COVID-19疫苗於中重症病人之效益評估

Abstract

背景：COVID-19疫情自2020年起對全球健康造成重大衝擊，疫苗接種被視為減少重症與死亡的關鍵策略。然而，關於疫苗在中重症族群中對於中長期死亡風險的影響，尤其是在真實世界資料下之實證仍相對稀少。本研究旨在運用時間依賴Cox模型與隨機生存森林模型（Random Survival Forest, RSF），探討COVID-19疫苗接種劑數與臨床特徵對中重症患者全因死亡風險的影響，並比較兩種模型預測效能。

方法：本研究採回溯性世代研究（retrospective cohort study），以台灣某縣市公共衛生資料庫為資料來源，納入自2021年3月22日至2023年3月31日期間通報之COVID-19中重症個案。探討的影響變項包含疫苗接種劑數、加護病房住院紀錄、臨床嚴重度分級等。分析方法包括時間依賴Cox比例風險模型與機器學習隨機生存森林模型，並比較其一致性指數（C-index）與準確性分數（Brier Score）之表現。

結果：本研究共收納通報期間之2,997位COVID-19中重症個案，並集合其疫苗接種狀態，透過時間依賴Cox模型顯示，接種共3劑與共4劑COVID-19疫苗者相較未接種者具較低之死亡風險（Hazard Ratio, HR < 1），但共1劑與共2劑疫苗接種組別與未接種組存活率相近。另一方面，隨機生存森林模型中，「中症」與「入住加護病房」為最重要全死因相關因子，顯示早期識別並介入之個體具有較佳預後；相較之下，「重症」變項之預測能力較弱，可能反映病情快速惡化或延遲治療的高風險。隨機生存森林模型整體預測效能優於時間依賴Cox模型，能成功辨識出疫苗接種間隔、臨床分級等非線性關係對死亡風險的重要性。隨機生存森林模型效能，最佳表現為RSF_200_5，其一致性指標與準確性分數（C-index=0.747，Brier Score=0.153），均優於時間依賴Cox比例風險模型（C-index=0.700, Brier Score=0.282）。

結論：針對中重症個案，COVID-19疫苗追加劑接種與早期臨床介入有助於降低中長期死亡風險，中症族群應視為高優先辨識與介入措施及早提供對象。隨機生存森林模型展現處理真實世界高維資料之潛力，未來可作為臨床照護決策參考。

Background: Since 2020, the COVID-19 pandemic has exerted a profound global health impact. Vaccination has been recognized as a key strategy for reducing severe illness and mortality. However, evidence on the effect of vaccination on medium- to long-term mortality risk among patients with moderate-to-severe COVID-19, particularly from real-world data, remains limited. This study aimed to examine the association between the number of COVID-19 vaccine doses, clinical characteristics, and all-cause mortality risk in moderate-to-severe cases, using a time-dependent Cox proportional hazards model and a Random Survival Forest (RSF) model, and to compare the predictive performance of these two approaches.

Methods: This retrospective cohort study utilized a public health database from a county in Taiwan, including all reported moderate-to-severe COVID-19 cases between March 22, 2021, and March 31, 2023. Variables of interest included vaccination dose number, intensive care unit (ICU) admission, and clinical severity classification. Analyses were performed using a time-dependent Cox proportional hazards model and an RSF machine learning approach. Model performance was evaluated and compared using the concordance index (C-index) and Brier score.

Results: A total of 2,997 moderate-to-severe COVID-19 cases reported during the study period were included. Using the time-dependent Cox model, receipt of three or four doses of COVID-19 vaccine was associated with a lower risk of all-cause mortality compared with no vaccination (hazard ratio [HR] < 1), whereas one- or two-dose recipients had survival rates comparable to the unvaccinated group. In the RSF model, moderate severity and ICU admission emerged as the most important predictors of mortality, suggesting that early identification and timely intervention may improve prognosis. In contrast, severe disease had weaker predictive power, possibly reflecting a high risk of rapid deterioration or delayed treatment. Overall, the RSF outperformed the time-dependent Cox model, successfully capturing nonlinear relationships between mortality risk and factors such as vaccination interval and clinical severity. The best-performing RSF configuration (RSF_200_5) achieved a C-index of 0.747 and a Brier score of 0.153, both superior to the Time-dependent Cox model (C-index = 0.700, Brier score = 0.282).

Conclusion: Among patients with moderate-to-severe COVID-19, booster vaccination and early clinical intervention were associated with a reduced medium- to long-term risk of all-cause mortality. Moderate cases should be prioritized for early identification and timely care. The RSF model demonstrated strong potential for handling high-dimensional real-world data and may serve as a valuable tool to support clinical decision-making.