高張食鹽水與甘露醇對神經重症病患住院死亡率影響之比較效果研究：雙資料庫目標試驗模擬分析

高張食鹽水與甘露醇對神經重症病患住院死亡率影響之比較效果研究：雙資料庫目標試驗模擬分析背景與目的：滲透治療是神經重症病患（創傷性腦損傷、自發性腦出血、蜘蛛網膜下腔出血）合併顱內壓升高之第一線藥物介入，常用藥物為 20% 甘露醇（Mannitol）與 3% 高張食鹽水（HTS）。既有隨機對照試驗樣本量不足，缺乏以住院死亡率為主要結果之高品質比較效果證據。本研究採目標試驗模擬（TTE）框架，比較 HTS 與 Mannitol 對神經重症病患住院死亡率之影響。方法：以新使用者主動比較藥物設計，利用 eICU-CRD（N = 508）為主分析資料庫，MIMIC-IV（N = 416）為外部驗證資料庫，遵循 TARGET 2025 報告準則。以重疊權重傾向分數加權（ATO-IPW）估計平均治療效應（ATE），輔以雙重穩健估計、Cox 模型及因果森林分析驗證穩健性，並進行預先規範之亞族群與敏感度分析。結果：eICU 主分析顯示 HTS 之住院死亡率風險差為 −5.2 個百分點（p = 0.260），未達顯著；MIMIC-IV 外部驗證效應更強（ATE = −12.1%，p = 0.031），達統計顯著且方向一致。TBI 子群效應最一致且最大（eICU −8.9%，MIMIC −22.2%，p = 0.021），兩庫合併達顯著（pooled RD = −0.136，p = 0.034）。因果森林顯示 MIMIC 中 100%、eICU 中 71.5% 個體之條件治療效應方向有利於 HTS，正對照與 E-value 分析支持研究效度。結論：雙資料庫目標試驗模擬顯示 HTS 優於 Mannitol 之一致方向性證據；整體族群未達顯著，但 TBI 子群（尤其重度病患）效應具臨床意義，支持未來設計大型隨機對照試驗以死亡率驗證其臨床效益。

Comparative Effectiveness of Hypertonic Saline versus Mannitol on In-Hospital Mortality in Neurocritical Care: A Two-Database Target Trial EmulationBackground and Purpose: Osmotherapy is the first-line intervention for elevated intracranial pressure in neurocritical care patients with traumatic brain injury (TBI), intracerebral hemorrhage, and subarachnoid hemorrhage. Evidence comparing 20% mannitol and 3% hypertonic saline (HTS) on in-hospital mortality remains limited to underpowered trials. This study applied a Target Trial Emulation (TTE) framework to compare HTS versus mannitol on in-hospital mortality among neurocritical care patients.Methods: Using a new-user active-comparator design, this study analyzed eICU-CRD (N = 508) as the primary database and MIMIC-IV (N = 416) as an external validation database, following the TARGET 2025 guideline. Average treatment effects (ATE) were estimated using overlap-weighted propensity scores (ATO-IPW), with doubly robust estimation, Cox models, and causal forest analysis as confirmatory approaches, alongside pre-specified subgroup and sensitivity analyses.Results: In eICU, HTS showed a risk difference of -5.2 percentage points versus mannitol (p = 0.260), not significant. MIMIC-IV validation showed a stronger, significant effect (ATE = -12.1%, p = 0.031), directionally consistent with eICU. The TBI subgroup showed the most consistent, largest effect (eICU -8.9%; MIMIC -22.2%, p = 0.021), with a significant pooled estimate (pooled RD = -0.136, p = 0.034). Causal forest analysis found individual treatment effects favoring HTS in 100% of MIMIC and 71.5% of eICU patients (M=20 imputations, simple average); positive-control and E-value analyses supported the study design.Conclusion: This dual-database target trial emulation provides consistent directional evidence that HTS may reduce in-hospital mortality versus mannitol. While the overall effect was not significant, the effect in the TBI subgroup, particularly among severe patients, was clinically meaningful, supporting a future randomized trial using mortality as the primary outcome.