Assessing quality standards of biomarker measurement in first‐trimester combined screening for preeclampsia in the Chinese population: A multicenter study

Abstract

Introduction

The first‐trimester combined screening test for preterm preeclampsia (preterm‐PE) has been implemented in Mainland China. This study aims, first, to evaluate the applicability of multiple of the median (MoM) models based on biomarkers developed and utilized in a European population for prospective preterm‐PE screening in Mainland China; second, to assess the criteria for biomarker adjustment; and third, to establish criteria for continuous quality assurance in the screening process.

Material and Methods

In this prospective multicenter cross‐sectional study, 22 066 singleton pregnancies undergoing preterm‐PE screening at 11 ^+ 0–13 ^+ 6 weeks were recruited from six tertiary hospitals in Mainland China. Mean arterial pressure (MAP) and placental growth factor (PlGF) were measured and converted to their equivalent MoM values using Fetal Medicine Foundation MoMing models incorporated in commercial software. PlGF levels were also determined using immunoassays developed in China. MoM and log₁₀MoM distributions in the initial screened subjects were assessed for (1) central tendency and dispersion and (2) whether biomarker MoM distributions needed to be recentered. Biomarker central tendency and temporal stability both before and after recentering were assessed using Cumulative Sum plots. Distribution central tendency was considered good if its median MoM was between 0.95 and 1.05 and acceptable if between 1.05 and 1.10.

Results

After excluding the initial 400 pregnancies from each center, all PlGF, but not all MAP MoM distributions, required recentering. Post‐recentering, the overall distribution median (95% confidence interval) MoM values for MAP and PlGF were 1.003 (1.001, 1.004) and 1.004 (0.996, 1.013), respectively.

Conclusions

Chinese centers performing combined PE screening should assess biomarker MoM distributions in their initial cases to ensure that they conform to the expected central tendencies assumed in the risk estimation models.

In 22 066 first‐trimester pregnancies from six Chinese hospitals, biomarker measurements required recalibration to align with the established risk model. Continuous quality assurance is essential to ensure accurate preeclampsia risk assessment when implementing combined screening in new populations.

Abbreviations

MAP

mean arterial pressure

MoM

multiples of the median

PE

preeclampsia

PlGF

placental growth factor

Key message.

This study underscores the importance of quality assurance in preterm‐preeclampsia screening, highlighting that multiple of the median ‐adjusted maternal serum biomarker values may not always align with expected trends. It emphasizes the need for Cumulative Sum plots to detect biases and structural breakpoints, ensuring accurate PE risk estimation and successful global implementation of first‐trimester combined screening.

1. INTRODUCTION

Preeclampsia (PE) is a multisystemic disorder that affects 2%–5% of pregnancies globally, contributing to a major cause of maternal and perinatal morbidity and mortality, ¹ , ² , ³ , ⁴ , ⁵ particularly in developing countries, such as Mainland China. The reported incidence of PE in Mainland China is 2.4%. ⁶ , ⁷ Current professional and national guidelines in Mainland China for identifying women at risk of PE are mainly based on maternal and obstetric history. Women at high‐risk are offered daily aspirin prophylaxis (50–150 mg daily) from 12 and 16 weeks until 26–28 weeks to improve both immediate and long‐term maternal and neonatal morbidity and mortality. ⁷ While this approach is relatively simple to implement and would identify some women at risk, its screening performance has been demonstrated as being less effective in comparison to screening using a multi‐biomarker‐based risk models as demonstrated by the ASPRE trial and related studies. ⁸

Following the ASPRE trial, the Fetal Medicine Foundation (FMF, London, UK) risk models are increasingly used internationally to screen for preterm preeclampsia (preterm‐PE). ⁹ The FMF risk models, first described in 2015–2017 and subsequently updated, ¹⁰ are analogous to that used to screen for Down syndrome. An a priori risk based on maternal past and current obstetric history is adjusted according to levels of specific biomarkers, which differentiate between those who did and did not develop PE in study populations, after standardizing them to multiples of the expected median (multiple of the median [MoM]). ⁹ The FMF biomarker MoMing models used to adjust measured biomarker levels for independent fixed and random maternal and obstetric factors are provided to commercial companies for incorporation into their risk software on the assumption that they are universally applicable. ¹¹

However, our previous study, despite using the same immunoanalysers as the FMF and MoMing models derived at the FMF, indicated significant shifts in the central tendency of placental growth factor (PlGF) MoM in both East and South Asians, even though adjustment had been made for East Asian and South Asian ethnicity. ¹² The implementation of the FMF screening test for preterm‐PE in Mainland China is hindered, first, by the fact that Mainland Chinese companies have developed immunoanalysers capable of measuring PlGF but are not currently used by the FMF and therefore cannot be utilized without adapting the FMF MoMing models used in the commercial risk calculation software. Second, unlike the United Kingdom, the centers in Mainland China seeking to routinely offer the FMF screening test for preterm‐PE using three biomarkers require collaboration of three separate departments: the obstetricians to measure mean arterial pressure (MAP) and document history, the biochemists to measure PlGF, and last ultrasound providers to measure uterine artery Dopplers in order to comply with Mainland China medical practice regulations. ¹³ The efficacy of prospective screening is dependent on, first, the concordance between data collected at the time of screening with the risk models used to estimate individual patient risk, and second, whether a screening test using only two biomarkers could be implemented in Mainland China to simplify the screening process.

The objective of the current study was therefore to firstly determine whether the FMF MoMing models can be readily adapted for prospective preterm‐PE screening in Mainland China using locally produced immunoassays; second, assess criteria for biomarker adjustment; and third, establish criteria for continuous quality assurance (QA) to ensure long‐term screening efficacy.

2. MATERIAL AND METHODS

2.1. Study population

This study was performed as a QA assessment in the Aspirin versus metformin in pregnancies at high‐risk of preterm pre‐eclampsia in China (AVERT): a multicenter, double‐blind, 3‐arm randomized controlled trial (clinicaltrials.gov NCT05580523). ¹⁴ This is an ongoing study initiated in June 2023 in Mainland China to assess the effectiveness of different aspirin dosages with or without metformin in preventing PE in Chinese women with singleton pregnancies screened between 11 ^+ 0 and 13 ^+ 6weeks of gestation. Consenting participants were screened for preterm‐PE using a combination of past and current maternal obstetric history combined with two biomarkers, MAP and PlGF. ¹⁴ Maternal past and current medical and obstetric history were recorded at the time of screening.

Inclusion criteria for the study were age ≥18 years, singleton pregnancy, and a live fetus at the 11–13 ^+ 6 weeks of gestation. Women were excluded if they had major fetal abnormalities detected in the first trimester, were unable to provide full informed consent due to learning difficulties or severe mental illness, or experienced pregnancy termination or miscarriage before 24 weeks of gestation.

2.2. Determination of MAP measurements and MoMs

MAP was determined from repeated bilateral blood pressure (BP) measurements using automated devices (BP3AQ1; Microlife, Taipei, Taiwan) ¹¹ , ¹⁵ according to the published standardized measurement protocol. ¹⁶ MAP was converted to its equivalent MoM value and adjusted for fixed and random maternal and obstetric independent factors using the MAP MoMing model published by Tan et al. ⁹ constructed using a multiethnic population of women having singleton pregnancies screened at the FMF.

2.3. Determination of PlGF concentrations and MoMs

Maternal blood samples were collected in 3 or 5 mL Z‐serum clot vacuum tubes sourced by the individual hospital, which were manufactured in Mainland China. Blood draw was performed in accordance with local hospital practice using either iodine solution (iodophor, povidone‐iodine [PVP‐I]) or alcohol wipes to cleanse the blood draw site. Collected blood was left to clot for 30 min at room temperature before undergoing centrifugation for 10 min. The separated serum was then either (1) analyzed immediately or (2) stored in a 4°C–8°C refrigerator for batch analysis within 24 h or (3) frozen and defrosted if immediate or batch analysis could not be performed within 24 h of the blood draw procedure.

Maternal serum PlGF concentrations were measured using two immunoassays on automated immunoanalysers manufactured in China, the SuperFlex (Revvity Inc., Shanghai, China) and iMAGIN 1800 (Ningbo Aucheer Biotechnology Co., Ltd., Ningbo, China). Measured PlGF levels were converted to their equivalent MoM values. The expected median level was estimated after adjusting for fixed and random maternal and obstetric independent factors using the PlGF MoMing model published by Tan et al., which was constructed using the multiethnic population of women with singleton pregnancies screened at the FMF. ⁹

2.4. Estimation of risk for preterm‐PE

Estimated risks for preterm‐PE at the time of screening in centers using Superflex were determined using PreCare 1.0 software (Revvity Inc., Shanghai, China) while those in centers using iMAGIN 1800 were determined using Astraia software (Astraia Gmbh). Individuals with an estimated risk of 1:80 or higher were identified as high‐risk in accordance with the study protocol. ¹⁴ The same MoMing models were used in all centers, irrespective of which risk estimation software was employed at each center.

2.5. Adjustments made for the use of Chinese immunoanalysers

In order to use the Tan et al. MoMing model, the risk estimation software was modified as it is based on only three PlGF assays and immunoanalysers, namely the Roche Cobas, Revvity DELFIA Xpress, and Thermo Fisher Scientific B·R·A·H·M·S KRYPTOR. The expected median PlGF levels in all centers using Revvity SuperFlex immunoanalyser were estimated assuming that the measured PlGF levels on SuperFlex would be comparable to those determined using the Revvity DELFIA Xpress. For centers using the iMAGIN 1800 analyzer, we calculated the expected median PlGF levels by referencing the performance of Thermo Fisher Scientific B·R·A·H·M·S KRYPTOR, an instrument not available in Mainland China. A pretrial comparative study conducted in Hong Kong SAR indicated that both SuperFlex and iMAGIN 1800 measured PlGF were correlated with measured PlGF levels determined using the Thermo Fisher Scientific B·R·A·H·M·S KRYPTOR and Roche Cobas e411, as shown in Figures S1 and S2.

Our previous study using Roche Cobas, Revvity DELFIA Xpress, and Thermo Fisher Scientific B·R·A·H·M·S KRYPTOR immunoanalysers indicated that the Tan et al. PlGF MoMing model resulted in PlGF MoM distributions in Asian pregnant women with central tendency that were significantly deviated from 1 MoM. This was due to higher estimated expected median levels determined using the Tan et al. MoMing model after adjusting for fixed and random maternal and pregnancy factors. ¹² This required all PlGF MoM distributions to be recentered even though they were already adjusted for East and South Asian ethnicity. ⁹

Estimation of the number of biomarker measurements required to prospectively determine whether recentering was necessary in the present study and by how much was based on log₁₀ PlGF MoM distribution variance in unaffected pregnancies reported by O'Gorman et al. ¹⁷ Assuming that the PlGF measurement coefficient of variation of quality control samples on the immunoanalysers was 5% or higher, then to determine whether the central tendency was not within 5% of the expected mean of 0 with 95% confidence would require a minimum of 320 consecutive measurements, or 400 when rounded up to the nearest hundred.

The decision whether a specific biomarker MoM distribution showed evidence of a systematic bias and required recentering depended on whether the 95% confidence interval (CI) limits for the median of the MoM distribution excluded 1 MoM or not. If the 95% CI limits excluded 1 MoM, then the biomarker was recentered using the expression Recentered_MoM = Calculated_MOM/10 ^X where “X” is the mean of biomarker log₁₀ MoM distribution and “Calculated_MOM” is the value determined using the MoMing models reported by Tan et al. ¹⁰ used in the risk calculation software. Software providers were informed and requested to modify their risk estimation software as necessary.

2.6. Assessment of measurement quality

Anonymized screening data were transferred monthly to a central study coordinator who consolidated the received data into a centralized database for the assessment of measurement quality. The quality of the measurement at each participating center was evaluated using statistical measures of central tendency, dispersion, and temporal stability by constructing target plots and Cumulative Sum (CUSUM) charts as previously described for the study cohort overall as well as for each individual anonymized center. ¹⁸

Observed measurement distributions were considered “ideal” and temporally “stable” if their median was between 0.95 and 1.05, and that the CUSUM plot remained between the 0.95 and 1.05 limits for most of the time. Observed measurement distributions were considered “acceptable” and temporally “stable” if their median was between 0.90 and 0.95 or between 1.05 and 1.10, and that the CUSUM plot remained between the 0.90 and 0.95 or between the 1.05 and 1.10 limits for most of the time. Observed measurement distributions were considered “unacceptable” and or “unstable” if their median of the MoM distribution was <0.90 or >1.10, and the CUSUM plot remained persistently below the 0.90 or exceeded 1.10 limits, or evidence of structural breaks and change points in the time series measurement data suggestive of non‐adherence to measurement protocols, change in PlGF reagent lot characteristics, undetected immunoanalyser instrument deterioration, or inaccurate PlGF lot calibration.

As this was a prospective study with unknown pregnancy outcomes at the time of screening and to be consistent with the biomarker MoM truncation practice adopted in both the original FMF risk model and our prior large‐scale Asian multicenter study, ¹² all biomarker values were truncated at predefined cutoffs of ≤0.4 MoM or ≥2.5 MoM. These thresholds were previously validated for first‐trimester PE screening in Asian populations to exclude implausible outliers and minimize their disruptive effects on CUSUM and target plots. ¹² Each center included in the analysis was assigned an alphabetic code (A–F) to maintain anonymity and confidentiality.

2.7. Statistical analysis

Statistical analyses were performed using SPSS for Windows, version 29 (IBM Corp., Armonk, NY, USA), while central tendency and dispersion charts were generated with Python 3.13.1 (Python Software Foundation, DE, USA), and CUSUM were generated using Microsoft Excel. Inter‐center comparisons of patient demographics were performed using t‐test and chi‐squared test, with p values <0.05 being considered statistically significant. 95% CI for the MoM medians were estimated assuming a normal approximation using the formula Median ± Z _α/2 × IQR/√(n), where Z _α/2 is the critical value from the standard normal distribution table corresponding to 95% CI (=1.96), and interquartile range (IQR) is the difference between the third and first quartiles of the ordered data.

Analysis of intercenter differences in maternal characteristics of participants was performed. Based on our previous studies, ¹¹ the prevalence of maternal history of SLE, APS, diabetes mellitus, and smoking was expected to be low, but these indicators were still included in the assessment of intercenter differences. Intercenter differences in method of conception were assessed as IVF versus non‐IVF, as the number of ovulation inductions performed was also expected to be low. No assessment of ethnicity was performed as all women undergoing screening were of the same ethnicity.

3. RESULTS

Of the 44 554 women approached, 22 066 (49.5%) provided informed consent to undergo screening across the six centers, with a minimum of 1523 at center F and a maximum of 7105 from center B. The baseline characteristics of the 22 066 pregnancies screened are summarized in Table 1.

TABLE 1.

Characteristics of the study population.

Characteristic
Maternal age (age)	32.1 (29.7–35.0)
Maternal weight (kg)	57.0 (52.0–63.0)
Maternal height (cm)	161.0 (158.0–165.0)
BMI (kg/m2)	21.8 (20.0–24.0)
GA at screening (weeks)	87.0 (84.0–90.0)
Ethnicity
East Asian	22 066 (100%)
Method of conception
Spontaneous	19 826 (84.7)
Ovulation drugs	151 (0.6)
IVF	2089 (8.9)
Cigarette smoker	58 (0.2)
Chronic hypertension	105 (0.4)
SLE/APS	80 (0.4)
Diabetes mellitus
Type 1	8 (0.0)
Type 2	67 (0.3)
Parity
Nulliparous	17 444 (74.5)
Parous, no previous PE	4576 (19.5)
Parous, previous PE	46 (0.2)
Family history of PE	56 (0.2)
Interpregnancy interval (years)	4.9 (2.8–7.7)

Note: Data are shown as median (interquartile range), n (%) or n/N (%).

Abbreviations: APS, antiphospholipid syndrome; BMI, body mass index; GA, gestational age; IVF, in vitro fertilization; SLE, systemic lupus erythematosus.

Table S1 summarizes the maternal baseline demographics in the initial 400 women screened at each center. The intercenter assessment demonstrated that women recruited at Center F were significantly older than those at other centers. They also had higher rates of multiparous women, IVF conception, and a history of PE in previous pregnancies. Additionally, participants at Center A had higher gestational age at screening and a higher rate of smoking compared to other centers. Table 2 and Figure 1 summarize the central tendency of the MAP and PlGF MoM distributions and CUSUM plots in the first 400 screened pregnancies at the six centers. The MAP and PlGF MoM distributions of all centers were, respectively, 0.980 (95% CI: 0.975, 0.986) and 0.970 (95% CI: 0.946, 0.994). Although the MAP MoM CUSUM plots for all six centers were within the 0.95–1.05 limit lines and the central tendency was within the acceptable range (0.95–1.05), it was decided that the MAP MoM distributions at centers A, C, and F required recentering. PlGF MoM CUSUM and target plots indicated evidence of a systematic bias at all centers, with all but one requiring recentering. In addition, visual inspection of the PlGF MoM CUSUM plot indicated structural breakpoints at centers A and F, suggestive of sudden changes in the way PlGF measurements were determined. A structural breakpoint denotes an abrupt lasting shift in the CUSUM curve triggered by assay‐related changes, visually identified when the curve permanently diverges from its original stable trend. All preliminary 400 cases from each center were excluded from the final pooled analysis to avoid such interference, and MoM recentering was performed on the remaining formal study data for bias correction.

TABLE 2.

Biomarker median multiples of the median (MoM) values for the pretrial singleton pregnancies according to the European‐derived biomarker MoM formulae.

Study center	N	MAP MoM (95%CI)	PlGF MoM (95% CI)
Center A	400	0.960 (0.949–0.971)	0.870 (0.801–0.939)
Center B	400	0.999 (0.989–1.009)	0.887 (0.832–0.942)
Center C	400	0.965 (0.954–0.976)	0.976 (0.925–1.027)
Center D	400	0.994 (0.980–1.008)	0.947 (0.899–0.995)
Center E	400	1.010 (0.998–1.022)	0.935 (0.887–0.983)
Center F	400	0.950 (0.939–0.961)	1.195 (1.121–1.269)
Total	2400	0.980 (0.975–0.986)	0.970 (0.946–0.994)

Abbreviations: MAP, mean arterial pressure; MoM, multiples of the median; PlGF, placental growth factor.

FIGURE 1.

Cumulative sum (CUSUM) charts of log₁₀ mean arterial blood pressure (MAP) multiples of the median (MoM) (A), central tendency and dispersion charts of log₁₀ MAP MoM (B), CUSUM charts of log₁₀ placental growth factor (PlGF) MoM (C), and central tendency and dispersion charts of log₁₀ PlGF MoM (D) in women with singleton pregnancy from China, based on the pretrial dataset. Expected MoM value of 1.0, and ideal (0.95–1.05) (‐‐) and acceptable (0.90–1.10) (…) MoM limits. Center A, —; Center B, —; Center C, —; Center D, —; Center E, —; Center F, —; Each point represents every 100 consecutively screened pregnancies. Center A, •; Center B, ▲; Center C, ■; Center D, ◆; Center E, ▼; Center F, ╳.

Baseline characteristics of pregnancies screened at each center after excluding the first 400 women are summarized in Table S2. The intercenter differences in patient characteristics, although statistically significant, were less pronounced, with rates of IVF conception and previous history of PE affected pregnancy at center F now similar to that of other study sites.

Table 3 and Figure 2 summarize the MAP and PlGF MoM central tendency and CUSUM plots after excluding the first 400 screened pregnancies. Two centers (D and F) still had MAP MoM median 95% CI limits which excluded 1 MoM, despite Center F having its MoM measurement recentered unlike center D. Despite recentering, PlGF MoM distributions at centers A, E, and F still indicated systematic biases in measurements, with the bias target plot summary indicating periods when the measurements remained “unstable” while the CUSUM plot indicated structural breakpoints.

TABLE 3.

Biomarker profile of Median MoM values for 19 666 singleton pregnancies from the Chinese population, according to the corrected quality assurance assessment.

Study center	N	Corrected MAP MoM (95% CI)	Corrected PlGF MoM (95% CI)
Center A	1549	1.000 (0.995–1.006)	1.060 (1.028–1.092)
Center B	6705	1.008 (1.006–1.011)	1.024 (1.008–1.040)
Center C	6116	1.005 (1.003–1.008)	1.025 (1.009–1.040)
Center D	2548	0.986 (0.981–0.992)	0.984 (0.962–1.005)
Center E	1625	1.020 (1.015–1.025)	0.920 (0.898–0.942)
Center F	1123	0.970 (0.963–0.977)	0.920 (0.884–0.956)
Total	19 666	1.003 (1.001–1.004)	1.004 (0.996–1.013)

Abbreviations: MAP, mean arterial pressure; MoM, multiples of the median; PlGF, placental growth factor.

FIGURE 2.

Cumulative sum (CUSUM) charts of log₁₀ mean arterial blood pressure (MAP) multiples of the median (MoM) (A), central tendency and dispersion charts of log₁₀ MAP MoM (B), CUSUM charts of log₁₀ placental growth factor (PlGF) MoM (C) and central tendency and dispersion charts of log₁₀ PlGF MoM (D) in women with singleton pregnancy from China, after recenter biomarkers distribution. Expected MoM value of 1.0, and ideal (0.95–1.05) (‐‐) and acceptable (0.90–1.10) (…) MoM limits. Center A, —; Center B, —; Center C, —; Center D, —; Center E, —; Center F, —; Each point represents every 100 consecutively screened pregnancies. Center A, •; Center B, ▲; Center C, ■; Center D, ◆; Center E, ▼; Center F, ╳.

Table 4 summarizes the screen‐positive rate at each center in the initial 400 women who were screened versus that observed after recentering MoM distributions. The overall screen‐positive rate for preterm‐PE in the first 2400 women initially screened across the 6 centers was 11.5% (95% CI, 10.3%–12.9%) compared to 12.3% (95% CI, 11.8%, 12.8%) after recentering. The screen‐positive rate after recentering remained unchanged in two centers (B and E), decreased in two centers (A and D), and increased in the remaining two centers (C and F) (Figure 3).

TABLE 4.

Comparison of the screen‐positive rate of the pretrial dataset and after adjustment in each center.

Study center	SPR (95% CI) initial cases	Screen‐positive rate (95% CI) after adjustment	p value
Center A	10.5% (7.86%–13.89%)	10.07% (8.67%–11.67%)	0.800
Center B	12.75% (9.83%–16.38%)	12.74% (11.96%–13.56%)	0.994
Center C	12.75% (9.83%–16.38%)	13.29% (12.47%–14.17%)	0.756
Center D	12.75% (9.83%–16.38%)	10.12% (9.01%–11.36%)	0.111
Center E	13.00% (10.05%–16.65%)	13.23% (11.67%–14.97%)	0.903
Center F	7.50% (5.30%–10.50%)	10.77% (9.09%–12.72%)	0.060
Total	11.54% (10.32%–12.88%)	12.29% (11.84%–12.76%)	0.290

FIGURE 3.

Central tendency and dispersion charts of log₁₀ mean arterial blood pressure (MAP) multiples of the median (MoM) (A) and log₁₀ placental growth factor (PlGF) MoM (B) in women with singleton pregnancy from China after quality adjustment. Expected MoM value of 1.0, and ideal (0.95–1.05) and acceptable (0.90–1.10) MoM limits are shown (dashed lines). Each point represents every 400 consecutively screened pregnancies. Center A, •; Centre B, ▲; Centre C, ■; Center D, ◆; Center E, ▼; Center F, ╳.

4. DISCUSSION

Consistent screening for PE is reliant on reproducible measurements and concordance between assessed biomarker measurements and their distributions after normalization with those expected in the risk estimation models. Early identification of when biomarker(s) measurements are discordant with risk estimation is crucial to avoid inconsistencies in screening results. Our current study yields three key findings. First, it highlights the necessity to establish a dedicated quality QA protocol for pregnancy screening biomarkers and to perform routine QA assessments. Second, transposing MoMing equations—such as the model published by Tan et al. for maternal blood biomarkers—cannot guarantee that the biomarker MoM distribution is centered around 1.0, even after ethnicity adjustment. Third, researchers should use CUSUM plots to detect both fixed systematic bias and structural breakpoints. This helps identify unexpected changes in biomarker measurement procedures and enables timely corrective actions.

The AVERT trial makes use of two simple tools, target and CUSUM plots, to assess each center's biomarker measurement quality and adherence to measurement protocols on the assumption that those screened are from the general obstetric population. The CUSUM plot measures the cumulative deviation from an expected value, while the target plot compares both the central tendency and measurement variance versus that assumed for the unaffected population in the risk estimation model, namely that the median of the biomarker distribution is 1 MoM. The interpretation of both plots depends on whether structural breakpoints are evident in the CUSUM plot. If there are no structural breakpoints, the target plot and CUSUM can be used interchangeably to determine if there is any systematic measurement bias, and if so to what extent. In the present study, MAP MoM measurements of AVERT trial participants at the six centers showed evidence of a fixed systematic bias as the 95% CI of the median excluded 1 MoM, while the CUSUM trend plot appeared linear with no structural breakpoints. In comparison, the measurement of PlGF was more variable with evidence of structural breakpoints within the first 400 women screened at two centers, whereas the measurements at the other centers indicated evidence of a systematic bias as the CUSUM plots trends at these four centers were predominantly linear. The assessment of PlGF MoM distribution central tendency in these two centers is therefore misleading and should only be assessed once the cause of the structural breakpoints has been determined. PlGF measurements, unlike MAP, require more procedural steps and depend on pre‐ and postanalytical procedures. Anticipatable pre‐ and postanalytical factors when measuring PlGF include the method of blood draw site cleansing, the type of blood collection tube used, the interval and the storage of collected samples pending laboratory processing, the minimum permitted time to allow blood to clot prior to centrifugation, and the choice of MoMing model on the assumption that a laboratory would be able to correctly calibrate the PlGF immunoassay and that there is minimal inter‐lot reagent variation. Establishing standardized operating procedures with respect to biomarkers determination and their quality therefore remains imperative to ensure precise and consistent biomarker evaluation if screening is to be efficacious. ¹⁹ , ²⁰

The AVERT trial makes use of the MAP and PlGF MoMing models published by Tan et al., which were derived from women screened at the FMF, who were predominantly White, Afro‐Caribbean, and South Asian. We postulate that one reason for the negative systematic bias in MAP MoM distributions is due to inherent physical size differences between East Asian Chinese women and their non‐East Asian counterparts. The MAP is determined by cardiac output (CO), systemic vascular resistance, and central venous pressure. ²¹ The median body mass index (BMI) of Chinese women screened in the AVERT trial was 21.8 kg/m² significantly lower than the 24.5 kg/m² observed in the FMF population used to construct the MAP MoMing model. ¹⁷ Previous studies have demonstrated that a higher BMI is associated with higher BPs due to increased CO. ²² , ²³ Screening centers deciding to use the Tan et al. MAP MoMing model should therefore assess whether their own populations' physical characteristics significantly differ from that of the FMF and if so whether MAP distributions need to be recentered and if so by how much. The alternative would be for China‐based screening centers to combine and generate their own national MAP MoMing model.

One aspect of the AVERT trial was the requirement to use PlGF immunoassays developed and manufactured in China. There is currently no FMF procedural protocol with respect to how FMF PlGF MoMing models should be adapted to permit use of non‐FMF assessed PlGF immunoassays. We therefore adopted a pragmatic approach and assumed that the only difference between the Chinese immunoassays and those currently assessed by the FMF would be a simple level difference, which could be estimated from the systematic bias observed in the first 400 women screened at each center. We assumed that the observed level difference, if significant, was center‐immunoanalyser model‐specific rather than immunoanalyser manufacturer‐specific due to differing preanalytical and analytical practices at each AVERT trial site. Such differentiation arises from inconsistent preanalytical and analytical workflows across sites, as supported by cross‐method comparison results in Figures S1 and S2.

One option to assist screening centers in China in detecting, rectifying, and ensuring stable biomarker measurements would be to establish a centralized monitoring system similar to the one used in the United Kingdom to assess the quality of biomarker measurement for Down syndrome screening. Laboratories in the United Kingdom, without exception, are required to submit relevant measurement information for assessing screening quality and, in return, are advised what, if any, remedial action is required. The evaluation of measurement quality within the AVERT trial indicates that it would likely be beneficial, as it would enable screening centers in China to undertake remedial actions if necessary, which would, in turn, help enhance screening effectiveness.

The current study presents several notable strengths. First, the large sample size enhances the statistical power of the findings, ensuring that the results are robust and provide meaningful insights into the variability of biomarker measurements within the target population. Secondly, we examined variations in biomarker measurements, a crucial consideration prior to the external validation and performance evaluation of this screening method across different clinical settings. This approach is vital for establishing the foundation for future external validation studies in other countries. Additionally, the study's adherence to a well‐established protocol and comprehensive prestudy training to standardize measurements ensured that the median biomarker MoM values across the cohort remained within the acceptable range of 0.90–1.10, thereby further reinforcing the reliability of the findings.

The primary limitation of this study is the assumption that all screened pregnancies were unaffected. This reflects the reality of prospective first‐trimester screening, where pregnancy outcome data are typically unavailable at the time of analysis and were not included in this study. We used median MoM values for biomarkers without accounting for variations in pregnancies affected by PE. Given the global incidence of PE is approximately 2%–5%, this assumption is unlikely to have a substantial impact on the overall findings. In addition, only 49.5% of eligible women accepted screening. Although such uptake rates are not uncommon in large prospective studies, this may introduce selection bias, and variation across centers may reflect differences in local pregnancy booking procedures and demographic characteristics. Nevertheless, the level of screening acceptance is expected to have minimal impact on biomarker QA, as these assessments rely primarily on analytical performance rather than participation rates.

5. CONCLUSION

This study emphasizes the need for QA protocols in preterm‐PE screening, highlighting the importance of regular assessments.

Our findings demonstrate that when maternal serum biomarker values used in PE screening are expressed as MoM, even with ethnic adjustments, the resulting values may not always correspond to the expected central tendency. Additionally, CUSUM plots should be used to detect both structural breakpoints and biases, enabling timely corrective actions. For centers using MoM values, particularly in China, it is essential to assess initial biomarker distributions to ensure alignment with expected values for accurate PE risk estimation. QA and biomarker normalization are crucial for the successful worldwide implementation of first‐trimester combined screening for PE.

AUTHOR CONTRIBUTIONS

L.C.P. contributed to the trial design, supervised trial implementation. L.C.P., J.L., and D.S.S. supervised data collection, directly accessed the underlying data reported in the article. L.C.P., D.S.S., L.N.‐H., and J.L. designed and reviewed the manuscript. J.L. and L.C.P. analyzed the data and wrote the first draft of the manuscript. J.L. drafted the tables and figures. J.L., L.N.‐H., X.L., H.Y., Q.Z., H.Y., R.Z., Y.Z., W.X., H.X., X.T., L.K., J.Q., C.C.W., X.L., D.S.S., and L.C.P. reviewed the article and approved the final draft of the manuscript.

FUNDING INFORMATION

AVERT trial is supported by the National Key Research and Development Program of China (nos. 2021YFC2701600 and 2021YFC2701604). Reagents and equipment for the measurement of serum PlGF were provided by Revvity Inc., Shanghai, China and Ningbo Aucheer Biotechnology Co., Ltd., Ningbo, China. The PE risk calculators, including Precare, and Astraia software were provided by Revvity Inc., Shanghai, China, and Astraia Gmbh, respectively.

CONFLICT OF INTEREST STATEMENT

L.C. Poon has received speaker fees and consultancy payments from Roche Diagnostics and Ferring Pharmaceuticals and in‐kind contributions from Roche Diagnostics, Revvity Inc. (formerly PerkinElmer Life and Analytical Sciences), Thermo Fisher Scientific, Ningbo Aucheer Biological Technology Co., Ltd., and GE HealthCare. In addition, she serves as the CEO of PregnaSense Co. Limited. D.S. Sahota has received in‐kind contributions from Revvity Inc. (formerly PerkinElmer Life and Analytical Sciences), Thermo Fisher Scientific, Roche Diagnostics, Diabetomics, and Ningbo Aucheer Biological Technology Co., Ltd. Other authors report no conflict of interest.

ETHICS STATEMENT

A favorable ethical opinion was obtained on August 18, 2021 from the Joint Chinese University of Hong Kong‐New Territories East Cluster Clinical Research Ethics Committee (Reference number 2021.406). Subsequent approval by individual ethics committees and competent authority was granted.

Supporting information

Figure S1. Passing‐Bablok charts showing regression line (solid line) and 95% confidence bands for the regression line (—) when comparing measured concentrations of placental growth factor (PlGF) in 400 samples determined using Revvity Superflex immunoassay and immunoanalyser versus those determined using (a) Thermo Fisher Scientific B·R·A·H·M·S KRYPTOR immunoassay and immunoanalyser, (b) Roche Cobas Elecsys immunoassay and immunoanalyser, (c) Revvity DELFIAXpress immunoassay and immunoanalyser.

Figure S2. Passing‐Bablok charts showing regression line (solid line) and 95% confidence bands for the regression line (—) when comparing measured concentrations of placental growth factor (PlGF) in 400 samples determined using Ningbo Aucheer iMAGIN 1800 immunoassay and immunoanalyser and versus those determined using (a) Thermo Fisher Scientific B·R·A·H·M·S KRYPTOR immunoassay and immunoanalyser, (b) Roche Cobas Elecsys immunoassay and immunoanalyser, (c) Revvity DELFIAXpress immunoassay and immunoanalyse.

Table S1. Comparison of screened pregnancies in each center.

Table S2. Comparison of screened pregnancies in each center after applying biomarker MoM distribution recentering.

DATA AVAILABILITY STATEMENT

The individual participant data will not be shared according to the Data Security Law and the Personal Information Protection Law of the government of the People's Republic of China. The summary data could be shared.