Abstract:

This review systematically reviews existing research on risk prediction for acute exacerbations of chronic obstructive pulmonary disease （AECOPD）， clarifies the inherent evolutionary logic of relevant research paradigms transitioning from macroscopic clinical scoring systems to microscopic biomarkers， and elaborates on the trade-offs between prediction accuracy， clinical accessibility and methodological quality across different prediction strategies. Based on comprehensive literature retrieval， this paper systematically sorts out and summarizes AECOPD risk prediction-related research， covering prognostic scoring systems， inflammatory immune markers， radiomics， and machine learning-based methods. Among multidimensional scoring systems， the DOSE index is primarily used to predict the risk of future acute exacerbations and the frequency of hospitalization， and it is positively correlated with the number of outpatient visits and hospitalization costs （all P < 0.001）. The ADO index is mainly applied to predict 1-3 year all-cause mortality， with a C-statistic of 0.730 for 1-year mortality in external validation. Its predictive performance is further improved after integrating the ADO index with anemia， nutritional status and other indicators. The BODE index achieves an AUC of 0.82 for predicting 3-year mortality， but its high dependence on cardiopulmonary exercise testing limits its popularization in primary medical institutions. In terms of inflammatory and immune markers， the combined AUC of SII， SIRI， NLR， PLR and MLR for the diagnosis of moderate-to-severe AECOPD reaches 0.837， and the nomogram constructed based on these predictors yields an AUC of 0.806 - 0.839 for predicting AECOPD-related respiratory failure. The combined model of radiomics and machine learning achieves an AUC of up to 0.957 for AECOPD identification， but existing systematic reviews show that the pooled AUC of existing AI/ML models only ranges from 0.73 to 0.77， and 57.9% of included studies are assessed as having a high risk of bias. The BLISS score is designed to predict the risk of respiratory-related hospitalization within the next two years， while the PRECISE‑X model predicts the first severe acute exacerbation， enabling a prospective paradigm shift from “post-event assessment” to “diagnosis-based risk stratification”. AECOPD risk prediction is developing towards a multi-modal， dynamic and early-stage direction. The core priority for future research is not to develop more complex individual models， but to construct a hierarchical and progressive validation system： establishing unified testing standards for core indicators， implementing rigorous multi-center external validation， and integrating machine learning methods on the premise of ensuring model interpretability.

Key words: acute exacerbation of chronic obstructive pulmonary disease, predictive models, inflammatory biomarker, imaging, machine learning