Over the past 4 decades, trends in cardiovascular disease (CVD) have declined, with the most notable improvements observed for CVD-related mortality.¹ Nonetheless, these trends have not been consistent across all subgroups of the population. In young women, the incidence of acute myocardial infarction continues to rise largely because of the increased prevalence of comorbidities (eg, diabetes mellitus) and traditional risk factors for CVD (eg, hypertension and obesity). In addition, these women are more likely to report a history of hypertensive disorders of pregnancy than the general population of women.²

Studies have consistently shown an association between a history of pregnancy complications, including hypertensive disorders of pregnancy, gestational diabetes, stillbirth, preterm birth, placental abruption, and intrauterine growth restriction, and subsequent CVD in women.² The potential importance of these complications for the early identification of women at risk of future CVD has been acknowledged by professional societies, including the American Heart Association, the American College of Obstetricians and Gynecologists, and the Society of Obstetricians and Gynaecologists of Canada. A more recent presidential advisory from the American Heart Association/American College of Obstetricians and Gynecologists and the American College of Obstetricians and Gynecologists Task Force, and a perspective piece have highlighted the need for targeted screening and long-term follow-up in these women as part of routine practice across all levels of care (eg, general practitioners to specialists). This call for early risk stratification in women of reproductive age highlights the need for high-quality research to develop prediction models that incorporate relevant information in a relevant population.

Despite these recommendations, and the evidence that women with these complications are at increased risk, currently available tools to predict the risk of CVD lack the accuracy to identify high-risk women <40 years of age, because these tools were developed and validated in older populations (>50 years of age). Three studies to date have examined the incremental value of hypertensive disorders of pregnancy, preterm birth, small-for-gestational-age birth, and low birth weight above well-established risk factors for the prediction of CVD in women and found no added value.^3–5 These studies used population-based registries and cohorts from Norway, Sweden, and the United States to create sample populations of parous women ≥40 years of age with no prior history of CVD. The risk prediction models in these studies varied slightly, including risk factors such as age, systolic blood pressure, total and high-density lipoprotein cholesterol, smoking, use of antihypertensive medications, and family history of myocardial infarction. The proportion of women experiencing a CVD event across these studies ranged from 1% to 8%. Although these studies used population-based cohorts, the primacy of women beyond their reproductive years in the sample populations may in part explain the lack of positive findings. Guidance documents for the development of prognostic models emphasize the importance of choosing sample populations that reflect the target population intended for screening. Therefore, if the objective is early detection and management of risk factors for CVD soon after a pregnancy affected by a complication, women would ideally be screened in the postpartum period with a target population for development of prognostic models that includes women of reproductive age (ie, 15–45 years). Moreover, the sample population used to develop these models should consist of nulliparous and parous women to increase the generalizability of the model to the target population intended for screening. In addition, the prevalence of predictors in the sample population is an important determinant of the relative contribution and subsequent inclusion of predictors in prognostic models. There are notable differences in the prevalence of traditional risk factors for CVD in women above the age of 50 years relative to younger women. For example, hypertension is present in 29.4% of women aged 40 to 59 years versus only 5.6% in women under the age of 40 years. A similar trend occurs with other comorbidities, such as diabetes mellitus, obesity, and dyslipidemia, because their prevalence increases with advanced maternal age. Sampling older women may have contributed to the limited incremental value of hypertensive disorders of pregnancy or low birth weight seen in these previous studies, and led to the underestimation of the importance of pregnancy complications to the early detection of women at long-term risk for CVD. This variation in prevalence of predictors also affects the generalizability of prediction models such that models developed in older women may not be applicable to younger women and vice versa.

In contrast to causal models, the choice of variables for inclusion in prediction models should favor risk factors that are known to be associated with or highly predictive of the outcome. Because the etiology of several pregnancy complications, in particular, preeclampsia, intrauterine growth restriction, and stillbirth, is unknown, it may be difficult to disentangle the relative contribution of these individual factors to the development of CVD. Moreover, previous studies have hypothesized that these complications may not be causal of CVD but rather markers of an underlying etiology that itself is associated with the development of CVD. Many of these complications may also not be independent of each other and may act as intermediates on the causal pathway to CVD. However, if the goal is to stratify women at risk of CVD, then the objective should be to include risk factors that are highly predictive of CVD independent of their causal relationship with CVD. Therefore, if the estimates from previous studies are accurate with respect to the association of pregnancy- and female-specific risk factors (eg, preeclampsia, depression, infertility, rheumatoid arthritis) and CVD, then these factors have the potential to contribute to the early identification of women who would benefit from management and treatment of cardiovascular risk factors.

Researchers developing prognostic models also need to be mindful of the candidate predictors considered for inclusion in the models and the complexity of models. To facilitate the integration of prediction models in clinical practice, it is essential that researchers only consider predictors readily available to clinicians at the time of screening. Although researchers may be inclined to include well-established risk factors for CVD (eg, dyslipidemia, atrial fibrillation) when building prognostic models, the low prevalence, in general, and the high probability of missingness of these predictors in women of reproductive age hinder the utility of such a model and raise concerns regarding the representativeness of the sample of measured values. In addition, complex models may restrict the utility of prognostic models in clinical practice despite the potential gains in the predictive accuracy of these models. Therefore, careful consideration should be taken when developing prognostic models for implementation in clinical practice to improve the transportability to a wide range of clinical settings.

To realize the recommendations from professional societies for early screening and management of risk factors for CVD in women with a history of pregnancy complications, clinicians need a tool that is able to provide accurate risk stratification in this patient population. Although assessing the incremental value of pregnancy complications above traditional risk factors seems like the logical starting point to develop prognostic models in this patient population, the above-mentioned limitations mask the relative contribution of these predictors for risk prediction. This therefore highlights the need to develop a prognostic model in a sample of women of reproductive age to facilitate the early identification of women who would benefit from targeted treatment and continual follow-up.

Drs Smith and Platt and S.M. Grandi report a grant from the Canadian Institutes of Health Research as part of the submitted work.

The opinions expressed in this article are not necessarily those of the editors or of the American Heart Association.

https://www.ahajournals.org/journal/circ

在过去的40年中，心血管疾病（CVD）的趋势有所下降，其中与CVD相关的死亡率得到了最明显的改善。¹尽管如此，这些趋势在人口的所有亚组中并不一致。在年轻妇女中，急性心肌梗塞的发生率继续增加，主要是由于合并症（例如糖尿病）和传统的CVD危险因素（例如高血压和肥胖）的患病率增加。此外，这些妇女比一般妇女更容易报告妊娠高血压疾病史。^2个

研究一直显示妊娠并发症的历史与妊娠高血压病，妊娠糖尿病，死产，早产，早产，胎盘早剥和宫内生长受限以及随后的女性CVD之间存在关联。^2个这些并发症对于及早发现可能患CVD的女性的潜在重要性已经得到了专业协会的认可，这些协会包括美国心脏协会，美国妇产科学院以及加拿大妇产科医师学会。美国心脏协会/美国妇产科学院和美国妇产科学院工作队的最新总统咨询和观点文章强调了对这些妇女进行针对性筛查和长期随访的必要性，因为所有护理水平的日常实践的一部分（例如，从全科医生到专家）。

尽管提出了这些建议，并且有证据表明患有这些并发症的妇女的风险更高，但目前可用的预测CVD风险的工具仍不足以识别<40岁的高风险妇女，因为这些工具的开发和验证年龄较大。人群（> 50岁）。迄今为止，已有三项研究检查了妊娠高血压疾病，早产，小胎龄妊娠和低出生体重的婴儿在确定心血管疾病的危险因素上的增加价值，并没有增加价值。^3–5这些研究使用了来自挪威，瑞典和美国的基于人口的登记处和队列，以创建≥40岁且没有CVD既往史的产妇的抽样人群。这些研究中的风险预测模型略有不同，包括年龄，收缩压，总和高密度脂蛋白胆固醇，吸烟，使用降压药的使用以及心肌梗塞家族史等风险因素。在这些研究中，发生CVD事件的女性比例为1％至8％。尽管这些研究使用的是基于人群的队列研究，但在样本人群中，超过生育年龄的妇女居于首位，这可能在一定程度上解释了缺乏阳性结果的原因。开发预后模型的指导文件强调选择样本人群的重要性，这些人群应能反映拟筛查的目标人群。因此，如果目标是在受并发症影响的怀孕后尽快发现和控制CVD的危险因素，那么理想的情况是，在产后阶段对目标人群进行筛查，以开发出包括育龄妇女（即，即女性）在内的预后模型。 ，15-45岁）。此外，用于开发这些模型的样本人群应包括未产妇和同卵妇女，以增加该模型对拟筛查目标人群的通用性。此外，样本人群中预测因子的普遍性是预测因子相对贡献以及随后将其纳入预后模型的重要决定因素。与年轻女性相比，年龄在50岁以上的女性中，传统的CVD危险因素的患病率存在​​显着差异。例如，在40至59岁的女性中，高血压占29.4％，而在40岁以下的女性中，高血压仅占5.6％。其他合并症也有类似的趋势，例如糖尿病，肥胖和血脂异常，因为它们的患病率随着孕妇年龄的增长而增加。在这些先前的研究中发现，对老年妇女进行抽样可能是导致妊娠高血压疾病或低出生体重的有限增量价值的原因，并导致低估了妊娠并发症对及早发现具有长期CVD风险的女性的重要性。预测变量盛行率的这种变化也影响了预测模型的可推广性，因此在老年妇女中开发的模型可能不适用于年轻妇女，反之亦然。

与因果模型相反，选择要包含在预测模型中的变量应优先考虑已知与结果相关或对结果具有高度预测性的风险因素。由于尚不清楚几种妊娠并发症的病因，尤其是先兆子痫，子宫内生长受限和死产，因此可能难以弄清这些个体因素对CVD发生的相对贡献。此外，以前的研究假设这些并发症可能不是CVD的原因，而是与CVD的发展有关的潜在病因的标志。这些并发症中的许多也可能不是彼此独立的，并且可能充当CVD的因果路径上的中间体。但是，如果目标是对有CVD危险的女性进行分层，那么目标应该是包括可以高度预测CVD的危险因素，而与它们与CVD的因果关系无关。因此，如果先前研究的估计值在特定于妊娠和女性的危险因素（例如先兆子痫，抑郁，不育，类风湿性关节炎）和CVD的关联方面是准确的，则这些因素有可能促成早期确定将从心血管危险因素的管理和治疗中受益的妇女。

开发预后模型的研究人员还需要注意考虑纳入模型的候选预测因素以及模型的复杂性。为了促进将预测模型整合到临床实践中，至关重要的是，研究人员仅应考虑筛查时临床医生容易获得的预测因子。尽管研究人员在建立预后模型时可能倾向于将已确定的CVD危险因素（例如血脂异常，心房颤动）包括在内，但总体上这些患病率低，育龄妇女缺乏这些预测因素的可能性较高，这阻碍了心血管疾病的发展。这种模型的实用性并引起人们对测量值样本代表性的担忧。此外，复杂模型可能会限制预后模型在临床实践中的应用，尽管这些模型的预测准确性可能会有所提高。因此，在开发在临床实践中实施的预后模型时应谨慎考虑，以改善在各种临床环境中的可运输性。

为了实现专业协会的建议，对有妊娠并发症史的女性进行CVD的早期筛查和管理，临床医生需要一种能够在此患者人群中提供准确风险分层的工具。尽管评估高于传统危险因素的妊娠并发症的增值价值似乎是在该患者人群中开发预后模型的逻辑起点，但上述限制掩盖了这些预测因素对风险预测的相对贡献。因此，这突出表明有必要在育龄妇女的样本中建立一种预后模型，以促进及早发现将受益于靶向治疗和持续随访的妇女。

Smith和Platt博士以及SM Grandi博士报告了加拿大卫生研究院的赠款，作为已提交工作的一部分。

本文表达的观点不一定是编辑者或美国心脏协会的观点。

https://www.ahajournals.org/journal/circ