Identification of Incident Injuries in Hospital Discharge Registers
Identification of Incident Injuries in Hospital Discharge Registers
复制标题
DOI：
10.1097/ede.0b013e318181319e
复制
发表时间：
2008-11-01
期刊：
影响因子：
5.4
通讯作者：
Michaelsson, Karl
中科院分区：
文献类型：
--
作者：
Gedeborg, Rolf;Engquist, Henrik;Michaelsson, Karl
背景：出院损伤数据是流行病学研究的潜在有力数据来源。然而，可靠的识别事故伤害入院是必要的。本研究的目的是建立一个预测模型，以确定医院住院事件，基于从医院出院登记的变量。方法：1998-2004年瑞典有743,022例外伤住院。其中，23,920人在乌普萨拉县，其中24%的人之前曾受伤。为了确定这些入院是新伤还是早伤再入院，我们回顾了817个随机选择的医院记录。基于患者年龄、入院类型（紧急或选择性）、上一次入院的时间间隔、主要诊断和科室类型，建立了意外伤害入院的预测模型。结果：最终预测模型判别性较好（c-statistic = 0.969）。该模型使用最佳截止水平应用于验证数据集，并根据每个损伤类别中既往损伤入院的比例调整结果的敏感性和特异性。再入院比例最高的损伤是髋关节挫伤（35%）。然而，使用预测模型，识别偶发性髋挫伤的灵敏度为94%(95%置信区间= 93%-95%)，特异性为95%（94%-97%）。其他损伤类别的准确率更高。结论：利用基于出院数据的预测模型，可以准确地将意外伤害入院与再入院区分开来。
Background: Hospital discharge data on injuries constitute a potentially powerful data source for epidemiologic studies. However, reliable identification of incident injury admissions is necessary. The objective of this study was to develop a prediction model for identifying incident hospital admissions, based on variables derived from a hospital discharge register.Methods: There were 743,022 hospital admissions for injury in Sweden 1998-2004. Of these, 23,920 were in the county of Uppsala and 24% of these people had previous injury admissions. To determine if these admissions were new injuries or readmissions for earlier injuries, we reviewed 817 randomly selected hospital records. A prediction model for incident injury admissions was developed on the basis of patient age, type of admission (urgent or elective), time interval from the previous injury admission, main diagnosis, and department type.Results: The final prediction model showed good discrimination (c-statistic = 0.969). This model was applied to the validation dataset using the optimal cut-off level, and the resulting sensitivity and specificity were adjusted according to the proportion with a previous injury admission in each injury category. The injury with the highest proportion of possible readmissions was hip contusion (35%). Nevertheless, using the prediction model, incident hip contusions were identified with a sensitivity of 94% (95% confidence interval = 93%-95%) and a specificity of 95% (94%-97%). The accuracy was higher for all other injury categories.Conclusions: Incident injury admissions can be accurately separated from readmissions using a prediction model based on information derived from hospital discharge data.