衰弱指數預測與其應用

Abstract

衰弱在老年醫學和醫療保健研究中是一個常見問題。本研究引入了一種新的方法，可應用於12歲以上的人群進而評估其健康狀況。這一方法旨在在準確性、可解釋性和運算效率之間取得平衡，從而建構出預測衰弱指數。我們主要利用全民健康保險資料和四種機器學習演算法建立預測模型。首先，我們使用2009年國民健康訪問調查(NHIS)的數據，建立了基於自我回答的衰弱指數(SFI)。接著，我們使用同一年份的全民健康保險資料庫(NHIRD)以機器學習方法建立基於看診記錄的衰弱指數(CFI)，這包括使用國際疾病分類(ICD)代碼以及醫療花費點數。最後，我們選擇了一個相對精確的模型，用於預測衰弱指數，從而幫助評估個人的健康狀況並提供早期治療的相關建議。根據實證結果，與非自我回答的受訪者相比，自我回答的受訪者其預測結果可顯著提高預測準確度，最大R2約為0.52，而平均絕對誤差約為0.03，衰弱指數在0到0.2之間時，預測能力更佳。在計算疾病時以二元法計算的模型比使用加總法的模型預測能力更好。當應用於預測死亡率時，若預測之衰弱指數高於0.2時，應考慮提早接受治療以延長生命。

Frailty is a common issue in geriatric medicine and healthcare research. This study introduces a novel method applicable to individuals aged 12 and above for assessing their health status. The aim is to balance accuracy, interpretability, and computational efficiency in constructing a predictive frailty index. The primary approach involves utilizing nationwide health insurance data and four machine learning algorithms to establish predictive models. A self-reported Frailty Index (SFI) is initially constructed based on the 2009 National Health Interview Survey (NHIS) data. Subsequently, a claim-based Frailty Index (CFI) is built using the same year's National Health Insurance Research Database (NHIRD) with medical records and International Classification of Diseases (ICD) codes along with healthcare expenditure. Finally, we select a relatively accurate model to predict the frailty index, which helps evaluate individual health conditions and provides relevant suggestions for early intervention. According to empirical results, compared to non-self-reporting respondents, those who self-reported showed significantly improved predictive accuracy, with a maximum R2 of approximately 0.52 and a mean absolute error of around 0.03. The frailty index demonstrates better predictive capability when it ranges between 0 and 0.2. The model using the binary methodology for disease calculation outperformed the one using the summation method for predictive capability. When applied to predict mortality rates, early intervention should be considered if the predicted frailty index exceeds 0.2, aiming to prolong life.