Adverse pregnancy outcomes in women with systemic lupus erythematosus: improving and validating risk prediction

摘要：系统性红斑狼疮（SLE）主要影响育龄期女性， 引起人们对怀孕期间孕产妇和胎儿健康的关注。尽管医生们不再统一 不鼓励SLE女性生育，患者面临20%的不良妊娠结局的可能性 (APO)包括先兆子痫、胎儿和新生儿死亡、生长受限和早产，即使在 临床疾病静止期。因为没有既定的工具来预测个体患者的载脂蛋白O， SLE妊娠被密切监测，患者的情感成本和社会的经济负担。的 在妊娠早期识别APO高危患者的能力将显著提高我们的能力， 临床管理患者。此外，需要验证的风险分层模型来设计和执行 预防APO的试验在PROMISSE中，最大的多中心，多民族和多种族的妊娠研究 迄今为止，SLE患者中有几个危险因素被确定为APO的重要预测因子。虽然主要 先进的，这些结果既没有外部验证，也没有显示出推广到独立的研究 人口。此外，风险因素是使用标准统计模型确定的，这些模型没有充分考虑到 多个预测变量的复杂效应。在这个项目中，一个国际团队的系统性红斑狼疮，产科和 由PROMISSE研究人员领导的生物统计学研究人员将严格开发和外部验证APO 通过利用来自PROMISSE（N=447）的数据以及五个独立的狼疮队列的预测模型 来自不同国家的患者（总N = 979）。在目标1中，将应用强大的机器学习算法 PROMISSE数据，以创建一个准确的和临床上有用的模型来预测SLE患者的APO。到 在真实的世界中最大限度地利用该模型，只有常规和 在临床护理期间对SLE患者进行准确评估将被视为潜在的预测因子。在目标2中， APO模型将在来自欧洲的妊娠狼疮患者前瞻性队列中进行外部验证（法国： N=246;德国：N=180;挪威：N=349）和未纳入PROMISSE的美国地区（南卡罗来纳州： N=82;布朗克斯，纽约：N=122）。这些群体在种族、民族、社会经济和文化背景方面是异质的。 分层，和SLE疾病活动，允许彻底调查的普遍性和可移植性， APO模型应用于不同的狼疮患者人群。在每个队列中，详细的基线和纵向临床， 实验室和妊娠结果数据已使用类似于PROMISSE的程序获得。 总体目标是开发一个在线风险计算器，这将显著改善真实的世界临床 为未来的APO预防试验进行风险分层。影响：一个准确的，经过验证的， 而易于使用的APO预测模型对于妊娠狼疮患者的有效临床护理是必要的， 医疗资源的最佳分配，以及未来临床试验的设计和招募， 实验性干预以预防APO。

ABSTRACT: Systemic lupus erythematosus (SLE) predominantly affects women during reproductive years, raising concerns regarding maternal and fetal health during pregnancy. Although physicians no longer uniformly discourage women with SLE from childbearing, patients face 20% likelihood of adverse pregnancy outcomes (APO), including preeclampsia, fetal and neonatal death, growth restriction, and preterm delivery, even during clinical disease quiescence. Because there are no established instruments to predict APO in individual patients, SLE pregnancies are intensely monitored at an emotional cost to patients and financial burden to society. The ability to identify, early in pregnancy, patients at high risk of APO would significantly enhance our capacity to clinically manage patients. Furthermore, validated risk stratification models are needed to design and execute trials to prevent APOs. In PROMISSE, the largest multi-center, multi-ethnic and multi-racial study of pregnant SLE patients to date, several risk factors were identified as significant predictors of APO. Although a major advance, these results have neither been externally validated nor shown to generalize to independent study populations. Moreover, risk factors were identified using standard statistical models that did not fully account for complex effects of multiple predictor variables. In this project, an international team of SLE, obstetric and biostatistics researchers, led by PROMISSE investigators, will rigorously develop and externally validate an APO prediction model by leveraging data from PROMISSE (N=447), as well as five independent cohorts of lupus patients from different countries (total N = 979). In Aim 1, powerful machine learning algorithms will be applied to PROMISSE data to create an accurate and clinically useful model to predict APOs in SLE patients. To maximize utility of this model in the real world, only clinical and laboratory features that are routinely and accurately assessed on SLE patients during clinical care will be considered as potential predictors. In Aim 2, the APO model will be externally validated in prospective cohorts of pregnant lupus patients from Europe (France: N=246; Germany: N=180; Norway: N=349) and regions in the US, not included in PROMISSE (South Carolina: N=82; Bronx, NY: N=122). These cohorts are heterogeneous with respect to race, ethnicity, socioeconomic strata, and SLE disease activity, allowing for a thorough investigation into generalizability and transportability of the APO model to diverse lupus patient populations. In each cohort, detailed baseline and longitudinal clinical, laboratory and pregnancy outcome data have been obtained using procedures similar to those in PROMISSE. The overarching goal is development of an online risk calculator that will significantly improve real world clinical decision making and enable risk stratification for future APO prevention trials. Impact: An accurate, validated, and user-friendly prediction model for APO is necessary for effective clinical care of pregnant lupus patients, optimal allocation of healthcare resources, and the design of and recruitment to future clinical trials of experimental interventions to prevent APO.