Warlé-van Herwaarden Margaretha F, Valkhoff Vera E, Herings Ron M C, Engelkes Marjolein, van Blijderveen Jan C, Rodenburg Eline M, de Bie Sandra, Alsma Jelmer, van de Steeg-Gompel Caroline, Kramers Cornelis, Meyboom Ronald H B, Sturkenboom Miriam C J M, De Smet Peter A G M
IQ Healthcare, Radboud University Medical Center (RUMC), Nijmegen, The Netherlands; Community Pharmacy Groesbeek, Groesbeek, The Netherlands.
Pharmacoepidemiol Drug Saf. 2015 May;24(5):495-503. doi: 10.1002/pds.3747. Epub 2015 Feb 12.
To develop a computerized prescreening procedure for the identification of possible/probably Hospital Admissions potential Related to Medications (HARMs).
Pairs of drugs and reasons for hospitalization (generated automatically from the PHARMO record linkage database by using two data mining techniques) were assessed manually to determine whether they represented pharmacologically plausible adverse drug events (PP-ADEs). Two crude samples of these PP-ADEs (from 2005 and 2008) were examined manually to establish causality and preventability on the basis of hospital discharge letters plus medication dispensing data. The results were used to calculate the positive predictive value (PPV) of the crude causality PP-ADEs, the net percentage of possible/probably HARMs, and their potential preventability.
Data mining by Gamma Poisson Shrinkage and trend analysis produced 1330 and 2941 significant drug-event pairs, respectively. After manual assessment, 307 different PP-ADEs remained. The annual prevalence of these PP-ADEs was stable at approximately 8% throughout 2000-2009. Manual assessment of two samples of crude PP-ADEs showed that their causality PPV was 53.7% (95%CI: 52.7%-54.7%) in 2005 and 47.9% (95%CI: 46.9%-49.0%) in 2008. The net contribution of possible/probably HARMs to all acute admissions was 4.6% (95%CI: 4.5%-4.8%) in 2005 and 3.9% (95%CI: 3.8%-4.0%) in 2008. The potential preventability of all possible/probably HARMs in the two samples was 19.3% (95%CI: 18.5-20.1).
Automated pre-selection of PP-ADEs is an efficient way to monitor crude trends. Further validation and manual assessment of the automatically selected hospitalizations is necessary to get a more detailed and precise picture.
开发一种计算机化预筛选程序，以识别可能/很可能与药物相关的住院潜在风险（HARMs）。
对通过两种数据挖掘技术从PHARMO记录链接数据库中自动生成的药物与住院原因对进行人工评估，以确定它们是否代表药理学上合理的药物不良事件（PP-ADEs）。对这些PP-ADEs的两个原始样本（来自2005年和2008年）根据医院出院信件和药物配药数据进行人工检查，以确定因果关系和可预防性。结果用于计算原始因果关系PP-ADEs的阳性预测值（PPV）、可能/很可能的HARMs的净百分比及其潜在可预防性。
Gamma泊松收缩法和趋势分析的数据挖掘分别产生了1330对和2941对显著的药物-事件对。经过人工评估后，仍有307种不同的PP-ADEs。在2000 - 2009年期间，这些PP-ADEs的年患病率稳定在约8%。对两个原始PP-ADEs样本的人工评估表明，其因果关系PPV在2005年为53.7%（95%CI：52.7% - 54.7%），在2008年为47.9%（95%CI：46.9% - 49.0%）。2005年可能/很可能的HARMs对所有急性住院的净贡献为4.6%（95%CI：4.5% - 4.8%），2008年为3.9%（95%CI：3.8% - 4.0%）。两个样本中所有可能/很可能的HARMs的潜在可预防性为19.3%（95%CI：18.5 - 20.1）。
PP-ADEs的自动预筛选是监测原始趋势的有效方法。对自动选择的住院病例进行进一步验证和人工评估，以获得更详细和精确的情况是必要的。
