Oral Microbiome, Serum Markers Improve Childhood Asthma Risk Prediction

New research found that combining oral microbiome and inflammatory mediator data with clinical history provides a more accurate way to identify children at risk of future severe asthma attacks.1

Children with persistent asthma who experienced ≥ 1 severe attack in the previous year have a 2- to 2.5-fold higher risk of future severe attacks, according to earlier studies.2 Detecting which children are at risk for asthma attacks early on can help clinicians tailor and optimize management.

Investigators sought to develop a prediction model for future asthma attacks in children that encompasses both epidemiological and clinical data, along with objective biomarkers.1 In their predictive model, the team aimed to incorporate salivary microbiome and serum inflammatory mediator profiles with asthma attack history. They assessed the model during a discovery (SysPharmPediA) and a replication (U-BIOPRED) phase.

“This approach aligns with current recommendations that predictive models of severe asthma attacks should include a combination of epidemiological/clinical data, as well as objective biomarkers,” wrote study investigator Shahriyar Shahbazi Khamas, PharmD, PhD candidate, from Amsterdam University Medical Center in the Netherlands, and colleagues.1

Investigators classified the 154 children, 121 in the SysPharmPediA phase and 33 in the U-BIOPRED phase, into at-risk and no-risk groups.1 The presence or absence of ≥ 1 severe attack over the 1-year follow-up served as the basis for classification. Children were between the ages of 6 and 17 years and were mostly Caucasian (SysPharmPediA: at risk (77.1%), no risk (49.1%); U-BIOPRED: at risk (88.9%), no risk (50%) and female SysPharmPediA: at risk (60%), no risk (30.9%); U-BIOPRED: at risk (59.3%), no risk (66.7%).

Using a random forest approach, the team trained prediction models on 70% of the dataset, incorporating past attack history, microbiome composition, and serum inflammatory mediators. Investigators then tested the models on the remaining 30% of the population.1

During the discovery phase, the model trained on past asthma attacks yielded an area under the receiver operating characteristic curve (AUROCC) of about 0.7. Models trained on 6 salivary bacteria or 6 inflammatory mediators achieved similar findings. However, models trained with all 7 features—past asthma attacks, Capnocytophaga, Corynebacterium, and Cardiobacterium, TIMP-4, VEGF, and MIP-3β—achieved the greatest accuracy with an AUROCC of ~0.87.1

“This model showed a substantial improvement in predictive accuracy compared to the individual models; meaning it might better inform physicians when assessing the risk of future asthma attacks in a child, potentially enabling more informed clinical decisions regarding preventive intervention,” investigators wrote.1

In the replication phase, they could not include all the markers from the original model because only 1 inflammatory mediator, VEGF, was available. Still, the combined model with data on VEGF, past asthma attack history, and the 3 salivary bacteria (Capnocytophaga, Corynebacterium, and Cardiobacterium) achieved an AUROCC of 0.84. This showed that the combined model had a strong predictive accuracy for future asthma attacks, even without the other inflammatory mediators.1

Investigators found that, apart from Capnocytophaga, all other predictive bacteria were more prevalent in children who remained attack-free during follow-up. These findings align with earlier research reporting greater sputum levels of Capnocytophaga in children experiencing asthma attacks, with a positive correlation to MIP-1β, a key eosinophil-recruiting mediator.

The final predictive model included three key serum mediators—TIMP-4, VEGF, and MIP-3β—which collectively drive inflammation, immune activation, and airway remodeling. Their interplay may create a cycle of inflammation and structural change that heightens the risk of severe asthma attacks.1

Bacterial products from Capnocytophaga, Corynebacterium, and Cardiobacterium—including lipopolysaccharides, short-chain fatty acids, hydrogen sulfide, and ammonia—may drive airway inflammation and remodeling by affecting VEGF, MMP/TIMP-4 balance, and MIP-3β–mediated immune responses. These pathways likely contribute to asthma progression and merit further investigation.1

“The integration of salivary bacterial composition and inflammatory profiles in serum showed promise for improving predictive accuracy and may offer insights into the mechanisms underlying asthma attacks,” investigators concluded.1 “Compared to conventional biomarkers like FENO and blood eosinophil count, which showed poor predictive performance in this study, the combined model demonstrated superior predictive capability… Notably, while a history of past asthma attacks emerged as a strong individual predictor…our results show that incorporating microbiome and inflammatory mediator data significantly enhances model performance.”

References

1. Shahbazi Khamas S, Brinkman P, Neerincx AH, et al. Complementary Predictors for Asthma Attack Prediction in Children: Salivary Microbiome, Serum Inflammatory Mediators, and Past Attack History. Allergy. Published online August 18, 2025. doi:10.1111/all.70004

2. Covar RA, Szefler SJ, Zeiger RS, et al. Factors associated with asthma exacerbations during a long-term clinical trial of controller medications in children. J Allergy Clin Immunol. 2008;122(4):741-747.e4. doi:10.1016/j.jaci.2008.08.021