Prediction of pre‐eclampsia: review of reviews

Primary studies and systematic reviews provide estimates of varying accuracy for different factors in the prediction of pre-eclampsia. The aim of this study was to review published systematic reviews to collate evidence on the ability of available tests to predict pre-eclampsia, to identify high-value avenues for future research and to minimize future research waste in this field. MEDLINE, EMBASE and The Cochrane Library including DARE (Database of Abstracts of Reviews of Effects) databases, from database inception to March 2017, and bibliographies of relevant articles were searched, without language restrictions, for systematic reviews and meta-analyses on the prediction of pre-eclampsia. The quality of the included reviews was assessed using the AMSTAR tool and a modified version of the QUIPS tool. We evaluated the comprehensiveness of search, sample size, tests and outcomes evaluated, data synthesis methods, predictive ability estimates, risk of bias related to the population studied, measurement of predictors and outcomes, study attrition and adjustment for confounding. From 2444 citations identified, 126 reviews were included, reporting on over 90 predictors and 52 prediction models for pre-eclampsia. Around a third (n = 37 (29.4%)) of all reviews investigated solely biochemical markers for predicting pre-eclampsia, 31 (24.6%) investigated genetic associations with pre-eclampsia, 46 (36.5%) reported on clinical characteristics, four (3.2%) evaluated only ultrasound markers and six (4.8%) studied a combination of tests; two (1.6%) additional reviews evaluated primary studies investigating any screening test for pre-eclampsia. Reviews included between two and 265 primary studies, including up to 25 356 688 women in the largest review. Only approximately half (n = 67 (53.2%)) of the reviews assessed the quality of the included studies. There was a high risk of bias in many of the included reviews, particularly in relation to population representativeness and study attrition. Over 80% (n = 106 (84.1%)) summarized the findings using meta-analysis. Thirty-two (25.4%) studies lacked a formal statement on funding. The predictors with the best test performance were body mass index (BMI) > 35 kg/m2, with a specificity of 92% (95% CI, 89–95%) and a sensitivity of 21% (95% CI, 12–31%); BMI > 25 kg/m2, with a specificity of 73% (95% CI, 64–83%) and a sensitivity of 47% (95% CI, 33–61%); first-trimester uterine artery pulsatility index or resistance index > 90th centile (specificity 93% (95% CI, 90–96%) and sensitivity 26% (95% CI, 23–31%)); placental growth factor (specificity 89% (95% CI, 89–89%) and sensitivity 65% (95% CI, 63–67%)); and placental protein 13 (specificity 88% (95% CI, 87–89%) and sensitivity 37% (95% CI, 33–41%)). No single marker had a test performance suitable for routine clinical use. Models combining markers showed promise, but none had undergone external validation. This review of reviews calls into question the need for further aggregate meta-analysis in this area given the large number of published reviews subject to the common limitations of primary predictive studies. Prospective, well-designed studies of predictive markers, preferably randomized intervention studies, and combined through individual-patient data meta-analysis are needed to develop and validate new prediction models to facilitate the prediction of pre-eclampsia and minimize further research waste in this field. Copyright © 2018 ISUOG. Published by John Wiley & Sons Ltd. Predicción de la preeclampsia: revisión de revisiones Los estudios primarios y las revisiones sistemáticas proporcionan estimaciones de precisión variable para diferentes factores en la predicción de la preeclampsia. El objetivo de este estudio fue revisar las revisiones sistemáticas publicadas para recopilar evidencia sobre la capacidad de las pruebas disponibles para predecir la preeclampsia, identificar avenidas de investigación futura valiosas y minimizar el desperdicio futuro de investigación en este campo. Se realizaron búsquedas de artículos relevantes en bibliografías sobre el tema y en las bases de datos MEDLINE, EMBASE y The Cochrane Library, incluida DARE (Database of Abstracts of Reviews of Effects), desde el inicio de cada base de datos hasta marzo de 2017, sin restricciones de idioma, para obtener revisiones sistemáticas y metaanálisis sobre la predicción de la preeclampsia. La calidad de las revisiones incluidas se evaluó utilizando la herramienta AMSTAR y una versión modificada de la herramienta QUIPS. Se evaluó la amplitud de la búsqueda, el tamaño de la muestra, las pruebas y los resultados evaluados, los métodos de síntesis de datos, las estimaciones de la capacidad de predicción, el riesgo de sesgo relacionado con la población estudiada, la medición de los predictores y los resultados, la deserción del estudio y el ajuste por confusión. De las 2444 citas identificadas, se incluyeron 126 revisiones, que informaron sobre más de 90 predictores y 52 modelos de predicción para la preeclampsia. Alrededor de un tercio (n=37 (29,4%)) de todas las revisiones investigaron únicamente marcadores bioquímicos para predecir la preeclampsia, 31 (24,6%) investigaron asociaciones genéticas con la preeclampsia, 46 (36,5%) informaron sobre las características clínicas, cuatro (3,2%) evaluaron sólo marcadores ecográficos y seis (4,8%) estudiaron una combinación de pruebas; dos (1,6%) revisiones adicionales evaluaron los estudios primarios que investigaron cualquier prueba de diagnóstico de la preeclampsia. Las revisiones incluyeron entre dos y 265 estudios primarios, que incluyeron hasta 25 356 688 mujeres en la revisión más grande. Sólo aproximadamente la mitad (n=67 (53,2%)) de las revisiones evaluaron la calidad de los estudios incluidos. En muchas de las revisiones incluidas hubo un alto riesgo de sesgo, particularmente en relación con la representatividad de la población y la deserción de los estudios. Más del 80% (n=106 (84,1%)) resumió los hallazgos utilizando el metaanálisis. Treinta y dos (25,4%) estudios carecían de una declaración formal sobre la financiación. Los predictores con el mejor rendimiento de la prueba fueron el índice de masa corporal (IMC) >35 kg.m–2, con una especificidad del 92% (IC 95%, 89–95%) y una sensibilidad del 21% (IC 95%, 12–31%); IMC >25 kg.m–2, con una especificidad del 73% (IC 95%: 64–83%) y una sensibilidad del 47% (IC 95%: 33–61%); índice de pulsatilidad de la arteria uterina en el primer trimestre o índice de resistencia >90° percentil (especificidad del 93% (IC 95%: 90–96%) y sensibilidad del 26% (IC 95%: 23–31%)); factor de crecimiento placentario (especificidad 89% (IC 95%, 89–89%) y sensibilidad 65% (IC 95%, 63–67%)); y proteína placentaria 13 (especificidad 88% (IC 95%, 87–89%) y sensibilidad 37% (IC 95%, 33–41%)). Ningún marcador por sí solo tuvo un rendimiento de la prueba adecuado para el uso clínico rutinario. Los modelos que combinan marcadores son prometedores, pero ninguno fue sometido a una validación externa. Esta revisión de revisiones ha puesto en duda la necesidad de un metaanálisis agregado adicional en esta área, dado el gran número de revisiones publicadas sujetas a las limitaciones comunes de los estudios predictivos primarios. Se necesitan estudios prospectivos bien diseñados de marcadores predictivos, preferiblemente en estudios de intervención aleatorios, y combinados mediante el metaanálisis de datos de pacientes individuales, para desarrollar y validar nuevos modelos predictivos que faciliten la predicción de la preeclampsia y minimicen el desperdicio de investigación adicional en este campo. 先兆子痫的预测:评论回顾 初步研究和系统评论根据预测先兆子痫的不同因素提供各种准确的估计值。本研究的目的是回顾已发表的系统评论，核对现有检测能力的证据以便预测先兆子痫，确定未来研究的高价值途径，并尽量减少该领域未来研究的浪费。 检索联机医学文献分析和检索系统(MEDLINE)、荷兰医学文摘数据库(EMBASE)和Cochrane图书馆包括DARE(效果评论摘要数据库)，从建立数据库到2017年3月，并检索了相关文章的参考文献，不受语言限制，对先兆子痫的预测进行系统回顾和meta分析。使用多系统评价膀胱问卷(AMSTAR)工具和QUIPS处理系统工具的修改版本评估纳入评论的质量。我们评估了搜索的全面性、样本大小、测试和评价结果、数据综合方法、预测能力估计、与研究人群相关的偏差风险、预测因素和结果的测量值、研究淘汰率和混杂设计的调整。 共确定2444篇引文，其中包括126篇评论，报道了90多个预测因素和52种先兆子痫的预测模型。约三分之一(n=37(29.4%)的评论只研究了预测先兆子痫的生化标记，31篇(24.6%)研究了与先兆子痫的遗传关联，46篇(36.5%)报告了临床特征，4篇(3.2%)仅评价了超声标志物，6篇(4.8%)研究了联合检测；另外两篇(1.6%)评论评估了对先兆子痫筛查试验的初步研究。评论包括两项和265项初步研究，其中最大型评论纳入了多达25,356,688名女性。只有大约一半(n=67(53.2%))的评论评估了纳入研究的质量。许多纳入的评论有很高的偏差风险，特别是在人群代表性和研究淘汰率方面。80%(n=106(84.1%)以上的评论采用Meta分析方法总结研究结果。32项研究(25.4%)缺乏关于资金的正式说明。最佳检测指标为体重指数(MBI)>35 kg.m-2，特异性92%(95% CI，89-95%)，敏感性21%(95% CI，12-31%)；BMI>25 kg.m-2，特异性73%(95% CI，64-83%)，敏感性47%(95% CI，33-61%)；妊娠早期子宫动脉搏动指数或阻力指数>90%(特异性93%(95% CI，90%-96%)和敏感性26%(95% CI, 23%–31%))；胎盘生长因子(特异性89%(95% CI，89%-89%)，敏感性65%(95% CI，63%-67%)；胎盘蛋白13(特异度88%(95% CI，87%-89%)，敏感性37%(95% CI, 33%–41%))。 没有一种单一标记物具有适合常规临床使用的测试性能。模型结合标记物显示有希望，但没有一种模型经过外部验证。 鉴于大量已发表的评论受制于初级预测研究的共同局限性，本评论回顾引发了对这一领域进行进一步汇总meta分析的必要性的质疑。为了开发和验证新的预测模型，促进对先兆子痫的预测，并最大限度地减少该领域的进一步研究浪费，需要对预测标志物进行前瞻性、精心设计的研究，最好是随机干预研究，并与个体患者数据的meta分析相结合。 Pre-eclampsia remains a major contributor to maternal and perinatal mortality and morbidity1, 2. Early treatment with aspirin reduces the risk of pre-eclampsia3, 4, so accurate screening tests for pre-eclampsia are a clinical priority. At present, clinical assessment of the risk of pre-eclampsia is based mainly on maternal history5, which has limited predictive ability6-8 and is not applicable to nulliparous women. Numerous primary studies have evaluated the predictive ability of various tests including clinical characteristics, biomarkers and ultrasound markers, individually or in combination, for predicting early-, late- and any-onset pre-eclampsia. Systematic reviews collate evidence and aim to provide meaningful summary estimates of the predictive ability of tests through meta-analysis. Despite the number of published studies on predictive factors and screening tests for pre-eclampsia, no consensus on the optimal strategy has been reached; neither clinicians nor national or international guidelines have implemented screening tests in routine clinical practice. This could be because no tests with adequate performance have been identified, but it could also be attributed to the variable quality of the reviews. Very few have validated existing prediction models9 or reported on test performance using various combinations of predictors for different thresholds and outcomes. There is a need to map and appraise critically the available evidence in this field in order to minimize research waste and prioritize robust investigation of high-yield predictive factors and models. The aim of this study was to collate systematically and evaluate critically the published systematic reviews on risk factors identified as predictors of pre-eclampsia and the reported ability of individual tests to predict pre-eclampsia. This review of reviews was based on a prospective protocol according to current recommendations10-12, and was reported using the PRISMA guidelines13. The study was registered in the PROSPERO database (CRD42015020386, http://www.crd.york.ac.uk/PROSPERO). MEDLINE, EMBASE and the Cochrane Library, including The Cochrane Database of Systematic Reviews (CDSR), Database of Abstracts of Reviews of Effects (DARE), The Cochrane Central Register of Controlled Trials (CENTRAL), Health Technology Assessment (HTA) Database and the NHS Economic Evaluation Database (NHS-EED) were searched from inception to March 2017. Combinations of the relevant medical subject heading (MeSH) terms, keywords and word variants for ‘pre-eclampsia’, ‘gestational hypertension’, ‘pregnancy-induced hypertension’ and ‘review’ were used (Appendix S1). No language restrictions were applied. Reference lists of relevant articles and reviews were searched manually to identify additional papers. Two reviewers (R.T., A.K.) reviewed independently all abstracts. Any discrepancies regarding the potential relevance of the papers were resolved by consensus. Full-text copies of reviews that met the inclusion criteria were obtained. Reviews that assessed clinical characteristics or biochemical, genetic or ultrasound-based variables as predictors or predictive tests for pre-eclampsia in the first, second or third trimester were included. Case reports, case series, individual observational or randomized studies, narrative reviews, rapid reviews, editorials and poster abstracts were excluded. Two reviewers (R.T., A.K.) extracted relevant data independently. Data were obtained on year of publication, number of databases searched, number of studies included, number of pregnancies/women included, screening tests evaluated and the performance of the tests or degree of association reported with the evaluated predictors. The authors' definitions of pre-eclampsia and hypertensive disorders were accepted, and, when reported, data were collected discriminating between early-onset pre-eclampsia (requiring delivery before 34 weeks' gestation), late-onset pre-eclampsia (delivery after 34 weeks) or delivery with pre-eclampsia at any time. Clinical characteristics included signs, symptoms, medical and obstetric histories and exposure to environmental factors, identified through maternal history or physical examination by the booking clinician at the first antenatal visit. Biochemical tests included any measurement of the concentration of a biochemical marker in biological fluids (e.g. serum and urine). Ultrasound factors included any characteristic identified on ultrasound examination at any gestational age. A predictor was defined as a clinical characteristic or biochemical, genetic or ultrasound marker with the potential to predict the outcome of interest (pre-eclampsia). A predictive model was defined as a combination of predictors obtained through logistic regression analysis to discriminate between populations. A review was defined as systematic if it included an explicit method for searching the literature, searched two or more data sources and provided well-defined inclusion and exclusion criteria for the studies. The rigor of the systematic reviews and risk of bias in the review findings were assessed independently by two reviewers (R.T., Y.P.) using the AMSTAR (A MeaSurement Tool to Assess systematic Reviews) tool14, 15 and a modified version of the QUIPS (QUality In Prognosis Studies) tool16, respectively (Table S1). In the AMSTAR assessment, we considered whether the review included the following: ‘a-priori’ design, a comprehensive literature search, the publication status (e.g. gray literature) used as an inclusion criterion, duplicate study selection and data extraction, list of the included and excluded studies, characteristics of the included studies, assessment and documentation of the quality of the included studies, appropriate use of the scientific quality of the studies in formulating conclusions, use of appropriate methods for combining the findings of studies, assessment of the likelihood of publication bias and reporting of any conflict of interest. Risk of bias reported in the included reviews was assessed according to the QUIPS domains that relate to the key methodological concerns of prognostic research. Whether the reviewers had assessed the representativeness of the patient sample, the impact of study attrition, prognostic factor and outcome measurements, important confounders and the quality of the statistical analysis in the primary studies were considered. When this information was reported, we considered whether the authors had assessed the degree of associated risk of bias. For the studies on genetic factors, the Venice criteria17 were used to assess the epidemiological credibility of the association based on the amount of evidence, replication and protection from bias in each study. Of the 2444 citations identified, 126 systematic reviews were included in this review18-143. Figure 1 provides details of the review identification and selection process. A list of the excluded studies is provided in Table S2. Figure 2a summarizes the findings of the quality assessment of the included reviews using the AMSTAR tool. Fewer than a quarter (n = 24 (19.0%)) of the included reviews followed a prospectively specified protocol. Most (n = 120 (95.2%)) of the reviews performed a comprehensive literature search, with the majority exploring more than two databases. The majority of reviews undertook duplicate study selection (n = 111 (88.1%)), provided the characteristics of the included studies (n = 109 (86.5%)) and assessed the likelihood of publication bias (n = 80 (63.5%)). However, fewer than a quarter provided a list of the included and excluded studies (n = 28 (22.2%)). Just over half of the reviews assessed the quality of the included studies (n = 67 (53.2%)), and only about a third took into account the quality of the studies in formulating their conclusions (n = 38 (30.2%)). About half (n = 71 (56.3%)) of the reviews performed their literature search without language restriction. The most commonly used tools for quality assessment were QUADAS (n = 17 (13.5%)) and the Newcastle–Ottawa scale (n = 31 (24.6%)), although neither is designed for predictive research. None of the reviews published since 2013 used the QUIPS tool, which was described in 2013 and is designed for quality assessment of studies on predictive factors16. Although only about half of the reviews assessed the quality of the included studies, many of the primary studies were potentially methodologically biased. They were often retrospective or case–control in design, and therefore, subject to bias. Examples include significant heterogeneity between primary studies, failure to blind those managing the pregnancy or the outcome assessors, nested case–control studies including only a subset of pre-eclampsia cases of the original cohort and failure to apply the screening test to all eligible participants in cohort studies. Furthermore, the included primary studies had numerous limitations including poor reporting of summary statistics, varying cut-offs of continuous variables, and variation in outcomes assessed and the adjustment factors used to calculate test performance144. Figure 2b shows the findings of the assessment of risk of bias of included studies using the modified QUIPS tool. Only one study reported on all domains. Of the included reviews, 80 (63.5%) reported on participants and representativeness of the population, with 55 (68.8%) of these reporting a high or moderate risk of bias in this area in the primary studies. Study attrition was considered in 31 (24.6%) reviews, with 20 (64.5%) of these reporting a high or moderate risk of bias. Measurement of predictors was evaluated in 101 (80.2%) reviews, of which 63 (62.4%) described a high or moderate risk of bias. Measurement of the outcome was well reported (n = 109 (86.5%) reviews), but 67/109 (61.5%) reviews found a high or moderate risk of bias, most commonly related to heterogeneity or lack of clarity in the definition of the outcomes in the primary studies. Confounding was considered in 84 (66.7%) reviews, of which 59 (70.2%) reported a high or moderate risk of bias relating to insufficient or inappropriate adjustment for important covariables. The included reviews reported on between two and 265 primary studies, with the majority including 10–50 primary studies and up to 25 356 688 pregnancies in the largest review133 (Figure 3). More than 90 predictors were evaluated in the included reviews (Table S3). The majority of reviews (n = 68 (54.0%)) investigated solely biochemical or genetic tests for predicting pre-eclampsia, while 46 (36.5%) investigated solely clinical characteristics. Ultrasound markers alone were reported in only four (3.2%) reviews and a combination of tests in six (4.8%) reviews (Figure 3). We also identified two (1.6%) broad systematic reviews of primary studies investigating all screening tests for pre-eclampsia, published in 200461 and 2008107. The most commonly reported clinical characteristics included body mass index (BMI) or weight (n = 9 reviews)58, 69, 90, 107, 113, 117, 124, 133, 139, maternal age (n = 2)69, 133, parity (n = 2)69, 133, blood pressure (n = 5)47, 57, 61, 69, 90 and oocyte donation (n = 3)42, 86, 106, and six reviews reported on multiple clinical characteristics23, 61, 69, 89, 107, 133. Of the biochemical markers, the following were most commonly studied: pregnancy-associated plasma protein-A (PAPP-A) (n = 7)24, 43, 61, 89, 93, 109, 110, placental growth factor (PlGF) (n = 7)21, 23, 24, 43, 88, 89, 93 and β-human chorionic gonadotropin (β-hCG) (n = 7)43, 61, 89, 93, 104, 107, 110. Over 30 additional biomarkers were reviewed. Ultrasound tests included uterine artery (UtA) Doppler (n = 8)47, 59, 61, 75, 90, 93, 107, 137 and placental vascularization indices (n = 1)71. Over 80% (n = 106 (84.1%)) summarized the findings using meta-analysis. Only two reviews90, 125 summarized the findings using an individual-participant data meta-analysis (IPD-MA). The details of the included reviews and their key findings are shown in Table S4, according to the type of predictor assessed. The majority (n = 67 (53.2%)) of the included reviews reported odds ratio (OR) as a single measure of predictor association with pre-eclampsia, rather than directly reporting predictive ability of the investigated predictors (Table S4). Only 31 (24.6%) studies reported measures of predictive ability, with 19 reporting sensitivity and specificity, six reporting area under the receiver–operating characteristics curve (AUC) and six reporting likelihood ratios. Twenty-one (16.7%) studies declared that no funding had been received, while 32 (25.4%) lacked a formal statement regarding funding. Of the remaining 73 studies, 14 (19.2%) declared multiple funding sources. The majority of these studies (51/73 (69.9%)) that declared their funding source had been sponsored by national or regional governmental bodies (e.g. National Institute for Health Research (NIHR), National Institutes of Health, Canadian Institutes of Health Research, Health Technology Assessment (HTA), National Health and Medical Research Council (NHMRC)). Nearly one quarter (21.9%) of the studies were funded through academic institutions, 19.2% by charitable bodies, 4.1% received funding from industry and 9.5% were funded by international bodies, chiefly by the World Health Organization (WHO). There was substantial variation in outcome reporting, including failure to report gestational age at delivery and severity of pre-eclampsia. Despite the fact that there has been a transition from a severity-based to a temporal classification of pre-eclampsia145, only 12 reviews reported early-onset pre-eclampsia, probably because the outcome was reported infrequently in primary studies. Some studies combined pre-eclampsia with hypertensive disorders, which limited the comparisons that could be made between studies. Considerable heterogeneity was noted in many of the included reviews, which precluded meta-analysis in 19 (15.1%) reviews. The included reviews reported on over 90 potential predictors for pre-eclampsia. For each predictor, we applied the Grades of Recommendation, Assessment, Development and Evaluation (GRADE) approach to prognostic studies146 to assess the quality of the evidence supporting the associations found (Table S5). The most robustly associated clinical, ultrasound and biochemical predictors for pre-eclampsia included maternal BMI or weight, blood pressure, UtA Doppler findings, PlGF, soluble fms-like tyrosine kinase-1 (sFlt-1) and alpha fetoprotein (AFP) (Table 1). Maternal BMI was analyzed as a continuous, binary or categorical variable, and was considered consistently to be a predictor of pre-eclampsia, with a number of studies demonstrating a biological gradient in which increasing BMI increased the risk of pre-eclampsia58, 107, 113, 117, 133, 139. Increased maternal blood pressure, evaluated alone57, 61 or in combination with other predictors47, 90 in the first or second trimester, was also associated consistently with an increased risk of pre-eclampsia, but the measurement of blood pressure varied between studies57. In 2008, Cnossen et al.57 compared the predictive ability of systolic and diastolic blood pressure and mean arterial pressure (MAP) measured at booking and found that MAP had a greater AUC (0.76 (95% CI, 0.70–0.82)) than did either diastolic or systolic blood pressure for all pre-eclampsia. Other evaluated clinical characteristics that demonstrated a consistent low or moderate association included donor oocyte use in assisted reproduction, sleep-disordered breathing, polycystic ovary syndrome, periodontal disease, air pollution, use of assisted reproductive technology and maternal infections (Table 1). First-trimester UtA Doppler appears to have high specificity (92.1% (95% CI, 88.6–94.6%)) but low sensitivity (47.8% (95% CI, 39.0–56.8%)) in predicting early-onset pre-eclampsia137. The sensitivity of UtA Doppler was even lower in predicting any pre-eclampsia, being 26.4% (95% CI, 22.5–30.8%)137. One review evaluated placental vascularization indices measured on three-dimensional ultrasound, and found that when measured in the first trimester they were predictive of later pre-eclampsia, with the most sensitive measure being the vascularization flow index (VFI)71. The authors reported an AUC for the prediction of early-onset pre-eclampsia by VFI of 0.89 (95% CI, 0.78–1.00) and an AUC for the prediction of any pre-eclampsia of 0.77 (95% CI, 0.69–0.84)71. The biochemical screening markers were grouped according to their mechanism of action (Table S4). Of the markers associated with angiogenesis, both PlGF and sFlt-1 were associated consistently with the risk of pre-eclampsia, with an OR of 9.0 (95% CI, 5.6–14.5) for PlGF tested before 30 weeks' gestation in one large study88. Another review reported no significant association between first-trimester PlGF and all pre-eclampsia (OR, 1.94 (95% CI, 0.81–4.67)), but there was an association between first-trimester PlGF and early-onset pre-eclampsia (OR, 3.41 (95% CI, 1.61–7.24))89. For sFlt-1, ORs from 1.3 (95% CI, 1.02–1.65) to 6.6 (95% CI, 3.1–13.7) were reported, the association being stronger when tested later in pregnancy88, 89. For a 5% false-positive rate, PlGF and sFlt-1 had sensitivities of 32% and 26%, respectively88. Soluble endoglin (sEng) and vascular endothelial growth factor were not found as consistently to be associated with pre-eclampsia88, 89, although one study reported that sEng had a sensitivity of 18% to detect pre-eclampsia at a 5% false-positive rate88. Of the markers tested routinely during aneuploidy screening in the first trimester, AFP had the highest specificity, at 96% (95% CI, 94–98%), but with a specificity of only 9% (95% CI, 5–16%)107. A wide number of gene mutations were considered to be associated with the development of pre-eclampsia, but no single polymorphism was identified as having a clinically useful predictive performance (Table S4). The most frequently investigated genes were methylenetetrahydrofolate reductase (MTHFR) (n = 7)25, 26, 38, 48, 91, 102, 140 and endothelial nitric oxide synthase (eNOS) (n = 7)34, 36, 48, 53, 65, 118, 129, and a number of genes relating to elements of the renin-angiotensin-aldosterone system (RAAS) were investigated. The credibility of the association between the MTHFR C677T mutation and pre-eclampsia was generally weak and the association was not large. The credibility of association with mutations of the eNOS gene was moderate, but again this was not a large effect. These patterns support an association between endothelial and RAAS function and pre-eclampsia, but are not at present useful for prediction of the disease. No screening marker, whether a clinical characteristic or ultrasound, genetic or biochemical marker, had both sensitivity and specificity > 90%. Six reviews opted for an approach using combinations of predictive markers (Table S4)47, 75, 78, 83, 90, 93 and reported results for 52 individually described models, while one study reported on an additional 70 models in groups labeled as ‘simple’ or ‘specialized’ based on the inclusion of ultrasound and biochemical tests78. Of these studies, only one reported calibration statistics for the described model90 and one found that, of the 14 primary model development studies assessed, only six reported model calibration78. The remaining prediction-modeling papers did not describe calibration of the models presented or assess calibration statistics in the primary studies reviewed. The detection rates (DR) of single markers (A-disintegrin and metalloprotease-12, β-hCG, inhibin A, activin A, placental protein 13 (PP-13), PlGF and PAPP-A) for early-onset pre-eclampsia ranged from 22% to 83% at a fixed false-positive rate of 10%93. These figures improved to between 38% and 100% when a combination of more than two markers was used93. The best results (DR, 100% (95% CI, 69–100%)) were achieved with a combination of three biochemical markers (inhibin A, PlGF, PAPP-A), UtA Doppler and maternal characteristics93. For early-onset pre-eclampsia, a model including only BMI was significantly improved by the addition of mean UtA resistance index and bilateral notching, with the AUC increasing from 0.66 to 0.92 (P < 0.001). The addition of mean UtA pulsatility index (UtA-PI) and bilateral notching improved the AUC from 0.62 to 0.95 (P < 0.001)90. The sensitivity of UtA-PI and MAP was 83% for early-onset pre-eclampsia47 but only 58.5% for late-onset pre-eclampsia. The improved performance of models including Doppler indices or biomarkers is consistent with the finding of one study that adding ultrasound markers or biomarkers to models based on maternal characteristics alone led to a median gain of 18% in sensitivity78. Our review identified 126 systematic reviews on over 90 potential predictors for pre-eclampsia, although only about a quarter of them reported directly predictive ability. No test was found to have both sensitivity and specificity above 90%. A high sensitivity and specificity are necessary to make screening more cost-effective than a ‘treat-all’ policy in clinical practice107. BMI > 34 kg/m2, AFP and bilateral UtA Doppler notching were reported to have specificity > 90% but low sensitivities, rendering them unsuitable for categorizing safely women as ‘low risk’107. Individual predictors most strongly correlated with pre-eclampsia were UtA Doppler indices and angiogenic biomarkers59, 90, 93. Prediction models combining maternal characteristics (particularly blood pressure) with UtA Doppler and biomarkers were able to achieve sensitivity and specificity > 80%47, 75, 90. Our search identified one prior systematic review of reviews on this topic147 and two broad systematic reviews of primary studies for predicting pre-eclampsia, one from the HTA in 2008107 and the other by WHO in 200461. These studies also identified BMI, UtA Doppler and AFP as high-performing variables but were limited by heterogeneity and inconsistent reporting in included primary studies107. A subsequently published ‘umbrella’ review of risk factors for pre-eclampsia, which did not examine UtA Doppler, also identified a number of maternal characteristics as important risk factors, including obesity, primiparity and smoking status, and additionally noted a strong association between assisted reproduction and pre-eclampsia that should be considered in the development of new prediction tools148. Several of these studies reported evidence that infrequently studied predictors, including kallikreinuria and fibronectin, might offer high sensitivity in the prediction of pre-eclampsia and require further research. No new reviews including these predictors were identified in our search nearly 10 years later, although new variables, including cell-free fetal DNA, can be added to the selection of variables that require further investigation. Previous reviews have also highlighted the need for the development of multivariable models. In this review, we identified over 50 models that have been reported in the last decade, but we also found none that had undergone external validation and could be recommended for routine practice. The strengths of this review include the thorough search strategy and critically evaluative approach. The analysis collates a wide variety of reviews, representing the state of research in this field. The findings of this review are limited by the quality of the included studies, compromised by limitations carried over from the primary studies and then the later conduct of the review analysis, especially when investigators did not address risks of bias specific to prediction research. Maternal characteristics at booking are currently recommended for use for pre-eclampsia screening by most guidelines5, 149, 150. An important characteristic, owing to its increasing prevalence, is maternal obesity151, 152. This review confirmed a plausible biological gradient in which maternal obesity is associated with pre-eclampsia, and found that the inclusion of BMI improved the performance of several models90, 93. It is probable that any clinically useful model would be improved by inclusion of a measurement of maternal obesity. In seeking to improve screening by maternal characteristics, many biomarkers were investigated. The angiogenic markers are most promising, particularly PlGF and sFlt-121, 24, 43, 88, 89, 93. Of the placental proteins, PP-13 and PAPP-A were associated most consistently with pre-eclampsia24, 43, 89, 109, 110, 126. Large prospective studies using biomarkers are expensive, and most existing data are on markers obtained routinely during fetal anomaly screening. There is evidence from smaller studies that supports the use of markers such as fibronectin97, 107, cell-free fetal DNA63, 105, 107 and urinary kallikrein61, 107, but this requires further investigation. This review further confirms the screening performance of UtA Doppler in the first and second trimesters. Using a model combining systolic blood pressure, UtA-PI and bilateral notching with BMI can achieve an AUC of 0.85 (95% CI, 0.67–1.00)90, but this model was still undergoing external validation at the time of writing, in the SPREE study comparing the National Institute for Health and Care Excellence and Fetal Medicine Foundation screening models153. While in previous years the search had been for a single marker to predict pre-eclampsia, recognition of the heterogeneity of the disease phenotype and complexity of prediction has led to a consensus that the best approach to pre-eclampsia screening is likely to be calculating individualized risk based on a combination of markers6. In this review, we have identified key predictors that could be used in developing such a prediction model and propose a solution to address the problems of inconsistent reporting and heterogeneity that have consistently affected the ability of previous reviews to make recommendations on screening61, 107, 147. As information on multiple predictors will be required, model development will optimally utilize individual level data, which can facilitate analysis to identify the predictors that explain most of the variance of the full model. The aim of this approach, already established in cardiovascular prediction modeling154, is to develop a model well balanced between optimal performance and parsimony of included predictors leading to the greatest ease of use in clinical practice. Using IPD-MA for model development could additionally address poor reporting and heterogeneity in primary studies. While resource intensive and still subject to publication bias, IPD-MA is becoming the gold standard for predictive meta-analysis155. The advantages of IPD-MA over conventional meta-analysis include use of all available data, flexibility to combine data uniformly and the use of original data allowing analysis of continuous variables and comparison between datasets156. Moreover, it permits comparison of multivariable prediction strategies and the possibility of time-to-event analysis, which is particularly relevant to pregnancies complicated by pre-eclampsia, in which gestational age is inextricably linked to maternal and fetal outcomes157. Research priorities should include prospectively registered predictive studies of promising markers, with results for each marker alone and in combination with other tests, and clear reporting of methods and timing of variable and outcome measurements. A particular focus should be high-performance tests in the first trimester, when the benefits of intervention are greatest. IPD-MA combining the most promising predictors could then be used to develop prediction models for external validation before introduction into clinical practice. Predictive variables by themselves do not improve outcome, it is the subsequent preventive interventions that do. Since it is not self-evident that a treatment has a consistent effect in women with different profiles, predictive markers should be evaluated in studies that evaluate the impact of predictive strategies158. The ideal predictor predicts not only pre-eclampsia, but also treatment modification, i.e. whether a treatment improves the outcome in a particular category of patient. In order to conduct effective primary studies and analyses, consensus on outcomes is needed. Identification of a core outcome set for pre-eclampsia studies is a key priority159. Such an approach would enable us to move beyond repeating small, low-quality prognostic-factor studies to investigating the clinical impact of the use of prediction models in clinical practice. B.W.M. is supported by a NHMRC Practitioner Fellowship (GNT1082548). B.W.M. reports consultancy for ObsEva, Merck and Guerbet. S.T. is the chief investigator of the NIHR-funded IPPIC IPD-MA to predict pre-eclampsia. J. Allotey1,2,3; K. Snell4; C. Chan2; R. Hopper2; J. Dodds1,2,3; E. Rogozinska1,2,3; K. Khan1,2,3; L. Poston5; L. Kenny6; J. Myers7; B. Thilaganathan8; L. Chappell9; B. W. Mol10; P. Von Dadelszen9; A. Ahmed11; M. Green12; A. Khalil8; K. Moons13; R. D. Riley4; S. Thangaratinam1,2,3 1Women's Health Research Unit, Barts and the London School of Medicine and Dentistry, Queen Mary University of London, London, UK. 2Pragmatic Clinical Trials Unit, Barts and the London School of Medicine and Dentistry, Queen Mary University of London, London, UK. 3Multidisciplinary Evidence Synthesis Hub (MESH), Queen Mary University of London, London, UK. 4Research Institute for Primary Care and Health Sciences, Keele University, Keele, UK. 5Division of Women's Health, Women's Health Academic Centre, King's College London, London, UK. 6Irish Centre for Fetal and Neonatal Translational Research (INFANT), University College Cork, Ireland. 7Maternal & Fetal Heath Research Centre, Manchester Academic Health Science Centre, University of Manchester, Central Manchester NHS Trust, Manchester, UK. 8Fetal Medicine Unit, St George's Hospital, St George's University of London, London, UK. 9Department of Women's and Children's Health, King's College London, London, UK. 10Department of Obstetrics and Gynaecology, School of Medicine, Monash University, Melbourne, Australia. 11Aston Medical School, Aston University, Birmingham, UK. 12Action on Pre-eclampsia (APEC) Charity, Worcestershire, UK. 13Julius Centre for Health Sciences and Primary Care, University Medical Centre Utrecht, Utrecht, The Netherlands. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.