Department of Medical Informatics, Academic Medical Center, University of Amsterdam, Amsterdam, The Netherlands.
Crit Care Med. 2012 Feb;40(2):373-8. doi: 10.1097/CCM.0b013e318232d7b0.
It has been postulated that prognostic models based on administrative data can provide valid adjusted mortality rates in specific patient populations. In this study we compared the performance and robustness of a model based on administrative data (customized hospital standardized mortality ratio) and a model based on clinical data (customized Simplified Acute Physiology Score II) in the Dutch intensive care unit population.
Cohort study of intensive care unit records from a national intensive care unit quality registry linked to administrative records from the Dutch National Medical Registration. The hospital standardized mortality ratio and Simplified Acute Physiology Score II models were first-level customized on the intensive care unit population.
Fifty-five Dutch intensive care units.
A total of 66,564 intensive care unit patients admitted from 2005 to 2008.
None.
Performance expressed by measures of discrimination, accuracy, and calibration (area under the receiver operating characteristic curve, Brier score, Hosmer-Lemeshow Ĉ-statistic, and calibration plots). Additionally, the robustness of the models was assessed by simulating changes in the population's severity of illness and analyzing the effect on the intensive care units' standardized mortality ratios.The area under the receiver operating characteristic curve and Brier score of the customized Simplified Acute Physiology Score II were significantly superior to that of the customized hospital standardized mortality ratio (0.85 and 0.11 vs. 0.77 and 0.13, respectively). Calibration plots showed good agreement between observed and predicted mortality for low-risk patients in both models, with more discrepancy in the high-risk patients when using the customized hospital standardized mortality ratio. Severity of illness had influence on the intensive care units' standardized mortality ratios in both models, but the customized Simplified Acute Physiology Score II showed more robustness.
The customized Simplified Acute Physiology Score II outperforms the customized hospital standardized mortality ratio in the Dutch intensive care unit population. Comparing institutions based on standardized mortality ratios can be unfavorable for those with a more severely ill intensive care unit population, especially when using the customized hospital standardized mortality ratio.
据推测，基于行政数据的预后模型可以为特定患者人群提供有效的校正死亡率。本研究比较了基于行政数据的模型（定制医院标准化死亡率）和基于临床数据的模型（定制简化急性生理学评分 II）在荷兰重症监护病房人群中的表现和稳健性。
一项全国性重症监护病房质量登记处的重症监护病房记录的队列研究，与荷兰国家医疗登记处的行政记录相链接。首先对重症监护病房人群进行医院标准化死亡率和简化急性生理学评分 II 模型的一级定制。
55 家荷兰重症监护病房。
2005 年至 2008 年期间收治的 66564 名重症监护病房患者。
无。
通过区分度、准确性和校准度的测量值（接受者操作特征曲线下的面积、Brier 评分、Hosmer-Lemeshow Ĉ 统计量和校准图）来表示性能。此外，通过模拟人群疾病严重程度的变化并分析其对重症监护病房标准化死亡率的影响来评估模型的稳健性。定制简化急性生理学评分 II 的接受者操作特征曲线下面积和 Brier 评分明显优于定制医院标准化死亡率（分别为 0.85 和 0.11 与 0.77 和 0.13）。校准图显示，在两个模型中，低危患者的观察死亡率与预测死亡率之间存在良好的一致性，而在高危患者中，使用定制医院标准化死亡率时则存在更多差异。在两个模型中，疾病严重程度都对重症监护病房的标准化死亡率有影响，但定制简化急性生理学评分 II 表现出更强的稳健性。
在荷兰重症监护病房人群中，定制简化急性生理学评分 II 优于定制医院标准化死亡率。基于标准化死亡率比较机构可能对重症监护病房患者病情较重的机构不利，尤其是使用定制医院标准化死亡率时。
