Identifying injection drug use and estimating population size of people who inject drugs using healthcare administrative datasets
Identifying injection drug use and estimating population size of people who inject drugs using healthcare administrative datasets
复制标题
DOI：
10.1016/j.drugpo.2018.02.001
复制
发表时间：
2018-05-01
影响因子：
4.4
通讯作者：
Krajden, Mel
中科院分区：
文献类型：
--
作者：
Janjua, Naveed Zafar;Islam, Nazrul;Krajden, Mel
背景大型链接的医疗保健管理数据集可用于监测为注射毒品者提供预防和治疗服务的计划。然而，行政数据集中的诊断代码并不能区分非注射和注射药物使用（IDU）。我们验证了基于诊断代码和处方记录的算法，这些诊断代码和处方记录代表了管理数据集中基于访谈的IDU数据。不列颠哥伦比亚省肝炎检测人员队列（BC-HTC）包括1990年至2015年期间在不列颠哥伦比亚省检测HCV/HIV或报告HBV/HCV/HIV/结核病病例的170万人，与包括医生就诊在内的行政数据集相关，住院和处方药记录。在BC-HTC（n = 6559）中纳入的病例子集中，通过访谈评估IDU，作为HIV或HCV/HBV诊断时加强监测的一部分，将其用作金标准。将IDU和注射相关感染（IRI）的ICD-9/ICD-10编码与阿片类药物替代治疗（OST）记录一起分组到管理数据集中的多个IDU算法中。我们通过计算灵敏度、特异性、阳性预测值和阴性预测值来评估IDU算法的性能。结果：基于IDU或IRI和药物滥用代码的算法灵敏度最高（90-94%），特异性最低（42-73%）。需要药物滥用和IRI的算法灵敏度较低（57-60%），特异性较高（90-92%）。最佳的敏感性和特异性组合被发现与两个医疗访问或一个单一的住院注射药物与OST（83%/82%）和无OST（78%/83%）分别。根据包括两次医疗访问，一次住院或OST记录的算法，（不列颠哥伦比亚11-65岁人群的1.2%）不列颠哥伦比亚最近的PWID，基于3年期间的健康接触结论：使用链接的管理数据中的诊断代码识别PWID的算法可以用于跟踪针对PWID的编程的进展。通过基于人口的数据集，该工具可用于为急需的PWID人口规模估计提供信息。
Background. Large linked healthcare administrative datasets could be used to monitor programs providing prevention and treatment services to people who inject drugs (PWID). However, diagnostic codes in administrative datasets do not differentiate non-injection from injection drug use (IDU). We validated algorithms based on diagnostic codes and prescription records representing IDU in administrative datasets against interview-based IDU data.Methods: The British Columbia Hepatitis Testers Cohort (BC-HTC) includes similar to 1.7 million individuals tested for HCV/HIV or reported HBV/HCV/HIV/tuberculosis cases in BC from 1990 to 2015, linked to administrative datasets including physician visit, hospitalization and prescription drug records. IDU, assessed through interviews as part of enhanced surveillance at the time of HIV or HCV/HBV diagnosis from a subset of cases included in the BC-HTC (n = 6559), was used as the gold standard. ICD-9/ICD-10 codes for IDU and injecting-related infections (IRI) were grouped with records of opioid substitution therapy (OST) into multiple IDU algorithms in administrative datasets. We assessed the performance of IDU algorithms through calculation of sensitivity, specificity, positive predictive, and negative predictive values.Results: Sensitivity was highest (90-94%), and specificity was lowest (42-73%) for algorithms based either on IDU or IRI and drug misuse codes. Algorithms requiring both drug misuse and IRI had lower sensitivity (57-60%) and higher specificity (90-92%). An optimal sensitivity and specificity combination was found with two medical visits or a single hospitalization for injectable drugs with (83%/82%) and without OST (78%/83%), respectively. Based on algorithms that included two medical visits, a single hospitalization or OST records, there were 41,358 (1.2% of 11-65 years individuals in BC) recent PWID in BC based on health encounters during 3- year period (2013-2015).Conclusion: Algorithms for identifying PWID using diagnostic codes in linked administrative data could be used for tracking the progress of programing aimed at PWID. With population-based datasets, this tool can be used to inform much needed estimates of PWID population size.