Prediction of Preeclampsia Using the Soluble fms-Like Tyrosine Kinase 1 to Placental Growth Factor Ratio

Supplemental Digital Content is available in the text.

The aim of the present analysis was to evaluate the effectiveness of the sFlt-1:PlGF ratio as a screening test for preeclampsia in unselected nulliparous women recruited to the POP study (Pregnancy Outcome Prediction). 7,8 Most of the participants were healthy because the cohort selection was solely based on nulliparity, singleton pregnancy, and the study catchment area. We analyzed the sFlt-1:PlGF ratio measured repeatedly at 20, 28, and 36 wkGA using the Roche Cobas e411 Elecsys immunoassay system, which has been certified by the Conformité Européenne mark for use as an in vitro medical device. Screen positive was defined on the basis of the previously described and validated cutoff of >38. 4 We studied the most clinically important manifestations of preeclampsia, namely any severity of the disease leading to preterm birth or preeclampsia with severe features.

Study Design
The POP study was conducted at the Rosie Hospital, Cambridge, United Kingdom, as previously described. 7,8 In brief, it was a prospective cohort study of nulliparous women attending the hospital for their dating ultrasound scan between January 14, 2008 and July 31, 2012 with a viable singleton pregnancy. The only clinical exclusion criterion was multiple pregnancy. The population was drawn from Cambridge and surrounding areas, with low rates of severe socioeconomic deprivation. Therefore, the cohort can be considered a population with low prior risk of disease and homogeneous from a socioeconomic perspective. Blood was obtained at the time of recruitment (not analyzed in this study). Study participants attended the NIHR (National Institute for Health Research) Cambridge Clinical Research Facility at ≈20, ≈28, and ≈36 wkGA for blood sampling and ultrasound scans. Ethical approval was given by the Cambridgeshire 2 Research Ethics Committee (reference number 07/H0308/163) and all participants provided written informed Consent.

Outcome Data
Outcome data were ascertained by review of each woman's paper case record by research midwives and by record linkage to clinical electronic databases of ultrasonography (Astraia, Munchen, Germany), delivery (Protos; iSoft, Banbury, United Kingdom), biochemical tests (Meditech, Westwood, MA), and neonatal intensive care (Badgernet; Clevermed Ltd, Edinburgh, United Kingdom). Where preeclampsia was suspected on the basis of these data, there was a second review of the clinical case record to confirm the diagnosis and classification (ie, with or without severe features) on the basis of the objective criteria of the 2013 ACOG Guideline (The American Congress of Obstetricians and Gynecologists; online-only Data Supplement). 9 Superimposed preeclampsia was defined as preeclampsia in women with preexisting renal disease or hypertension. Socioeconomic status was quantified using the Index of Multiple Deprivation, 10 and birth weight percentiles were calculated using a population-based United Kingdom reference. 11

Analysis Plan and Reporting
The definition of exposures, primary outcomes, secondary outcomes, and sensitivity analyses were agreed in an analysis plan (online-only Data Supplement) before performing any analysis of the sFlt-1 and PlGF data. Analyses which were not predefined are identified as such. The primary outcome for the 20 wkGA measurement was a composite of (1) preeclampsia with delivery before 28 wkGA or (2) preeclampsia with delivery before 37 wkGA where the onset of hypertension was before 28 wkGA. The primary outcome for the 28 wkGA measurement was preeclampsia with delivery before 37 wkGA. The primary outcome for the 36 wkGA measurement was preeclampsia with severe features (ie, preeclampsia with either severe hypertension or evidence of hepatic, renal, hematologic, cerebral, or pulmonary complications; online-only Data Supplement). Secondary outcomes are defined in the online-only Data Supplement. High risk of preeclampsia was defined as either (1) maternal characteristics, using the UK National Institute for Health and Care Excellence Guideline (online-only Data Supplement) 12 or (2) elevated 20 wkGA uterine artery Doppler, defined as a mean pulsatility index in the highest decile, as previously described. 8 Family history of preeclampsia was not included in the definition of risk status because this information was not available. The reporting of this study conforms to the STROBE (The Strengthening the Reporting of Observational Studies in Epidemiology) statement.

Samples and Immunoassays
Serum samples were collected as previously reported 6 and stored at −80°C. All samples used in the current analysis had not previously been thawed before the day of analysis. Researchers performing the assays were blinded to the patients' clinical information and pregnancy outcome. Maternal serum levels of sFlt-1 and PlGF were measured using Roche Elecsys assays on the electrochemiluminescence immunoassay platform, Cobas e411 (Roche Diagnostics). Using this system, the intra-assay coefficient of variation for human serum samples is <2% for sFlt-1 and PlGF, and the interassay coefficients of variation are 2.3% to 4.3% for the sFlt-1 assay and 2.7% to 4.1% for the PlGF assay. Screen positive was defined as sFlt-1:PlGF ratio of >38. 4 We also studied more severe elevation of the ratio, namely, >85 at 28 wkGA and >110 at 36 wkGA. 5

Statistical Analysis
Full details of the statistical analysis are described in the Analysis plan (online-only Data Supplement). In brief, standard screening summary statistics were calculated from 2×2 tables. In addition, sFlt-1:PlGF ratio was analyzed as a continuous variable using the area under the receiver operating characteristic curve (AUROCC). Timeto-event analysis was performed where delivery with the given outcome was the event and delivery without the outcome was treated as a competing risk, and this method was used to generate plots of the cumulative incidence of the outcome from the time of measurement. All analyses were performed using Stata 14.1.

Description of the Study Cohort
The study group consisted of 4099 women, of whom 3953 had sFlt-1:PlGF measurement available from the 20 wkGA visit, 3989 from the 28 wkGA visit, and 3776 from the 36 wkGA visit. The overall incidence of preeclampsia was 6.5% (265/4099). The incidence of preterm preeclampsia with onset before 28 wkGA after the 20 wkGA measurement was 0.10% (4/3953). The incidence of preeclampsia leading to preterm delivery following the 28 wkGA measurement was 0.65% (26/3989). The incidence of delivery with severe preeclampsia after the 36 wkGA measurement was 2.8% (106/3776). The characteristics of the cohort are tabulated according to their experience of hypertensive complications of pregnancy (Table 1). In normotensive women, the median sFlt-1:PlGF ratios were 6.5, 2.7, and 11.2 at 20, 28, and 36 wkGA, respectively. The median ratio at 28 wkGA

Screening Performance at 20 wkGA
None of the 4 women who experienced the primary outcome after the 20 wkGA measurement had an sFlt-1:PlGF ratio >38. The AUROCC for the sFlt-1:PlGF ratio was 0.70 (95% confidence interval, 0.43-0.97; Figure 1A). Ten women had a ratio >38, and 1 woman had a ratio >85 at 20 wkGA. Further analysis was not performed because of the small number of events.

Analysis of Severe Elevation of the sFlt-1:PlGF Ratio
We studied more severe elevation of the ratio, using predefined thresholds, namely, 85 at 28 wkGA and 110 at 36 Data are expressed as median (IQR) or n (column %) as appropriate. For fields where there is no category labeled missing, data were 100% complete. Maternal age was defined as age at recruitment. All other maternal characteristics were defined by self-report at the 20 wkGA interview, from examination of the clinical case record, or linkage to the hospital's electronic databases. Deprivation was quantified using the Index of Multiple Deprivation 2007, 10 which is based on census data from the area of the mother's postcode. Birth weight percentiles and z scores were calculated using a population-based UK reference. 11 Gestational hypertension was defined as development of hypertension in the second half of pregnancy in a previously normotensive woman who did not fulfill the diagnostic criteria for preeclampsia (online-only Data Supplement). DM indicates diabetes mellitus; FTE, full-time education; IQR, interquartile range; PI, pulsatility index; PlGF, placental growth factor; sFlt-1, soluble fms-like tyrosine kinase 1; UtA, uterine artery; and wkGA, weeks of gestational age.

Screening Performance at 36 wkGA Using the sFlt-1:PlGF Ratio Combined With Maternal Risk Factors
We then analyzed a composite definition of screen positive at 36 wkGA, namely, an sFlt-1:PlGF ratio of >110 irrespective of maternal risk factors or an sFlt-1:PlGF ratio >38 combined with maternal risk factors. A total of 195 (5.2%) women screened positive by this definition and 43% of them subsequently delivered with a diagnosis of preeclampsia: 41 women (PPV=21%) developed preeclampsia with severe features, and 43 (PPV=22%) developed preeclampsia without severe features. The characteristics and outcomes for the 195 women are summarized (online-only Data Supplement).

Time-to-Event Analysis
We plotted the cumulative incidence of the primary outcomes after the 28 and 36 wkGA measurements of the sFlt-1:PlGF ratio using a competing risks model (Figure 2A and B). The 36 wkGA measurements were stratified by maternal risk status ( Figure 2B). In both cases, the curves started to deviate at least 1 week after the time of measurement, and the proportional increase in risk was maintained over the 7 to 8 weeks after the test. We also plotted the cumulative incidence of preeclampsia for women at 36 wkGA with the composite definition of screen positive ( Figure 2C), that is, a ratio >38 plus risk factors or a ratio of >110 irrespective of risk factors. Delivery without the primary outcome was treated as a competing risk in all 3 analyses. In all 3 plots, it is evident that >90% of the deliveries in the highest risk group occurred >1 week from the time of measurement of the ratio.

Discussion
The main finding of this study is that, in a cohort of unselected, first singleton pregnancies, measurement of the sFlt-1:PlGF ratio identified women with a high absolute risk   of experiencing the clinically most important manifestations of preeclampsia. At 28 wkGA, an sFlt-1:PlGF ratio >38 identified women with a high risk (>30%) of subsequently delivering preterm with preeclampsia. Women who had a more severe elevation of the ratio (>85) had nearly 60% risk of delivering preterm with preeclampsia, whereas >99% of women who had a ratio of <38 did not develop the outcome. At 36 wkGA, ≈5% of women were identified as high risk on the basis of either an sFlt-1:PlGF ratio of >110 or an sFlt-1:PlGF ratio >38 plus maternal risk factors. Of this group, 43% were subsequently delivered with a diagnosis of preeclampsia, and about half of these cases were severe. Approximately 70% of women were identified as low risk at 36 wkGA, that is, they had no maternal risk factors and an sFlt-1:PlGF ratio ≤38; their risk of developing severe preeclampsia was <1%.
Screening is generally only conducted when there are evidence-based interventions which mitigate the risk. A key element in the study design was to make a measurement close to term (36 wkGA), the rationale being that delivery is the main intervention to prevent preeclampsia. This can more easily and more safely be performed at term. 13 Moreover, a randomized controlled trial has shown improved outcome after immediate induction of labor compared with expectant management in women with gestational hypertension or mild preeclampsia near term. 14 In this study, we evaluated the diagnostic effectiveness of the sFlt-1:PlGF ratio in identifying women at risk of developing preeclampsia. In light of our results, we hypothesize that one approach to reducing the burden of morbidity associated with preeclampsia could be to screen nulliparous women at 36 wkGA using maternal risk factors and sFlt-1:PlGF ratio, monitor screen positive women closely, and perform induction of labor before the development of severe disease. This hypothesis could be readily tested in a randomized controlled trial evaluating whether the introduction of the screening test improves pregnancy outcome (clinical effectiveness). Such an intervention is unlikely to cause harm, and we have recently demonstrated that routine induction of labor at 39 wkGA does not increase the risk of cesarean delivery or perinatal morbidity in another high-risk population of nulliparous women, namely, those aged ≥35. 15 The sFlt-1:PlGF ratio was also informative for the risk of preterm preeclampsia. Women with a ratio >38 at 28 wkGA had a 32% risk of preterm delivery with preeclampsia, and the ratio >38 had a high positive likelihood ratio (≈70). This may reflect the fact that this threshold represents much more significant elevation at 28 wkGA (99.5th percentile) than 36 wkGA (85th percentile). However, the sensitivity of the sFlt-1:PlGF ratio >38 at 28 wkGA was only 23%. Although the POP study included mostly healthy women, this is consistent with the findings in women clinically suspected to have preeclampsia, where an elevated sFlt-1:PlGF ratio between 24 and 37 wkGA was also associated with an increased risk of developing the disorder within 4 weeks after the measurement. 16 Hence, although the test provides clinically useful prediction of risk for a small proportion of women, the majority of women experiencing the disease would not be identified using the test at 28 wkGA. However, the AUROCC for the sFlt-1:PlGF ratio at 28 wkGA for preterm preeclampsia was 0.80. It is possible that combining the ratio with other measurements (clinical, biomarker, and ultrasonic) in a multivariable model might provide better risk prediction. This is an area of further investigation, which will be hopefully paralleled by studies aiming at the implementation of better treatment options. Currently, the main limitation of the clinical usefulness of the 28 wkGA measurement is the lack of a clearly effective intervention mitigating the risks for those who screen positive, other than close monitoring of the patient. Another important area for further study is to refine the estimation of risk in women whose 36 wkGA assessment identified them as being at intermediate risk (5%-6%) of severe preeclampsia, namely, women with an sFlt-1:PlGF ratio >38 and no risk factors or risk factors but a ratio of ≤38. Possible approaches include identifying other informative biomarkers or repeating measurement of the sFlt-1:PlGF ratio after 36 wkGA.
This study had many methodological strengths over previous studies. First, the size was sufficiently large that we were able to study the variants of preeclampsia associated with the most severe complications. Second, we used a clinically validated assay where the definition of screen positive was based on prior studies which had identified and validated the chosen threshold. Moreover, the sFlt-1:PlGF ratio can be calculated without modeling in relation to gestational age or maternal characteristics, and in this study, clinical care was provided without knowledge of the test result. Finally, the analyses in this study were planned and specified in advance.
A large-scale study on screening women using the sFlt-1: PlGF ratio has recently been reported. 17 However, their findings are difficult to compare to this study because they pooled the results from a wide range of gestational ages (30-37 wkGA). Moreover, almost half of their population consisted of multiparous women who had not previously experienced preeclampsia, and this is a group with a very low prior risk of disease. The PPV of a test depends both on the a priori risk and the positive likelihood ratio. It is difficult to interpret a summary estimate of PPV when almost half the population has a very low prior risk of the outcome. Another large study based on a multiethnic cohort of nulliparous women and high-risk parous women concluded that angiogenic biomarkers measured in the first half of pregnancy performed poorly for predicting later development of preeclampsia. 18 They also observed, however, that the measurements became more strongly predictive when made closer to disease onset, but the analyses of late pregnancy data were limited by high rates of missing biomarker data (20%-30%). The focus of this study on nulliparous women was purposeful. One of the strongest clinical predictors of the risk of preeclampsia is whether a woman has a previous history of pregnancy affected by the condition. This information dominates risk prediction in multiparous women but is necessarily absent among nulliparous women. Another purposeful feature of the design of this study was measurement of biomarkers throughout gestation. Results reported in the current and previous studies 13 suggest that screening tests for pregnancy complications have a better predictive value when performed close to disease onset.
Many studies have evaluated trying to predict preeclampsia solely using measurements made in the first trimester. Statistical models are used to determine prior risk. Further modeling is used to process values of first trimester uterine artery Doppler flow velocimetry and to convert first trimester protein concentrations into gestational age-corrected multiples of the median. Two models were externally evaluated in a prospective cohort study in Norway, where measurements were performed between 11 and 14 wkGA. 19 A study-derived threshold (10% false-positive rate) yielded PPVs of 5% to 12%, a sensitivity of 40% and positive likelihood ratios of 1.5 to 3.6 for all preeclampsia cases. Although the nature of that study does not allow direct comparison with this analysis, the current approach may be more likely to be clinically applicable, given that (1) the definition of screen positive was externally defined, (2) the PPVs were higher, (3) the outcome was confined to the clinically most significant cases, and (4) the handling of clinical and biochemical predictors is simpler.

Perspectives
We conclude that measurement of the sFlt-1:PlGF ratio at 36 wkGA, combined with maternal risk factors, provides clinically useful prediction of the risk of preeclampsia at term for ≈3 quarters of unselected nulliparous women, identifying 5% of them as high risk and 70% as low risk. Screening the pregnant nulliparous population in late pregnancy using this measurement could plausibly improve maternal and perinatal outcome when coupled with close monitoring and induction of labor, and this would be an appropriate focus for future randomized controlled trials. Women with extreme elevation of the ratio at 28 wkGA have high absolute risks of preterm disease, and the test may be useful to identify women for the evaluation of candidate disease-modifying therapies as they become available.