Ross Catherine E, Harrysson Iliana J, Goel Veena V, Strandberg Erika J, Kan Peiyi, Franzon Deborah E, Pageler Natalie M
1Division of Medicine Critical Care, Department of Medicine, Boston Children's Hospital, Boston, MA. 2Department of Pediatrics, Santa Clara Valley Medical Center, San Jose, CA. 3Division of Pediatric Hospital Medicine, Department of Pediatrics, Palo Alto Medical Foundation, Sutter Health, Palo Alto, CA. 4Biomedical Informatics, Stanford University School of Medicine, Stanford, CA. 5Statistical Unit, Department of Pediatrics, Stanford University School of Medicine, Stanford, CA. 6Division of Critical Care Medicine, Department of Pediatrics, University of California, San Francisco, San Francisco, CA. 7Division of Critical Care Medicine, Department of Pediatrics, Stanford University School of Medicine, Stanford, CA. 8Department of Clinical Informatics, Stanford Children's Health, Stanford, CA. 9Division of Systems Medicine, Department of Pediatrics, Stanford University School of Medicine, Stanford, CA.
Pediatr Crit Care Med. 2017 May;18(5):469-476. doi: 10.1097/PCC.0000000000001150.
Pediatric early warning systems using expert-derived vital sign parameters demonstrate limited sensitivity and specificity in identifying deterioration. We hypothesized that modified tools using data-driven vital sign parameters would improve the performance of a validated tool.
Retrospective case control.
Quaternary-care children's hospital.
Hospitalized, noncritically ill patients less than 18 years old. Cases were defined as patients who experienced an emergent transfer to an ICU or out-of-ICU cardiac arrest. Controls were patients who never required intensive care. Cases and controls were split into training and testing groups.
The Bedside Pediatric Early Warning System was modified by integrating data-driven heart rate and respiratory rate parameters (modified Bedside Pediatric Early Warning System 1 and 2). Modified Bedside Pediatric Early Warning System 1 used the 10th and 90th percentiles as normal parameters, whereas modified Bedside Pediatric Early Warning System 2 used fifth and 95th percentiles.
The training set consisted of 358 case events and 1,830 controls; the testing set had 331 case events and 1,215 controls. In the sensitivity analysis, 207 of the 331 testing set cases (62.5%) were predicted by the original tool versus 206 (62.2%; p = 0.54) with modified Bedside Pediatric Early Warning System 1 and 191 (57.7%; p < 0.001) with modified Bedside Pediatric Early Warning System 2. For specificity, 1,005 of the 1,215 testing set control patients (82.7%) were identified by original Bedside Pediatric Early Warning System versus 1,013 (83.1%; p = 0.54) with modified Bedside Pediatric Early Warning System 1 and 1,055 (86.8%; p < 0.001) with modified Bedside Pediatric Early Warning System 2. There was no net gain in sensitivity and specificity using either of the modified Bedside Pediatric Early Warning System tools.
Integration of data-driven vital sign parameters into a validated pediatric early warning system did not significantly impact sensitivity or specificity, and all the tools showed lower than desired sensitivity and specificity at a single cutoff point. Future work is needed to develop an objective tool that can more accurately predict pediatric decompensation.
使用专家得出的生命体征参数的儿科早期预警系统在识别病情恶化方面显示出有限的敏感性和特异性。我们假设使用数据驱动的生命体征参数的改良工具将提高经过验证的工具的性能。
回顾性病例对照研究。
四级护理儿童医院。
18岁以下住院的非危重症患者。病例定义为紧急转至重症监护病房（ICU）或发生ICU外心脏骤停的患者。对照为从未需要重症监护的患者。病例和对照被分为训练组和测试组。
通过整合数据驱动的心率和呼吸频率参数对床边儿科早期预警系统进行改良（改良床边儿科早期预警系统1和2）。改良床边儿科早期预警系统1使用第10和第90百分位数作为正常参数，而改良床边儿科早期预警系统2使用第5和第95百分位数。
训练集包括358例病例事件和1830例对照；测试集有331例病例事件和1215例对照。在敏感性分析中，原始工具预测了331例测试集病例中的207例（62.5%），改良床边儿科早期预警系统1预测了206例（62.2%；p = 0.54），改良床边儿科早期预警系统2预测了191例（57.7%；p < 0.001）。对于特异性，原始床边儿科早期预警系统识别出1215例测试集对照患者中的1005例（82.7%），改良床边儿科早期预警系统1识别出1013例（83.1%；p = 0.54），改良床边儿科早期预警系统2识别出1055例（86.8%；p < 0.001）。使用任何一种改良床边儿科早期预警系统工具，敏感性和特异性均无净增益。
将数据驱动的生命体征参数整合到经过验证的儿科早期预警系统中对敏感性或特异性没有显著影响，并且所有工具在单一临界值时均显示出低于预期的敏感性和特异性。未来需要开展工作来开发一种能够更准确预测儿科失代偿的客观工具。
