Ilkhanoff Leonard, Lewis James D, Hennessy Sean, Berlin Jesse A, Kimmel Stephen E
Department of Medicine, Cardiovascular Division, Hospital of the University of Pennsylvania, Philadelphia, PA, USA.
Pharmacoepidemiol Drug Saf. 2005 Aug;14(8):513-22. doi: 10.1002/pds.1129.
To determine whether specific limitations in electronic database studies may lead to biased estimates of the association between prescription, non-selective non-aspirin non-steroidal anti-inflammatory drugs (NANSAIDs) and myocardial infarction (MI) METHODS: Using our case-control study of NANSAIDs and first, non-fatal MI, we determined the odds ratio (OR) for prescription NANSAIDs and MI. In the 'Replicating Electronic Database Analysis,' we considered non-prescription NANSAID users to be 'non-users,' did not stratify by aspirin use, and did not adjust for confounders typically unavailable or incomplete in existing databases. In the 'Misclassification Assessment Analysis,' we removed non-prescription NANSAIDs from the 'non-user' category. In the 'Confounding Assessment Analysis #1,' we additionally adjusted for smoking, family history, and years of education. In the 'Confounding Assessment Analysis #2,' we also adjusted for body mass index (BMI) and physical activity. In the 'Interaction Assessment Analysis,' we stratified on aspirin use and repeated the latter analysis.
The prevalence of current NANSAID and aspirin use was higher in our controls than in electronic database studies, consistent with the fact that non-prescription NANSAIDs accounted for 81% of all NANSAID use. Education, physical activity, and BMI also were associated with prescription NANSAID use. When each potential source of bias was removed, the OR for NANSAIDs moved further from 1.0 (i.e., toward a protective association with MI): 'Replicating Electronic Database' analysis (OR 1.00, 95% confidence interval [CI]: 0.78--1.28); 'Misclassification Assessment Analysis' (OR 0.89, 95%CI: 0.70--1.14); 'Confounding Assessment Analysis #1' (OR 0.85, 95%CI: 0.66--1.10); 'Confounding Assessment Analysis #2' (OR 0.78, 95%CI: 0.60--1.01); 'Interaction Assessment Analysis' (OR 0.69, 95%CI: 0.51--0.95).
Limitations in electronic databases may be responsible for the lack of association of NANSAIDs on lower MI risk noted in these studies. Further studies-preferably randomized trials-are needed to address the risk-benefit ratio of NANSAID use.
确定电子数据库研究中的特定局限性是否可能导致对处方非选择性非阿司匹林非甾体抗炎药（NANSAIDs）与心肌梗死（MI）之间关联的估计产生偏差。方法：利用我们关于NANSAIDs与首次非致命性MI的病例对照研究，我们确定了处方NANSAIDs与MI的比值比（OR）。在“重复电子数据库分析”中，我们将非处方NANSAIDs使用者视为“非使用者”，未按阿司匹林使用情况进行分层，也未对现有数据库中通常无法获取或不完整的混杂因素进行调整。在“错误分类评估分析”中，我们将非处方NANSAIDs从“非使用者”类别中剔除。在“混杂因素评估分析#1”中，我们额外调整了吸烟、家族史和受教育年限。在“混杂因素评估分析#2”中，我们还调整了体重指数（BMI）和身体活动情况。在“交互作用评估分析”中，我们按阿司匹林使用情况进行分层并重复了后一项分析。结果：在我们的对照组中，当前使用NANSAIDs和阿司匹林的比例高于电子数据库研究中的比例，这与非处方NANSAIDs占所有NANSAIDs使用量的81%这一事实相符。教育程度、身体活动和BMI也与处方NANSAIDs的使用有关。当消除每个潜在的偏差来源时，NANSAIDs的OR值进一步偏离1.0（即朝着与MI的保护性关联方向）：“重复电子数据库”分析（OR 1.00，95%置信区间[CI]：0.78 - 1.28）；“错误分类评估分析”（OR 0.89，95%CI：0.70 - 1.14）；“混杂因素评估分析#1”（OR 0.85，95%CI：0.66 - 1.10）；“混杂因素评估分析#2”（OR
0.78，95%CI：0.60 - 1.01）；“交互作用评估分析”（OR 0.69，95%CI：0.51 - 0.95）。结论：电子数据库中的局限性可能是这些研究中未发现NANSAIDs与较低MI风险之间存在关联的原因。需要进一步的研究——最好是随机试验——来探讨使用NANSAIDs的风险效益比。
