在醫學領域中，有許多的研究人員開發出各種機器學習模型以支援臨床決策。然而， 電子健康記錄(electronic health records, EHR)的格式在不同機構間存在差異，使得將多 個模型整合到醫療資訊系統(hospital information system, HIS)中變得具有挑戰性且耗時。 為了解決這個問題，我們提出了 MoCab 框架，該框架使用 Fast Healthcare Interoperability Resources (FHIR)作為數據存儲和檢索標準，以將模型部署到多個醫療資 訊系統之中。MoCab 通過標準化的配置設定和導入已保存的模型來部署模型，並基於 FHIR 標準檢索患者數據並將其傳遞給所需的模型來進行預測。基於預測結果，MoCab 透過 Clinical Decision Support (CDS) Hooks 給予醫生警示或通知，並使用基於 SMART on FHIR 架構所開發的應用程式進一步顯示患者的檢驗結果趨勢。此外，MoCab 能夠使 用持續出現的新病患數據不斷訓練與優化模型，以提升已部署模型的預測效能。本文演 示了三種類型的模型如何在 MoCab 中部署以支援臨床決策。我們所提出的 MoCab 框架 可以增強模型在多個醫療資訊系統與電子健康記錄中的可重複利用性，並協助臨床決策 的過程。;Researchers have developed machine learning models to support clinical decision-making in the medical field. However, the format of electronic health records (EHRs) varies across institutions, making integrating these models into a health information system (HIS) challenging and time-consuming. To address this issue, we proposed MoCab, which uses fast healthcare interoperability resources (FHIR) as data storage and retrieval standards for deploying models to various HISs. MoCab deploys models by configuring and importing the saved model. MoCab makes the prediction by retrieving patients′ data from the FHIR server and passing it to the requested model. If a warning or alert is needed based on the prediction results, MoCab alerts physicians through Clinical Decision Support (CDS) Hooks and displays patient laboratory result trends using Substitutable Medical Apps Reusable Technologies (SMART) on FHIR. Moreover, MoCab can fine-tune and improve the prediction performance of deployed models over time with the endless stream of new data. Three types of models are presented to demonstrate how the models can be implemented in MoCab to support decision-making. The proposed MoCab framework can enhance the reusability of models in multiple EHRs and assist in the clinical decision-making process.
