{"title":"An introduction to analytical methods for the postmarketing surveillance of veterinary vaccines.","authors":"D Siev","doi":"10.1016/s0065-3519(99)80057-3","DOIUrl":null,"url":null,"abstract":"<p><p>Any analysis of spontaneous AER data must consider the many biases inherent in the observation and reporting of vaccine adverse events. The absence of a clear probability structure requires statistical procedures to be used in a spirit of exploratory description rather than definitive confirmation. The extent of such descriptions should be temperate, without the implication that they extend to parent populations. It is important to recognize the presence of overdispersion in selecting methods and constructing models. Important stochastic or systematic features of the data may always be unknown. Our attempts to delineate what constitutes an AER have not eliminated all the fuzziness in its definition. Some count every event in a report as a separate AER. Besides confusing the role of event and report, this introduces a complex correlational structure, since multiple event descriptions received in a single report can hardly be considered independent. The many events described by one reporter would then become inordinately weighted. The alternative is to record an AER once, regardless of how many event descriptions it includes. As a practical compromise, many regard the simultaneous submission of several report forms by one reporter as a single AER, and the next submission by that reporter as another AER. This method is reasonable when reporters submit AERs very infrequently. When individual reporters make frequent reports, it becomes difficult to justify the inconsistency of counting multiple events as a single AER when they are submitted together, but as separate AERs when they are reported at different times. While either choice is imperfect, the latter approach is currently used by the USDA and its licensed manufacturers in developing a mandatory postmarketing surveillance system for veterinary immunobiologicals in the United States. Under the proposed system, summaries of an estimated 10,000 AERs received annually by the manufacturers would be submitted to the USDA. In quantitative summaries, AERs received from lay consumers are usually weighted equally with those received from veterinary health professionals, although arguments have been advanced for separate classifications. The emphasis on AER rate estimation differentiates the surveillance of veterinary vaccines by the USDA CVB from the surveillance of veterinary drugs as practiced by the Food and Drug Administration (FDA) Center for Veterinary Medicine (CVM). The FDA CVM does, in fact, perform a retrodictive causality assessment for individual AERs (Parkhie et al., 1995). This distinction reflects the differences between vaccines and drugs, as well as the difference in regulatory philosophy between the FDA and the USDA. The modified Kramer algorithm (Kramer et al., 1979) used by the FDA relies on features more appropriate to drug therapy than vaccination, such as an ongoing treatment regimen which allows evaluation of the response to dechallenge and rechallenge. In tracking AERs, the FDA has emphasized the inclusion of clinical manifestations on labels and inserts, while the USDA has been reluctant to have such information appear in product literature or to use postmarketing data for this purpose. The potential for the misuse of spontaneous AER data is great. Disinformation is likely when the nature of this type of data is misunderstood and inappropriate analytical methods blindly employed. A greater danger lies in the glib transformation of AER data into something else entirely. Since approval before publication is not required, advertisements for veterinary vaccines appear with claims such as \"over 3 million doses, 99.9905% satisfaction rating,\" or \"11,500,000 doses, 99.98% reaction free.\" These claims, presumably based on spontaneous AERs, are almost fraudulent in their deceptiveness. Are we to suppose that 11.5 million vaccinations were observed for reactions? In comparing the two advertisements, we find the second presumed AER rate is double the first. (ABSTRACT TRU</p>","PeriodicalId":72111,"journal":{"name":"Advances in veterinary medicine","volume":"41 ","pages":"749-74"},"PeriodicalIF":0.0000,"publicationDate":"1999-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://sci-hub-pdf.com/10.1016/s0065-3519(99)80057-3","citationCount":"6","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Advances in veterinary medicine","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1016/s0065-3519(99)80057-3","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 6
Abstract
Any analysis of spontaneous AER data must consider the many biases inherent in the observation and reporting of vaccine adverse events. The absence of a clear probability structure requires statistical procedures to be used in a spirit of exploratory description rather than definitive confirmation. The extent of such descriptions should be temperate, without the implication that they extend to parent populations. It is important to recognize the presence of overdispersion in selecting methods and constructing models. Important stochastic or systematic features of the data may always be unknown. Our attempts to delineate what constitutes an AER have not eliminated all the fuzziness in its definition. Some count every event in a report as a separate AER. Besides confusing the role of event and report, this introduces a complex correlational structure, since multiple event descriptions received in a single report can hardly be considered independent. The many events described by one reporter would then become inordinately weighted. The alternative is to record an AER once, regardless of how many event descriptions it includes. As a practical compromise, many regard the simultaneous submission of several report forms by one reporter as a single AER, and the next submission by that reporter as another AER. This method is reasonable when reporters submit AERs very infrequently. When individual reporters make frequent reports, it becomes difficult to justify the inconsistency of counting multiple events as a single AER when they are submitted together, but as separate AERs when they are reported at different times. While either choice is imperfect, the latter approach is currently used by the USDA and its licensed manufacturers in developing a mandatory postmarketing surveillance system for veterinary immunobiologicals in the United States. Under the proposed system, summaries of an estimated 10,000 AERs received annually by the manufacturers would be submitted to the USDA. In quantitative summaries, AERs received from lay consumers are usually weighted equally with those received from veterinary health professionals, although arguments have been advanced for separate classifications. The emphasis on AER rate estimation differentiates the surveillance of veterinary vaccines by the USDA CVB from the surveillance of veterinary drugs as practiced by the Food and Drug Administration (FDA) Center for Veterinary Medicine (CVM). The FDA CVM does, in fact, perform a retrodictive causality assessment for individual AERs (Parkhie et al., 1995). This distinction reflects the differences between vaccines and drugs, as well as the difference in regulatory philosophy between the FDA and the USDA. The modified Kramer algorithm (Kramer et al., 1979) used by the FDA relies on features more appropriate to drug therapy than vaccination, such as an ongoing treatment regimen which allows evaluation of the response to dechallenge and rechallenge. In tracking AERs, the FDA has emphasized the inclusion of clinical manifestations on labels and inserts, while the USDA has been reluctant to have such information appear in product literature or to use postmarketing data for this purpose. The potential for the misuse of spontaneous AER data is great. Disinformation is likely when the nature of this type of data is misunderstood and inappropriate analytical methods blindly employed. A greater danger lies in the glib transformation of AER data into something else entirely. Since approval before publication is not required, advertisements for veterinary vaccines appear with claims such as "over 3 million doses, 99.9905% satisfaction rating," or "11,500,000 doses, 99.98% reaction free." These claims, presumably based on spontaneous AERs, are almost fraudulent in their deceptiveness. Are we to suppose that 11.5 million vaccinations were observed for reactions? In comparing the two advertisements, we find the second presumed AER rate is double the first. (ABSTRACT TRU
对自发性AER数据的任何分析都必须考虑到在观察和报告疫苗不良事件时固有的许多偏差。由于没有明确的概率结构,需要本着探索性描述的精神使用统计程序,而不是明确的确认。这种描述的范围应该是有节制的,不暗示它们延伸到亲本种群。在选择方法和构建模型时,认识到过度分散的存在是很重要的。数据中重要的随机或系统特征可能总是未知的。我们试图描述什么构成AER并没有消除其定义中的所有模糊性。有些人将报告中的每个事件都视为单独的AER。除了混淆事件和报告的角色之外,这还引入了复杂的关联结构,因为在单个报告中接收到的多个事件描述很难被认为是独立的。一个记者所描述的许多事件就会变得异常重要。另一种方法是记录一次AER,而不管它包含多少个事件描述。作为一种实际的妥协,许多人将一个记者同时提交的几份报告表格视为一个AER,而该记者下一次提交的报告表格则视为另一个AER。当记者很少提交AERs时,这种方法是合理的。当单个报告者频繁地进行报告时,很难证明将多个事件一起提交时作为单个AER计算,而在不同时间报告时作为单独AER计算的不一致性是正确的。虽然这两种选择都不完美,但后一种方法目前被美国农业部及其许可制造商用于制定美国兽医免疫生物制剂的强制性上市后监测系统。根据拟议的系统,制造商每年收到的大约10,000份AERs的摘要将提交给美国农业部。在定量总结中,从非专业消费者那里收到的不良反应通常与从兽医卫生专业人员那里收到的不良反应加权相同,尽管有人提出了单独分类的论点。对AER率估计的强调区分了USDA CVB对兽药疫苗的监督和FDA兽药中心(CVM)对兽药的监督。事实上,FDA的CVM确实对个别AERs进行了回溯性因果评估(Parkhie et al., 1995)。这种区别反映了疫苗和药物之间的差异,以及FDA和USDA之间监管理念的差异。FDA使用的改良Kramer算法(Kramer et al., 1979)依赖于比疫苗接种更适合药物治疗的特征,例如持续的治疗方案,可以评估对挑战和再挑战的反应。在跟踪AERs时,FDA强调在标签和说明书上包括临床表现,而USDA不愿在产品文献中出现此类信息或为此目的使用上市后数据。滥用自发AER数据的可能性很大。当这类数据的性质被误解和盲目使用不适当的分析方法时,就可能产生虚假信息。更大的危险在于将AER数据巧妙地转换成完全不同的东西。由于不需要在发布前批准,兽医疫苗的广告出现了诸如“超过300万剂,99.9905%的满意率”或“1150万剂,99.98%无反应”之类的说法。这些说法,大概是基于自发的aer,在其欺骗性上几乎是欺诈性的。我们是否应该假设观察到1150万次接种疫苗的反应?在比较两个广告时,我们发现第二个假定的AER率是第一个的两倍。(抽象TRU