Cohort profile: Improved Pregnancy Outcomes via Early Detection (IMPROvED), an International Multicentre Prospective Cohort

Cohort setting, location, and key dates

The IMPROvED cohort contains hospital-based maternity data from Republic of Ireland (University College Cork), United Kingdom (Keele University, University of Liverpool, and University Centre Shrewsbury), The Netherlands (Erasmus MC, University Medical Center, Rotterdam), and Sweden (Karolinska Institutet, Stockholm). Recruitment took place between 29th November 2013 and 3rd August 2017. While work on the use of metabolite biomarkers for preeclampsia risk assessment is ongoing, the data and biobank are now available for international collaboration to conduct high quality research into the cause and prevention of adverse pregnancy outcomes.

All centres obtained ethical approval for the IMPROvED consortium project from their respective ethic committees (Clinical Research Ethics Committee of the Cork Teaching Hospitals: ECM5(3)06/08/13 in August 2013; West Midlands - Solihull Research Ethics Committee: 13/WM/0268 in July 2013; Medical Ethics Committee Erasmus MC of Rotterdam: NL44426.078.13 in July 2013, and EPN – Stockholm Regional Ethics Review Board: 2013-306-31-2 in April 2013). Informed consent was signed by participants at the first study visit (11+0 to 13+6 weeks’ gestation).

Eligibility criteria and inclusion

Eligibility criteria included females aged 16 years or older, nulliparous, singleton pregnancy, and signed informed consent. Full exclusion criteria have been published previously² and included the following: unsure of last menstrual period and unwilling to have ultrasound scan at ≤20 weeks’ gestation; ≥3 miscarriages; ≥3 terminations; known or suspected major foetal anomaly/abnormal karyotype; essential hypertension treated pre-pregnancy; moderate-severe hypertension at booking (BP >160/100 mmHg); diabetes mellitus; renal disease; systemic lupus erythematosus; anti-phospholipid syndrome; sickle cell disease; HIV positive; major uterine anomaly; cervical suture in situ; knife cone biopsy; long-term glucocorticosteroids; treatment with low-dose aspirin; or treatment with heparin/low molecular weight heparin².

Initially recruitment was planned in five European countries with the following recruitment targets: Republic of Ireland (N=1,000), United Kingdom (N=1,500), The Netherlands N=1,000, Sweden (N=750), and Germany (N=750)². However, the study was not feasible at the German site, and they subsequently withdrew from the study at an early stage⁸. The following samples were enrolled in each remaining countries: Republic of Ireland N=1,501; United Kingdom N=1,108; The Netherlands N=810, and Sweden N=619 (Figure 1 and Table 1).

Figure 1. Flow diagram of study participants for main study visits in the IMPROvED cohort.

Data collection and follow-up

Only those who consented to sampling procedures at the second (~15 weeks’ gestation) and third (~20 weeks’ gestation) time-points were eligible for recruitment. While participation at the first (~11 weeks’ gestation) and fourth (~34 weeks’ gestation) time-points were desirable, these were not mandatory. Informed written consent was obtained from all participants at their first study visit. Participants were interviewed by a research midwife at each sampling time-point and all data, including data on storage details of specimens, were entered directly into the IMPROvED database. Participants were instructed to contact the research midwife if delivery occurred before the final study visit or if they developed preeclampsia, had a spontaneous preterm birth or delivered a small for gestational age (SGA) baby².

First sampling (optional visit): The first study visit took place at 11+0 to 13+6 weeks’ gestation. Maternal measurements were performed for height, weight, blood pressure, pulse, urinary protein, and blood glucose for a maximum of 1,076 participants. Specimens including non-fasting 30 ml blood, 10 ml mid-stream sample of urine, and sample of hair were also collected for 1,076 participants.

Second sampling: The second study visit took place at 14+0 to 16+6 weeks’ gestation, resulting in a total sample size of 4,038 participants (including the 1,076 participants from the first non-mandatory study visit). If maternal measurements were not taken at the first study visit, these were taken at the second study visit. Specimens including non-fasting 30 ml blood (n=3,992), 10 ml mid-stream sample of urine (n=3,992), and sample of hair (n=3,081) were also collected. Information on demographics, current pregnancy details, and lifestyle factors were collected during this visit through interview with a research midwife. Demographic information included maternal age, marital status, ethnicity, country of birth, education, occupation, living situation, household income and type of maternity care. Current pregnancy details included information such as gravidity and history of pregnancy complications, for example, infertility, hypertensive disorders of pregnancy and stillbirth, as well as any medical conditions. Lifestyle factors included data on smoking, alcohol use, as well as multivitamin use during pre-pregnancy, during the first trimester, and by the first study visit (Table 2).

Third sampling: The third study visit took place at 19+0 to 21+6 weeks’ gestation. Maternal measurements were performed for weight, blood pressure, pulse, urinary protein, and blood glucose for a maximum of 3,794 participants. Specimens including non-fasting 30 ml blood (n=3,794), 10 ml mid-stream sample of urine (n=3,794), and sample of hair (n=3,083) were also collected

Fourth sampling (optional visit): The fourth study visit took place at 32+0 to 34+6 weeks’ gestation. Maternal measurements were performed for weight, blood pressure, pulse, urinary protein, and blood glucose for a maximum 1,313 participants. Specimens including non-fasting 30 ml blood (n=1,313), 10 ml mid-stream sample of urine (n=1,313), and sample of hair (n=1,024) were also collected.

At birth (optional visit): Placental samples (n=59) were taken shortly after delivery. At centres in Republic of Ireland and Sweden, blood from the umbilical cord and a sample of the cord itself were taken for 974 and 41 participants, respectively.

Postpartum: Within 72 hours following delivery, information about the pregnancy, delivery and the baby were obtained by a research midwife through interview/reviewing medical records. The total sample size at this visit was 3,951 participants (including those lost to follow-up but for whom outcome data could be obtained from medical records, N=168). This data included, but was not limited to, information on mode of delivery, infant sex, hypertensive disorders of pregnancy, placental abruption, rupture of membranes, birthweight and gestational age of baby, Apgar scores at 1 and 5 minutes, any severe neonatal morbidity, whether infant was admitted to neonatal unit as well as reason for neonatal unit admission (Table 2). If possible, the baby’s measurements were also taken at this time. If not, these were obtained from medical records. All information was confirmed by reviewing medical records. Information about complications of pregnancy, including the primary outcomes were also recorded. Any participant who developed preeclampsia, experienced spontaneous preterm birth, or delivered an SGA baby had detailed clinical, laboratory, and outcome data collected².

Database and biobank development

An IMPROvED customised clinical data and biobank management database was developed in Sweden. This database was specifically designed for data management in clinical trials and cohort studies. Comprehensive clinical data, blood, urine, and hair samples were collected and recorded in this database at each study visit.

In addition, IMPROvED established a high calibre pregnancy biobank containing samples from participants at each study visit. The IMPROvED pregnancy biobank is housed at University of Cork, Ireland. Both the epidemiological data and biobank samples can be used by the scientific community to conduct high quality research into maternal and child health.

Data analysis

All descriptive statistics for the current study were performed using Stata MP 14.2 (RRID:SCR_012763) (free alternative, RStudio).

Baseline characteristics

Baseline characteristics of study participants are outlined in Table 1. Briefly, most participants were in the 28–37 years age bracket (66.7% overall). The United Kingdom had a higher proportion (45.0%) of younger participants enrolled (i.e., 18 to 27 year-olds) compared to other countries (30.0% overall). A slightly higher proportion of obese participants were enrolled in the United Kingdom (16.5%), while a lower proportion were enrolled in Sweden (7.0%), compared to 11.9% overall. The majority of participants had a third level of education (60.9% overall). However, this was less pronounced in the United Kingdom (51.7%) and The Netherlands (45.8%). Most participants were in paid employment across all countries (89.4% overall). Similarly, most participants were married or in a stable relationship (94.7% overall). The majority of participants were non-smokers (77.8% overall). However, a slightly higher proportion of participants smoked at time of first study visit in Republic of Ireland (7.7%) and the United Kingdom (8.7%) compared to 6.1% overall. There were fewer non-drinkers of alcohol (defined as zero alcohol intake in the three months prior to pregnancy) enrolled in Republic of Ireland (8.9%) and Sweden (19.9%) compared to the United Kingdom (39.1%) and The Netherlands (34.4%). Overall, 74.5% stopped drinking alcohol before the first visit, while 2.5% of participants in the Republic of Ireland were still drinking alcohol at time of first study visit compared to 1.5% overall. Missing data was <1% at baseline.

Findings to date

Data on 87 participants were lost to follow-up between enrolment and the final study visit (within 72 hours following delivery), resulting in 3,951 participants with postpartum outcome data. Of these, 74.0% (N=2922) had an uncomplicated pregnancy. Similar to estimates reported elsewhere^5,9, approximately 3.0% (N=122) of participants developed the primary study outcome of preeclampsia. This is subdivided by country as follows: Republic of Ireland N=57 (3.9%); United Kingdom N=30 (2.9%); The Netherlands N=24 (3.0%); and Sweden N=11 (1.8%). Other outcomes of interest including spontaneous preterm birth occurred in 3.6% (N=143) and SGA in 10.5% (N=416). Missing data was minimal (<1%) for key study outcomes, while there was ≥1% missing data for mode of delivery (1.0%) and Apgar score at 1 minute and 5 minutes (1.6% and 1.8%, respectively) (Table 3).

Conducted analyses

ProTest: The original protein biomarker study identified, verified, and validated novel panels of protein biomarkers for the prediction of preeclampsia at ~20 weeks’ gestation⁷. From this study, a panel of five protein biomarkers was selected for development into a clinical assay. These proteins were: insulin-like growth factor acid labile subunit (IGFALS), serine peptidase inhibitor Kunitz type 1 (SPINT1), melanoma cell adhesion molecule (MCAM), and the angiogenic factors placental growth factor (PlGF) and soluble endoglin (sENG), by now well-established markers for preeclampsia¹⁰. Throughout the protein biomarker study, various mass spectrometric techniques were applied to firstly identify¹¹ and then quantify the proteins of interest in a targeted fashion^12,13. However, in 2012, mass spectrometry based multiplex protein analyses were not well established in clinical laboratory routine. A technology transfer from mass spectrometry based analyses to another multiplexing technology¹⁴ compatible with antibody based immunoassay technologies for protein analyses was therefore deemed strategic for any future market acceptance of ProTest. As part of this transfer, specific antibody pairs for the proteins in the ProTest panel needed to be generated and evaluated for technical feasibility. With PlGF technology already available in the market, efforts were focused on developing a multiplex protein assay for the four other protein biomarkers. Unfortunately, severe technical issues were encountered. First, the IGFALS detection antibody cross reacted with capture antibodies against MCAM, SPINT1, and sENG, resulting in false positive results. Despite extensive further assay optimization efforts to resolve antibody cross-reactivity, the IGFALS assay could not be incorporated in a multiplex assay; a decision not to progress de novo antibody development for IGFALS was taken. Second, inadequate analytical sensitivity was obtained for SPINT1. Although dedicated assay parameter optimisation yielded sufficient analytical sensitivity, these parameters were not compatible with the parameters applicable for other protein assays, rendering SPINT analysis incompatible with multiplexing techniques. By that time, the clinical use case for a 20 weeks preeclampsia screening solution was put into question by meta-analyses indicating that aspirin prophylaxis to prevent (preterm) preeclampsia needed to start before <16 weeks’ gestation^15,16, this finding was corroborated by the results of the ASPRE trial¹⁷. Confronted with the multitude of technical hurdles and changes in screening test requirements, MyCartis (Belgium) developing ProTest abandoned the project.

Ongoing analyses

MetTest: The original metabolite biomarker study used a single LC-MS experimental set-up to analyse deproteinised metabolite extracts from patient blood specimens (EDTA plasma) collected at ~15 weeks’ gestation in a discovery/validation metabolism profiling study. This resulted in a multivariate predictive model combining 14 putatively identified metabolites⁶. In contrast to proteins, mass spectrometry based multiplex metabolite analyses were well established in clinical laboratory settings at the start of IMPROvED project; notably newborn screening for inborn errors of metabolism has a worldwide clinical application^18,19. Hence, there was no commercial imperative to migrate the metabolite analyses to another analytical technology platform. Instead, the IMPROvED translational research planning for MetTest focused on converting the results of the metabolism profiling study into targeted LC-MS analyses based on the use of reference materials, as required for future application in clinical laboratories. Early in the IMPROvED project it was found that for many of the metabolite biomarkers in the original MetTest multivariate model no reference materials were readily available, thus the original metabolite panel was not amenable to further clinical and commercial development. In response, Metabolomic Diagnostics (Ireland) developing MetTest was compelled to establish an in-house LC-MS translational research workflow centring on multiplexing targeted LC-MS assays for 10s of putative metabolite biomarkers, whereby the availability of reference materials was a selection criterion for biomarker inclusion²⁰.

In parallel, MetTest researchers looked into formalising screening targets for preeclampsia screening in low-risk nulliparous women. This led to novel methodology to assess predictive values, statistics relevant to clinical practitioners, directly from receiver operating characteristic curves, used by test developers to summarise a test’s diagnostic performances in function of test sensitivity and test specificity²¹. Using this methodology, it was proposed that a preeclampsia risk stratification test for nulliparous should ideally mimic the preeclampsia risk information as available for a second-time pregnant woman²¹. The IMPROvED Consortium used these screening targets to perform an early cost-effectiveness analysis to assess both costs and health outcomes of a new screening test that would deliver such risk stratification²².

During the lifetime of the IMPROvED project, PlGF gained wide-spread acceptance as an important preeclampsia risk biomarker, and early pregnancy preterm preeclampsia risk screening evolved into a separate clinical application. In response, MetTest was re-envisioned as a test that combined PlGF and metabolite biomarkers to deliver improved preterm preeclampsia as well as preeclampsia screening in nulliparous women. Using the purpose-developed LC-MS translational research workflow, candidate metabolite biomarkers were analysed in a case-control study. Models were evaluated in function of two pre-defined clinical use scenarios: (1) identify women at risk of developing preterm preeclampsia and (2) identify women at risk of developing preeclampsia at any stage of the pregnancy. It was found that combining dilinoleoyl-glycerol with PlGF effectively predicted increased preterm preeclampsia risk at ca. 15 weeks’ gestation. The further addition of heptadecanoyl-2-hydroxy-sn-glycero-3-phosphocholine expanded the capacity to also identify pregnant women at decreased risk of developing any form of preeclampsia²⁰.

In a further evolution of MetTest, it was shown that metabolite biomarkers can differentially predict preterm preeclampsia across body mass index classes²³, supporting the existence of distinct maternal risk profiles, a contemporary understanding in preeclampsia research^24–26. Using machine learning methodology, these findings led to the development of novel prediction algorithms for preterm preeclampsia prediction in all pregnant women. Metabolite biomarkers augmented the established biomarkers PlGF, mean arterial pressure (MAP), and uterine artery pulsatility index (UTA-PI). Three novel prediction models were developed for three scenarios reflecting different levels of screening resources available; in each scenario use of metabolite biomarkers improved preterm preeclampsia prediction over the comparator models without metabolites. Classification of the pregnant women according to the maternal characteristics body mass index and/or race proved instrumental in achieving improved prediction²⁷. The latest iteration of MetTest is currently being developed into a clinical test.

Publications

A recent publication²⁸ by members of the current study used IMPROvED data to examine the association between socioeconomic status and pregnancy and neonatal outcomes. We did not find strong evidence of associations between individual-level socioeconomic factors and pregnancy and neonatal outcomes overall, with only few significant associations observed among pregnancy outcomes. It is anticipated that IMPROvED data will be used in further maternal and child health secondary analysis research in the future.

Other publications arising as part of the IMPROvED Consortium include a systematic review and meta-analysis examining early pregnancy biomarkers in preeclampsia²⁹, as well as a cost-effectiveness analysis of screening for preeclampsia in nulliparous women²².

Strengths and limitations and further details

The IMPROvED Study has some limitations that should be noted. First, at time of enrolment, the majority of participants recruited had a third level of education, were in paid employment, and were married or in a stable relationship. Therefore, those with lower-level socioeconomic indicators may be underrepresented in the current cohort. Second, while recruitment was initially planned in five European countries (i.e., Republic of Ireland, United Kingdom, The Netherlands, Sweden, and Germany), the German site withdrew from the study at an early stage and were therefore not included in the IMPROvED cohort⁸. Third, target recruitment numbers were not met resulting in 212 fewer participants than what was originally anticipated among participating countries. However, there was little data lost (~2%) between enrolment (N=4038) and the final study visit (N=3951) reducing the potential for selection bias driven by attrition from the cohort. Finally, participants were not universally screened for the presence of gestational diabetes mellitus (GDM), potentially underestimating the incidence of GDM in the study.

There are also several strengths. First, it contains both epidemiological data and biobank data that utilised numerous aliquots on multiple media enabling high quality research into the cause and prevention of adverse pregnancy outcomes. Second, all data were collected in a standardised manner by trained research midwives following detailed study specific standard operating procedures. This was to ensure standardised processes across recruitment centres to minimise bias during recruitment and data collection. Third, there was minimal missing data (<1%) among baseline characteristics and key outcomes, therefore maintaining the statistical power and representativeness of the cohort³⁰. Finally, this cohort profile increases awareness among the scientific community of the potential to access IMPROvED data and biobank samples. This, in turn, could foster collaborations and encourage researchers to obtain funding and ethical approval for evidence-based studies they would not normally be in a position to perform because of lack of access to large patient cohorts.