; PhA, phase angle; VFA, visceral fat area.

Inflammatory markers

Regarding inflammatory markers, IL-6 levels were significantly lower in the poorly controlled asthma group [5.4 (IQR, 3.2, 8.4) vs. 9.1 (IQR, 6.3, 12.4) pg/mL, P<0.001]. There were no significant differences in CRP, FeNO, or eosinophil counts between the two groups (P>0.05).

Univariable logistic regression analysis

Univariable logistic regression analysis revealed that among the 12 candidate variables, 6 variables were significantly associated with poorly controlled asthma (P<0.05) (Table 2).

Body composition indices

ECW/TBW demonstrated the highest discriminatory value (OR =4.20, 95% CI: 2.33–7.56, P<0.001, AUC =0.824), associated with increased risk for poorly controlled asthma. PhA was inversely associated with poorly controlled asthma (OR =0.34, 95% CI: 0.20–0.58, P<0.001, AUC =0.756). SMM was also a protective factor (OR =0.64, 95% CI: 0.41–0.98, P=0.03, AUC =0.627). VFA did not reach statistical significance (OR =1.34, 95% CI: 0.89–2.00, P=0.15, AUC =0.575). The forest plot showed that the 95% CIs of both ECW/TBW and PhA did not cross 1 (Figure 2B).

Pulmonary function parameters

FEV₁/FVC was a significant protective factor against poorly controlled asthma (OR =0.45; 95% CI: 0.28–0.71; P<0.001; AUC =0.706). Each one SD increase in FEV₁/FVC corresponded to a 55% reduction in the risk of suboptimal asthma control. Conversely, decreased FEV₁/FVC was linked to inadequate asthma control. Percent predicted FVC was a risk factor for poorly controlled asthma (OR =1.58; 95% CI: 1.04–2.40; P=0.03; AUC =0.626); each one SD increase in percent predicted FVC resulted in a 58% increase in the risk of suboptimal control. Percent predicted FEV₁ was not significantly associated with poorly controlled asthma (OR =0.72; 95% CI: 0.48–1.08; P=0.11; AUC =0.594).

Inflammatory markers

IL-6 was a protective factor (OR =0.40, 95% CI: 0.25–0.65, P<0.001, AUC =0.725). CRP approached significance (OR =0.65, 95% CI: 0.43–1.00, P=0.051, AUC =0.580).

Demographic characteristics

Age (OR =1.29, 95% CI: 0.86–1.92, P=0.221), sex (OR =1.02, 95% CI: 0.69–1.52, P=0.90), and BMI (OR =1.04, 95% CI: 0.70–1.55, P=0.83) did not show significant associations (P>0.05).

Analysis of two multivariable logistic regression models

VIF analysis showed that all variables had VIF values <2, indicating no multicollinearity among discriminator variables, ensuring the stability and reliability of the regression models. The maximum VIF value in the base model was 1.11, and in the enhanced model was 1.15 (Table S1).

Base model

In the multivariable analysis of the base model, four traditional clinical variables were included: age, sex, FEV₁ % predicted, and FEV₁/FVC ratio. Only FEV₁/FVC showed independent discriminatory value (adjusted OR =0.48, 95% CI: 0.30–0.75, P=0.001), while age (adjusted OR =1.22, 95% CI: 0.79–1.88, P=0.36), sex (adjusted OR=1.03, 95% CI: 0.67–1.57, P=0.90), and FEV₁ % predicted (adjusted OR =0.81, 95% CI: 0.52–1.25, P=0.33) were not statistically significant (Table S2).

Enhanced model

The enhanced model additionally incorporated three body composition indices—VFA, PhA, and ECW/TBW—on top of the base model. Multivariable logistic regression analysis, after adjusting for age, sex, and pulmonary function parameters (Table 3), showed that ECW/TBW was the most significant independent risk factor (adjusted OR =9.83, 95% CI: 3.53–27.37, P<0.001); each 1-SD increase was associated with approximately a ninefold increase in the risk of poorly controlled. PhA, defined as phase angle, was significantly inversely associated (adjusted OR =0.17, 95% CI: 0.06–0.45, P<0.001); each 1-SD increase was associated with an 83% reduction in the risk of poorly controlled. VFA showed a trend toward increased risk of poorly controlled (adjusted OR =1.68, 95% CI: 0.90–3.13) but did not reach statistical significance (P=0.10). FEV₁/FVC maintained significant independent discriminatory value (adjusted OR =0.25, 95% CI: 0.11–0.56, P< 0.001); each 1-SD increase was associated with a 75% reduction in the risk of poorly controlled. Predicted FEV₁ percent lost statistical significance as an independent discriminator after adjustment (adjusted OR =0.78, 95% CI: 0.37–1.63, P=0.50). Age (adjusted OR =0.81, 95% CI: 0.39–1.70, P=0.58) and sex (adjusted OR =0.65, 95% CI: 0.33–1.27, P=0.20) were not statistically significant, indicating good baseline comparability between the two groups.

Sensitivity analysis for ICS dose

After additional adjustment for ICS dose in the enhanced model, all three key discriminators maintained statistical significance and consistent direction of effect: ECW/TBW remained a significant risk factor (adjusted OR =31.83, 95% CI: 5.24–193.44, P<0.001), PhA remained a significant protective factor (adjusted OR =0.07, 95% CI: 0.02–0.33, P<0.001), and FEV₁/FVC retained significance (adjusted OR =0.12, 95% CI: 0.03–0.46, P=0.002). However, the substantial shifts in point estimates and markedly widened CIs compared with the primary enhanced model reflect the instability inherent to the further reduced EPV ratio (EPV ≈5.1 with eight covariates), and the magnitude of these adjusted ORs should be interpreted with caution. The consistency of the direction and significance of associations, rather than the precise point estimates, supports the robustness of the primary findings (Tables S3,S4).

Performance comparison of two logistic regression models

Discriminatory ability

The AUC of the base model was 0.618 (95% CI: 0.505–0.721). The ROC curve showed that the base model curve was close to the diagonal line, which represents chance-level performance, indicating weak discriminatory ability (Table 4, Figure 3A).

Figure 3 Discrimination and calibration performance of the baseline and enhanced models. (A) ROC curves comparing the baseline model and the enhanced model. The enhanced model shows markedly improved discrimination, with the ROC curve approaching the upper left corner. (B) Calibration curve of the baseline model, demonstrating a deviation between predicted and observed probabilities. (C) Calibration curve of the enhanced model, showing close agreement with the ideal diagonal line, indicating good calibration performance. AUC values shown represent five-fold cross-validated estimates. AUC, area under the curve; CI, confidence interval; ROC, receiver operating characteristic.

The AUC of the enhanced model significantly improved to 0.893 (95% CI: 0.820–0.949), with the ROC curve approaching the upper left corner and clearly outperforming the base model curve (Figure 3A), indicating improved discriminatory ability. Compared with the base model, the enhanced model showed an AUC increase of 0.275 (relative change: 44.5%, P<0.001).

DeLong’s test confirmed that the discriminatory ability of the enhanced model was significantly superior to that of the base model (Z statistic =4.030, P<0.001). The optimal cutoff value was determined by the Youden index (sensitivity + specificity −1). The cutoff value of the enhanced model (0.430) provided better discriminatory ability (Youden index =0.610), which is superior to that of the base model cutoff value (0.460, Youden index =0.164). Children with predicted probability ≥0.430 should be classified as high-risk for poorly controlled asthma (Table S6).

Classification performance

The base model had a sensitivity of 39.0%, a specificity of 77.4%, and an accuracy of 62.1%. The PPV was 53.3%, and the NPV was 65.8%. This indicates that the model had limited ability to identify children with poorly controlled, correctly identifying fewer than 40% of such cases. In contrast, the sensitivity of the enhanced model improved to 70.7%, an increase of 31.7 percentage points. Specificity improved to 90.3%, an increase of 12.9 percentage points, and accuracy improved to 82.5%, an increase of 20.4 percentage points. The PPV improved to 82.9%, and the NPV improved to 82.4%. This indicates that the enhanced model could correctly identify over 70% of children with poorly controlled, while maintaining specificity above 90%.

Model calibration

The calibration curve showed that the predicted probabilities of the base model deviated somewhat from the observed probabilities, especially in the intermediate risk range (Figure 3B). In contrast, the calibration curve for the enhanced model demonstrated that the predicted probabilities were highly consistent with the observed probabilities, with the curve lying close to the ideal diagonal line (Figure 3C). This indicates that the model demonstrated both adequate discriminatory ability and satisfactory calibration performance.

To further assess model calibration, the Hosmer-Lemeshow test was performed. Both the base model (χ²=8.39, P=0.39) and the enhanced model (χ²=3.83, P=0.87) demonstrated adequate calibration. Both models showed good calibration (P>0.05), indicating that predicted probabilities were consistent with actual outcomes (Table S6).

Internal validation and model robustness

Bootstrap validation of the enhanced model

Bootstrap internal validation (1,000 resamples) of the enhanced model yielded an apparent AUC of 0.932 and a mean optimism of 0.026, resulting in an optimism-corrected AUC of 0.905 (bootstrap 95% CI: 0.897–0.935). The bootstrap-derived calibration slope was 0.737, indicating moderate overfitting of the regression coefficients, consistent with the low EPV ratio (5.9). The five-fold cross-validated AUC was 0.893, closely aligned with the optimism-corrected estimate (Table S5).

Parsimonious model

To mitigate overfitting, a parsimonious model was constructed using only the three strongest discriminators: ECW/TBW, PhA, and FEV₁/FVC (EPV =13.7). In this model, ECW/TBW remained the most significant risk factor (adjusted OR =7.52, 95% CI: 3.08–18.35, P<0.001), PhA remained protective (adjusted OR =0.21, 95% CI: 0.09–0.50, P<0.001), and FEV₁/FVC retained significance (adjusted OR =0.28, 95% CI: 0.13–0.61, P=0.001) (Table S7). The parsimonious model achieved an apparent AUC of 0.920 and an optimism-corrected AUC of 0.910 (bootstrap 95% CI: 0.902–0.920), with a calibration slope of 0.887 and a five-fold cross-validated AUC of 0.898. Notably, the parsimonious model demonstrated lower optimism (0.009 vs. 0.026) and superior calibration compared with the full enhanced model, while maintaining comparable discriminatory performance.

LASSO sensitivity analysis

LASSO penalized regression retained 4 of the 7 enhanced model variables: ECW/TBW, PhA, FEV₁/FVC, and VFA. Age, sex, and FEV₁ % predicted were shrunk to zero, consistent with their non-significance in multivariable analysis. The five-fold cross-validated AUC of the LASSO model was 0.873. The LASSO-selected variables closely matched those in the parsimonious model, providing independent confirmation that the body composition indices ECW/TBW and PhA are robust discriminators that survive penalization (Table S5).

Incremental value assessment of body composition indices

The ΔAUC of the enhanced model compared with the base model was 0.275, exceeding the clinical significance threshold (0.10), indicating that body composition indices provided substantial additional discriminatory information. The NRI was 0.446, indicating that 44.6% of children were correctly reclassified. The reclassification table (Table 5) showed that among 41 children with poorly controlled, the enhanced model reclassified 17 children from “well-controlled” to “poorly controlled”, thereby improving sensitivity. Among 62 children with well-controlled, the enhanced model reclassified 14 children from “poorly controlled” to “well-controlled”, improving specificity. The IDI was 0.394, further confirming significantly improved concordance between predicted probabilities and actual outcomes. DCA (Figure 4) showed that across a wide range of risk thresholds from 10% to 70%, the net benefit of the enhanced model was consistently higher than both the base model and the “treat-all” strategy, indicating that clinical decisions based on the enhanced model could bring practical benefits to more children.

Figure 4 Decision curve analysis of the baseline and enhanced models. Decision curve analysis comparing the clinical net benefit of the baseline model and the enhanced model across a range of threshold probabilities. The enhanced model provides a consistently higher net benefit than the baseline model and the “treat-all” strategy over a wide threshold probability range (10–70%), indicating superior clinical utility for decision-making.

Feature importance analysis in enhanced model

The Random Forest algorithm was used to evaluate the relative importance of each variable in the enhanced model. Random Forest analysis (Figure 5) showed that ECW/TBW had the highest importance (31.7%), indicating the greatest importance in the model, followed by PhA (22.3%), FEV₁/FVC (17.2%), and VFA (11.0%). FEV₁% predicted, age, and sex had importances of 10.3%, 5.8%, and 1.7%, respectively. The combined importance of ECW/TBW and PhA, the two key body composition indices, reached 54.0%, far exceeding that of traditional pulmonary function parameters (27.5%). This ranking was highly consistent with the results of univariable and multivariable analyses, further confirming the central role of body composition indices in the logistic regression model.

Figure 5 Feature importance analysis of the enhanced model using random forest algorithm. Feature importance was calculated using the Random Forest algorithm with 1,000 trees. The horizontal axis represents the relative importance of each variable, measured by the mean decrease in Gini impurity. Body composition indicators (purple bars) are ECW/TBW (31.7%) and PhA (22.3%), which collectively contribute 54.0% to the model. Base variables (blue bars) include FEV₁/FVC (17.2%), VFA (11.0%), FEV₁% predicted (10.3%), age (5.8%), and sex (1.7%). The ranking is highly consistent with univariable and multivariable analyses, confirming the central role of body composition indicators in asthma control discrimination. ECW/TBW, extracellular water-to-total body water ratio; FEV₁, forced expiratory volume in 1 second; FEV₁_pred_pct, FEV₁ % predicted; FVC, forced vital capacity; PhA, phase angle; VFA, visceral fat area.

Discussion

This study systematically evaluated the value of body composition indices in discriminating poorly controlled asthma in children. The main findings include: (I) incorporating body composition indices in addition to traditional clinical assessment significantly improved discriminatory accuracy, with the AUC increasing from 0.618 to 0.893 (ΔAUC =0.275, P<0.001); (II) ECW/TBW and PhA emerged as the most significant independent discriminator, with discriminatory value exceeding that of traditional pulmonary function parameters alone; and (III) the enhanced model’s sensitivity improved from 39.0% to 70.7%, specificity improved from 77.4% to 90.3%, and the NRI was 44.6%. Taken together, these findings suggest that body composition abnormalities may be indicative of underlying systemic inflammation and metabolic disturbances in children with poorly controlled asthma. This provides a novel perspective for precision assessment and individualized management of childhood asthma.

Value of body composition indices in asthma discrimination

Previous systematic reviews and model validation studies have demonstrated that asthma control logistic regression models based primarily on pulmonary function parameters and symptom scores have a generally limited ability to discriminate between well-controlled and poorly controlled patients. Even when evaluated using metrics such as the AUC or c-statistic, most models show only moderate discriminatory performance, with relevant indicators typically in the range of about 0.6 to 0.8, suggesting that the discriminatory accuracy of traditional single-dimensional control indicators remains limited across different populations (14-17). This highlights the need to explore more comprehensive, multidimensional assessment strategies for asthma control (14,15). The base model in this study (AUC =0.618) aligns with findings reported in the literature, further confirming the limitations of traditional assessment methods. However, the introduction of body composition indices resulted in substantial improvements, with the enhanced model achieving an AUC of 0.893, representing a significant improvement over previous single-discriminator models. Previous studies have primarily focused on the association between obesity and asthma control (11,18,19). In this study, we conducted a preliminary evaluation of the independent discriminatory roles of more refined body composition indices, such as ECW/TBW and PhA, in discriminating childhood asthma control. This provides a novel perspective for understanding the relationship between body composition and asthma.

Mechanistic discussion of ECW/TBW and poorly controlled asthma

This study found that elevated ECW/TBW is an important risk factor for poorly controlled asthma. From a physiological perspective, elevated ECW/TBW typically reflects extracellular fluid volume expansion and abnormal fluid distribution. This phenomenon occurs in various inflammatory states and may involve mechanisms such as increased capillary permeability and plasma component extravasation (20). BIA is a commonly used noninvasive method for assessing fluid distribution and can, under certain conditions, distinguish between intracellular and extracellular water (21,22). Body composition analysis demonstrated that patients with COPD exhibited a significantly higher ECW/TBW ratio compared to the control group. This elevation was associated with both disease severity and systemic inflammation (23,24). In contrast, to date, there is a lack of systematic reviews or high-quality evidence examining the ECW/TBW ratio in relation to systemic inflammatory markers in individuals with asthma. Nonetheless, existing literature suggests that the PhA is sensitive to changes in body fluid distribution and correlates with inflammatory processes (24,25). Therefore, further targeted research is warranted to confirm whether this association exists specifically in the context of asthma. Obesity, particularly central obesity, is frequently associated with systemic low-grade inflammation and metabolic disturbances, which are closely linked to alterations in body fluid distribution. In individuals with central obesity, the ECW/TBW ratio, as measured by BIA, is significantly elevated. This ratio shows a positive correlation with both fat accumulation and fluid compartment imbalances. These findings indicate its potential as a diagnostic biomarker of adiposity-related inflammatory processes and metabolic dysregulation (8,11). In obesity-related asthma, visceral fat, as an active endocrine organ, secretes adipokines and proinflammatory cytokines that are key drivers of systemic and airway low-grade inflammation (12). This study found that abnormally elevated ECW/TBW was significantly associated with poorly controlled asthma. This finding may represent an early, quantifiable manifestation of adipose tissue-derived inflammation affecting fluid balance, which warrants further investigation in future studies.

Association of PhA with poorly controlled asthma

PhA reflects the capacitive properties of cell membranes and is a sensitive indicator for assessing cell membrane integrity, cell quality, and nutritional status (7,9,26). Research has shown that decreased PhA is generally associated with increased inflammatory burden, malnutrition, and adverse outcomes in chronic diseases (27). In COPD, decreased PhA is closely associated with worsening airflow limitation, decreased exercise tolerance, deteriorating quality of life, and reduced survival (10,28). For children with asthma, decreased PhA may reflect cellular damage mediated by chronic inflammation, mitochondrial dysfunction, malnutrition, or metabolic disturbances. This study’s results showed that PhA was significantly inversely associated with poorly controlled asthma. Each 1-SD increase in PhA was associated with an 83% reduction in the risk of poorly controlled, consistent with previous studies. As a noninvasive and rapid assessment tool, PhA may have advantages in identifying children at high risk for poorly controlled asthma and monitoring disease progression. This study showed that SMM was significantly decreased in the poorly controlled group, further supporting the association between nutritional status and asthma control.

Interpretation of complex pathophysiological phenomena

The paradoxically lower IL-6 levels in the poorly controlled group in this study require cautious interpretation. It is noteworthy that although there were statistical differences in IL-6 levels between the two groups, the IL-6 values in both groups were within the normal reference range. Therefore, this difference may not be clinically meaningful. IL-6 exerts bidirectional regulatory effects through “classical” and “trans-signaling” pathways, typically promoting inflammation in the acute phase but potentially participating in regulation and repair during the chronic phase (29). The changes in IL-6 levels in children with poorly controlled asthma in this study may be related to the suppressive effect of their long-term glucocorticoid therapy on the production of proinflammatory factors such as IL-6 (30).

Furthermore, the paradoxically higher FVC % predicted and PEF % predicted observed in the poorly controlled group warrant careful interpretation. Several factors may account for this unexpected finding. First, the between-group differences in FVC % predicted (82.4% vs. 79.4%) and PEF % predicted (79.3% vs. 74.0%), while statistically significant, were modest in absolute magnitude and both groups’ mean values remained within the clinically impaired range (<85% predicted). Given the relatively small sample size (n=103), these differences may partly reflect sampling variability rather than a true pathophysiological phenomenon. Second, in obstructive lung disease, the FEV₁/FVC ratio is generally considered a more reliable indicator of airflow limitation than FEV₁ or FVC alone, as FVC can be preserved or even relatively maintained in the presence of significant airway obstruction (31). Our data are consistent with this interpretation, as FEV₁/FVC showed the expected pattern of being significantly lower in the poorly controlled group. Third, it is important to distinguish between asthma severity and asthma control; these are related but distinct constructs, and spirometric parameters do not always align consistently with symptom-based control assessments (32). We acknowledge that without body plethysmography data to measure residual volume (RV) and total lung capacity (TLC), we cannot definitively determine the contribution of air trapping to the observed FVC patterns. This represents a limitation of the current study.

Together, these phenomena suggest that the pathophysiological mechanisms of poorly controlled asthma are complex and require comprehensive assessment across multiple dimensions, including clinical evaluation and biomarker analysis. Therefore, interpretation of single biomarkers should always be combined with clinical context and reference ranges.

Strengths and innovations of this study

This study has the following strengths and innovations. In terms of study design, strict inclusion and exclusion criteria were applied, and age and sex matching ensured comparability between groups, effectively minimizing the influence of confounding factors. Methodologically, incremental discriminatory value assessment methods—including ΔAUC, NRI, and IDI—as well as DCA were employed, combined with cross-validation for internal model validation, thereby enhancing the robustness of the results (33-35). In this study, we used the Random Forest algorithm to assess feature importance (36). The results showed that the combined importance of ECW/TBW and PhA reached 54.0%, significantly higher than traditional pulmonary function parameters (27.5%), further supporting the potential role of body composition indices in asthma control assessment from a machine learning perspective. Furthermore, in terms of theoretical innovation, a framework associating body composition, inflammation, and asthma control was proposed, which aligns with current research trends in precision medicine and systems biology.

Clinical application value

The enhanced model demonstrated improved sensitivity and specificity compared with the base model in this cohort, suggesting potential utility for identifying children at higher risk for poorly controlled asthma, pending external validation. The NRI indicated that nearly half of the children were more accurately stratified using body composition assessment. To further evaluate the clinical utility of the model, we used DCA for verification (34). The results showed that across a risk threshold range of 10–70%, the enhanced model incorporating body composition indices consistently provided superior clinical net benefit compared with the baseline model containing only traditional indicators. These results suggest that incorporating body composition indices may enhance the clinical utility of asthma control assessment in this population. BIA is noninvasive, rapid, reproducible, and low-cost, making it well suited for use in pediatric clinical settings.

Study limitations and future directions

Several limitations should be acknowledged. First, the single-center cross-sectional design precludes causal inference, and the relatively small sample size resulted in a low EPV ratio (~5.9) in the enhanced model, which increases the risk of overfitting. The exceptionally high adjusted OR for ECW/TBW (9.83) and the wide CIs (95% CI: 3.53–27.37) reflect this statistical instability. To address this concern, bootstrap internal validation was performed, which yielded an optimism-corrected AUC of 0.905 (optimism =0.026) and a calibration slope of 0.737, confirming the presence of moderate overfitting in the coefficient estimates. Importantly, a parsimonious model including only ECW/TBW, PhA, and FEV₁/FVC with an adequate EPV ratio (EPV =13.7) achieved an optimism-corrected AUC of 0.910 with a calibration slope of 0.887, demonstrating that comparable discriminatory performance can be achieved with substantially reduced overfitting. LASSO penalized regression independently confirmed the importance of these three variables. However, the wide CIs around key estimates reflect inherent imprecision that warrants confirmation in larger cohorts. External validation in geographically and ethnically diverse populations is essential before these models can be recommended for clinical use. Second, single time-point measurements could not capture dynamic changes in body composition in relation to disease progression and treatment response. Third, BIA measurements may be influenced by hydration status, food intake, and physical activity, and standardized measurement conditions are essential to minimize individual variability. Fourth, the potential confounding effect of ICS therapy warrants consideration, as children with poorly controlled asthma received significantly higher ICS doses. Sensitivity analysis with additional adjustment for ICS dose demonstrated that the key body composition indices maintained independent statistical significance, although the widened CIs after this adjustment underscore the need for larger studies to more precisely estimate these associations. Fifth, this study relied solely on spirometry without body plethysmography; therefore, absolute lung volumes such as RV and TLC were unavailable, limiting the ability to assess air trapping. Sixth, the inflammatory markers examined were limited and did not include potential mediators such as adipokines. Seventh, asthma control was assessed using symptom-based scores, which carry inherent subjectivity; future studies should integrate objective monitoring indicators for more comprehensive evaluation.

Future research should prioritize multicenter prospective cohort studies with larger sample sizes and external validation to confirm the discriminatory value of body composition indices for asthma control across diverse pediatric populations. Longitudinal follow-up data would help establish whether dynamic changes in body composition are causally associated with asthma control trajectories, exacerbation risk, and long-term outcomes. Intervention studies evaluating whether targeted modification of body composition—such as reduction of visceral fat or improvement of nutritional status—can improve asthma control would provide direct evidence for clinical translation. Additionally, incorporating body plethysmography alongside BIA and spirometry in future studies would enable a more comprehensive assessment of the interplay between body composition, lung volumes, and airway physiology. Finally, investigating the differential roles of body composition indices across distinct asthma phenotypes may contribute to more refined phenotypic classification and individualized management strategies.

Conclusions

This cross-sectional study suggested that BIA-derived body composition indices, particularly ECW/TBW and PhA, may provide incremental discriminatory value for identifying poorly controlled asthma in children beyond traditional clinical and spirometric assessments. A parsimonious model incorporating ECW/TBW, PhA, and FEV₁/FVC achieved comparable discriminatory performance with improved statistical stability relative to the full enhanced model. These body composition abnormalities may reflect underlying systemic inflammation and metabolic disturbances associated with poor asthma control. As BIA is noninvasive, rapid, and reproducible, these findings highlight its potential as a complementary assessment tool in pediatric asthma, although confirmation through multicenter prospective studies with external validation is required before clinical implementation.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the STROBE and TRIPOD reporting checklists. Available at https://tp.amegroups.com/article/view/10.21037/tp-2026-1-0102/rc

Data Sharing Statement: Available at https://tp.amegroups.com/article/view/10.21037/tp-2026-1-0102/dss

Peer Review File: Available at https://tp.amegroups.com/article/view/10.21037/tp-2026-1-0102/prf

Funding: This work was supported by the Tianshan Elite High-Level Medical Talent Development Program administered by the Health Commission of Xinjiang Uygur Autonomous Region (No. TSYC202301B003) and the State Key Laboratory of Pathogenesis, Prevention, Treatment of Central Asian High Incidence Diseases Fund (No. SKL-HIDCA-2024-GB9).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tp.amegroups.com/article/view/10.21037/tp-2026-1-0102/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. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of The First Affiliated Hospital of Xinjiang Medical University (No. K202307-28) and informed consent was obtained from legal guardians of all individual participants.

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. Dharmage SC, Perret JL, Custovic A. Epidemiology of Asthma in Children and Adults. Front Pediatr 2019;7:246. [Crossref] [PubMed]
2. Rawat SS. Global Initiative for Asthma (GINA) 2025: A Revolutionary Document for Management of Asthma in Children. J Pediatr Pulmonol 2025;4:29-30.
3. Kothalawala DM, Kadalayil L, Weiss VBN, et al. Prediction models for childhood asthma: A systematic review. Pediatr Allergy Immunol 2020;31:616-27. [Crossref] [PubMed]
4. Smit HA, Pinart M, Antó JM, et al. Childhood asthma prediction models: a systematic review. Lancet Respir Med 2015;3:973-84. [Crossref] [PubMed]
5. Colicino S, Munblit D, Minelli C, et al. Validation of childhood asthma predictive tools: A systematic review. Clin Exp Allergy 2019;49:410-8. [Crossref] [PubMed]
6. Kyle UG, Earthman CP, Pichard C, et al. Body composition during growth in children: limitations and perspectives of bioelectrical impedance analysis. Eur J Clin Nutr 2015;69:1298-305. [Crossref] [PubMed]
7. Lukaski HC, Kyle UG, Kondrup J. Assessment of adult malnutrition and prognosis with bioelectrical impedance analysis: phase angle and impedance ratio. Curr Opin Clin Nutr Metab Care 2017;20:330-9. [Crossref] [PubMed]
8. Harimawan AIW, Prabandari AASM, Wihandani DM, et al. Association between phase angle and ECW/TBW ratio with body composition in individuals with central obesity: a cross-sectional study. Front Nutr 2025;12:1638075. [Crossref] [PubMed]
9. Norman K, Stobäus N, Pirlich M, et al. Bioelectrical phase angle and impedance vector analysis--clinical relevance and applicability of impedance parameters. Clin Nutr 2012;31:854-61. [Crossref] [PubMed]
10. Maddocks M, Kon SS, Jones SE, et al. Bioelectrical impedance phase angle relates to function, disease severity and prognosis in stable chronic obstructive pulmonary disease. Clin Nutr 2015;34:1245-50. [Crossref] [PubMed]
11. Peters U, Dixon AE, Forno E. Obesity and asthma. J Allergy Clin Immunol 2018;141:1169-79. [Crossref] [PubMed]
12. Sideleva O, Suratt BT, Black KE, et al. Obesity and asthma: an inflammatory disease of adipose tissue not the airway. Am J Respir Crit Care Med 2012;186:598-605. [Crossref] [PubMed]
13. Global Initiative for Asthma. Global Strategy for Asthma Management and Prevention, 2024. Updated May 2024. Available from: https://ginasthma.org/
14. Loymans RJB, Debray TPA, Honkoop PJ, et al. Exacerbations in Adults with Asthma: A Systematic Review and External Validation of Prediction Models. J Allergy Clin Immunol Pract 2018;6:1942-1952.e15. [Crossref] [PubMed]
15. Bridge J, Blakey JD, Bonnett LJ. A systematic review of methodology used in the development of prediction models for future asthma exacerbation. BMC Med Res Methodol 2020;20:22. [Crossref] [PubMed]
16. Xiong S, Chen W, Jia X, et al. Machine learning for prediction of asthma exacerbations among asthmatic patients: a systematic review and meta-analysis. BMC Pulm Med 2023;23:278. [Crossref] [PubMed]
17. Liu A, Zhang Y, Yadav CP, et al. An Updated Systematic Review on Asthma Exacerbation Risk Prediction Models Between 2017 and 2023: Risk of Bias and Applicability. J Asthma Allergy 2025;18:579-89. [Crossref] [PubMed]
18. Dixon AE, Poynter ME. Mechanisms of Asthma in Obesity. Pleiotropic Aspects of Obesity Produce Distinct Asthma Phenotypes. Am J Respir Cell Mol Biol 2016;54:601-8.
19. Lang JE, Hossain J, Dixon AE, et al. Does age impact the obese asthma phenotype? Longitudinal asthma control, airway function, and airflow perception among mild persistent asthmatics. Chest 2011;140:1524-33.
20. Kyle UG, Bosaeus I, De Lorenzo AD, et al. Bioelectrical impedance analysis--part I: review of principles and methods. Clin Nutr 2004;23:1226-43. [Crossref] [PubMed]
21. Kyle UG, Bosaeus I, De Lorenzo AD, et al. Bioelectrical impedance analysis-part II: utilization in clinical practice. Clin Nutr 2004;23:1430-53. [Crossref] [PubMed]
22. Earthman C, Traughber D, Dobratz J, et al. Bioimpedance spectroscopy for clinical assessment of fluid distribution and body cell mass. Nutr Clin Pract 2007;22:389-405. [Crossref] [PubMed]
23. Jin X, Yang Y, Chen G, et al. Correlation between body composition and disease severity in patients with chronic obstructive pulmonary disease. Front Med (Lausanne) 2024;11:1304384. [Crossref] [PubMed]
24. Xie AN, Huang WJ, Ko CY. Extracellular Water Ratio and Phase Angle as Predictors of Exacerbation in Chronic Obstructive Pulmonary Disease. Adv Respir Med 2024;92:230-40. [Crossref] [PubMed]
25. Ceolin J, de Borba EL, Mundstock E, et al. Phase angle of bioimpedance as a marker of inflammation in cardiovascular diseases: A systematic review. Nutrition 2023;112:112064. [Crossref] [PubMed]
26. Barbosa-Silva MC, Barros AJ. Bioelectrical impedance analysis in clinical practice: a new perspective on its use beyond body composition equations. Curr Opin Clin Nutr Metab Care 2005;8:311-7. [Crossref] [PubMed]
27. Norman K, Wirth R, Neubauer M, et al. The bioimpedance phase angle predicts low muscle strength, impaired quality of life, and increased mortality in old patients with cancer. J Am Med Dir Assoc 2015;16:173.e17-22. [Crossref] [PubMed]
28. Kobayashi T, Murakami T, Ono H, et al. Segmental phase angle can predict incidence of severe exacerbation in male patients with COPD. Nutrition 2025;132:112681. [Crossref] [PubMed]
29. Scheller J, Chalaris A, Schmidt-Arras D, et al. The pro- and anti-inflammatory properties of the cytokine interleukin-6. Biochim Biophys Acta 2011;1813:878-88. [Crossref] [PubMed]
30. Barnes PJ. Corticosteroid effects on cell signalling. Eur Respir J 2006;27:413-26. [Crossref] [PubMed]
31. Pellegrino R, Viegi G, Brusasco V, et al. Interpretative strategies for lung function tests. Eur Respir J 2005;26:948-68. [Crossref] [PubMed]
32. Reddel HK, Taylor DR, Bateman ED, et al. An official American Thoracic Society/European Respiratory Society statement: asthma control and exacerbations: standardizing endpoints for clinical asthma trials and clinical practice. Am J Respir Crit Care Med 2009;180:59-99. [Crossref] [PubMed]
33. Pencina MJ, D'Agostino RB Sr, D'Agostino RB Jr, et al. Evaluating the added predictive ability of a new marker: from area under the ROC curve to reclassification and beyond. Stat Med 2008;27:157-72; discussion 207-12. [Crossref] [PubMed]
34. Vickers AJ, Elkin EB. Decision curve analysis: a novel method for evaluating prediction models. Med Decis Making 2006;26:565-74. [Crossref] [PubMed]
35. Steyerberg EW, Harrell FE Jr, Borsboom GJ, et al. Internal validation of predictive models: efficiency of some procedures for logistic regression analysis. J Clin Epidemiol 2001;54:774-81. [Crossref] [PubMed]
36. Greener JG, Kandathil SM, Moffat L, et al. A guide to machine learning for biologists. Nat Rev Mol Cell Biol 2022;23:40-55. [Crossref] [PubMed]