Abstract: Monitoring inflammation's presence is crucial in both preventive medicine and disease management. However, conventional inflammatory biomarkers, levels of C-reactive protein (CRP), require invasive blood tests for assessment, which can only be conducted in hospitals. In this study, we develop machine learning models to detect inflammation and track its progression from nocturnal breathing signals. The performance was evaluated across multiple datasets comprising 1,174 nights of recordings from 950 individuals. Results demonstrate that our model offers promising accuracy in predicting CRP levels: achieving a Pearson's correlation of $r=0.54$ using breathing signals alone, and $r=0.63$ when incorporating additional data on Body Mass Index (BMI) and demographics. It is the first time that decent accuracy in predicting CRP levels from breathing signals has been achieved. Notably, the model exhibits proficiency in identifying high inflammatory states (CRP $>$ 10 mg/l) with an impressive area-under-the-curve value of $0.84$. The results demonstrates the potential for non-invasive, longitudinal, in-home inflammation monitoring through breathing-based biomarkers.