Salivary microbiome as a prognostic biomarker for asthma attacks in children: results from a machine-learning approach in the SysPharmPediA study

<bold>Introduction:</bold> The composition of oral microbiota has been previously linked to immunomodulation, asthma risk and severity. We aimed to evaluate the salivary microbiome's potential as a biomarker for predicting future asthma attacks in children. <bold>Methods:</bold> In this prospective pan-European cohort study, saliva samples were obtained from 143 children (6 to 17 years) with moderate-to-severe asthma at inclusion [SysPharmPediA study]. Children were followed-up for 12 months and were classified into risk (those with at least one severe attack) and no-risk groups. Their microbiome was compared using global (alpha/beta) diversity. Statistical modeling to classify the risk group involved: 1) splitting the cohort into training (65%) and validation (35%) sets, 2.1) building a random forest (RF) machine learning model for taxa selection; 2.2) building final predictive model using selected taxa, 3) area under the receiver operating characteristic curves (AUROCCs) calculation <bold>Results:</bold> Complete data was available for 101 patients (11.7±2.6 years, 40.6% female, 36/101 exacerbators). Global diversity did not significantly differ between the groups. RF selected 4 genera, including <italic>Corynebacterium</italic>, <italic>Capnocytophaga</italic>, <italic>Neisseria</italic>, <italic>Streptobacillus</italic> differentiating the groups with AUROCCs of 0.70 (95% CI 0.55-0.84) for the training and 0.70 (95% CI 0.51-0.88) for the validation set. <bold>Conclusion:</bold> The salivary microbiota composition shows promise as a potential biomarker for predicting attacks risk. Our findings supports the need for further exploration of oral microbiota to uncover novel pathophysiologic insights and may open avenues for innovative treatment approaches.