Department of Clinical Informatics, Sydney Adventist Hospital, Sydney, NSW, Australia.
Department of Cardiothoracic Surgery, Royal North Shore Hospital, Sydney, NSW, Australia.
Crit Care Med. 2021 Oct 1;49(10):e961-e967. doi: 10.1097/CCM.0000000000005064.
To determine whether a statistically derived, trend-based, deterioration index is superior to other early warning scores at predicting adverse events and whether it can be integrated into an electronic medical record to enable real-time alerts.
Forty-three variables and their trends from cases and controls were used to develop a logistic model and deterioration index to predict patient deterioration greater than or equal to 1 hour prior to an adverse event.
Two large Australian teaching hospitals.
Cases were considered as patients who suffered adverse events (unexpected death, unplanned ICU transfer, urgent surgery, and rapid-response alert) between August 1, 2016, and April 1, 2019.
The logistic model and deterioration index were tested on historical data and then integrated into an electronic medical record for a 6-month prospective "silent" validation.
Data were acquired from 258,732 admissions. There were 8,002 adverse events. The addition of vital sign and laboratory trend values to the logistic model increased the area under the curve from 0.84 to 0.89 and the sensitivity to predict an adverse event 1-48 hours prior from 0.35 to 0.41. A 48-hour simulation showed that the logistic model had a higher area under the curve than the Modified Early Warning Score and National Early Warning Score (0.87 vs 0.74 vs 0.71). During the silently run prospective trial, the sensitivity of the deterioration index to detect adverse event any time prior to the adverse event was 0.474, 0.369 1 hour prior, and 0.327 4 hours prior, with a specificity of 0.972.
A deterioration prediction model was developed using patient demographics, ward-based observations, laboratory values, and their trends. The model's outputs were converted to a deterioration index that was successfully integrated into a live hospital electronic medical record. The sensitivity and specificity of the tool to detect inpatient deterioration were superior to traditional early warning scores.
确定基于统计学的趋势恶化指数是否优于其他早期预警评分，以预测不良事件，以及是否可以将其整合到电子病历中以实现实时警报。
使用病例和对照组中的 43 个变量及其趋势来开发逻辑模型和恶化指数，以预测不良事件发生前 1 小时以上的患者恶化。
两家澳大利亚大型教学医院。
病例被视为 2016 年 8 月 1 日至 2019 年 4 月 1 日期间发生不良事件（意外死亡、非计划性 ICU 转科、紧急手术和快速反应警报）的患者。
对逻辑模型和恶化指数进行了历史数据测试，然后将其整合到电子病历中进行为期 6 个月的前瞻性“静默”验证。
共获取 258732 次入院数据。发生 8002 例不良事件。将生命体征和实验室趋势值添加到逻辑模型中，将曲线下面积从 0.84 增加到 0.89，将预测不良事件 1-48 小时前的灵敏度从 0.35 增加到 0.41。48 小时模拟显示，逻辑模型的曲线下面积高于改良早期预警评分和国家早期预警评分（0.87 对 0.74 对 0.71）。在静默运行的前瞻性试验中，恶化指数检测任何时间点不良事件前的灵敏度为 0.474，1 小时前为 0.369，4 小时前为 0.327，特异性为 0.972。
使用患者人口统计学数据、病房观察、实验室值及其趋势开发了一种恶化预测模型。模型的输出被转换为恶化指数，并成功整合到实时医院电子病历中。该工具检测住院患者恶化的灵敏度和特异性优于传统的早期预警评分。
