Liu Anqi, Zhang Yue, Yadav Chandra Prakash, Chen Wenjia
Saw Swee Hock School of Public Health, National University of Singapore, Singapore.
J Asthma Allergy. 2025 Apr 19;18:579-589. doi: 10.2147/JAA.S509260. eCollection 2025.
Accurate risk prediction of exacerbations in asthma patients promotes personalized asthma management.
This systematic review aimed to provide an update and critically appraise the quality and usability of asthma exacerbation prediction models which were developed since 2017.
In the Embase and PubMed databases, we performed a systematic search for studies published in English between May 2017 and August 2023, and identified peer-reviewed publications regarding the development of prognostic prediction models for the risk of asthma exacerbations in adult patients with asthma. We then applied the Prediction Risk of Bias Assessment tool (PROBAST) to assess the risk of bias and applicability of the included models.
Of 415 studies screened, 10 met eligibility criteria, comprising 41 prediction models. Among them, 7 (70%) studies used real-world data (RWD) and 3 (30%) were based on trial data to derive the models, 7 (70%) studies applied machine learning algorithms, and 2 (20%) studies included biomarkers like blood eosinophil count and fractional exhaled nitric oxide in the model. PROBAST indicated a generally high risk of bias (80%) in these models, which mainly originated from the sample selection ("Participant" domain, 6 studies) and statistical analysis ("Analysis" domain, 7 studies). Meanwhile, 5 (50%) studies were rated as having a high concern in applicability due to model complexity.
Despite the use of big health data and advanced ML, asthma risk prediction models from 2017-2023 had high risk of bias and limited practical use. Future efforts should enhance generalizability and practicality for real-world implementation.
准确预测哮喘患者的病情加重情况有助于实现个性化哮喘管理。
本系统评价旨在更新并严格评估2017年以来开发的哮喘病情加重预测模型的质量和实用性。
在Embase和PubMed数据库中，我们系统检索了2017年5月至2023年8月期间发表的英文研究，并确定了关于成年哮喘患者哮喘病情加重风险预后预测模型开发的同行评审出版物。然后，我们应用预测偏倚风险评估工具（PROBAST）来评估纳入模型的偏倚风险和适用性。
在筛选的415项研究中，10项符合纳入标准，包括41个预测模型。其中，7项（70%）研究使用了真实世界数据（RWD），3项（30%）基于试验数据推导模型；7项（70%）研究应用了机器学习算法，2项（20%）研究在模型中纳入了血液嗜酸性粒细胞计数和呼出一氧化氮分数等生物标志物。PROBAST表明这些模型普遍存在较高的偏倚风险（80%），主要源于样本选择（“参与者”领域，6项研究）和统计分析（“分析”领域，7项研究）。同时，5项（50%）研究因模型复杂性而被评为适用性方面存在高度担忧。
尽管使用了大量健康数据和先进的机器学习，但2017 - 2023年的哮喘风险预测模型存在较高的偏倚风险且实际应用有限。未来的努力应提高其在现实世界中的可推广性和实用性。
