In Reply We would like to thank Pennap and colleagues for their thoughtful letter. We completely agree that the use of diagnostic data are likely to greatly underestimate the prevalence of suicide attempts, and we should have stated that explicitly in our article¹ as a possible limitation. To emphasize this point, we cite a recent study of postpartum women that demonstrated that detection of suicidal behavior through natural language processing of clinical notes resulted in an 11-fold greater estimate of suicide attempts than in those based solely on diagnostic data.² However, as Pennap and colleagues note in their Letter, while this underestimate most assuredly affects an estimate of the risk difference, it may not affect the relative risk of suicide attempts. Therefore, we believe that it is safe to conclude that while the diagnostic data reported in our study¹ most likely underestimate the true prevalence of suicide attempts and the risk difference, it is probably a reasonable estimate of the relative risk of suicide attempts in the children of parents who use opioids on a regular basis. We also note that our sample was based on parents and children with private health insurance and the rate is likely to be lower than in a general emergency department sample, which includes patients with public health insurance and no health insurance. For example the background rate of acute liver injury based on International Classification of Diseases, Ninth Revision code 570 was 4 times higher in Medicaid claims data relative to claims based on people with private health insurance (ie, the same MarketScan data used in our article).³ Future work that studies suicidal behavior using medical records may benefit from augmenting estimates of the prevalence of suicidal behavior from diagnostic codes with the use of natural language processing to identify cases of suicidal behavior that are documented in the medical record but are not reflected in the diagnostic code.

在答复中，我们要感谢Pennap和同事们的深思熟虑。我们完全同意，使用诊断数据可能会大大低估自杀未遂的发生率，我们应该在第¹条中明确指出这是可能的限制。为了强调这一点，我们引用了一项关于产后妇女的近期研究，该研究表明，通过自然语言处理临床笔记对自杀行为的检测导致自杀企图的估计数比仅基于诊断数据的自杀企图高出11倍。^2个但是，正如Pennap及其同事在信中指出的那样，虽然这一低估确实可以确保风险差异的估计，但它可能不会影响自杀未遂的相对风险。因此，我们相信可以得出结论，尽管我们的研究报告了¹的诊断数据最有可能低估了自杀未遂的真实发生率和风险差异，这可能是对定期使用阿片类药物的父母子女的自杀未遂相对风险的合理估计。我们还注意到，我们的样本基于拥有私人健康保险的父母和孩子，而且这一比率很可能低于普通急诊部门的样本，后者包括拥有公共健康保险但没有健康保险的患者。例如，根据国际疾病分类，第九修订版570的急性肝损伤背景发生率，医疗补助理赔数据比基于私人健康保险的理赔人数高4倍（即，与本文中使用的MarketScan数据相同） 。³ 使用病历研究自杀行为的未来工作可能会受益于使用自然语言处理来识别病历中记录但未反映在诊断中的自杀行为病例，从而通过诊断代码增加对自杀行为发生率的估计代码。