Pediatric Anesthesia Fellow, Starship Children's Health, 2 Park Road, Grafton, Auckland 1023, New Zealand.
PhD Candidate, Department of Statistics, The University of Auckland, Auckland 1023, New Zealand.
J Pediatr Surg. 2023 Mar;58(3):524-531. doi: 10.1016/j.jpedsurg.2022.07.017. Epub 2022 Jul 28.
Risk prediction models are well established as an adjunct to perioperative decision making, but few exist for pediatric surgical outcomes. The majority of risk tools do not feature Australasian data and do not estimate mortality risk beyond 30-days. Our aim was to develop and validate a model for mortality risk prediction in children (age <18yrs) at 30-days, 90-days and 1 year following all types of surgery using a national database.
The New Zealand Ministry of Health National Minimum Dataset was accessed to obtain clinical and demographic data for all children having surgery between June 1st 2011 and July 1st 2016. Three quarters of the data were used to derive 3 models to predict 30-day, 90-day and 1-year mortality risk, and the remaining data used for validation. We constructed 3 models using data from 135 217 patients, validating a total of 11 covariates for risk prediction. Included were neonate, prematurity, ASA-PS status, heart and lung disease, active malignancy, sepsis, surgical type, surgical severity score, surgical urgency, ethnicity and socioeconomic deprivation. All models showed excellent discrimination (area under the receiver operating characteristic curve (AUROC) values of 0.947, 0.933 and 0.908 respectively) and calibration statistics (calibration slopes of 0.778, 1.125, 1.153, Brier scores of 0.001, 0.002, 0.003 respectively).
Combining objective data with severity indices, NZRISK-Paed presents a risk stratification model which is intuitive and practical. Application of 30-day, 90-day and 1-year percentage mortality risk aids in longer-term planning, shared decision-making and allocation of resource to the individual and to high needs populations. Risk prediction tools add an objective measure to pre-operative assessment but few exist for pediatric surgery and none predict mortality beyond 30-days.
风险预测模型是围手术期决策的重要辅助手段，但针对儿科手术结果的模型却很少。大多数风险工具都没有澳大利亚数据，也无法预测 30 天以上的死亡率。我们的目的是利用全国性数据库，为所有类型手术的儿童（<18 岁）建立并验证一个用于预测 30 天、90 天和 1 年死亡率的模型。
我们访问了新西兰卫生部国家最小数据集，以获取 2011 年 6 月 1 日至 2016 年 7 月 1 日期间所有接受手术的儿童的临床和人口统计学数据。四分之三的数据用于推导 3 个模型，以预测 30 天、90 天和 1 年的死亡率，其余数据用于验证。我们使用来自 135217 名患者的数据构建了 3 个模型，验证了 11 个风险预测的协变量。包括新生儿、早产、ASA-PS 状态、心脏和肺部疾病、活动性恶性肿瘤、脓毒症、手术类型、手术严重程度评分、手术紧急程度、种族和社会经济剥夺。所有模型均具有出色的区分度（接受者操作特征曲线下面积（AUROC）值分别为 0.947、0.933 和 0.908）和校准统计数据（校准斜率分别为 0.778、1.125、1.153，Brier 分数分别为 0.001、0.002、0.003）。
将客观数据与严重程度指数相结合，NZRISK-Paed 提供了一种直观实用的风险分层模型。应用 30 天、90 天和 1 年的死亡率风险百分比有助于进行更长期的规划、共同决策以及为个体和高需求人群分配资源。风险预测工具为术前评估提供了客观衡量标准，但针对儿科手术的工具很少，并且没有一个可以预测 30 天以上的死亡率。
