School of Health Studies, Western University, London, Ontario, Canada.
Institute for Clinical Evaluative Sciences, Ontario, Canada; Department of Epidemiology and Biostatistics, Western University, London, Ontario, Canada.
Can J Diabetes. 2017 Jun;41(3):322-328. doi: 10.1016/j.jcjd.2016.11.003. Epub 2017 Mar 3.
To determine the positive predictive value and sensitivity of an International Statistical Classification of Diseases and Related Health Problems, 10th Revision, coding algorithm for hospital encounters concerning hypoglycemia.
We carried out 2 retrospective studies in Ontario, Canada. We examined medical records from 2002 through 2014, in which older adults (mean age, 76) were assigned at least 1 code for hypoglycemia (E15, E160, E161, E162, E1063, E1163, E1363, E1463). The positive predictive value of the algorithm was calculated using a gold-standard definition (blood glucose value <4 mmol/L or physician diagnosis of hypoglycemia). To determine the algorithm's sensitivity, we used linked healthcare databases to identify older adults (mean age, 77) with laboratory plasma glucose values <4 mmol/L during a hospital encounter that took place between 2003 and 2011. We assessed how frequently a code for hypoglycemia was present. We also examined the algorithm's performance in differing clinical settings (e.g. inpatient vs. emergency department, by hypoglycemia severity).
The positive predictive value of the algorithm was 94.0% (95% confidence interval 89.3% to 97.0%), and its sensitivity was 12.7% (95% confidence interval 11.9% to 13.5%). It performed better in the emergency department and in cases of more severe hypoglycemia (plasma glucose values <3.5 mmol/L compared with ≥3.5 mmol/L).
Our hypoglycemia algorithm has a high positive predictive value but is limited in sensitivity. Although we can be confident that older adults who are assigned 1 of these codes truly had a hypoglycemia event, many episodes will not be captured by studies using administrative databases.
确定国际疾病分类第十版(ICD-10)编码算法对医院低血糖就诊的阳性预测值和敏感度。
我们在加拿大安大略省进行了两项回顾性研究。我们检查了 2002 年至 2014 年的病历,其中老年人(平均年龄 76 岁)至少被分配了一个低血糖代码(E15、E160、E161、E162、E1063、E1163、E1363、E1463)。该算法的阳性预测值使用金标准定义(血糖值<4mmol/L 或医生诊断低血糖)进行计算。为了确定算法的敏感度,我们使用链接的医疗保健数据库来识别 2003 年至 2011 年期间在医院就诊时血糖值<4mmol/L 的老年人(平均年龄 77 岁)。我们评估了低血糖代码出现的频率。我们还检查了该算法在不同临床环境下的性能(例如,住院患者与急诊患者,按低血糖严重程度)。
该算法的阳性预测值为 94.0%(95%置信区间 89.3%至 97.0%),敏感度为 12.7%(95%置信区间 11.9%至 13.5%)。它在急诊科和更严重的低血糖情况下表现更好(血糖值<3.5mmol/L 与≥3.5mmol/L 相比)。
我们的低血糖算法具有较高的阳性预测值,但敏感度有限。虽然我们可以确信被分配这些代码之一的老年人确实发生了低血糖事件,但许多事件不会被使用行政数据库的研究捕捉到。