用于估计哮喘患者12个月再入院风险的实用预测模型（READMIT评分）的开发与验证

文献信息

中文摘要

背景： 哮喘急性发作后的再入院给医疗系统带来了沉重负担。为减少再入院而部署资源密集型干预措施需要进行风险分层，但专门针对哮喘相关再入院的预测模型很少。
目的： 开发并验证一个预测模型，用于估计出院后12个月内哮喘相关再入院的风险。
方法： 我们对2018年至2021年间在樟宜综合医院因哮喘入院的1451名患者进行了回顾性队列研究。多变量模型的预测因子选择采用最小绝对收缩和选择算子逻辑回归对候选预测因子进行筛选。模型的性能在同一医院2022-2023年的时间验证队列以及新加坡中央医院2015-2020年的外部队列中进行评估。从最终模型中推导出一个简化的基于评分的风险（READMIT）评分。
结果： 最终模型纳入了九个预测因子——慢性鼻炎、急诊就诊、既往住院、抑郁、多病共存（阻塞性睡眠呼吸暂停、胃食管反流、支气管扩张）、吸入长效毒蕈碱拮抗剂、全身性皮质类固醇治疗。在开发队列中，模型显示出良好的区分度（AUC 0.748，95% CI 0.711-0.785），校准令人满意。在外部队列中区分度仍然良好（AUC 0.737，95% CI 0.712-0.762）。尽管外部队列中绝对风险被高估，但针对基线再入院率进行重新校准提高了预测风险的准确性。简化的READMIT评分表现与完整模型相当，且观察到的风险分层清晰。
结论： 我们开发并外部验证了READMIT评分，这是一个实用的九因素预测工具，用于对12个月哮喘相关再入院风险进行分层。

英文摘要

BACKGROUND: Hospital readmissions after asthma exacerbations contribute substantially to healthcare burden. Deployment of resource-intensive interventions to reduce readmission requires risk stratification, but few prediction models have been developed specifically for asthma-related hospital readmission.
OBJECTIVE: To develop and validate a prediction model to estimate the risk of asthma-related hospital readmission within 12 months of discharge.
METHODS: We conducted a retrospective cohort study of 1451 patients admitted for asthma to Changi General Hospital between 2018 and 2021. Predictor selection for the multivariable model was performed using least absolute shrinkage and selection operator logistic regression of candidate predictors. Performance was evaluated in a temporal-validation cohort from the same hospital during 2022-2023 and an external cohort from Singapore General Hospital during 2015-2020. A simplified points-based risk (READMIT) score was derived from the final model.
RESULTS: The final model incorporated nine predictors - chronic Rhinitis, Emergency Department visit, prior hospital Admission, Depression, Multimorbidity (obstructive sleep apnoea, gastro-oesophageal reflux, bronchiectasis), Inhaled long-acting muscarinic antagonist, systemic corticosteroid Treatment. In the development cohort, the model demonstrated good discrimination (AUC 0.748, 95% CI 0.711-0.785) with satisfactory calibration. Discrimination remained good in the external cohort (AUC 0.737, 95% CI 0.712-0.762). Although absolute risk was overestimated in the external cohort, recalibration for baseline readmission rate improved the accuracy of predicted risks. The simplified READMIT score performed comparably to the full model with clear separation of observed risks.
CONCLUSIONS: We developed and externally validated the READMIT score, a pragmatic nine-factor prediction tool for stratifying 12-month asthma-related hospital readmission risk.