Ultrasound and biochemical predictors of pregnancy outcome at diagnosis of early-onset fetal growth restriction

51 52

Severe early-onset fetal growth restriction (FGR) causes significant fetal and neonatal 53 mortality and morbidity. Predicting the outcome of affected pregnancies at the time of diagnosis is 54 difficult, preventing accurate patient counselling. We investigated the use of maternal serum protein 55 and ultrasound measures at diagnosis to predict fetal or neonatal death and three secondary 56 outcomes: fetal death or delivery ≤28+0 weeks; development of abnormal umbilical artery Doppler 57 velocimetry; slow fetal growth. 58

Women with singleton pregnancies (n=142, estimated fetal weights [EFWs] <3rd centile, 59 <600g 20+0-26+6 weeks of gestation, no known chromosomal, genetic or major structural 60 abnormalities), were recruited from four European centres. Maternal serum from the discovery set 61 (n=63) was analysed for seven proteins linked to angiogenesis, 90 additional proteins associated 62 with cardiovascular disease and five proteins identified through pooled liquid chromatography 63 tandem mass spectrometry. Patient and clinician stakeholder priorities were used to select models 64 tested in the validation set (n=60), with final models calculated from combined data. 65

The most discriminative model for fetal or neonatal death included EFW z-score (Hadlock 3 66 formula/Marsal chart), gestational age and umbilical artery Doppler category (AUC 0.91, 95%CI 0.86-67 0.97) but was less well calibrated than the model containing only EFW z-score (Hadlock3/Marsal). 68 The most discriminative model for fetal death or delivery ≤28+0 weeks included maternal serum 69 placental growth factor (PlGF) concentration and umbilical artery Doppler category (AUC 0.89, 95%CI 70 0.83-0.94). 71

Ultrasound measurements and maternal serum PlGF concentration at diagnosis of 72 severe early-onset FGR predict pregnancy outcomes of importance to patients and clinicians. 73 Trial registration: ClinicalTrials.gov NCT02097667 74 Funding: European Union, Rosetrees Trust, Mitchell Charitable Trust. 75 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 4 27/01/2023

76 The survival and growth of a fetus depends on placental provision of nutrients and waste exchange 77 with the mother. When this system is impaired, by inadequate transformation of the uteroplacental 78 circulation or deficits in the structure or function of the placenta, the fetus fails to reach their 79 growth potential (1, 2). The resulting fetal growth restriction (FGR) may be diagnosed antenatally on 80 the basis of an ultrasound determined low estimated fetal weight (EFW) for gestational age; either 81 <3rd centile or <10th centile with abnormal Doppler ultrasound indices in the uterine (UtA) and/or 82 umbilical (UmA) arteries (3) (4, 5). Early-onset FGR, occurring before 32 weeks of gestation, carries 83 significant risks of stillbirth, neonatal morbidity and mortality, neurodevelopmental impairment, and 84 long-term health problems (6-13). There is currently no treatment that can improve fetal growth in 85 utero; instead, management involves monitoring the pregnancy and timing delivery to balance the 86 risks of stillbirth and prematurity (4, 14-16). 87 88 An important question when developing novel therapies for early-onset FGR is which pregnancies to 89 include in early-phase clinical trials. There is a balance to be struck between identifying pregnancies 90 that are sufficiently severely affected to justify the possible risks of the intervention but not so 91 severely affected that there is no potential to determine efficacy. Numerous studies have 92 investigated predictive markers for the development of FGR (17-22), but far fewer have studied the 93 prediction of pregnancy outcome when early-onset FGR is diagnosed. Lack of knowledge about 94 pregnancy outcome in FGR makes it difficult to optimise the inclusion criteria for clinical trials, but it 95 also leaves pregnant patients and their partners with a considerable burden of uncertainty (23, 24). 96 97 The EVERREST Project aims to carry out a phase I/IIa trial of maternal vascular endothelial growth 98 factor (VEGF) gene therapy for early-onset FGR (25). The greatest potential for benefit is in 99 pregnancies at the threshold of viability, for whom our current management option, preterm 100 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 5 27/01/2023 delivery, is not possible or is very high risk. In preparation for a clinical trial of a novel therapeutic, 101 we established a multicentre prospective study to characterise the natural history of early-onset 102 FGR, choosing an extreme phenotype in which the estimated fetal weight (EFW) was <3rd centile and 103 <600g between 20+0 and 26+6 weeks of gestation (henceforth referred to as ‘severe’ early-onset 104 FGR) (26). 105 106 The aim of this work was to prospectively identify and validate ultrasound and serum biochemical 107 factors that could be used to predict fetal or neonatal death in pregnancies affected by severe early-108 onset FGR. These could subsequently be used to select the most appropriate women for a first-in-109 human study of a novel therapeutic to treat FGR, and to better counsel women and their partners 110 about pregnancy outcome. To this end, we asked patients and clinicians to assess the value of our 111 primary and secondary outcomes, based on which we then selected models for validation. 112 Unsupervised parenclitic network analysis by pregnancy outcome and functional network analysis of 113 proteins associated with pregnancy outcomes were also performed to maximise the utility of the 114 proteomic data with the aim of providing insights into the underlying pathophysiology. 115

116 The discovery set, recruited between March 2014 and September 2016, comprised 63 pregnant 117 participants (Figure 1 & Table 1; supplementary data Table 1). Follow-up for the ascertainment of 118 study outcomes was completed in November 2016. The validation set, recruited between October 119 2016 and January 2020, comprised 60 pregnant participants, with follow-up for the ascertainment of 120 study outcomes completed in March 2020. There were no significant differences in maternal 121 demographics, pregnancies characteristics or pregnancy outcomes between the discovery and 122 validation sets (Table 1; supplementary data Table 2 & Figure 1). Overall, 42 (34%) of the 123 pregnancies ended in the primary outcome of fetal or neonatal death (within the first 28 days of 124 life). For the three secondary outcomes, only fetal death or delivery ≤28+0 weeks of gestation could 125 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 6 27/01/2023 be ascertained for all pregnancies, occurring in 58 cases (47%). The UmA Doppler velocimetry was 126 normal (<95th centile for gestation (27)) at enrolment in 46 participants, of whom 21 (46%) 127 subsequently developed abnormal UmA Doppler measurements. Fetal growth trajectory (based on 128 change in percentage weight deviation over a period or two weeks or more (28)) could be assessed 129 for 104 pregnancies (85%), with the remaining pregnancies ending in fetal death or delivery before a 130 2-week interval was reached. Forty-one of these 104 fetuses (39%) demonstrated slow fetal growth 131 (worsening of weight deviation of >10 percentage points). A smaller proportion of fetuses 132 demonstrated slow fetal growth in the validation set (31%) than in the discovery set (47%). 133 134 135 Figure 1: Flow diagram of participant eligibility and enrolment across the four EVERREST Prospective 136 Study centres from 10th March 2014 to 30th January 2020 for the discovery and validation sets. 137 138 139 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 7 27/01/2023 Table 1: Characteristics and outcomes of the discovery, validation and combined participant sets. 140 Discovery and validation sets were compared using 2-sided t tests for symmetrical continuous 141 variables, Mann-Whitney U tests for skewed continuous variables and chi square or Fisher’s exact 142 tests (where specified) for categorical variables. 143 Discovery n=63 Validation n=60 p value Combined n=123 Maternal characteristics Maternal age, mean years (SD) 33.7 (6.0) 33.3 (6.7) 0.72 33.5 (6.3) Primiparous, n (%) 43 (68) 37 (62) 0.44 80 (65) BMI, median (IQR) 24.9 (22.7-28.6) 26.4 (22.8-31.0) 0.36 25.7 (22.8-30.0) Ethnicity, n (%)1 White 42 (67) 27 (46) 0.16 69 (57) Black 10 (16) 15 (25) 25 (20) Asian 10 (16) 15 (25) 25 (20) Other 1 (2) 2 (3) 3 (2) Essential hypertension, n (%) 10 (16) 6 (10) 0.33 16 (13) Pre-eclampsia at enrolment, n (%)# 4 (6) 5 (8) 0.47 9 (7) Enrolment ultrasound measurements Gestational age at enrolment, median weeks+days (IQR) [range] 23+6 (22+3-25+1) [20+1-26+5] 23+5 (22+3-24+4) [20+4-26+4] 0.48 23+5 (22+3-24+5) [20+1-26+5] EFW at enrolment, median grams (IQR) (29) 392 (281-503) 387 (280-448) 0.61 389 (281-484) EFW z-score at enrolment, median (IQR) (29, 30) -3.0 (-3.6 to -2.5) -3.2 (-3.8 to -2.7) 0.17 -3.1 (-3.7 to -2.6) Mean UtA PI >95th centile at enrolment, n (%)2 (27) 49 (79) 49 (82) 0.71 98 (80) UmA PI >95th centile at enrolment, n (%)3 (27) 32 (51) 36 (60) 0.26 78 (55) Absent or reversed UmA end- diastolic flow at enrolment, n (%)3 18 (29) 25 (42) 0.11 43 (35) MCA PI <5th centile at enrolment, n (%)4 (27) 10 (16) 7 (12) 0.41 17 (15) DV a wave absent or reversed at enrolment, n (%)5# 5 (8) 4 (7) 0.51 9 (8) Pregnancy outcomes Pre-eclampsia at any point in pregnancy, n (%)6 24 (39) 17 (33) 0.51 41 (36) Gestational age at diagnosis of stillbirth or delivery of live birth, median weeks+days (IQR) [range] 28+2 (26+3-34+0) [21+4-39+3] 28+2 (26+4-33+2) [22+2-39+6] 0.90 28+2 (26+3-33+2) [22+2-39+6] Female fetus / infant, n (%)7 36 (57) 24 (43) 0.19 60 (50) Caesarean delivery, n (%) 42 (67) 37 (63) 0.65 79 (65) Live births, n (%) 47 (75) 43 (72) 0.71 90 (73) n=47 n=43 n=90 Live births ≤28+0 weeks, n (%) 15 (32) 14 (33) 0.95 29 (32) Live births >37 weeks, n (%) 8 (17) 10 (23) 0.46 18 (20) Caesarean delivery for live births, n (%) 42 (89) 37 (88) 0.85 79 (89) Neonatal deaths, n (%)# 5 (11) 4 (9) 0.56 9 (10) All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 8 27/01/2023 Birth weight z-score for live births, mean (SD) (30) -3.5 (1.1) -3.5 (0.9) 0.97 -3.5 (1.0) Study outcomes n=63 n=60 n=123 Fetal or neonatal death, n (%) 21 (33) 21 (35) 0.85 42 (34) Death or delivery ≤28+0 weeks of gestation, n (%) 30 (48) 28 (47) 0.92 58 (47) n=55 n=49 n=104 Slow fetal growth, n (%) 26 (47) 15 (31) 0.083 41 (39) n=26 n=20 n=46 Development of UmA PI >95th centile, n (%) 12 (46) 9 (45) 0.94 21 (46) 1n=1 missing from validation, 2n=1 missing from discovery, 3n=1 missing from validation, 4n=5 144 missing from discovery and n=1 missing from validation, 5n=3 missing from discovery and n=1 145 missing from validation, 6n=1 missing from discovery and n=8 missing from validation, 7n=5 missing 146 from validation, #Fisher’s exact test. DV=ductus venosus, EFW=estimated fetal weight, MCA=middle 147 cerebral artery, PI=pulsatility index, UmA=umbilical artery, UtA=uterine artery. 148 149 Ultrasound measurements as predictors of fetal or neonatal death and death or delivery ≤28+0 150 weeks of gestation in the discovery set: The best ultrasound predictor of fetal or neonatal death 151 was EFW z-score, either as calculated using the Hadlock 3 formula and Marsal chart (EFWHM: AUC 152 0.81, 95% CI 0.69-0.93) or the Intergrowth formula and chart (EFWInt: AUC 0.83, 95% CI 0.71-0.95). 153 UmA category (95th centile with positive EDF; absent EDF; reversed EDF. AUC 0.75, 154 95% CI 0.62-0.88) and slow fetal growth (AUC 0.70, 95% CI 0.56-0.83) were also fair predictors. UmA 155 category was the best predictor of death or delivery ≤28+0 weeks (AUC 0.80, 95% CI 0.70-0.91), with 156 mean UtA PI (AUC 0.77, 95% CI 0.65-0.89) and Intergrowth EFW z-score (AUC 0.73, 95% CI 0.60-0.85) 157 also fair predictors (supplementary data Tables 3 & 4). 158 Proteomics: Mass spectrometry (MS) profiling of pooled samples gave quantitative information for 159 200 protein groups (sets of proteins that cannot be distinguished based on peptide sequences), from 160 which human placental lactogen (HPL), fibronectin, pregnancy-specific beta-1 glycoprotein (PSG1), 161 serum amyloid A (SAA) and leucyl-cystinyl aminopeptidase (LNPEP) were selected for individual 162 validation, based on the scoring system outlined in the methods. 163 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 9 27/01/2023 Univariate associations between maternal serum protein concentrations and outcomes in the 164 discovery set: Four proteins were undetectable in most of the samples: VEGF-A, natriuretic peptides 165 B (BNP), melusin, poly[ADP-ribose] polymerase 1 (PARP1). These were excluded from further 166 prediction analyses. The associations between the remaining 98 proteins and the four pregnancy 167 outcomes are summarised in Figure 2. Placental growth factor (PlGF) and HPL concentration were 168 significantly associated with fetal or neonatal death (after Benjamini-Hochberg correction), with 169 fold-changes of 0.52 in pregnancies ending in fetal or neonatal death compared with pregnancies 170 ending in neonatal survival. The concentrations of nine proteins were significantly associated with 171 death or delivery ≤28+0 weeks (after correction). The greatest magnitudes of fold-changes were 172 seen for PlGF (0.28), HPL (0.45), and PSG1 (0.48). 173 174 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 10 27/01/2023 175 176 177 Figure 2: Volcano plots showing the statistical significance and magnitude of the associations 178 between the 98 proteins and four pregnancy outcomes in the discovery set. Associations tested with 179 2-sided t tests for symmetrical data and Mann-Whitney U tests for skewed data. Dotted line 180 indicates a p value of 0.05. Dashed line indicates the Benjamini-Hochberg cut-off with a 5% false 181 discovery rate (p=0.0015 for fetal or neonatal death, p=0.005 for death or delivery ≤28+0 weeks). 182 See supplementary data Table 5 for full protein names. 183 184 185 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 11 27/01/2023 Parenclitic network analysis of the discovery set: Both the networks for fetal or neonatal death and 186 death or delivery ≤28+0 weeks contained clusters centred around HPL (clusters 6 and 4, Figure 3). 187 These clusters also contained pentraxin-related protein PTX3, spondin-2 (SPON2) and 188 thrombomodulin (TM) and contained or were linked to decorin (DCN). The network for death or 189 delivery ≤28+0 weeks also contained a cluster centred around PlGF (cluster 2). For all three of the 190 networks that included fetal sex, similar clusters emerged that contained renin (REN), angiopoietin 1 191 (ANG1), dickkopf-related protein 1 (DKK1) and platelet-derived growth factor subunit beta (PDGFB), 192 contained or were linked to pregnancy-associated plasma protein A (PAPP-A) and in two of the three 193 networks included lectin-like oxidized LDL receptor 1 (LOX1). Networks and associated dendrograms 194 for the development of abnormal umbilical artery Dopplers and slow fetal growth are provided in 195 supplementary data Figures 4 & 5. 196 197 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 12 27/01/2023 198 199 Figure 3: Parenclitic networks for clusters generated based on: (A) fetal or neonatal death (B) death 200 or delivery ≤28+0 weeks. See supplementary data Table 5 for full protein names and supplementary 201 data Figures 2 & 3 for associated dendrograms. 202 203 (A) (B) Cluster 1 Cluster 2 Cluster 3 Cluster 4 Cluster 5 Cluster 6 Cluster 7 Cluster 2 Cluster 3 Cluster 4 Cluster 1 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 13 27/01/2023 Model selection by stakeholders: None of the single variable or multivariable models for predicting 204 slow fetal growth performed well enough to warrant validation (AUCs <0.70). For the three 205 remaining outcomes, an online survey was performed to ascertain the priorities of clinicians and 206 patients in predicting outcomes. Forty-five clinicians from 18 countries (of 173 contacted, 26%) and 207 seven patients from the UK who had experienced a pregnancy complicated by severe early-onset 208 FGR (of 36 contacted, 19%) responded (supplementary data Table 6). The prediction of fetal or 209 neonatal death and death or delivery ≤28+0 weeks were considered important or very important by 210 all patients and were also rated highly by clinicians for the purposes of patient counselling and 211 clinical management (Figure 4). Patients and clinicians marginally prioritised sensitivity over 212 specificity for most outcomes (supplementary data Figure 6). For the prediction of the development 213 of abnormal UmA PI, patients universally prioritised sensitivity while clinicians marginally prioritised 214 specificity for patient counselling. 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 Figure 4: The proportion of patients who ranked our three pregnancy outcomes as ‘important’ or 230 ‘very important’ and the proportion of clinicians who ranked them as ‘important’ or ‘very important’ 231 for clinical management and for patient counselling. UmA PI=umbilical artery pulsatility index. 232 233 Key Very important Important Patient priorities Clinical management Patient counselling 14% 16% 57% 70% 50% 71% 77% 59% 71% 23% 32% 14% 27% 43% 29% 23% 27% 29% % 20% 40% 60% 80% 100% Percentage of respondents Prediction of fetal or neonatal death Prediction of death or delivery <28+0 weeks Prediction of developing abnormal UmA PI All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 14 27/01/2023 Based on the survey results, the model performance metrics and the assay reliability, models 234 containing the variables listed in Tables 2 and 3 were selected for validation. For the prediction of 235 death or delivery ≤28+0 weeks, models including HPL marginally outperformed models including 236 PlGF. However, possibly related to the short processing time, the commercial HPL ELISA had high 237 intra-assay variability in our hands (mean coefficient of variation 8.0%, SD 7.3%, 28% requiring 238 repeat analysis for coefficient of variation >10%). Because of this, and the existence of clinically 239 approved tests for PlGF, models including PlGF were selected for validation. 240 Model validation: Five of the seven protein models (Table 2) and all five of the models containing 241 ultrasound measurements (Table 3), generated in the discovery set, were successfully validated, 242 with AUCs included in the AUC 95% CIs generated from the discovery cross-validation estimates. 243 Addition of pregnancy characteristics: Validated models were not significantly improved by the 244 addition of maternal BMI, maternal age, maternal ethnicity or fetal sex. Adding gestational age at 245 enrolment significantly improved the models containing EFWHM alone (LR test p=0.0001) and EFWHM 246 with UmA category (LR test p<0.00005) to predict fetal or neonatal death (supplementary data Table 247 7). The addition of ‘pre-eclampsia at enrolment’ significantly improved all validated models 248 predicting fetal death or delivery ≤28 weeks of gestation (supplementary data Table 7). 249 PlGF values for maximum likelihood ratios: Receiver operating characteristic (ROC) curves are 250 shown in Figures 5 and 6. Model constants and coefficients, along with optimal cut points for 251 positive and negative likelihood ratios and correct classification are provided in supplementary data 252 Tables 8 and 9. Serum PlGF concentration <14.2 pg/ml predicted fetal or neonatal death with a 253 positive likelihood ratio of 18.3, sensitivity of 45% and specificity of 98% and correctly classified 80% 254 of participants. Serum PlGF concentration <14.5 pg/ml predicted death or delivery ≤28+0 weeks with 255 a positive likelihood ratio of 24.7, sensitivity of 38% and specificity of 98% and correctly classified 256 70% of participants. 257 258 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 15 27/01/2023 Table 2: Model validation for predicting adverse pregnancy outcomes: the seven pre-specified 259 models containing maternal serum proteins. The model not validated is shaded. Validated models 260 include estimates from the combined discovery and validation sets. 261 Outcomes Variable(s) Discovery (with LOOCV) Validation Combined AUC 95% CI AUC 95% CI AUC 95% CI Fetal or neonatal death PlGF 0.75 0.62-0.88 0.83 0.72-0.95 0.81 0.73-0.89 PlGF & lymphotactin 0.84 0.73-0.95 0.75 0.62-0.88 0.83 0.75-0.91 PlGF, lymphotactin & fibronectin 0.85 0.74-0.96 0.69 0.55-0.83 Death or delivery ≤28+0 weeks PlGF 0.86 0.76-0.96 0.76 0.64-0.88 0.82 0.75-0.89 PlGF & pre- eclampsia 0.84 0.77-0.91 PlGF & PSG1* 0.91 0.82-0.99 0.80 0.69-0.91 0.86 0.80-0.93 PlGF, PSG1 & pre- eclampsia 0.88 0.82-0.94 Development of abnormal UmA PI PlGF 0.84 0.50-0.95 0.64 0.38-0.89 0.78 0.64-0.91 PlGF & fibronectin 0.88 0.74-1.00 0.65 0.39-0.90 0.80 0.67-0.93 *Discovery AUC 0.906, validation 95% CI 0.688-0.912. LOOCV=leave-one-out cross-validation, 262 PI=pulsatility index, PlGF=placental growth factor, PSG1=pregnancy-specific beta-1 glycoprotein, 263 UmA=umbilical artery. Models were generated and tested using the natural log of PlGF in pg/ml and 264 centered and scaled values for lymphotactin normalised protein expression (NPX) on a log2 scale. 265 266 267 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 16 27/01/2023 Table 3: Model validation for predicting adverse pregnancy outcomes: models containing ultrasound 268 measurements, maternal serum protein concentrations and pregnancy characteristics, and their 269 final AUCs from the combined discovery and validation sets. 270 Outcomes Variable(s) Discovery (with LOOCV) Validation Combined AUC 95% CI AUC 95% CI AUC 95% CI Fetal or neonatal death EFWHM z-score 0.78 0.66-0.91 0.89 0.80-0.97 0.85 0.78-0.92 EFWHM z-score & GA 0.90 0.84-0.96 EFWInt z-score 0.80 0.67-0.93 0.91 0.83-0.99 0.87 0.80-0.94 EFWHM z-score & UmA category1 0.78 0.64-0.91 0.88 0.80-0.97 0.86 0.79-0.93 EFWHM z-score, UmA category & GA1 0.91 0.86-0.97 Death or delivery ≤28+0 weeks UmA category1 0.78 0.67-0.89 0.79 0.68-0.91 0.80 0.72-0.88 UmA category & pre- eclampsia1 0.84 0.77-0.91 UmA category & PlGF1 0.89 0.81-0.97 0.85 0.75-0.94 0.89 0.83-0.94 UmA category, PlGF & pre-eclampsia1 0.90 0.85-0.95 1n=1 missing from validation set. EFWHM=estimated fetal weight calculated using Hadlock 3 formula 271 (29) with z-score calculated using Marsal reference chart (30), EFWInt=estimated fetal weight and z-272 score calculated using Intergrowth formulae (31), GA=gestational age at enrolment in days, 273 LOOCV=leave-one-out cross-validation, PlGF=placental growth factor, UmA=umbilical artery. 274 275 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 17 27/01/2023 276 Figure 5: Comparison of the receiver operating characteristic (ROC) curves for the models predicting 277 fetal or neonatal death. EFW-HM=estimated fetal weight calculated using Hadlock 3 formula (29) 278 with z-score calculated using Marsal reference chart (30), EFW-Intergrowth=estimated fetal weight 279 and z-score calculated using Intergrowth formula and reference chart (31), GA=gestational age at 280 enrolment, PlGF=placental growth factor concentration, UmA=umbilical artery Doppler category 281 (0=pulsatility index 95th centile, 2=absent end-diastolic flow, 282 3=reversed end-diastolic flow). 283 284 Figure 6: Comparison of the receiver operating characteristic (ROC) curves for the models predicting 285 fetal death or delivery ≤28+0 weeks of gestation. PET=pre-eclampsia at enrolment, PlGF=placental 286 growth factor concentration, PSG1=pregnancy-specific glycoprotein 1 normalised protein 287 expression, UmA=umbilical artery Doppler category (0=pulsatility index 95th centile, 2=absent end-diastolic flow, 3=reversed end-diastolic flow). 289 290 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 18 27/01/2023 Alternative EFW formulae: Although the EFW z-score calculated using the Intergrowth formula and 291 chart gave the highest AUC for predicting fetal or neonatal death, the Intergrowth formula for 292 estimating fetal weight performed poorly in our sample, especially at lower fetal weights. For the 21 293 livebirths with a birthweight <600g and an EFW performed within seven days of delivery, the 294 Intergrowth formula overestimated birthweight by a mean of 47% (SD 14%), in contrast to the 295 Hadlock 3 formula which overestimated birthweight by a mean of 25% (SD 10%, supplementary data 296 Figure 8). For all 67 livebirths with an EFW performed within seven days of delivery, the Intergrowth 297 formula overestimated birthweight by a mean of 29% (SD 20%) and the Hadlock 3 formula 298 overestimated birthweight by a mean of 15% (SD 13%). As might be expected, using the EFW 299 calculated from one formula in the model derived from the other had a substantial negative impact 300 on calibration (supplementary data Figure 9). 301 Re-analysis of the combined sets: Combining the centred and scaled data from the discovery and 302 validation sets, the strongest associations with both fetal or neonatal death, and death or delivery 303 ≤28+0 weeks were the previously observed negative associations with PlGF (p=1.4x10-8 and 304 p=3.0x10-11) and HPL (p=1.3x10-7 and p=1.4x10-10) (Figure 7). The evidence for the negative 305 association between PSG1 and both outcomes was strengthened, as was the evidence for negative 306 associations between matrix metalloproteinase-12 (MMP12) and programmed cell death 1 ligand 2 307 (PDL2) and death or delivery ≤28+0 weeks. None of the proteins showed an association with the 308 development of abnormal UmA Dopplers or slow fetal growth at a Benjamini-Hochberg 5% false 309 discovery rate (supplementary data Figure 10). 310 311 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 19 27/01/2023 312 Figure 7: Volcano plots showing the statistical significance and magnitude of associations between 313 fetal and neonatal death and death or delivery ≤28+0 weeks and the centred and scaled 314 concentrations of the 93 proteins from the discovery and validation sets combined. Associations 315 tested with 2-sided t tests. Dotted line indicates p=0.05, short-dashed line indicates Benjamini-316 Hochberg cut-off with a 5% false discovery rate (A p=0.0048, B p=0.012), long-dashed line indicates 317 Benjamini-Hochberg cut-off with a 1% false discovery rate (A p=0.00032, B p=0.0013). See 318 supplementary data Table 10 for full protein names and individual -log10 p values. 319 320 Functional analysis of the proteins associated with fetal or neonatal death at a 5% false discovery 321 rate demonstrated co-expression of HPL and GH. Expanding the network to include intervening 322 proteins resulted in clusters sharing GO biological processes of growth hormone receptor signalling, 323 VEGF signalling and calcitonin family receptor signalling, with proteins in the latter two clusters also 324 involved in angiogenesis and regulation of angiogenesis (Figure 8). Proteins associated with death or 325 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 20 27/01/2023 delivery ≤28+0 weeks showed multiple interactions, predominantly centred on fibronectin. Shared 326 GO biological processes included those relating to growth (regulation of angiogenesis, cellular 327 response to growth factors and growth hormone receptor signalling pathway via jak-stat), immune 328 function (leucocyte migration, inflammatory response, and positive regulation of T cell activation) or 329 both (regulation of cell adhesion, positive regulation of nik/NF-kappaB signalling) (Figure 9). 330 331 332 Figure 8: An expanded functional network demonstrating interactions and shared GO biological 333 processes of the proteins associated with fetal or neonatal death in the combined data set at a 334 Benjamini-Hochberg false discovery rate of 5%. Red=growth hormone receptor signalling pathway, 335 yellow=vascular endothelial growth factor signalling pathway, green=calcitonin family receptor 336 signalling pathway, light blue=regulation of angiogenesis, dark blue=angiogenesis. See 337 supplementary data Table 5 for full protein names. Analysis, graphic and legend from STRING (Swiss 338 Institute of Bioinformatics)(32). 339 340 Key Increased in pregnancies ending in fetal or neonatal death Decreased in pregnancies ending in fetal or neonatal death All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 21 27/01/2023 341 Figure 9: Functional interactions and shared GO biological processes of the proteins associated with 342 death or delivery ≤28+0 weeks in the combined data set at a Benjamini-Hochberg false discovery 343 rate of 5%. Red=growth hormone receptor signalling pathway via jak-stat, orange=inflammatory 344 response, yellow=leukocyte migration, light green=extracellular matrix, dark green=regulation of cell 345 adhesion, light blue=regulation of angiogenesis, dark blue=cellular response to growth factor 346 stimulus, pink=positive regulation of nik/NF-kappaB signalling, purple=positive regulation of T cell 347 activation. See supplementary data Table 5 for full protein names. Analysis, graphic and legend from 348 STRING (Swiss Institute of Bioinformatics)(32). 349 350 The three best performing LOOCV models using the combined centred and scaled data all included 351 pro-adrenomedullin (ADM) for predicting fetal or neonatal death, PlGF and HPL for predicting death 352 or delivery ≤28+0 weeks and Platelet-derived growth factor subunit B (PDGFB) for predicting the 353 development of abnormal UmA Doppler measurements (supplementary data Table 12). The 354 emergence of ADM in the models predicting fetal or neonatal death was consistent with the 355 significant association present in the combined but not the discovery sets. In contrast, PDGFB did 356 Increased in pregnancies ending in fetal or neonatal death Decreased in pregnancies ending in fetal or neonatal death Key All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 22 27/01/2023 not show significant univariate associations with the development of abnormal UmA Dopplers in the 357 discovery, validation, or combined data sets. 358 Predicting gestational age of livebirth or diagnosis of fetal death and interval from enrolment to 359 livebirth or diagnosis of fetal death: Twelve protein and ultrasound measurements showed an 360 association with the gestational age at which the pregnancies ended in livebirth or fetal death, at a 361 1% Benjamini-Hochberg false discovery rate (supplementary data Table 13). The best model to 362 predict gestational age at livebirth or fetal death included PlGF and sflt1 concentrations, MMP12 and 363 IL1RL2 NPX and UmA category at enrolment (Figure 10). Eight protein and ultrasound measurements 364 showed an association with the interval between enrolment and either livebirth or the diagnosis of 365 fetal death at a 1% Benjamini-Hochberg false discovery rate (supplementary data Table 13). The best 366 model to predict the interval between enrolment and livebirth or fetal death included PlGF and sflt1 367 concentrations, MMP12 and decorin NPX, UmA category and gestational age at enrolment (Figure 368 10). Both models accounted for 68% of the variation in the outcomes they were predicting but had 369 95% prediction intervals of 40 days, limiting their clinical utility. Sparser models, including PlGF and 370 sflt1 concentrations and UmA category to predict gestational age at livebirth or fetal death, and 371 these same variables plus gestational age at enrolment to predict interval to livebirth or fetal death, 372 had only slightly wider 95% prediction intervals of 42 days (supplementary data Figure 11). 373 374 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 23 27/01/2023 375 Figure 10: (A) Predicted versus actual gestational age of either livebirth or diagnosis of fetal death, 376 based on the model containing PlGF and sflt1 concentrations, decorin and matrix metalloproteinase 377 12 normalised protein expression and umbilical artery Doppler category. (B) Predicted versus actual 378 interval from enrolment to either livebirth or diagnosis of fetal death, based on the model 379 containing PlGF and sflt1 concentrations, decorin and matrix metalloproteinase 12 normalised 380 protein expression, umbilical artery Doppler category and gestational age at enrolment. Green filled 381 circles=pregnancies ending in livebirth, red hollow circles=pregnancies ending in fetal death, dotted 382 lines=95% prediction intervals. 383 384 Placental histological classification: Placental samples for histological examination were available 385 for 55 pregnancies (45%); these had similar characteristics and outcomes to the pregnancies without 386 available samples (supplementary data Tables 14 & 15). The only statistically significant difference 387 was a higher proportion of female fetuses among the pregnancies that had placental samples than 388 those that did not (63% vs 41%, p=0.016). Forty-five (82%) placentas showed evidence of placental 389 pathology, with 39 (71%) classified as maternal vascular malperfusion (MVM), three (5%) as villitis of 390 unknown aetiology (VUE), one (2%) as fetal vascular malperfusion (FVM) and two (4%) as non-391 specific dysmorphic villi. Twelve of the 14 placental samples from pregnancies ending in stillbirth 392 showed MVM, while the three available samples from pregnancies ending in neonatal death showed 393 (A) (B) All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 24 27/01/2023 VUE, FVM and dysmorphic villi. Mean UtA PI, maternal serum PlGF and maternal serum PAPP-A at 394 enrolment all differed significantly between pregnancies with subsequent MVM and pregnancies 395 without MVM. In contrast, none of the umbilical artery parameters studied (UmA category at 396 enrolment; the occurrence of UmA PI >95th centile at any point before delivery; the occurrence of 397 absent or reversed UmA end-diastolic flow at any point before delivery) showed evidence of an 398 association with placental histological classification of MVM (supplementary data Table 16). 399

400 Principle findings and significance 401 To our knowledge this is the first study to use a discovery science approach, combining ultrasound 402 and biochemical parameters, to identify and validate prognostic markers at the time of diagnosis of 403 severe early-onset FGR. These findings can be used to inform personalised counselling and 404 management of affected pregnancies with outcomes of importance to patients and clinicians. 405 Furthermore, by providing alternative thresholds that prioritise positive and negative likelihood 406 ratios and maximum correct categorisation, eligibility criteria for clinical trials of novel therapeutics 407 can be adapted depending on the perceived risk: benefit ratio of the intervention. 408 Our secondary analyses, including parenclitic network analysis, functional enrichment analysis and 409 triangulation with placental histological classification, provide a deeper characterisation of this 410 unique case series. Some of these findings support and enhance our existing understanding of 411 placental FGR, such as the interplay between angiogenesis, immune cells, and the extracellular 412 matrix (33-35). Other findings offer new avenues for investigation, such as the parenclitic network 413 cluster around fetal sex, which includes proteins related to pericyte function (36). 414 415 Findings in the context of existing literature 416 Given that ultrasound assessment of biometry and Doppler velocimetry forms the mainstay of 417 identifying and monitoring FGR, it is unsurprising that EFW z-score and umbilical artery category 418 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 25 27/01/2023 were validated as predictors of fetal or neonatal death and fetal death or delivery ≤28+0 weeks 419 respectively (37-39). A secondary analysis of 105 pregnancies from the UK placebo-controlled trial of 420 sildenafil citrate for early-onset FGR (STRIDER; EFW or AC <10th centile with absent or reversed UmA 421 end-diastolic flow at 22+0-29+6 weeks) identified EFW as an independent predictor of livebirth (OR 422 per 100g 4.3, 95%CI 2.3-8.0, p<0.001) and overall survival (OR per 100g 2.9, 95%CI 1.8-4.4, 423 p<0.001)(40). What is less expected is that absent or reversed ductus venosus a wave was a poor 424 predictor of fetal or neonatal death in our participants (AUC 0.59, 95% CI 0.53-0.66, see 425 supplementary material Table 4), in contrast to the results of previous studies (41, 42). This may 426 reflect a change in clinical practice since these studies were published. Their findings led to ductus 427 venosus waveform becoming an important factor in timing of delivery in extremely pre-term FGR 428 (14, 43), which may have altered the natural history of the disease by prompting delivery before 429 stillbirth could occur. 430 A limitation of using ultrasound parameters is their potential for variation. In the case of Doppler 431 velocimetry this includes interobserver variability, temporal variation due to factors such as 432 maternal and fetal movement, and variation in umbilical artery waveforms between arteries and 433 along the length of the cord (44-47). There is also considerable variation between different Doppler 434

ranges, both in terms of the values of their ‘normal’ ranges and their methodological 435 quality(48). In the case of EFW z-score, variation arises from interobserver variability in measuring 436 biometry, variation in formulae used to generate the EFW and variation in charts used to determine 437 the z-score for gestational age (49-51). Despite the methodological limitations of the Hadlock 3 438 formula, a recent study of 65 pregnancies with early-onset FGR delivered within seven days of 439 ultrasound assessment found it gave a better combination of systematic and random error than 20 440 other formulae tested (52). 441 Several recent studies have highlighted the potential utility of PlGF concentration to predict 442 outcomes in SGA and FGR pregnancies. In a case series of 173 singleton pregnancies with a 443 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 26 27/01/2023 customised EFW <10th centile between 20+0 and 31+6 weeks, the sflt1/PlGF ratio at diagnosis was 444 an excellent predictor of delivery 0.99) and <34 weeks (AUC 445 0.94, 95% CI 0.89-0.98) and a good predictor of a composite adverse perinatal outcome (AUC 0.83, 446 95% CI 0.77-0.90) (53). Similarly, in 116 singleton pregnancies with early-onset FGR (customised EFW 447 <3rd centile or customised EFW <10th centile with abnormal UmA and/or UtA Doppler velocimetry; 448 <32+0 weeks) and positive UmA end-diastolic flow ending in livebirth, women with an sflt1/PlGF 449 ratio ≥85 were significantly more likely to deliver within one, two, three and four weeks from ratio 450 measurement than women with an sflt1/PlGF ratio <85 (54). Composite neonatal morbidity and 451 neonatal admission were also significantly higher following pregnancies with a sflt1/PlGF ratio ≥85 452 (53.8% vs 28.6% p=0.04; 97.5% vs 67.9% p<0.01). More strikingly, in a series of 130 singleton 453 pregnancies with SGA (AC or EFW <10th centile), fetal demise only occurred in pregnancies with a 454 PlGF <10th centile for gestational age at any time between 16 and 36 weeks (12/65 vs 0/65, 455 p<0.0001) (55). 456 While these studies revealed the PlGF results to the managing clinicians, similar results have been 457 found in studies where PlGF was not revealed. The secondary analysis of STRIDER UK trial 458 participants, mentioned above, found significant associations between pregnancy outcomes and 459 both the sflt1/PlGF ratio and PlGF alone (40). Higher PlGF concentrations and a lower sflt1/PlGF 460 ratios were associated with greater overall survival (PlGF coefficient 3.67, p<0.001; ratio coefficient 461 0.51, p=0.002) and later gestation at birth (PlGF coefficient 1.4, p<0.001; ratio coefficient -0.99, 462 p<0.001). Similarly, in a multinational case series of 411 pregnancies, PlGF <5th centile at the time of 463 suspected FGR (AC <10th centile from 20+0 weeks) had an 87.5% sensitivity and 62.8% specificity for 464 predicting stillbirth (56). PlGF 5th centile (13.0 vs 29.5 days, p<0.0001). 466 Our finding that placental histological classification of maternal vascular malperfusion was 467 significantly associated with lower maternal PlGF concentration and higher mean uterine artery PI at 468 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 27 27/01/2023 diagnosis of early-onset FGR, but not with UmA Doppler measurements, was in keeping with the 469

of previous studies (56-59). Agrawal et al. have recently reported that MVM, unlike other 470 placental pathologies, is characterised by raised mean uterine artery PI and a gradual decline in PlGF 471 as the pregnancy progresses (57). Triunfo et al. found in SGA pregnancies (EFW <10th) identified 472 between 30 and 34 weeks of gestation, a pattern of placental histopathology they termed ‘placental 473 underperfusion’, was most strongly associated with lower PlGF, measured at the time of diagnosis 474 (58). Benton et al. also found low PlGF to be a better predictor of placental pathology than umbilical 475 artery resistance index or abdominal circumference centile (56). 476 477 Strengths and limitations 478 The strengths of this multicentre study are that it was carried out prospectively in academic health 479 science centres with fetal medicine experts trained on a common ultrasound protocol, and level 3 480 perinatal care available for delivery. Participants and their fetuses/neonates were extensively 481 phenotyped at study entry, for the duration of the pregnancy and postnatally, and we report 482 temporally validated results. All pregnancies were managed according to local guidelines, although 483 these were broadly consistent, in line with national and international guidelines (4, 16, 60, 61)and 484 current RCT evidence (e.g. the TRUFFLE trial). This introduces variation, but potentially better 485 reflects real-world practice and hence adds external validity. All serum analysis was carried out after 486 pregnancy outcomes were obtained, using a proteomic discovery science approach which did not 487 assume associations with outcome, but also included additional analysis of proteins anticipated to 488 be related to pregnancy outcome in placental insufficiency. Placental histological classification was 489 blinded to pregnancy outcomes and included control and non-FGR preterm placental samples, to 490 remove some potential bias. Finally, our inclusion of stakeholders to guide model selection means 491 that their predictive value is most important to patients and clinicians. 492 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 28 27/01/2023 Our relatively narrow inclusion criteria are both a strength and a limitation, in that they have 493 allowed us to focus on a specific clinical group but have limited our sample size and the 494 generalisability of our findings. The sample size means our study was underpowered to demonstrate 495 small or medium effects and our estimates have wider confidence intervals than larger studies (62). 496 The exclusion of pregnancies 600g limits the number of pregnancies from 24+6 497 weeks of gestation to which our findings can be applied and the exclusion of pregnancies with 498 known genetic, chromosomal, and structural differences means our findings cannot be applied to 499 the whole spectrum of FGR. Generalisability is also limited to healthcare settings with comparable 500 neonatal care provision and outcomes, given their impact on neonatal survival and decision making 501 for iatrogenic preterm delivery. Clinicians managing the pregnancies were not blinded to ultrasound 502 measurements, and indeed many management decisions will have been influenced by ultrasound 503 findings. This could have biased the apparent associations between ultrasound variables and 504 pregnancy outcomes, either artificially strengthening or weakening them. 505 506 Future directions 507 Ideally, our findings should be independently and externally validated. Given the incidence of FGR 508 ≤28+0 weeks this would require another multicentre study. Further research is also needed to 509 determine whether the use of these models would have benefit in practice, both on the 510 psychological wellbeing of parents and on the use of health resources. Future studies to identify and 511 validate predictive models in early-onset SGA (EFW <10th centile <32+0 weeks of gestation) would 512 allow application to a wider population. This would complement the work currently being done in 513 the PLANES study, which is investigating the impact of revealed PlGF in SGA from 32+0 weeks (63). 514 Finally, our primary outcome of fetal or neonatal death provides only short-term information, and 515 data collection for 2-year neurodevelopmental outcomes is on-going. 516 517 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 29 27/01/2023

518 In conclusion, our study provides validated models for predicting fetal or neonatal death and fetal 519 death or delivery ≤28+0 weeks of gestation based on ultrasound and maternal serum protein 520 measurements at the time of diagnosis of severe early-onset FGR. The EFW z-score and umbilical 521 artery Doppler velocimetry were the best performing ultrasound parameters, but are vulnerable to 522 inter-rater variability, variation in formulae and reference ranges and temporal variation. The 523 biomarker PlGF was the best performing maternal serum protein for predicting both pregnancy 524 outcomes and maternal vascular malperfusion. This identification of a specific pathological 525 phenotype may be useful for targeting future potential therapies. 526 527 528 529 530 531 532 533 534 535 536

537 This study is reported according to the ‘strengthening the reporting of observational studies in 538 epidemiology’ (STROBE) guidelines (64) for cohort studies and the ‘transparent reporting of a 539 multivariable prediction model for individual prognosis or diagnosis’ (TRIPOD) guidelines (65). 540 Study design and setting: The EVERREST Prospective Study was a multicentre prospective cohort 541 study recruiting pregnant women from four European tertiary referral centres: University College 542 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 30 27/01/2023 London Hospital, UK; University Medical Centre Hamburg-Eppendorf, Germany; Maternal-Fetal Unit 543 Hospital Clinic Barcelona, Spain; Skane University Hospital, Lund, Sweden. 544 Study population: Full details of the protocol have been published previously (26). In brief, pregnant 545 women were eligible if they had a singleton fetus with an ultrasound estimated fetal weight (EFW) 546 <600g and <3rd centile according to local criteria between 20+0 and 26+6 weeks of gestation. 547 Exclusion criteria were a known abnormal karyotype or major fetal structural abnormality at 548 enrolment (66), indication for immediate delivery, preterm rupture of membranes before 549 enrolment, maternal HIV or hepatitis B or C infection, maternal age under 18 years, any medical or 550 psychiatric condition which compromised a woman’s ability to participate and lack of capacity to 551 consent. Pregnancies with a known congenital infection were not recruited and for the purposes of 552 this analysis pregnancies ending in termination were excluded. 553 Outcomes: The primary outcome was fetal or neonatal death (≤28 days of life). Secondary outcomes 554 were: fetal death or delivery ≤28+0 weeks of gestation; slow fetal growth, defined as a worsening of 555 weight deviation of ≥10 percentage points over a two-week interval (including before and after 556 enrolment) or equivalent trajectory over a longer period (28); and the development of abnormal 557 UmA Dopplers, defined as development of UmA PI >95th centile in pregnancies where UmA PI was 558 ≤95th centile at enrolment (27). Ascertainment for outcomes of this study was possible, at the latest, 559 by 29 days of life. Follow-up for neonatal morbidity and infant health and neurodevelopment to the 560 age of 2 years continues. 561 All pregnancies were managed according to the local fetal medicine unit protocols. Pre-eclampsia 562 was defined according to International Society for the Study of Hypertension in Pregnancy (ISSHP) 563 criteria (67), meaning that, given the presence of FGR, any woman developing new onset 564 hypertension after 20+0 weeks of gestation was classified as having pre-eclampsia rather than 565 pregnancy-induced hypertension. Formalin-fixed placental samples were classified according to 566 Amsterdam consensus criteria by a single assessor (NS) (68). To minimise bias, study placental 567 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 31 27/01/2023 samples were mixed with placental samples from healthy term pregnancies and pregnancies 568 delivering spontaneously preterm, with NS blinded to pregnancy phenotype and outcome during the 569 assessment. 570 Ultrasound measurements: All ultrasound examinations were performed by staff trained and 571 validated to the common EVERREST prospective study protocol (26). At each ultrasound scan 572 Doppler velocimetry of the umbilical artery (UmA), uterine artery (UtA), middle cerebral artery 573 (MCA), ductus venosus (DV) and umbilical vein was performed (69). Local EFW formulae and centile 574 charts were used to determine study eligibility, but for consistency all EFWs were recalculated using 575 the Hadlock 3 formula (incorporating head circumference, abdominal circumference and femur 576 length), with z-scores recalculated using the Marsal chart for descriptive data (supplementary data 577 Equations 1-3) (29, 30). EFWs and z-scores were also recalculated using Intergrowth formulae for 578 analysis (supplementary data Equations 4 & 5) (31). The effect of alternative Doppler reference 579 charts was explored, with similar results to those presented (70-74). 580 Blinding: Maternal serum protein concentrations were not available to clinicians, participants or 581 researchers during the pregnancy, as all samples were analysed after complete primary outcome 582 data had been ascertained. Serum PlGF and sflt-1 concentrations were not used as part of clinical 583 care in any of the study centres during the period of recruitment. 584 Sample collection: Maternal blood was collected at study enrolment in BD Vacutainer® serum 585 separating tubes and processed according to the manufacturer’s instructions. 500µl serum aliquots 586 were frozen and stored at -80°C. Placental samples, for Amsterdam criteria categorisation, were 587 collected from two areas of each placenta, midway between the cord insertion and margin in areas 588 free from macroscopic infarcts or lesions. Samples were rinsed in PBS, formalin-fixed, wax-589 embedded, sectioned and stained with H&E. 590 Measurement of a priori candidate biomarkers in maternal serum: PlGF and sflt-1 concentrations 591 were measured using Elecsys® electrochemiluminescence immunoassays on a Cobas® e411 analyser 592 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 32 27/01/2023 (Roche Diagnostics). The normalised protein expression (NPX) of 90 additional proteins associated 593 with cardiovascular disease was measured using the Olink® Cardiovascular II proximity extension 594 assay (full list of proteins in supplementary data Table 17). In the discovery set but not the validation 595 set VEGFA, VEGFD, VEGFR2, neuropilin 1 and endoglin were measured in triplicate using Quantikine® 596 colorimetric sandwich ELISAs (R&D Systems). 597 Identification of novel candidate biomarkers in maternal serum using liquid chromatography and 598 tandem mass spectrometry: Five pooled serum samples were created on the following basis: (1) 599 pregnancies ending in fetal or neonatal death (2) pregnancies ending in neonatal survival with 600 delivery <37+0 weeks of gestation (3) pregnancies ending in neonatal survival with delivery 37+0 601 weeks of gestation or more (4) slow fetal growth trajectory (5) normal fetal growth trajectory. 602 Pooled serum samples were depleted of 12 high-abundance proteins using Proteome PurifyTM 12 603 resin, as per the manufacturer’s instructions (R&D Systems), concentrated using Vivaspin© 500 5kDa 604 Molecular Weight Cut-Off columns (GE Healthcare), reduced with 10mM tris(2-605 carboxyethyl)phosphine hydrochloride then alkylated with 7.5mM iodoacetamide. Pooled samples 606 were digested using a trypsin/Lys-C mix, labelled with Tandem Mass TagsTM (Thermo Fisher 607 Scientific) and combined(75). The combined sample underwent two-dimensional high-performance 608 reverse-phase liquid chromatography and tandem mass spectrometry. In the first dimension, 609 samples were fractionated into 30 at high pH using a Poroshell 300 Extend C18 column (Agilent), 610 following which fractions 1-4 were combined with fractions 27-30 respectively due to low 611 abundance in the first four fractions. The second fractionation was performed on the Ultimate 3000 612 nano-liquid chromatography system using AcclaimTM PepMapTM 100 C18 pre-columns and AcclaimTM 613 PepMapTM 100 C18 Nano-LC columns run in tandem with analysis on the LTQ (linear trap 614 quadrupole) Orbitrap XLTM 2.5.5 (all Thermo Fisher Scientific). A blank calibration sample was run 615 after every three fractions, and a standard sample of known mass run after every six fractions, for 616 quality control. 617 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 33 27/01/2023 Proteins were identified using Proteome Discover V1.4 software (ThermoFisher Scientific) to search 618 the human Swiss-Prot database with the Mascot search engine (Matrix Science Ltd.). Proteins were 619 scored on variability, peptide count, ubiquity, ratio between pools and consistent trend across pools 620 (supplementary data Tables 18 & 19). Expression pattern clusters, based on standardised and raw 621 quantification ratios, were generated using the Graphical Proteomics Data Explorer (GPRoX) 622 platform. Based on their scores and expression clusters, five candidate proteins were selected and 623 measured in individual samples using ELISAs. Fibronectin, PSG1 (both R&D Systems) and HPL (DRG 624 International) were measured in the discovery and validation sets while SAA (R&D Systems) and 625 LNPEP (Cloud-clone) were measured in the discovery set only. 626 Priority survey and model selection: An online survey was sent to patients and clinicians asking their 627 opinion on the importance of different pregnancy outcomes and, for each outcome, whether they 628 would prioritise sensitivity or specificity. Models were selected on the basis of the survey results and 629 the model performance metrics described below. Protein models were published online prior to the 630 validation data analysis. 631 Sample size: Since this work involved discovery of novel biomarkers, a formal a priori sample size 632 calculation was not possible. Before analysing the discovery set it was determined that this sample 633 of n=63 with 21 fetal or neonatal deaths gave an 80% power to detect a standardised effect size of 634 0.9 (large) to a significance level of 0.05 (76). 635 Statistics: Data analysis was performed using STATA/MP 16.1 software (StataCorp LLC, College 636 Station, USA) unless otherwise specified. Descriptive and investigative variables were tested for 637 skew and kurtosis (77, 78) and handled as symmetrical if there was no evidence of either. PlGF, sflt1, 638 endoglin, VEGFD, NP1, HPL, SAA and LNPEP were transformed to their natural logs and multiplex 639 data was analysed as provided, on a log2 scale. Characteristics of the discovery and validation sets 640 were compared using chi-square tests (categorical data), Fisher’s exact tests (binary data with sparse 641 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 34 27/01/2023 outcomes), 2-sided t tests (symmetrical continuous data) and Mann-Whitney U tests (skewed 642 continuous data). 643 Missing data for BMI (n=5) were imputed using chain equations. Umbilical artery (UmA) PI at 644 enrolment was systematically missing (n=16), with most missing cases having absent or reversed 645 end-diastolic flow (EDF, n=15). Umbilical artery Doppler velocimetry was therefore handled as an 646 interval variable, ‘UmA PI category’, , where 0=UmA PI ≤95th centile, 1=UmA >95th centile with 647 positive EDF, 2=absent EDF, 3=reversed EDF. Where uterine artery (UtA) PI at enrolment was missing 648 (n=10) a mean UtA PI below or above the 95th centile could be inferred in nine cases where the 649 mean UtA PI was consistently normal (n=1) or abnormal (n=8) respectively at scans prior to and after 650 enrolment and UtA PI values were imputed using multiple imputation. Associations between 651 ultrasound measurements and both fetal or neonatal death and death or delivery ≤28+0 weeks of 652 gestation, were analysed using logistic regression. Univariate associations between protein 653 concentrations or NPX and outcomes were assessed using 2-sided t tests, Mann-Whitney U tests and 654 logistic regression, with Benjamini-Hochberg procedures to account for multiple comparisons. 655 Model development: Two-protein models for the development of abnormal UmA Dopplers and 656 two- and three-protein models for the other three pregnancy outcomes, with internal validation 657 using leave-one-out cross-validation (LOOCV), were compared based on AUC, specificity for 90% 658 sensitivity, sensitivity for 90% specificity, F1 score, Matthews correlation coefficient (MCC) and 659 precision-recall characteristics (PRROC) AUC. ROC curves were generated with the pROC R package 660 (version 1.18.0, https://cran.r-project.org/web/packages/pROC/index.html ). 95% confidence intervals 661 for AUCs were determined by stratified bootstrapping. PRROC curves were generated with the 662 MLmetrics R package (version 1.1.1, https://cran.r-project.org/web/packages/MLmetrics/index.html). 663 Models with variance inflation factors of five or more were excluded using the following R package: 664 https://cran.r-project.org/web/packages/car/index.html , version 3.1-0. Two-variable models 665 predicting fetal or neonatal death and death or delivery ≤28+0 weeks of gestation containing 666 ultrasound parameters with or without PlGF or HPL (as the proteins showing the strongest 667 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 35 27/01/2023 associations with these outcomes) were compared in the same way. Outcomes and protein models 668 to be validated were published on the study registry prior to analysis of the validation data. 669 Parenclitic network analysis: Parenclitic networks of the 102 proteins were generated for each of 670 the four pregnancy outcomes. For each outcome, two-dimensional kernel density estimations were 671 generated for every pair-combination of variables in ‘controls’ (pregnancies without the outcome). 672 Individual networks were then generated for each ‘case’, with linkages created if a pair-wise 673 relationship of variables differed from the control distribution by more than a given threshold(79). 674 These individual case networks were then combined. VEGF-A, BNP, PARP1 and melusin were 675 included as binary variables of ‘detectable’ or ‘not detectable’. Booking body mass index (BMI) and 676 fetal sex were included as variables in the networks, except for ‘development of abnormal UmA 677 Dopplers’, where the sample was not large enough to accommodate them. 678 Model validation: Concentrations of HPL and PSG1, as measured by ELISA, and NPX values for the 679 Olink multiplex proteins showed substantial variation in centrality and spread between the discovery 680 and validation sets. To account for this, values of each protein were centred to a mean of 0 and 681 scaled to an SD of 1 in the discovery set and validation set separately. These centred and scaled 682 values were used for subsequent analyses, including model validation. Concentrations of PlGF, sflt1 683 and fibronectin did not require transformation. 684 Models generated from the discovery set were run on data from the validation set and were 685 considered validated if the 95% CI for the validation estimate of AUC included the LOOCV AUC 686 estimate from the discovery set. For validated models, data from both sets were combined to give 687 final test characteristics. Likelihood ratio tests were used to determine whether the addition of 688 pregnancy characteristics (maternal BMI, maternal age, maternal ethnicity, fetal sex, gestational age 689 at enrolment and pre-eclampsia at enrolment) significantly improved the validated models. Model 690 calibration was assessed by plotting predicted probability against observed frequency of outcome. 691 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 36 27/01/2023 Functional interactions: Centred and scaled data from the discovery and validation sets were 692 combined to retest univariate associations with the primary and secondary outcomes. Proteins 693 showing a significant association at a 5% Benjamini-Hochberg false discovery rate were explored for 694 physical and functional interactions and for enrichment of GO biological processes, relative to the 695

of all proteins measured, using STRING (Swiss Institute of Bioinformatics)(32). Where no 696 enrichment was detected shared GO biological processes were identified through comparison to the 697 whole genome. 698 Modelling pregnancy duration: Protein and ultrasound measurements from the combined discovery 699 and validation sets were tested for their association with gestational age at livebirth or diagnosis of 700 fetal death and interval from enrolment to livebirth or diagnosis of fetal death using linear 701 regression. Variables showing a significant association at a 1% Benjamini-Hochberg false discovery 702 rate were used to create linear models predicting these outcomes in a stepwise fashion. Maternal 703 age, BMI, ethnicity, gestational age at enrolment, pre-eclampsia at enrolment and fetal sex were 704 tested for model improvement. Model fit was tested by assessing variance inflation factors for 705 multicollinearity, assessing the distribution of the residuals for heteroscedasticity and outliers, and 706 looking for observations with high leverage. UmA and UtA Doppler velocimetry, PlGF concentration, 707 HPL concentration and PAPP-A NPX were tested for their associations with placental histological 708 classification using logistic regression. 709 Study approval: Ethical approval was provided by the National Research Ethics Service Committee 710 London - Stanmore in the UK (REC reference: 13/LO/1254), the Hospital Clinic of Barcelona’s Clinical 711 Research Ethics Committee in Spain (Reg: HCB/2014/0091), the Regional Ethical Review Board in 712 Lund for Sweden, (DNr 2014/147) and the Ethics Committee of Hamburg Board of Physicians in 713 Germany (PV4809). This study was conducted according to the Declaration of Helsinki principles and 714 written informed consent was given by all participants before enrolment. 715 716 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 37 27/01/2023

717 The research leading to these results has received funding from the European Union Seventh 718 Framework Programme (FP7/2007-2013) under grant agreement no. 305823, the Rosetrees Trust 719 and the Mitchell Charitable Trust in memory of Shoshana Mitchell Glynn. This research has been 720 supported by the National Institute for Health Research University College London Hospitals 721 Biomedical Research Centre (RS, NM, ALD). NRN would like to thank the support from Cancer 722 Research UK (C12077/A26223). 723 724 This work would not have been possible without the contribution of the late Professor John Timms, 725 Elizabeth Garrett Anderson Institute for Women’s Health, University College London. His 726 contribution to the design, conduct and analysis of the study was invaluable, and we would have 727 been honoured to have him as a co-author. We would also like to thank all of the women and 728 families who took part in this study as well as staff at the UCL Comprehensive Clinical Trials Unit 729 (Anna Morka, Jade Dyer, Helen Knowles, Steve Hibbert, Kate Maclagan), Gina Buquis, Jade Okell, Dr 730 Mark Lees, Dr Carlo Rossi, Dr Tara Krishnan, Dr Roberta Morris, Dr Sarah Guillon, Richard Gunu, Dr 731 Eva Sedlak, Professor Alexei Zaikin and Dr Oleg Blyuss. 732 Data access statement: The full data set will not be made publicly available because the degree of 733 detailed phenotyping could allow individual patient identification. Limited data sharing may be 734 possible, with the agreement of the consortium, on request to RS or ALD. 735

736 1. Zur RL, Kingdom JC, Parks WT, Hobson SR. The Placental Basis of Fetal Growth Restriction. 737 Obstetrics and gynecology clinics of North America. 2020;47(1):81-98. 738 2. Sun C, Groom KM, Oyston C, Chamley LW, Clark AR, James JL. The placenta in fetal growth 739 restriction: What is going wrong? Placenta. 2020;96:10-8. 740 3. Gordijn SJ, Beune IM, Thilaganathan B, Papageorghiou A, Baschat AA, Baker PN, et al. 741 Consensus definition for placental fetal growth restriction: a Delphi procedure. Ultrasound Obstet 742 Gynecol. 2016;48(3):333-9. 743 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 38 27/01/2023 4. Lees CC, Stampalija T, Baschat A, da Silva Costa F, Ferrazzi E, Figueras F, et al. ISUOG Practice 744 Guidelines: diagnosis and management of small-for-gestational-age fetus and fetal growth 745 restriction. Ultrasound Obstet Gynecol. 2020;56(2):298-312. 746 5. Society for Maternal-Fetal Medicine, Martins JG, Biggio JR, Abuhamad A. Society for 747 Maternal-Fetal Medicine Consult Series #52: Diagnosis and management of fetal growth restriction. 748 Am J Obstet Gynecol. 2020;223(4):B2-B17. 749 6. Ganzevoort W, Thornton JG, Marlow N, Thilaganathan B, Arabin B, Prefumo F, et al. 750 Comparative analysis of 2-year outcomes in GRIT and TRUFFLE trials. Ultrasound Obstet Gynecol. 751 2020;55(1):68-74. 752 7. Sharp A, Cornforth C, Jackson R, Harrold J, Turner MA, Kenny LC, et al. Maternal sildenafil for 753 severe fetal growth restriction (STRIDER): a multicentre, randomised, placebo-controlled, double-754 blind trial. Lancet Child Adolesc Health. 2018;2(2):93-102. 755 8. Villar J, Restrepo-Méndez MC, McGready R, Barros FC, Victora CG, Munim S, et al. 756 Association Between Preterm-Birth Phenotypes and Differential Morbidity, Growth, and 757 Neurodevelopment at Age 2 Years: Results From the INTERBIO-21st Newborn Study. JAMA Pediatr. 758 2021;175(5):483-93. 759 9. Pels A, Beune IM, van Wassenaer-Leemhuis AG, Limpens J, Ganzevoort W. Early-onset fetal 760 growth restriction: A systematic review on mortality and morbidity. Acta Obstet Gynecol Scand. 761 2020;99(2):153-66. 762 10. Rock CR, White TA, Piscopo BR, Sutherland AE, Miller SL, Camm EJ, et al. Cardiovascular and 763 Cerebrovascular Implications of Growth Restriction: Mechanisms and Potential Treatments. 764 International journal of molecular sciences. 2021;22(14). 765 11. Arigliani M, Stocco C, Valentini E, De Pieri C, Castriotta L, Ferrari ME, et al. Lung function 766 between 8 and 15 years of age in very preterm infants with fetal growth restriction. Pediatr Res. 767 2021;90(3):657-63. 768 12. Mappa I, Maqina P, Bitsadze V, Khizroeva J, Makatsarya A, Arduini D, et al. Cardiac function 769 in fetal growth restriction. Minerva Obstet Gynecol. 2021;73(4):423-34. 770 13. Fernandez-Twinn DS, Hjort L, Novakovic B, Ozanne SE, Saffery R. Intrauterine programming 771 of obesity and type 2 diabetes. Diabetologia. 2019;62(10):1789-801. 772 14. Melamed N, Baschat A, Yinon Y, Athanasiadis A, Mecacci F, Figueras F, et al. FIGO 773 (international Federation of Gynecology and obstetrics) initiative on fetal growth: best practice 774 advice for screening, diagnosis, and management of fetal growth restriction. Int JGynaecol Obstet. 775 2021;152 Suppl 1:3-57. 776 15. Groom KM, David AL. The role of aspirin, heparin, and other interventions in the prevention 777 and treatment of fetal growth restriction. Am J Obstet Gynecol. 2018;218(2S):S829-S40. 778 16. Royal College of Obstetricians and Gynaecologists. Green-top Guideline No. 31: The 779 Investigation and Management of the Small-for-Gestation-Age Fetus. London: RCOG; 2014. 780 17. Ormesher L, Warrander L, Liu Y, Thomas S, Simcox L, Smith GCS, et al. Risk stratification for 781 early-onset fetal growth restriction in women with abnormal serum biomarkers: a retrospective 782 cohort study. Scientific reports. 2020;10(1):22259. 783 18. Sovio U, Goulding N, McBride N, Cook E, Gaccioli F, Charnock-Jones DS, et al. A maternal 784 serum metabolite ratio predicts fetal growth restriction at term. Nat Med. 2020;26(3):348-53. 785 19. Tong S, Joy Kaitu'u-Lino T, Walker SP, MacDonald TM. Blood-based biomarkers in the 786 maternal circulation associated with fetal growth restriction. Prenat Diagn. 2019;39(11):947-57. 787 20. Romero Infante XC, Uriel M, Porras Ramírez A, Rincón Franco S. Comparison of preeclampsia 788 and fetal growth restriction screenings at first trimester in a high-risk population. J Obstet Gynaecol 789 Res. 2021;47(2):765-73. 790 21. Asvold BO, Vatten LJ, Romundstad PR, Jenum PA, Karumanchi SA, Eskild A. Angiogenic 791 factors in maternal circulation and the risk of severe fetal growth restriction. Am J Epidemiol. 792 2011;173(6):630-9. 793 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 39 27/01/2023 22. Cnossen JS, Morris RK, ter Riet G, Mol BW, van der Post JA, Coomarasamy A, et al. Use of 794 uterine artery Doppler ultrasonography to predict pre-eclampsia and intrauterine growth restriction: 795 a systematic review and bivariable meta-analysis. CMAJ. 2008;178(6):701-11. 796 23. Harvey ME, David AL, Dyer J, Spencer R. Pregnant women's experiences and perceptions of 797 participating in the EVERREST prospective study; a qualitative study. BMC Pregnancy Childbirth. 798 2019;19(1):144. 799 24. Blakeley C, Smith DM, Johnstone ED, Wittkowski A. Women's lived experiences of a prenatal 800 diagnosis of fetal growth restriction at the limits of viability: An interpretative phenomenological 801 study. Midwifery. 2019;76:110-7. 802 25. David AL. Maternal uterine artery VEGF gene therapy for treatment of intrauterine growth 803 restriction. Placenta. 2017;59 Suppl 1:S44-S50. 804 26. Spencer R, Ambler G, Brodszki J, Diemert A, Figueras F, Gratacos E, et al. EVERREST 805 prospective study: a 6-year prospective study to define the clinical and biological characteristics of 806 pregnancies affected by severe early onset fetal growth restriction. BMC Pregnancy Childbirth. 807 2017;17(1):43. 808 27. Schaffer AC, Staudach A. Doppler-Referenzkurven. 1997. 809 28. Marsal K. Obstetric management of intrauterine growth restriction. Best Pract Res Clin 810 Obstet Gynaecol. 2009;23:857-70. 811 29. Hadlock FP, Harrist RB, Sharman RS, Deter RL, Park SK. Estimation of fetal weight with the 812 use of head, body, and femur measurements--a prospective study. Am J Obstet Gynecol. 813 1985;151(3):333-7. 814 30. Marsal K, Persson PH, Larsen T, Lilja H, Selbing A, Sultan B. Intrauterine growth curves based 815 on ultrasonically estimated foetal weights. Acta Paediatr. 1996;85(7):843-8. 816 31. Stirnemann J, Villar J, Salomon LJ, Ohuma E, Ruyan P, Altman DG, et al. International 817 estimated fetal weight standards of the INTERGROWTH-21st Project. Ultrasound Obstet Gynecol. 818 2017;49(4):478-86. 819 32. Szklarczyk D, Gable AL, Nastou KC, Lyon D, Kirsch R, Pyysalo S, et al. The STRING database in 820 2021: customizable protein-protein networks, and functional characterization of user-uploaded 821 gene/measurement sets. Nucleic Acids Res. 2021;49(D1):D605-D12. 822 33. Mahajan D, Sharma NR, Kancharla S, Kolli P, Tripathy A, Sharma AK, et al. Role of Natural 823 Killer Cells during Pregnancy and Related Complications. Biomolecules. 2022;12(1). 824 34. Thomas JR, Appios A, Zhao X, Dutkiewicz R, Donde M, Lee CYC, et al. Phenotypic and 825 functional characterization of first-trimester human placental macrophages, Hofbauer cells. J Exp 826 Med. 2021;218(1). 827 35. Raguema N, Moustadraf S, Bertagnolli M. Immune and Apoptosis Mechanisms Regulating 828 Placental Development and Vascularization in Preeclampsia. Front Physiol. 2020;11:98. 829 36. Schrimpf C, Teebken OE, Wilhelmi M, Duffield JS. The role of pericyte detachment in vascular 830 rarefaction. J Vasc Res. 2014;51(4):247-58. 831 37. Alfirevic Z, Stampalija T, Gyte GM. Fetal and umbilical Doppler ultrasound in high-risk 832 pregnancies. Cochrane Database Syst Rev. 2010(1):CD007529. 833 38. Morris RK, Malin G, Robson SC, Kleijnen J, Zamora J, Khan KS. Fetal umbilical artery Doppler 834 to predict compromise of fetal/neonatal wellbeing in a high-risk population: systematic review and 835 bivariate meta-analysis. Ultrasound Obstet Gynecol. 2011;37(2):135-42. 836 39. Baião AER, de Carvalho PRN, Moreira MEL, de Sá RAM, Junior SCG. Predictors of perinatal 837 outcome in early-onset fetal growth restriction: A study from an emerging economy country. Prenat 838 Diagn. 2020;40(3):373-9. 839 40. Sharp A, Jackson R, Cornforth C, Harrold J, Turner MA, Kenny L, et al. A prediction model for 840 short-term neonatal outcomes in severe early-onset fetal growth restriction. Eur J Obstet Gynecol 841 Reprod Biol. 2019;241:109-18. 842 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 40 27/01/2023 41. Baschat AA, Gembruch U, Reiss I, Gortner L, Weiner CP, Harman CR. Relationship between 843 arterial and venous Doppler and perinatal outcome in fetal growth restriction. Ultrasound Obstet 844 Gynecol. 2000;16(5):407-13. 845 42. Bilardo CM, Wolf H, Stigter RH, Ville Y, Baez E, Visser GH, et al. Relationship between 846 monitoring parameters and perinatal outcome in severe, early intrauterine growth restriction. 847 Ultrasound Obstet Gynecol. 2004;23(2):119-25. 848 43. The investigation and management of the small-for gestational age fetus. Green–top 849 Guideline No 31. London: Royal College of Obstetricians and Gynaecologists; 2013. 850 44. Thomas RL, Peng TC, Eglinton GS, Strobino DM, Johnson TR. Precision of umbilical artery 851 Doppler studies. Intraobserver, interobserver, and biologic variability of fetal Doppler velocimetry. J 852 Ultrasound Med. 1991;10(4):201-4. 853 45. Bruin CM, Ganzevoort W, Schuit E, Mensing van Charante NA, Wolf H. Inter- and intra-854 observer variability in fetal ductus venosus blood flow measurements in high-risk fetuses at 26-32 855 weeks. Eur J Obstet Gynecol Reprod Biol. 2019;243:67-71. 856 46. Pasquini L, Marchi L, Gaini C, Franchi C, Mecacci F, Bilardo CM. Intra- and Interobserver 857 Reproducibility of Third Trimester Middle Cerebral Artery Pulsatility Index Measurement: A 858 Prospective Cross-Sectional Study. Fetal Diagn Ther. 2020;47(3):214-9. 859 47. Steller JG, Driver C, Gumina D, Peek E, Harper T, Hobbins JC, et al. Doppler velocimetry 860 discordance between paired umbilical artery vessels and clinical implications in fetal growth 861 restriction. Am J Obstet Gynecol. 2022;227(2):285 e1- e7. 862 48. Oros D, Ruiz-Martinez S, Staines-Urias E, Conde-Agudelo A, Villar J, Fabre E, et al. Reference 863 ranges for Doppler indices of umbilical and fetal middle cerebral arteries and cerebroplacental ratio: 864 systematic review. Ultrasound Obstet Gynecol. 2019;53(4):454-64. 865 49. Sarris I, Ioannou C, Chamberlain P, Ohuma E, Roseman F, Hoch L, et al. Intra- and 866 interobserver variability in fetal ultrasound measurements. Ultrasound Obstet Gynecol. 867 2012;39(3):266-73. 868 50. Anderson NG, Jolley IJ, Wells JE. Sonographic estimation of fetal weight: comparison of bias, 869 precision and consistency using 12 different formulae. Ultrasound Obstet Gynecol. 2007;30(2):173-9. 870 51. Choi SKY, Gordon A, Hilder L, Henry A, Hyett JA, Brew BK, et al. Performance of six birth-871 weight and estimated-fetal-weight standards for predicting adverse perinatal outcome: a 10-year 872 nationwide population-based study. Ultrasound Obstet Gynecol. 2021;58(2):264-77. 873 52. Warrander LK, Ingram E, Heazell AEP, Johnstone ED. Evaluating the accuracy and precision of 874 sonographic fetal weight estimation models in extremely early-onset fetal growth restriction. Acta 875 Obstet Gynecol Scand. 2020;99(3):364-73. 876 53. Mendoza M, Hurtado I, Bonacina E, Garcia-Manau P, Serrano B, Tur H, et al. Individual risk 877 assessment for prenatal counseling in early-onset growth-restricted and small-for-gestational-age 878 fetuses. Acta Obstet Gynecol Scand. 2021;100(3):504-12. 879 54. Quezada MS, Rodríguez-Calvo J, Villalaín C, Gómez-Arriaga PI, Galindo A, Herraiz I. sFlt-880 1/PlGF ratio and timing of delivery in early-onset fetal growth restriction with antegrade umbilical 881 artery flow. Ultrasound Obstet Gynecol. 2020;56(4):549-56. 882 55. Shinar S, Tigert M, Agrawal S, Parks WA, Kingdom JC. Placental growth factor as a diagnostic 883 tool for placental mediated fetal growth restriction. Pregnancy hypertension. 2021;25:123-8. 884 56. Benton SJ, McCowan LM, Heazell AE, Grynspan D, Hutcheon JA, Senger C, et al. Placental 885 growth factor as a marker of fetal growth restriction caused by placental dysfunction. Placenta. 886 2016;42:1-8. 887 57. Agrawal S, Parks WT, Zeng HD, Ravichandran A, Ashwal E, Windrim RC, et al. Diagnostic 888 utility of serial circulating placental growth factor levels and uterine artery Doppler waveforms in 889 diagnosing underlying placental diseases in pregnancies at high risk of placental dysfunction. Am J 890 Obstet Gynecol. 2022. 891 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 41 27/01/2023 58. Triunfo S, Lobmaier S, Parra-Saavedra M, Crovetto F, Peguero A, Nadal A, et al. Angiogenic 892 factors at diagnosis of late-onset small-for-gestational age and histological placental underperfusion. 893 Placenta. 2014;35(6):398-403. 894 59. Schiffer V, Borghans CWJ, Arts N, Bons JAP, Severens-Rijvers CAH, van Kuijk SMJ, et al. The 895 association between first trimester placental biomarkers and placental lesions of maternal vascular 896 malperfusion. Placenta. 2021;103:206-13. 897 60. Lausman A, Kingdom J, Gagnon R, Basso M, Bos H, Crane J, et al. Intrauterine growth 898 restriction: screening, diagnosis, and management. J Obstet Gynaecol Can. 2013;35(8):741-57. 899 61. Perinatal Society of Australia and New Zealand and Centre of Research Excellence Stillbirth. 900 Position statement: detection and management of fetal growth restriction in singleton pregnancies. 901 Brisbane, Australia: Centre of Research Excellence in Stillbirth.; 2019. 902 62. Zeisler H, Llurba E, Chantraine F, Vatish M, Staff AC, Sennström M, et al. Predictive Value of 903 the sFlt-1:PlGF Ratio in Women with Suspected Preeclampsia. N Engl J Med. 2016;374(1):13-22. 904 63. Gent J, Bullough S, Harrold J, Jackson R, Woolfall K, Andronis L, et al. The PLANES study: a 905 protocol for a randomised controlled feasibility study of the placental growth factor (PlGF) blood 906 test-informed care versus standard care alone for women with a small for gestational age fetus at or 907 after 32 + 0 weeks' gestation. Pilot Feasibility Stud. 2020;6(1):179. 908 64. Vandenbroucke JP, von Elm E, Altman DG, Gotzsche PC, Mulrow CD, Pocock SJ, et al. 909 Strengthening the Reporting of Observational Studies in Epidemiology (STROBE): explanation and 910 elaboration. PLoS medicine. 2007;4(10):e297. 911 65. Collins GS, Reitsma JB, Altman DG, Moons KG. Transparent reporting of a multivariable 912 prediction model for individual prognosis or diagnosis (TRIPOD): the TRIPOD statement. BJOG. 913 2015;122(3):434-43. 914 66. Carvalho MHB, Brizot ML, Lopes LM, Chiba CH, Miyadahira S, Zugaib M. Detection of fetal 915 structural abnormalities at the 11–14 week ultrasound scan. Prenat Diagn. 2002;22(1):1-4. 916 67. Brown MA, Magee LA, Kenny LC, Karumanchi SA, McCarthy FP, Saito S, et al. The 917 hypertensive disorders of pregnancy: ISSHP classification, diagnosis & management 918 recommendations for international practice. Pregnancy hypertension. 2018;13:291-310. 919 68. Khong TY, Mooney EE, Ariel I, Balmus NC, Boyd TK, Brundler MA, et al. Sampling and 920 Definitions of Placental Lesions: Amsterdam Placental Workshop Group Consensus Statement. Arch 921 Pathol Lab Med. 2016;140(7):698-713. 922 69. Bhide A, Acharya G, Bilardo CM, Brezinka C, Cafici D, Hernandez-Andrade E, et al. ISUOG 923 practice guidelines: use of Doppler ultrasonography in obstetrics. Ultrasound Obstet Gynecol. 924 2013;41:233-9. 925 70. Arduini D, Rizzo G. Normal values of Pulsatility Index from fetal vessels: a cross-sectional 926 study on 1556 healthy fetuses. J Perinat Med. 1990;18(3):165-72. 927 71. Gómez O, Figueras F, Fernández S, Bennasar M, Martínez JM, Puerto B, et al. Reference 928 ranges for uterine artery mean pulsatility index at 11–41 weeks of gestation. Ultrasound Obstet 929 Gynecol. 2008;32(2):128-32. 930 72. Hofstaetter C, Dubiel M, Gudmundsson S, Marsal K. Uterine artery color Doppler assisted 931 velocimetry and perinatal outcome. Acta Obstet Gynecol Scand. 1996;75(7):612-9. 932 73. Gudmundsson S, Marsal K. Umbilical artery and uteroplacental blood flow velocity 933 waveforms in normal pregnancy--a cross-sectional study. Acta Obstet Gynecol Scand. 934 1988;67(4):347-54. 935 74. Mari G, Deter RL. Middle cerebral artery flow velocity waveforms in normal and small-for-936 gestational-age fetuses. Am J Obstet Gynecol. 1992;166(4):1262-70. 937 75. Whitwell HJ, Worthington J, Blyuss O, Gentry-Maharaj A, Ryan A, Gunu R, et al. Improved 938 early detection of ovarian cancer using longitudinal multimarker models. Br J Cancer. 939 2020;122(6):847-56. 940 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint IMPPICT_JCI_version_1 Page 42 27/01/2023 76. Cohen J. Statistical Power Analysis for Behavioural Sciences (2nd edition): Lawrence Erlbaum 941 Associates; 1988. 942 77. D’Agostino R, Belanger J, D’Agostino RJ. A suggestion for using powerful and informative 943 tests of normality. American Statistician. 1990;44:316-21. 944 78. Royston P. Comment on sg3.4 and an improved D’Agostino test. Stata Technical Bulletin. 945 1991;2:16-7. 946 79. Whitwell HJ, Blyuss O, Menon U, Timms JF, Zaikin A. Parenclitic networks for predicting 947 ovarian cancer. Oncotarget. 2018;9(32):22717-26. 948 949 All rights reserved. No reuse allowed without permission. perpetuity. preprint (which was not certified by peer review) is the author/funder, who has granted medRxiv a license to display the preprint in The copyright holder for thisthis version posted January 28, 2023. ; https://doi.org/10.1101/2023.01.27.23285087doi: medRxiv preprint