Development of a clinical prediction score for recurrent wheezing following acute lower respiratory tract infection in children

Abstract

BACKGROUND

Recurrent wheezing is a significant complication following acute lower respiratory tract infection (ALRTI) in children and is a known precursor to asthma development. Identifying high-risk children remains challenging, particularly in resource-limited settings.

AIM

To develop and evaluate a simplified clinical prediction score for recurrent wheezing following ALRTI in children.

METHODS

A retrospective cohort study was conducted at Naresuan University Hospital, Thailand, involving children under 15 years diagnosed with ALRTI from July 2020 to December 2021. Participants were followed for 12 months. Recurrent wheezing was defined as ≥ 2 physician-diagnosed episodes. Demographic, clinical, environmental, and laboratory variables were extracted from electronic medical records. Independent predictors were identified using multivariable logistic regression and used to construct a point-based clinical prediction score. Model performance was evaluated using discrimination, calibration, sensitivity, specificity, positive predictive value, negative predictive value and overall accuracy.

RESULTS

Of 280 eligible children, 52 (18.6%) developed recurrent wheezing. Independent predictors included allergic rhinitis, first-degree family history of asthma, eosinophilia, history of multiple LRTIs (≥ 2 prior episodes), food/drug allergy, and thalassemia. A point-based score (0-9) stratified patients into low-, moderate-, and high-risk groups. The model demonstrated good discrimination (area under the receiver operating characteristic curve: 0.79; 95% confidence interval: 0.71-0.88) and adequate calibration. Sensitivity decreased across increasing risk categories (48.57% for low, 34.29% for moderate, and 17.14% for high-risk groups), whereas specificity increased (12.75%, 87.25%, and 100%, respectively). Positive predictive values were 16.04%, 48.0%, and 100%, and negative predictive values were 41.94%, 79.46%, and 77.86%.

CONCLUSION

This clinical prediction score, based on readily available clinical variables, provides a practical tool for early risk stratification of recurrent wheezing in children following ALRTI. External validation is required to ensure the model’s robustness before widespread clinical adoption.

Key Words: Asthma; Children; Respiratory tract infection; Wheezing; Acute lower respiratory tract infection

Core Tip: Recurrent wheezing is a common condition in children with lower respiratory tract infections. This retrospective cohort study developed a predictive tool to estimate the risk of recurrent episodes, aiming to support treatment planning and clinical monitoring, and to identify children at increased risk of developing asthma later in life.

INTRODUCTION

Wheezing in early childhood is a common clinical manifestation, frequently precipitated by acute lower respiratory tract infections (ALRTI), such as bronchiolitis and pneumonia[1,2]. Viral pathogens, predominantly respiratory syncytial virus (RSV) and rhinovirus, are major etiological agents leading to hospitalization in infants and young children and have been strongly implicated in the subsequent development of recurrent wheezing[2,3]. This condition not only impairs the quality of life for affected children and their families but also poses a substantial economic burden on healthcare systems globally. Progression from post-ALRTI wheezing to persistent wheezing and a definitive diagnosis of asthma in later childhood is a significant clinical concern, underscoring the importance of early identification of children at increased risk[4-7].

A substantial body of literature has focused on identifying the risk factors for recurrent wheezing[8-10]. Established predictors include a family history of atopy, infantile eczema, allergic sensitization to aeroallergens, male gender, lower respiratory tract infections (LRTIs), and exposure to environmental tobacco smoke[8-10]. More recent investigations have explored the predictive value of laboratory-based biomarkers, such as peripheral blood eosinophil counts and the composition of the nasopharyngeal microbiota[11]. However, the reported significance and consistency of these factors vary across studies, reflecting heterogeneity in study populations, methodologies, and definitions of wheezing phenotypes. Existing predictive models, such as the Asthma Predictive Index, demonstrate clinical utility but are often constrained by their reliance on specific laboratory investigations (e.g., allergen-specific immunoglobulin E)[12,13]. As a result, these models may not be universally applicable or feasible in all clinical settings, particularly in resource-limited environments.

These limitations highlight a critical research gap: The absence of a simple yet robust clinical prediction tool that can be readily applied in routine practice to identify children at high risk of developing recurrent wheezing following an episode of ALRTI. Although epidemiological studies addressing this issue have been conducted, data from specific populations, including children in Thailand, remain limited, and the risk of wheezing recurrence after ALRTI has not been well characterized. For example, Chinratanapisit et al[14] conducted a longitudinal study involving 236 children aged 6 months to 5 years who were hospitalized with acute wheezing in four hospitals located in Bangkok and its surrounding provinces. After a one-year follow-up period, the prevalence of recurrent wheezing was reported to be 23.1%. However, the study did not investigate risk factors associated with recurrent wheezing. Similarly, Chantawarangul et al[15] reported a recurrent wheezing prevalence of 45.2%, although factors predictive of recurrence were not comprehensively evaluated.

Consequently, there is a compelling need for a pragmatic predictive score derived from readily obtainable demographic and clinical parameters to facilitate the early identification of children at increased risk of recurrent wheezing and to support timely preventive and management strategies.

The objective of this study was to develop and evaluate a clinical prediction score, based on readily available clinical and demographic risk factors, to estimate the risk of recurrent wheezing within 12 months following an episode of ALRTI in children, with the aim of supporting early identification of high-risk individuals and informing timely clinical management.

MATERIALS AND METHODS

Study design, setting, and participants

This study employed a retrospective cohort design to develop a clinical prediction score for recurrent wheezing in children following (ALRTI). The study population consisted of pediatric patients aged < 15 years who were diagnosed with ALRTI by a physician and received treatment at Naresuan University Hospital, a university-affiliated tertiary care center located in Phitsanulok, Thailand.

Participants were identified through the hospital’s electronic medical record system. Data were extracted for patients presenting from July 1, 2020 to December 31, 2021, which constituted the accrual period. All participants were followed for 12 months from the date of the index ALRTI diagnosis to assess the occurrence of recurrent wheezing. The overall study period thus encompassed both the accrual and the follow-up phases and concluded on December 31, 2022.

Inclusion criteria: (1) Children aged < 15 years at the index diagnosis; (2) Physician-confirmed diagnosis of ALRTI (e.g., bronchiolitis, pneumonia, tracheobronchitis) in either inpatient or outpatient visits; and (3) Clinical presentation within the defined accrual period.

Exclusion criteria: (1) Pre-existing asthma, bronchopulmonary dysplasia, cystic fibrosis, or other chronic pulmonary diseases; (2) A documented history of wheezing episodes prior to the index ALRTI; (3) Known congenital heart disease, immunodeficiency, or significant anatomical airway abnormalities; and (4) Incomplete electronic medical records, specifically regarding baseline demographics or 12-month follow-up.

Outcome

The primary outcome of this study was the occurrence of recurrent wheezing within 12 months following the initial diagnosis of ALRTI. Recurrent wheezing was defined as two or more episodes of physician-diagnosed wheezing occurring after the initial ALRTI episode, with each episode separated by a symptom-free interval of at least four weeks.

Outcomes were assessed through a comprehensive review of electronic medical records, including outpatient visits, emergency department encounters, and hospital admissions during the follow-up period. In cases where documentation was insufficient to confirm recurrent wheezing, direct verification was performed via telephone interviews with caregivers, using a structured questionnaire administered by trained research staff.

To minimize assessment bias, outcome assessors were blinded to the predictor variables used in model development. These assessors had no involvement in patient care and were unaware of baseline clinical characteristics at the time of outcome evaluation.

Predictors

All candidate predictors were obtained from electronic medical records at the time of the initial ALRTI diagnosis and prior to outcome occurrence. Data collection was retrospective and included demographic, clinical, environmental, and laboratory variables. The following variables were analyzed: Birth history: Gestational age, prematurity, mode of delivery, and birth weight. Demographics: Age, sex, and place of residence (urban or rural). Medical history: Underlying disease, history of eczema, personal history of allergy, and family history of asthma. Environmental exposures: Daycare attendance, exposure to passive smoking, maternal smoking during pregnancy, and presence of household pets. Laboratory investigations: Complete blood count and nasopharyngeal swab results for RSV, influenza virus, and severe acute respiratory syndrome coronavirus 2 (coronavirus disease 2019).

All predictor variables were measured and recorded during the initial ALRTI episode to ensure temporal precedence relative to the primary outcome. To minimize potential bias, data extraction was performed by trained research personnel using standardized abstraction forms. A robust double-blinding protocol was implemented: Data extractors remained blinded to the outcome status (development of recurrent wheezing), while outcome assessors were blinded to all baseline predictor data during follow-up evaluation phase. Follow-up information included the time to first wheezing recurrence and the total number of recurrent wheezing episodes within 12 months.

Study size estimation

The sample size was determined based on preliminary data from a pilot study of 20 pediatric patients treated for ALRTI at Naresuan University Hospital. This pilot cohort was equally distributed between patients with recurrent wheezing (n = 10) and those without (n = 10). Among the evaluated variables, residential status was identified as the most influential factor in determining the maximum required sample size and was therefore used for sample size estimation.

Based on the pilot data, the prevalence of recurrent wheezing was estimated at 20%. Utilizing the Riley RD formula for clinical prediction models - with a type I error of 0.05, a statistical power of 80%, and the inclusion of 6 parameters - the required sample size was calculated to be 264 patients divided in recurrent wheezing. This cohort requirement was stratified into 53 patients with recurrent wheezing and 211 patients without recurrence.

Missing data

Data on eosinophilia counts were missing in 137 cases (48.9%), primarily because laboratory testing was not routinely performed in some outpatient settings. A complete case analysis was conducted; consequently, only participants with complete datasets for all predictors in the final multivariable logistic regression model included 137 participants with complete data for all candidate predictors. An additional sample size assessment based on the Riley’s framework confirmed that the available complete-case sample was sufficient for prediction model development, satisfying the criterion for small overfitting (expected shrinkage ≤ 10%).

Statistical analysis

Categorical variables were summarized using n (%) and compared using the exact probability test. Continuous variables were assessed for normality; normally distributed were presented as mean ± SD and compared using the independent t-test. Statistical uncertainty was expressed using 95% two-sided confidence intervals (CIs), and P < 0.05 was considered statistically significant.

Multivariable binary logistic regression was performed to identify independent predictors of recurrent wheezing within 12 months. Clinically and biologically plausible variables were included in the initial model, including thalassemia, allergic rhinitis, food or drug allergy, first-degree family history of asthma, history of ≥ 2 previous LRTIs, and eosinophilia. Adjusted odds ratios (aORs) with 95%CIs were calculated.

Predictor selection was based on clinical relevance and statistical significance, with variables retained in the final model using backward stepwise elimination and a retention threshold of P < 0.05. Model performance was assessed in accordance with prediction model reporting standards, including discrimination assessed using the area under the receiver operating characteristic curve (AUC), calibration assessed using the Hosmer-Lemeshow goodness-of-fit test and calibration plots, and overall model accuracy evaluated using the Brier score. All statistical analyses were conducted using Stata version 18 (StataCorp, College Station, TX, United States).

Scoring system and risk stratification

A point-based scoring system was derived by transforming the regression coefficients (β) of the final model into weighted integers. Total risk scores were calculated for each participant and stratified into three categories based on the predicted probability of recurrent wheezing: Low risk: Predicted probability < 25%. Moderate risk: Predicted probability 25%-85%. High risk: Predicted probability > 85%.

The discriminative performance of each risk group was evaluated using the AUC and observed recurrence rates were examined to confirm consistency across risk strata. This stratification facilitates clinical decision-making regarding follow-up intensity, preventive strategies, and early referral.

Ethical considerations

The study protocol received ethical approval from the Naresuan University Institutional Review Board (approval No. P3-0102/2566). All procedures were conducted in accordance with the Declaration of Helsinki.

RESULTS

Participants

During the study period, 314 children were diagnosed with ALRTI at Naresuan University Hospital. Of these, 34 patients were excluded due to congenital heart disease (n = 14), pre-existing asthma (n = 15), or bronchopulmonary dysplasia (n = 5), resulting in 280 eligible participants included in the analysis. All eligible participants were followed for 12 months after the initial diagnosis of ALRTI. Among the 280 patients, 52 children (18.6%) developed recurrent wheezing during follow-up, while 228 (81.4%) did not. The mean time to the first recurrent wheezing episode among affected participants was 98 days. The participant flow is summarized in Figure 1.

Figure 1 Flow diagram of participant enrollment. ALRTI: Acute lower respiratory tract infection.

Baseline characteristics of the study population are detailed in Table 1. The mean age was 31.2 ± 2.3 months to in the recurrent wheezing group and 30.7 ± 1.1 months in the non-recurrent wheezing group (P = 0.843). Gender distribution was comparable between groups, with males comprising 57.7% of the recurrent group and 54.4% of the non-recurrent group (P = 0.758). In the comparison between the recurrent wheezing and non-recurrent wheezing groups, several clinical factors were found to be significantly different. Thalassemia disease was more frequently observed in the recurrent wheezing group than in the non-recurrent wheezing group (14.5% vs 3.5%, respectively; P = 0.028). The prevalence of allergic rhinitis was also significantly higher among patients with recurrent wheezing (23.1%) compared with those without recurrent wheezing (9.2%; P = 0.009). A family history of asthma in first-degree relatives was more common in the recurrent wheezing group than in the non-recurrent wheezing group (9.6% vs 2.2%; P = 0.022). In addition, a history of multiple LRTIs was significantly associated with recurrent wheezing, with 59.6% of patients in the recurrent wheezing group reporting multiple LRTIs compared with 21.1% in the non-recurrent wheezing group (P < 0.001). Eosinophilia, defined as an absolute eosinophil count ≥ 500 cells/mm², was significantly more prevalent in the recurrent wheezing group than in the non-recurrent wheezing group (25.7% vs 3.9%, respectively; P = 0.001). Other variables showed no statistically significant differences between the two groups (Table 1).

Table 1 Baseline characteristics of the study population by recurrent wheezing status, n (%).

Characteristics/risk factor	Recurrent wheezing (n = 52)	No recurrent wheezing (n = 228)	P value
Age (months), mean ± SD	31.2 ± 2.3	30.7 ± 1.1	0.843
Male	30 (57.7)	124 (54.4)	0.758
Preterm	4 (7.7)	27 (11.8)	0.472
Low birth weight	5 (9.6)	33 (14.5)	0.501
Delivery by C/S	24 (46.2)	126 (55.3)	0.281
Urban	37 (71.2)	171 (75.0)	0.599
Underlying disease	22 (42.3)	64 (28.1)	0.066
Thalassemia disease	6 (14.5)	8 (3.5)	0.028
G6PD deficiency	2 (3.9)	3 (1.3)	0.233
Allergic rhinitis	12 (23.1)	21 (9.2)	0.009
Atopic dermatitis	8 (15.4)	31 (13.6)	0.824
Food or drug allergy	6 (11.5)	13 (5.7)	0.136
Exposure smoking during pregnancy	7 (13.5)	19 (8.3)	0.288
Exposure smoking	13 (25.0)	41 (18.0)	0.248
First degree related asthma	5 (9.6)	5 (2.2)	0.022
Day care attending	36 (69.2)	134 (57.8)	0.208
Multiple URTI	32 (61.5)	121 (53.1)	0.284
Multiple LRTI	31 (59.6)	48 (21.1)	< 0.001
Eosinophilia (AEC ≥ 500 cell/mm2)	9 (25.7)	4 (3.9)	0.001
Pets	10 (19.2)	39 (17.1)	0.238
RSV infection	19 (36.5)	98 (43.0)	0.438

URTI: Upper respiratory tract infection; LRTI: Lower respiratory tract infection; AEC: Absolute eosinophil count; RSV: Respiratory syncytial virus.

Model development

All candidate predictors were evaluated for their univariable association with recurrent wheezing using logistic regression. For most predictors, analyses were conducted using data from the full cohort of 280 participants. However, eosinophilia was analyzed in a subset of 137 participants because eosinophil count data were unavailable for the remaining participants. The results of the univariable analyses are presented in Table 2. Thalassemia [odds ratio (OR): 3.59; 95%CI: 1.19-10.83; P = 0.023], allergic rhinitis (OR: 2.96; 95%CI: 1.35-6.49; P = 0.007), first-degree family history of asthma (OR: 4.74; 95%CI: 1.32-17.05; P = 0.017), history of ≥ 2 prior LRTIs (OR: 5.54; 95%CI: 2.92-10.49; P < 0.001), and eosinophilia (≥ 500 cells/mm³; OR: 8.48; 95%CI: 2.42-29.74; P = 0.001) were significantly associated with recurrent wheezing. Other variables, including food or drug allergies, daycare attendance, RSV infection, and hematologic values (hematocrit, hemoglobin), were not significantly associated with the outcome (Table 2).

Table 2 Univariable logistic regression analysis of risk factors for recurrent wheezing.

Risk factor	Odd ratio	95%CI	P value
Thalassemia disease	3.59	1.19-10.83	0.023
Allergic rhinitis	2.96	1.35-6.49	0.007
Food or drug allergy	1.84	0.68-5.01	0.230
Maternal smoking during pregnancy	1.71	0.68-4.31	0.255
Passive smoke exposure	1.52	0.75-3.10	0.249
First-degree related family history of asthma	4.74	1.32-17.05	0.017
Day care attendance	1.58	0.83-3.00	0.166
Multiple LRTIs	5.54	2.92-10.49	< 0.001
Eosinophilia1	8.48	2.42-29.74	0.001
RSV infection	0.76	0.41-1.42	0.396

¹Eosinophilia was analyzed in 137 participants with available eosinophil count data; all other variables were analyzed in the full cohort of 280 participants.

LRTI: Lower respiratory tract infection; RSV: Respiratory syncytial virus; CI: Confidence interval.

Model specification and scoring system

A multivariable logistic regression analysis was performed to construct the final prediction model. After adjusting for potential confounders, six variables were retained in the model (Table 3). While some factors did not reach a P < 0.05, they were included based on their clinical relevance and their contribution to the model’s overall predictive performance: Thalassemia disease (aOR = 2.66; 95%CI: 0.53-13.28; P = 0.234), allergic rhinitis (aOR = 4.41; 95%CI: 1.09-17.85; P = 0.037), food or drug allergy (aOR = 3.57; 95%CI: 0.50-25.44; P = 0.204), first-degree family history of asthma (aOR = 5.25; 95%CI: 0.48-57.36; P = 0.174), multiple prior LRTIs (aOR = 3.83; 95%CI: 1.51-9.73; P = 0.005), and eosinophilia (aOR = 5.43; 95%CI: 1.31-22.53; P = 0.020).

Table 3 Multivariable logistic regression analysis of risk factors and predictors of recurrent wheezing, and assigned item scores.

Predictors	OR	95%CI	P value	Coefficient	Score
Thalassemia disease	2.66	0.53-13.28	0.234	0.98	1
Allergic rhinitis	4.41	1.09-17.85	0.037	1.48	2
Food or drug allergy	3.57	0.50-25.44	0.204	1.27	1
First degree related asthma	5.25	0.48-57.36	0.174	1.65	2
Multiple LRTIs	3.83	1.51-9.73	0.005	1.34	1
Eosinophilia	5.43	1.31-22.53	0.020	1.69	2
Total score		Risk category		Recommended clinical follow-up
0-1 point		Low risk		Routine care; no follow up
2-3 point		Moderate risk		Follow-up within 6 months
4-9 point		High risk		Follow-up within 3 months

OR: Odds ratio; CI: Confidence interval; LRTI: Lower respiratory tract infection.

The full model specification on a logit scale is as follows: Logit (P) = -2.17 + 0.98 (thalassemia) + 1.48 (allergic rhinitis) + 3.57 (food or drug allergy) + 5.25 (first-degree related asthma) + 3.83 (multiple LRTIs) + 5.43 (eosinophilia). This equation allows for the estimation of the probability of recurrent wheezing within 12 months by coding each binary predictor as 1 (present) or 0 (absent).

To enhance clinical utility, a point-based scoring system was developed based on by transforming the regression coefficients into weighted integers (Table 3). The primary drivers of the score were first-degree related family history of asthma, allergic rhinitis, and eosinophilia, each assigned 2 points. A history of multiple LRTIs (≥ 2), thalassemia, and food or drug allergies were each assigned 1 point.

Risk categorization was determined by analyzing the relationship between the cumulative score and the predicted probability of recurrence (Figure 2). Patients were stratified into three distinct risk tiers: Low risk (score 0-1; < 25% probability of recurrent wheezing), moderate risk (score 2-3; 25%-85% probability of recurrent wheezing), and high risk (score 4-9; > 85% probability of recurrent wheezing).

Figure 2  Relationship between total risk score and predicted probability of recurrent wheezing.

Model performance

The final prediction model demonstrated robust discriminative capacity, with an AUC of 0.79, indicating a strong ability to differentiate between patients who developed recurrent wheezing and those who did not (Figure 3). Model calibration was evaluated using the Hosmer-Lemeshow goodness-of-fit test, which yielded no significant lack of fit (P > 0.05). This suggests excellent agreement between the predicted probabilities and the observed clinical outcomes (Figure 4). Additionally, the Brier score was 24.0%, further supporting the model’s overall predictive accuracy.

Figure 3 Receiver operating characteristic curve of the prediction model. ROC: Receiver operating characteristic curve.

Figure 4 Calibration plot of the prediction model. CITL: Calibration-in-the-large; AUC: Area under the receiver operating characteristic curve; CI: Confidence interval.

The clinical predictor score - comprising thalassemia disease, allergic rhinitis, food or drug allergy, first-degree related family history of asthma, history of multiple LRTIs, and eosinophilia - was utilized to stratify patients into low-, moderate-, and high-risk categories. This stratification is intended to guide clinical judgment and long-term follow-up planning following an initial ALRTI (Table 3).

Risk group-specific diagnostic performance is summarized in Table 4. Sensitivity decreased across increasing risk categories [48.57% (95%CI: 31.38-66.01) in low-risk, 34.29% (95%CI: 19.13-52.21) in moderate-risk, and 17.14% (95%CI: 6.56-33.65) in high-risk groups], whereas specificity increased [12.75% (95%CI: 6.96-20.81), 87.25% (95%CI: 79.19-93.04), and 100% (95%CI: 96.45-100), respectively]. Positive predictive values (PPV) were 16.04% (95%CI: 11.88-21.31), 48.0% (95%CI: 31.77-64.66), and 100% (95%CI: 54.07-100), and negative predictive values were 41.94% (95%CI: 28.36-56.85), 79.46% (95%CI: 75.07-83.25), and 77.86% (95%CI: 75.16-80.35) across low-, moderate-, and high-risk groups, respectively. The positive likelihood ratios for the low-, moderate-, and high-risk groups were 0.56 (95%CI: 0.39-0.79), 2.69 (95%CI: 1.36-5.33), and infinite (reflecting 100% specificity), respectively. Correspondingly, the negative likelihood ratios were 4.04 (95%CI: 2.21-7.36), 0.75 (95%CI: 0.59-0.97), and 0.83 (95%CI: 0.71-0.96).

Table 4 Performance of the clinical score and risk classification for recurrent wheezing.

Score (points)	Risk classification	PPV (95%CI)	NPV (95%CI)	Sensitivity (95%CI)	Specificity (95%CI)	PLR (95%CI)	NLR (95%CI)
0-1	Low	16.04% (11.88-21.31)	41.94% (28.36-56.85)	48.57% (31.38-66.01)	12.75% (6.96-20.81)	0.56 (0.39-0.79)	4.04 (2.21-7.36)
2-3	Moderate	48.0% (31.77-64.66)	79.46% (75.07-83.25)	34.29% (19.13-52.21)	87.25% (79.19-93.04)	2.69 (1.36-5.33)	0.75 (0.59-0.97)
4-9	High	100% (54.07-100)	77.86% (75.16-80.35)	17.14% (6.56-33.65)	100% (96.45-100)	∞ (due to 100% specificity)	0.83 (0.71-0.96)

PPV: Positive predictive value; CI: Confidence interval; NPV: Negative predictive value; PLR: Positive likelihood ratio; NLR: Negative likelihood ratio.

DISCUSSION

Wheezing is a common clinical manifestation in early childhood, with approximately half of children experiencing at least one episode before the age of six[16]. Early-onset wheezing is associated with an increased likelihood of persistent recurrences, greater healthcare utilization, and an increased risk of developing chronic conditions, such as asthma and other respiratory infections, later in life[17-19]. In this retrospective study, we developed a clinical prediction model to estimate the risk of recurrent wheezing within 12 months following ALRTI using routinely available clinical and laboratory variables.

In our cohort, 18.6% of children developed recurrent wheezing during follow-up. The final multivariable model demonstrated acceptable discriminative ability (AUC = 0.79; 95%CI: 0.71-0.88) and good calibration, indicating reasonable performance in distinguishing children who did and did not develop recurrent wheezing. Allergic rhinitis, a history of multiple prior LRTIs (≥ 2 episodes), and eosinophilia (≥ 500 cells/mm³) remained independently associated with recurrent wheezing after adjustment. Thalassemia, food or drug allergy, and first-degree family history of asthma were retained in the final model based on established clinical relevance, despite not reaching statistical significance in the adjusted analysis.

These findings are consistent with previous studies identifying family history of asthma, allergic rhinitis, and eosinophilia as strong predictors of recurrent wheezing and progression to asthma in young children[20-23]. A family history of asthma in first-degree relatives reflects genetic susceptibility and shared environmental influences that shape immune responses and airway reactivity[24,25]. Although family history did not remain an independent predictor in the adjusted model, its inclusion in the prediction score acknowledges its well-recognized role in pediatric wheezing disorders.

Peripheral blood eosinophilia emerged as a strong predictor of recurrent wheezing, supporting its role as a biomarker of Th2-driven inflammation and early subclinical airway involvement, even in children without a formal diagnosis of asthma[26,27]. Elevated eosinophil levels may contribute to airway epithelial injury, mucus hypersecretion, and structural airway remodeling, thereby increasing susceptibility to recurrent wheezing episodes[28].

The association between allergic rhinitis and recurrent wheezing observed in this study aligns with prior evidence supporting the “united airway” hypothesis, in which inflammation of the upper and lower airways coexist and interact[29,30]. Chronic nasal inflammation and impaired mucociliary clearance in children with allergic rhinitis may predispose them to lower airway involvement and recurrent respiratory symptoms. Food and drug allergies, although not independently associated with recurrent wheezing after adjustment, were included in the prediction model as markers of systemic atopy and early immune dysregulation within the atopic march.

Interestingly, thalassemia was retained as a predictor in the final model, highlighting a potential non-atopic contributor to recurrent wheezing risk. Children with thalassemia often experience chronic anemia, recurrent infections, and systemic inflammation, which may adversely affect pulmonary function and airway reactivity[31,32]. While the biological mechanisms underlying this association remain unclear, its inclusion highlights the potential role of systemic comorbidities in respiratory vulnerability following ALRTI. Regarding the possible mechanism of thalassemia in children with higher risk of asthma and wheezing, stated that the hemorheological changes in thalassemia include greater rigidity of red blood cells in the capillary bed, which can contribute to changes in bronchial circulation and bronchial hyperactivity yet without solid evidence[33]. The association between asthma-like symptoms and major hemoglobinopathies, including thalassemia and sickle cell disease, reveals a complex interplay between these conditions. Studies demonstrate a significant correlation between thalassemia and increased vulnerability to asthma, particularly in children. The underlying mechanisms involve dysregulated inflammatory pathways and shared immunological consequences, suggesting overlapping pathophysiological mechanisms[34]. Further research is essential to clarify the relationship between recurrent wheezing and hemoglobinopathies.

A history of multiple LRTIs was independently associated with wheezing recurrence, consistent with previous studies[8,35]. Recurrent infections may reflect both impaired host defense and a driver of persistent airway inflammation and remodeling. Viral pathogens, particularly RSV and rhinovirus, have been shown to induce long-term changes in airway immune responses, thereby increasing the risk of chronic wheezing phenotypes in susceptible children[36,37].

Altogether, these findings underscore the complex interplay between genetic predisposition, immune dysregulation, infectious burden, and systemic comorbidities in the pathogenesis of recurrent wheezing in early childhood. By integrating these factors into a single clinical prediction score, the model captures multiple mechanistic pathways contributing to wheezing recurrence and provides a structured approach to early risk stratification. This multidimensional framework reflects real-world clinical complexity and supports identification of children who may benefit from closer monitoring following ALRTI.

The model demonstrated distinct performance characteristics across increasing risk categories. Sensitivity progressively decreased from the low-risk group (48.57%) to the moderate-risk (34.29%) and high-risk groups (17.14%), whereas specificity increased substantially (12.75%, 87.25%, and 100%, respectively). This inverse relationship indicates that the model becomes more conservative in classifying positive cases as risk level increases. Notably, the high-risk group exhibited perfect specificity (100%) and PPV (100%), indicating that all children classified as high risk subsequently developed recurrent wheezing. This finding highlights the model’s strength in reliably identifying truly high-risk individuals and minimizing false-positive classifications at this level.

Conversely, the reduced sensitivity observed in the high-risk group indicates that a substantial proportion of patients who eventually develop recurrent wheezing may be classified into lower risk strata. This trade-off implies that while the model is highly reliable when recognizing high-risk individuals, it may under-detect some patients who are genuinely at risk. Therefore, the model functions effectively as a confirmatory “rule-in” tool rather than a broad screening instrument for high-risk classification.

The moderate-risk group showed a more balanced performance profile, characterized by moderate specificity (87.25%), a PPV of 48.0%, a negative predictive value of 79.46%, a positive likelihood ratio of 2.69 (95%CI: 1.36-5.33), and a negative likelihood ratio 0.75 (95%CI: 0.59-0.97), suggesting reasonable discriminative capability but with an inherent degree of clinical uncertainty. In contrast, the low-risk group exhibited higher sensitivity but poor specificity and PPV, indicating a tendency toward the over-classification of positive cases. This profile remains acceptable in early screening contexts, where minimizing missed cases is prioritized over false positives.

Overall, these findings demonstrate that the model’s risk stratification is most reliable within the high-risk category, defined by its absolute specificity and PPV. Future model refinements may focus on improving should aim to enhance sensitivity within the higher-risk strata without substantially compromising specificity, potentially through the recalibration of decision thresholds or integration of additional predictive biomarkers.

Implications

The clinical prediction model developed in this study has potential utility for the early identification and management of children at increased risk of recurrent wheezing following an episode of ALRTI. By integrating readily available clinical variables - such as allergic rhinitis, family history of asthma, eosinophilia, and prior LRTIs - the model offers a pragmatic approach to risk stratification that can be applied during routine clinical care.

In clinical practice, this tool may assist pediatricians and general practitioners in distinguishing children who require closer follow-up, intensified monitoring, or early preventive interventions from those at lower risk who may be safely managed with standard care, early recognition of high-risk patients could facilitate timely implementation of targeted strategies, such as optimized management of allergic comorbidities, caregiver education, environmental control measures, and referral to pediatric pulmonology or allergy specialists to reduce recurrent healthcare utilization and mitigate progression toward persistent wheezing or asthma.

The model’s reliance on routinely available clinical and laboratory parameters supports its feasibility for implementation, particularly in resource-limited settings where access to advanced diagnostic tools may be limited. However, before routine clinical adoption, external validation in larger, multicenter, and more diverse populations is required to confirm its predictive performance and generalizability. Until such validation is conducted in independent hospitals or populations, the score should be interpreted and applied with caution. Furthermore, the development of a digital clinical decision-support tool, such as a mobile application or web-based calculator, may enhance usability and facilitate integration into routine clinical practice.

Future research should explore the inclusion of other potentially relevant predictors, including specific environmental exposures, viral subtypes, advanced immunological biomarkers, and polygenic risk scores, which may further augment the model’s discriminative accuracy. Moreover, prospective cohort studies utilizing standardized outcome assessments and extended longitudinal follow-up are necessary to strengthen the evidence base and support broader clinical adoption.

Ultimately, this model represents a significant advancement toward personalized risk assessment in pediatric respiratory care. By providing a structured framework for early identification, this tool can contribute to better-informed clinical decision-making and improved long-term outcomes for children presenting with early-life wheezing illnesses.

Limitations

Several limitations of this study warrant consideration. First, the study was conducted at a single tertiary care center, which may limit the generalizability of the findings to broader pediatric populations or healthcare settings, particularly those in rural or primary care contexts. Second, while the total sample size (n = 280) met the calculated requirements for model development, the number of observed outcome events (n = 52) is relatively limited in relation to the number of candidate predictors. This carries an inherent risk of model overfitting, potentially limiting the reliability of the score when applied to external datasets. Third, missing data were present for several predictors, especially laboratory variables such as peripheral blood eosinophil counts, which were unavailable for approximately 49% of participants. To address this, a complete-case analysis was performed; however, this approach may have introduced selection bias and reduced statistical power. Fourth, although several predictors were retained in the final model based on their clinical relevance, some variables demonstrated wide 95%CIs, likely reflecting the limited number of outcome events and reduced sample size resulting from complete-case analysis. Consequently, the estimated effect sizes for these predictors should be interpreted with caution, as their precision may be limited. External validation in larger cohorts is warranted to confirm the stability and generalizability of the model. Fifth, although the prediction model demonstrated excellent specificity in the high-risk group, its sensitivity was relatively low (17.14%). Consequently, a substantial proportion of children who subsequently developed recurrent wheezing (82.86%) were not identified as high risk by the model. Therefore, the tool should be interpreted primarily as a rule-in rather than a rule-out instrument and should not be used as the sole screening method for identifying children at risk of recurrent wheezing. Finally, the retrospective cohort design is inherently subject to limitations associated with medical record review, including variability in documentation, potential misclassification of exposures and outcomes, and incomplete capture of relevant factors. Certain variables, such as environmental exposures, viral subtypes, and severity of the initial infection, may not have been systematically recorded and therefore could not be fully assessed. Future studies with larger, multicenter cohorts and prospective data collection, with external validation of the prediction model, are warranted to confirm its robustness, generalizability, and clinical utility across diverse settings.

CONCLUSION

This study developed a clinical prediction model to identify children at increased risk of recurrent wheezing within 12 months following an episode of ALRTI. Thalassemia, allergic rhinitis, family history of asthma, prior LRTIs, and eosinophilia emerged as key predictors of wheezing recurrence. The model demonstrated acceptable discriminative performance and provides a practical framework for early risk stratification using routinely available clinical data. With external validation, this tool may support proactive management of high-risk children in routine clinical practice.

ACKNOWLEDGEMENTS

The authors would like to express their gratitude to Miss Daisy Gonzales of the International Relations Section, Faculty of Medicine, Naresuan University, for her valuable assistance in editing the manuscript.