Ultrasound Obstet Gynecol 2013; 42: 257267 Published online 6 August 2013 in Wiley Online Library (wileyonlinelibrary.com).

DOI: 10.1002/uog.12435

Value of adding second-trimester uterine artery Doppler to patient characteristics in identication of nulliparous women at increased risk for pre-eclampsia: an individual patient data meta-analysis
C. E. KLEINROUWELER*, P. M. M. BOSSUYT, B. THILAGANATHAN, K. C. VOLLEBREGT*, REZ, A. OHKUCHI, K. L. DEURLOO**, M. MACLEOD, A. E. DIAB, J. ARENAS RAMI H. WOLF*, J. A. M. VAN DER POST*, B. W. J. MOL* and E. PAJKRT*
*Department of Obstetrics and Gynaecology, Academic Medical Center, Amsterdam, The Netherlands; Department of Clinical Epidemiology, Biostatistics and Bioinformatics, Academic Medical Center, Amsterdam, The Netherlands; St Georges Healthcare NHS Trust, Department of Obstetrics and Gynaecology, London, UK; Hospital de Cabuenes, Department of Obstetrics and Gynaecology, Gijon, Spain; Jichi Medical University School of Medicine, Department of Obstetrics and Gynecology, Shimotsuke-shi, Tochigi-ken, Japan; **VU University Medical Center, Department of Obstetrics and Gynaecology, Amsterdam, The Netherlands; University of Dundee, Division of Clinical & Population Sciences & Education, Academic Clinical Practice, Ninewells Hospital & Medical School, Dundee, UK; Zagazig Faculty of Medicine, Department of Obstetrics and Gynaecology, Zagazig, Ash Sharqiyah, Egypt

K E Y W O R D S: Doppler ultrasound; individual patient data meta-analysis; prediction; pre-eclampsia; uterine artery

ABSTRACT
Objective To investigate the value of adding secondtrimester uterine artery Doppler ultrasound to patient characteristics in the identication of nulliparous women at risk for pre-eclampsia. Methods For this individual patient data meta-analysis, studies published between January 1995 and December 2009 were identied in MEDLINE and EMBASE. Studies were eligible in which Doppler assessment of the uterine arteries had been performed among pregnant women and in which gestational age at ultrasound, Doppler ultrasound ndings and data on the occurrence of pre-eclampsia were available. We invited corresponding authors to share their original datasets. Data were included of nulliparous women who had had a secondtrimester uterine artery Doppler ultrasound examination. Shared data were checked for consistency, recoded to acquire uniformity and merged into a single dataset. We constructed random intercept logistic regression models for each of the patient and Doppler characteristics in isolation and for combinations. We compared goodness of t, discrimination and calibration. Results We analyzed eight datasets, reporting on 6708 nulliparous women, of whom 302 (4.5%) developed preeclampsia. Doppler ndings included higher, lower and mean pulsatility index (PI) and resistance index (RI) and

any or bilateral notching. Of these, the best predictors were combinations of mean PI or RI and bilateral notching, with areas under the receiveroperating characteristics curve (AUC) of 0.75 (95% condence interval (CI), 0.560.95) and 0.70 (95% CI, 0.660.74), respectively. Addition of Doppler ndings to the patient characteristics blood pressure or body mass index (BMI) signicantly improved discrimination. A model with blood pressure, PI and bilateral notching had an AUC of 0.85 (95% CI, 0.671.00). Conclusions The addition of Doppler characteristics of mean PI or RI and bilateral notching to patient characteristics of blood pressure or BMI improves the identication of nulliparous women at risk for preeclampsia. Copyright 2013 ISUOG. Published by John Wiley & Sons Ltd.

INTRODUCTION
Pre-eclampsia is a multisystem disorder of pregnancy, dened as the presence of hypertension and proteinuria, that contributes substantially to maternal and perinatal morbidity and mortality worldwide1 . The obstetric management of women with pre-eclampsia is complicated by the fact that by the time symptoms occur, the only denitive treatment of the underlying disorder

Correspondence to: Ms E. Kleinrouweler, Academic Medical Center, Department of Obstetrics and Gynaecology, Room H4-232, Meibergdreef 9, 1105 AZ Amsterdam, The Netherlands (e-mail: c.e.kleinrouweler@amc.uva.nl) Accepted: 1 February 2013

Copyright 2013 ISUOG. Published by John Wiley & Sons Ltd.

M E T A-A N A L Y S I S

258

Kleinrouweler et al. Center. This meta-analysis was conducted according to the Meta-analysis Of Observational Studies in Epidemiology (MOOSE) guidelines13 .

is delivery. Management focuses on safe prolongation of pregnancy through intensive monitoring, such that maternal and fetal complications can be prevented1 . Accurate prediction of this condition in early pregnancy would allow for timely allocation of monitoring resources, with the prospect of improving maternal and perinatal outcomes2,3 . Preventive treatment such as aspirin seems more likely to be benecial if started earlier in pregnancy4 . Vascular supply to the placenta depends on trophoblast invasion and remodeling of the spiral arteries during the late rst and early second trimesters. A deciency in this process is associated with fetal growth restriction and the development of pre-eclampsia. Doppler ultrasound can be used to assess blood ow velocity in the maternal uterine arteries and thus potentially identify pregnancies at increased risk for pre-eclampsia and, indeed, numerous studies have investigated screening strategies for pre-eclampsia based on uterine artery Doppler ultrasound. Combination screening strategies based on patient characteristics and multiple markers or tests, such as Doppler ultrasound, seem more promising than tests in isolation5 11 . There is uncertainty about the prognostic accuracy of Doppler ultrasound ndings, when combined with more readily available patient characteristics such as blood pressure and being overweight. In clinical practice, a test will always be performed when there is prior suspicion, the strength of which will vary between patients. The question is not whether Doppler ultrasound in isolation can distinguish between future pre-eclampsia or uncomplicated pregnancy, but whether Doppler ultrasound results add relevant information to what clinicians have already inferred from the clinical characteristics of the pregnant woman. Conventional meta-analyses of prognostic studies only summarize data on the accuracy of a single test under consideration and do not make use of all available information. In contrast, individual patient data (IPD) meta-analysis allows one to compare multivariable prediction strategies and also allows the possibility of time-to-event analysis12 . We report here on a project to investigate the added value of uterine artery Doppler measurements in the identication of nulliparous women at risk for preeclampsia, based on analyses of individual patient data from previously published studies.

Literature search, study selection, data collection and quality assessment

We searched MEDLINE and EMBASE between 1995 and 2009 to identify eligible studies. Studies were eligible if Doppler assessment of the uterine arteries had been performed in pregnant women, at any gestational age, in any healthcare setting, at any level of risk for preeclampsia and in which gestational age at ultrasound, Doppler ultrasound ndings and the occurrence of pre-eclampsia had been recorded. Acceptable reference standards for pre-eclampsia were persistent high systolic ( 140 mmHg) or diastolic ( 90 mmHg) blood pressure and proteinuria ( 0.3 g/24 h or a dipstick result of 1+, equivalent to 30 mg/dL in a single urine sample) of new onset after 20 weeks gestation, according to the International Society for the Study of Hypertension in Pregnancy criteria14 . More details of the search strategy and selection of eligible studies can be found in Appendix S1 online. Over a period of 3 years (20082010), the corresponding authors of eligible studies were contacted by email and invited to participate in the project and share their original datasets, i.e. those that had been used for the analyses published in the study reports. We only contacted authors of studies published from 1995, as we assumed that data collected earlier would possibly not be stored in an electronic format still useful today, and because Doppler ultrasound techniques have changed over time. When authors replied positively they were provided with a detailed project protocol and were asked to send their original, complete and anonymized data. Additional information was extracted from the original papers (see also Appendix S1). The methodological quality of the studies was assessed from the original publication using the Quality Assessment of Diagnostic Accuracy Studies (QUADAS) criteria15 . In addition, we checked for consistency between received data and the results in the published paper. Shared data were reformatted or recoded if necessary to ensure uniformity, and merged into a single dataset.

Data analysis METHODS

The IPD-POPULAR project (Prediction Of Pre-eclampsia from Doppler ULtrasonography, Anthropometric parameters and maternal Risk factors) relies on a systematic search of the literature, invitations to share data and a comparison of multivariable prediction models for preeclampsia in these data. Under Dutch law, this project did not require formal approval of an ethics committee or institutional review board, as was conrmed by the Medical Ethics Committee of the Academic Medical For this analysis16,17 , we restricted ourselves to nulliparous women who had had a second-trimester uterine artery Doppler ultrasound examination. The relationship between each patient characteristic or Doppler ultrasound parameter and pre-eclampsia was evaluated by univariable logistic regression analysis. We modeled continuous variables with quadratic, cubic and logarithmic terms to assess potential non-linearity. Two-level-random intercept logistic regression models were then constructed for each of the available Doppler

Copyright 2013 ISUOG. Published by John Wiley & Sons Ltd.

Ultrasound Obstet Gynecol 2013; 42: 257267.

IPD meta-analysis of UtA for PE parameters in isolation and in combinations, using the occurrence of pre-eclampsia at the individual patient level as the outcome variable. Stratication at the study level was used to control for remaining between-study heterogeneity18 . We identied the best predictive patient characteristic or combination of patient characteristics, and also the best predictive Doppler parameter, or combination of Doppler parameters, in terms of model t. Then we compared discrimination with a model consisting of only patient characteristics, a model with Doppler parameters only, and a model containing both patient characteristics and Doppler ultrasound ndings. Improvement in model t was evaluated using the likelihood ratio test statistic for nested submodels (comparing models with one or more additional variables). For nonnested submodels, testing whether one submodel was better than the other was done by calculating the difference in log likelihood between the two models in 10 000 bootstrap samples drawn from the original dataset. In each bootstrap, as many observations were drawn from each shared dataset as there were women included in that dataset. P-values were calculated assuming a normal distribution of the differences in log likelihood with a true mean of 0 and the observed standard deviation in the 10 000 bootstrap samples. For the univariable analyses of the patient characteristics and Doppler parameters we used data of all the women in whom that variable had been recorded. For the analyses in which variables were compared or evaluated in combination, we used data of a subset of all women for whom all studied variables were available. As not all factors were present in all datasets, the numbers of women studied in each step of the analysis differed. Model discrimination was assessed by calculating the area under the curve (AUC) in a receiveroperating characteristics (ROC) plot, also known as the c-statistic. Differences in AUCs between models were tested with the comproc function in R, which estimates the difference between model AUCs in 10 000 bootstrap samples and uses the Wald test for model comparisons based on the bootstrap standard errors. Calibration of the models was presented in calibration plots. The analyses were repeated with pre-eclampsia requiring delivery at < 34 weeks as the outcome variable. In Appendix S1 more details are given on the methods used for data analysis. In addition, we compared time to delivery between women ranked in the highest centiles of predicted probabilities of pre-eclampsia (> p75p80, > p80p85, > p85p90, > p90p95 and > p95) and women with probabilities of p75, according to the model containing both patient characteristics and Doppler parameters, and expressed the results in KaplanMeier plots. All analyses were performed using R version 2.12.1 (R Foundation, Vienna, Austria) and PASW statistics 18 (IBM Inc., New York, NY, USA).

259

RESULTS Study selection and data collection

The searches in MEDLINE and EMBASE (1995 to 2009) resulted in a total of 3199 citations after removal of duplicates, of which 176 study reports, written by 111 different corresponding authors, were deemed eligible for inclusion. We were able to contact 107 of them, of whom 49 (46%) replied that they were interested in the project and willing to share data. Twenty-two authors did not share data despite an expressed intention to do so (n = 17) or informed us that the data were no longer available (n = 3) or that they had not been given institutional review board approval for data sharing (n = 2). Eventually, 27 authors shared their datasets of 30 studies. From the shared datasets, we were able to use eight for this project, as these had data on nulliparous women who had had a second-trimester Doppler ultrasound examination19 26 . In total, data were available on 6708 unselected nulliparous women, of whom 302 (4.5%) developed pre-eclampsia. Table 1 summarizes study details, baseline patient characteristics and additional Doppler measurements in these datasets. Table S1 (in Appendix S1 online) describes the study groups in the datasets included in our analyses, and shows the number of nulliparous women in the original studies. The results of the quality assessment are described in Appendix S1. Evaluation with the QUADAS instrument showed overall good study quality. Table 2 illustrates the availability of the variables in each included dataset. As expected, not all variables were available in all datasets. The results of the investigation and selection of patient characteristics can also be found in Appendix S1. We found that body mass index (BMI) and systolic blood pressure contributed most to the prediction of pre-eclampsia among the available patient characteristics. The combination of BMI and systolic blood pressure predicted pre-eclampsia better than BMI alone, but as well as systolic blood pressure alone.

Selection of Doppler ultrasound predictors

In ve datasets combined, containing data from 3116 women, we were able to investigate the relationship between the higher, lower and mean (of the measured left and right uterine artery) pulsatility index (PI). Each was signicantly associated with pre-eclampsia. Of these three, the mean PI had the best predictive performance, but this was not signicantly better than that of the lower PI (P = 0.57) or higher PI (P = 0.06). The same was found to hold for higher, lower and mean resistance index (RI). In four combined datasets with data from 6271 women, all were signicantly associated with pre-eclampsia, but the mean RI had the best predictive performance. Mean RI performed signicantly better than lower RI (P = 0.005) but not signicantly better than higher RI (P = 0.33). The prognostic value of bilateral and any notching could be assessed in six combined datasets containing

Copyright 2013 ISUOG. Published by John Wiley & Sons Ltd.

Ultrasound Obstet Gynecol 2013; 42: 257267.

260

Table 1 Characteristics of the eight datasets and included nulliparous women who underwent second-trimester uterine artery (UtA) Doppler ultrasound examination Study 2 Deurloo The Netherlands Not reported 123 32 (2041) 22.1 (17.730.1) NA NA 20 + 1 (17 + 6 to 24 + 4) 0.91 (0.372.05) 0.56 (0.310.84) 4 (3.3) 5 (4.1) NA NA Yes Yes Yes No* No* No* No* Yes Yes 14 (43.8) 7 (7.0) 2 (1.4) Yes Yes Yes 1.70 (1.452.34)* NA* 23 (71.9)* 15 (46.9) 5 (15.6) NA* NA* 44 (44.0)* 14 (14.0) 4 (4.0) NA 0.55 (0.380.81) 21 (14.4) 6 (4.1) 0 (0.0) 0.80 (0.272.89) 0.53 (0.231.32) 524 (9.0) 238 (4.1) 33 (0.6) 71 (1.2) Yes Yes Yes Diab Egypt 20052006 32 22 (1827) 22.0 (19.028.0) 0 (0.0) NA 23 Macleod UK 20012003 100 NA 24.6 (18.046.1) 24 (24.0) 110 (90140) 1820 Ohkuchi Japan 19931998 146 27 (2042) NA NA NA 20 + 2 (16 + 0 to 23 + 6) Thilaganathan UK 19962006 5835 31 (1555) 22.7 (11.857.6) 457 (11.8) NA 21 + 3 (18 + 0 to 24 + 0) 3 4 5 6 7 Vollebregt The Netherlands 20042006 258 30 (1842) 22.7 (17.941.2) NA NA 19 + 3 (16 + 6 to 28 + 3) 0.90 (0.324.40) NA 34 (13.4) 13 (5.1) 4 (1.6) 8 (3.1) Yes Yes Yes 8 Wolf The Netherlands Not reported 21 29 (2437) 21.6 (18.730.9) 0 (0.0) 112 (102140) 20

Parameter

Copyright 2013 ISUOG. Published by John Wiley & Sons Ltd. 0.83 (0.572.22) NA 3 (14.3) 1 (4.8) 0 (0.0) 0 (0.0) Yes Yes Yes

Arenas Spain 20002001 193 29 (1643) 24.0 (17.325.8) 60 (31.1) 110 (90159) 20 + 1 (18 + 1 to 22 + 3)

0.80 (0.112.21) 0.52 (0.330.83) 17 (8.8) 10 (5.2) 1 (0.5)

7 (3.6)

Yes

Yes

Corresponding author Country Inclusion years Number of women Maternal age (years) BMI (kg/m2 ) Smoker SBP (mmHg) GA at ultrasound (weeks) UtA Doppler Mean PI Mean RI Bilateral notching All PE PE requiring delivery 34 weeks PE requiring delivery 37 weeks Used for patient factor model Used for Doppler model Used for combined model

Yes

Ultrasound Obstet Gynecol 2013; 42: 257267.

Data are presented as n (%) or median (range). *Women in these studies were excluded from analyses of Doppler ultrasound measurements, as they were selected based on abnormal Doppler ndings. Data on smoking available in 3874 cases. BMI, body mass index; GA, gestational age; NA, not available in this dataset; PE, pre-eclampsia; PI, pulsatility index; RI, resistance index; SBP, systolic blood pressure.

Kleinrouweler et al.

IPD meta-analysis of UtA for PE

Table 2 Availability of patient characteristics and Doppler ultrasound measurements per dataset included in this study Study Parameter Corresponding author Nulliparous women Age Height Weight BMI Obesity (BMI 30 kg/m2 ) Smoker Alcohol consumption Ethnicity SBP DBP MAP Lower PI Higher PI Mean PI Lower RI Higher RI Mean RI Bilateral notching Any notching Pre-eclampsia 1 Arenas 193 193 (100) 193 (100) 193 (100) 193 (100) 193 (100) 193 (100) 0 (0) 0 (0) 193 (100) 193 (100) 193 (100) 193 (100) 193 (100) 193 (100) 193 (100) 193 (100) 193 (100) 193 (100) 193 (100) 193 (100) 2 Deurloo 123 122 (99.2) 0 (0) 0 (0) 115 (93.5) 115 (93.5) 0 (0) 0 (0) 0 (0) 0 (0) 0 (0) 0 (0) 97 (78.9) 97 (78.9) 97 (78.9) 97 (78.9) 97 (78.9) 97 (78.9) 122 (99.2) 122 (99.2) 123 (100) 3 Diab 32 32 (100) 0 (0) 0 (0) 32 (100) 32 (100) 32 (100) 0 (0) 0 (0) 0 (0) 32 (100) 0 (0) 0 (0)* 0 (0)* 32 (100)* 0 (0)* 0 (0)* 0 (0)* 32 (100)* 0 (0)* 32 (100) 4 Macleod 100 0 (0) 100 (100) 100 (100) 100 (100) 100 (100) 100 (100) 0 (0) 0 (0) 99 (99) 99 (99) 99 (99) 0 (0)* 0 (0)* 0 (0)* 0 (0)* 0 (0)* 0 (0)* 100 (100)* 0 (0)* 100 (100) 5 Ohkuchi 146 146 (100) 0 (0) 0 (0) 0 (0) 0 (0) 0 (0) 0 (0) 146 (100) 0 (0) 0 (0) 0 (0) 0 (0) 0 (0) 0 (0) 146 (100) 146 (100) 146 (100) 146 (100) 146 (100) 146 (100) 6 Thilaganathan 5835 5833 (99.97) 4381 (75.1) 4326 (74.1) 4249 (72.8) 4249 (72.8) 3874 (66.4) 3707 (63.5) 5305 (90.9) 0 (0) 0 (0) 0 (0) 2572 (44.1) 2572 (44.1) 2572 (44.1) 5835 (100) 5835 (100) 5835 (100) 5835 (100) 5835 (100) 5835 (100) 7 Vollebregt 258 258 (100) 258 (100) 258 (100) 258 (100) 258 (100) 0 (0) 0 (0) 258 (100) 0 (0) 0 (0) 0 (0) 234 (90.7) 234 (90.7) 234 (90.7) 0 (0) 0 (0) 0 (0) 253 (98.1) 241 (93.4) 257 (99.6) 8 Wolf 21 21 (100) 21 (100) 21 (100) 21 (100) 21 (100) 21 (100) 0 (0) 0 (0) 21 (100) 21 (100) 21 (100) 21 (100) 21 (100) 21 (100) 0 (0) 0 (0) 0 (0) 21 (100) 21 (100) 21 (100)

261

Total 6708 (100) 6605 (98.5) 4953 (73.8) 4898 (73) 4968 (74.1) 4968 (74.1) 4220 (62.9) 3707 (55.3) 5709 (85.1) 313 (4.7) 345 (5.1) 313 (4.7) 3117 (46.5) 3117 (46.5) 3149 (46.9) 6271 (93.5) 6271 (93.5) 6271 (93.5) 6702 (99.9) 6558 (97.8) 6707 (100)

Data are presented as n (%). *Women in these studies were excluded from analyses of Doppler ultrasound measurements, as they were selected based on abnormal Doppler ndings. BMI, body mass index; DBP, diastolic blood pressure; MAP, mean arterial pressure; PI, pulsatility index; RI, resistance index; SBP, systolic blood pressure.

data from 6569 and 6557 women, respectively. Both bilateral and any notching were signicantly associated with pre-eclampsia (both P < 0.001). In the same datasets, data on both variables were available in a subset of 6556 women. In these women, bilateral notching performed slightly but not signicantly better than any notching (P = 0.94). The addition of bilateral notching to the models containing mean PI alone (in 3116 women from ve combined datasets) or mean RI alone (in 6271 women from four combined datasets) signicantly improved model t (P = 0.03 and P = 0.01, respectively). The combination of PI or RI and bilateral notching also predicted pre-eclampsia better than bilateral notching

alone (both P < 0.001). Addition of an interaction term between PI and bilateral notching (P = 0.11) to the model did not improve performance (P = 0.09). Addition of an interaction term between RI and bilateral notching (P = 0.22) did not signicantly improve model performance either (P = 0.20). The predictive models with mean PI alone and the model with mean PI and bilateral notching performed slightly better than the models with mean RI alone and mean RI and bilateral notching, though not signicantly so (P = 0.24 and P = 0.34, respectively), in three combined datasets containing data from 2862 women.

Copyright 2013 ISUOG. Published by John Wiley & Sons Ltd.

Ultrasound Obstet Gynecol 2013; 42: 257267.

262

Kleinrouweler et al.

Table 3 Calculation of probabilities of pre-eclampsia from various predictors and combinations of predictors, including second-trimester uterine artery Doppler Predictors SBP BMI SBP + BMI PI + bilateral notching RI + bilateral notching SBP + PI + bilateral notching SBP + RI + bilateral notching BMI + PI + bilateral notching BMI + RI + bilateral notching SBP + BMI + PI + bilateral notching SBP + BMI + RI + bilateral notching n 313 4967 313 3116 6271 214 193 2963 Probability of pre-eclampsia 1/(1 + EXP((7.28164 + (0.04192 SBP)))) 1/(1 + EXP((9.123706 + (0.404313 BMI) (0.005169 BMI2 )))) 1/(1 + EXP((3.766467 (0.295400 BMI) + (0.005813 BMI2 ) + (0.043142 SBP)))) 1/(1 + EXP ((2.7260 + (0.5991 bilatnotch) (4.2688 PI) + (4.5923 PI2 ) (1.0550 PI3 )))) 1/(1 + EXP((12.4237 + (0.4714 bilatnotch) (89.5798 RI) + (154.5781 RI2 ) (80.5996 RI3 )))) 1/(1 + EXP((1.72198 + (0.04097 SBP) (29.85228 PI) + (25.86898 PI2 ) (6.50349 PI3 ) + (1.93245 bilatnotch)))) 1/(1 + EXP((30.75249 + (0.02442 SBP) (192.07371 RI) + (312.40119 RI2 ) (158.64051 RI3 ) + (1.58368 bilatnotch)))) 1/(1 + EXP((9.454603 + (0.446135 BMI) (0.006185 BMI2 ) + (0.487445 bilatnotch) (5.209416 PI) + (5.375934 PI2 ) (1.241653 PI3 )))) 1/(1 + EXP(( 8.576 + (0.4140 BMI) (0.005355 BMI2 ) + (0.3521 bilatnotch) (104.6 RI) + (178.5 RI2 ) (92.20 RI3 )))) 1/(1 + EXP((0.8684 + (0.05955 BMI) (0.0004479 BMI2 ) + (0.03792 SBP) + (1.967 bilatnotch) (29.97 PI) + (26.06 PI2 ) (6.567 PI3 )))) 1/(1 + EXP((21.86637 + (0.68734 BMI) (0.01232 BMI2 ) + (0.02220 SBP) + (1.66428 bilatnotch) (191.65240 RI) + (308.49609 RI2 ) (154.51456 RI3 ))))

4533 214

193

n indicates number of women available for estimation of regression models, which was not always equal to number of women available for comparison of multiple models as shown in Figures 13. EXP(x) = ex . bilatnotch, bilateral notching (yes/no); BMI, body mass index (kg/m2 ); PI, mean pulsatility index; RI, mean resistance index; SBP, systolic blood pressure (mmHg).

At this point in the analyses, we concluded that the best models with Doppler characteristics included mean PI and bilateral notching, or mean RI and bilateral notching. In the next step, we investigated the added value of both combinations to the previously selected patient characteristics.

Added value of Doppler ultrasound measurements to patient characteristics

As data on all patient characteristics and Doppler parameters were not available for all the women, we investigated the added value of mean PI and bilateral notching or mean RI and bilateral notching to BMI or systolic blood pressure in isolation or in combination. The probabilities of pre-eclampsia were calculated for individual patients as illustrated in Table 3. Figures 13 show the discriminative ability for all combinations made. For all three combinations, the discriminative ability of the model including both patient characteristics and Doppler parameters was better than that of a model with patient characteristics or Doppler alone. The AUC of the model containing systolic blood pressure, mean PI and bilateral notching was 0.85 (95% condence interval (CI), 0.671.00), higher than for the model with systolic blood pressure only (AUC 0.64 (95% CI, 0.450.84), P < 0.01) or mean PI and bilateral notching only (AUC 0.75 (95% CI, 0.550.95), P = 0.33) in two combined datasets from 214 women. The model with BMI, mean PI and bilateral notching (AUC 0.73 (95% CI, 0.680.79)) outperformed the

models with BMI alone (AUC 0.64 (95% CI, 0.590.69)) and mean PI and bilateral notching alone (AUC 0.67 (95% CI, 0.610.73)) in ve combined datasets from 2963 women, with P < 0.0001 and P < 0.01, respectively. The model with systolic blood pressure, BMI, mean PI and bilateral notching also performed better (AUC 0.85 (95% CI, 0.671.00)) than did the models with systolic blood pressure and BMI alone (AUC 0.65 (95% CI, 0.450.84), P < 0.01) or mean PI and bilateral notching alone (AUC 0.75 (95% CI, 0.560.95), P = 0.34) in two combined datasets from 214 women. The results for all models involving mean RI were similar to those of the models involving mean PI. The AUC of the model containing systolic blood pressure, mean RI and bilateral notching was 0.83 (95% CI, 0.641.00), compared with 0.64 (95% CI, 0.420.85; P < 0.01) for the model with systolic blood pressure alone or 0.78 (95% CI, 0.610.96; P = 0.52) for the model with mean RI and bilateral notching alone in one dataset from 193 women. The model with BMI, mean RI and bilateral notching (AUC 0.76 (95% CI, 0.730.80)) outperformed the models with BMI alone (AUC 0.67 (95% CI, 0.630.71), P < 0.001) and mean RI and bilateral notching alone (AUC 0.70 (95% CI, 0.660.74), P < 0.001) in three combined datasets from 4533 women. The model with systolic blood pressure, BMI, mean RI and bilateral notching also performed better (AUC 0.85 (95% CI, 0.681.00)) than the models with systolic blood pressure and BMI alone (AUC 0.64 (95% CI, 0.430.86), P = 0.01) or mean RI and bilateral notching alone (AUC

Copyright 2013 ISUOG. Published by John Wiley & Sons Ltd.

Ultrasound Obstet Gynecol 2013; 42: 257267.

IPD meta-analysis of UtA for PE

(a) 1.00 0.90 0.80 0.70 Sensitivity Sensitivity 0.60 0.50 0.40 0.30 0.20 0.10 0.00 1.00 0.90 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 Specificity (b) 1.00 0.90 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10

263

0.00 1.00 0.90 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 Specificity

Figure 1 Receiveroperating characteristics curves showing discriminative abilities of systolic blood pressure (SBP), uterine artery Doppler ultrasound measurements and their combinations in the identication of women at risk for pre-eclampsia. (a) SBP combined with mean , SBP; , mean PI and bilateral notching; , SBP, mean PI and bilateral notching. pulsatility index (PI) and bilateral notching. , SBP; , mean RI and bilateral notching; , SBP, (b) SBP combined with mean resistance index (RI) and bilateral notching. mean RI and bilateral notching.
(a) 1.00 0.90 0.80 0.70 Sensitivity Sensitivity 0.60 0.50 0.40 0.30 0.20 0.10 0.00 1.00 0.90 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 Specificity (b) 1.00 0.90 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 1.00 0.90 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 Specificity

Figure 2 Receiveroperating characteristics curves showing discriminative abilities of body mass index (BMI), uterine artery Doppler ultrasound measurements and their combinations in the identication of women at risk for pre-eclampsia. (a) BMI combined with mean , BMI; , mean PI and bilateral notching; , BMI, mean PI and bilateral notching. pulsatility index (PI) and bilateral notching. , BMI; , mean RI and bilateral notching; , BMI, (b) BMI combined with mean resistance index (RI) and bilateral notching. mean RI and bilateral notching.

0.78 (95% CI, 0.610.96), P = 0.39) in one dataset from 193 women. Figure 4 shows two calibration plots. The model with BMI, mean RI and bilateral notching in Figure 4a has good calibration; women in the two deciles with highest calculated probabilities of pre-eclampsia can be easily distinguished from women with lower probabilities. The model with systolic blood pressure, mean PI and bilateral notching in Figure 4b shows the same trends, but CIs are larger because of a much smaller sample size.

Pre-eclampsia requiring delivery before 34 weeks

Because not all studies had recorded the variables preeclampsia requiring delivery before 34 weeks gestation

or gestational age at delivery, only the added value of mean RI and bilateral notching or mean PI and bilateral notching to BMI could be investigated. In two datasets combined, containing data from 4442 women, of whom 32 developed pre-eclampsia requiring delivery before 34 weeks, discrimination with a model containing only BMI was signicantly improved by the addition of mean RI and bilateral notching; the AUC increased from 0.66 (95% CI, 0.570.76) to 0.92 (95% CI, 0.870.98) (P < 0.001). The added value of mean PI and bilateral notching to BMI could be assessed in four combined datasets with data on 2872 women, of whom 38 developed pre-eclampsia requiring delivery before 34 weeks. By addition of mean PI and bilateral notching, the AUC greatly increased from

Copyright 2013 ISUOG. Published by John Wiley & Sons Ltd.

Ultrasound Obstet Gynecol 2013; 42: 257267.

264
(a) 1.00 0.90 0.80 0.70 Sensitivity Sensitivity 0.60 0.50 0.40 0.30 0.20 0.10 0.00 1.00 0.90 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 Specificity (b) 1.00 0.90 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10

Kleinrouweler et al.

0.00 1.00 0.90 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 Specificity

Figure 3 Receiveroperating characteristics curves showing discriminative abilities of systolic blood pressure (SBP), body mass index (BMI), uterine artery Doppler ultrasound measurements and their combinations in the identication of women at risk for pre-eclampsia. (a) SBP , SBP and BMI; , mean PI and bilateral notching; and BMI combined with mean pulsatility index (PI) and bilateral notching. , SBP, BMI, mean PI and bilateral notching. (b) SBP and BMI combined with mean resistance index (RI) and bilateral notching. , , mean RI and bilateral notching; , SBP, BMI, mean RI and bilateral notching. SBP and BMI;
(a) 0.25 (b) 0.55 0.50 0.20 0.45 0.40 Observed rate 0.15 Observed rate 0.05 0.10 0.15 Calculated probability 0.20 0.25 0.35 0.30 0.25 0.20 0.15 0.05 0.10 0.05 0.00 0.00 0.00 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55 Calculated probability

0.10

Figure 4 Calibration plots of prediction models based on: (a) body mass index, mean uterine artery resistance index and bilateral notching and (b) systolic blood pressure, mean uterine artery pulsatility index and bilateral notching.

0.62 (95% CI, 0.480.75) to 0.95 (95% CI, 0.920.98) (P < 0.001). Figure S3 (in Appendix S1 online) shows the ROC plots.

Time to delivery
To illustrate the benets of IPD meta-analysis for time-toevent data, we compared time to delivery between women ranked in the highest centiles of calculated probabilities (> p75p80, > p80p85, > p85p90, > p90p95 and > p95) and all other women with probabilities of p75, with probabilities calculated with the two models described earlier: the model with BMI, mean RI and bilateral notching (4533 women) and the model with

blood pressure, mean PI and bilateral notching (214 women). Figure 5 shows that women ranked in the highest centiles of predicted probabilities from both models delivered earlier than women with lower probabilities of pre-eclampsia.

DISCUSSION
This paper describes the added value of uterine artery Doppler ultrasound to patient characteristics in the identication of nulliparous women at risk for preeclampsia. We found that the combination of Doppler ultrasound parameters PI or RI and bilateral notching signicantly improves the prediction of pre-eclampsia

Copyright 2013 ISUOG. Published by John Wiley & Sons Ltd.

Ultrasound Obstet Gynecol 2013; 42: 257267.

IPD meta-analysis of UtA for PE

(a) (b) 1.0 Cumulative proportion of women not delivered

265

1.0

Cumulative proportion of women not delivered

0.8

0.8

0.6

0.6

0.4

0.4

0.2

0.2

0.0 28 30 34 36 38 40 32 Gestational age at delivery (weeks) 42

0.0 28 30 32 34 36 38 40 Gestational age at delivery (weeks) 42

Figure 5 Survival curves showing time to delivery in women grouped by percentiles (p) of predicted risk for pre-eclampsia based on (a) body mass index, mean uterine artery resistance index and bilateral notching and (b) systolic blood pressure, mean uterine artery pulsatility index , p75; , > p75p80; , > p80p85; , > p85p90; , > p90p95; , > p95. and bilateral notching.

based on the patient characteristics BMI and systolic blood pressure, alone or combined. The calibration plots show that women at the highest risk for pre-eclampsia can be easily differentiated from women with lower risk predictions. This is supported by the time-to-event analyses, in which women with the highest calculated risk of pre-eclampsia were shown to deliver earlier than women at lower risk. This difference in time to delivery could be caused by the higher incidence of preeclampsia, fetal growth restriction or intrauterine fetal death and induced delivery in the group with higher predicted probabilities of pre-eclampsia27,28 . However, it is important that our results be externally validated in another large dataset. The methodology of IPD meta-analysis has several advantages over conventional meta-analysis. IPD metaanalysis can use all available data from a study, including information that was not published, and is more exible in combining data into a uniform format. In addition, the use of original data and contact with authors allow for superior quality checks and better interpretation of the results29 . We approached over 95% of the authors of eligible studies, as identied in an extensive search, for participation in this project. The fact that only a minority of eligible datasets were available for data sharing is a potential drawback of this project, but an interesting nding as well. Worthy of remark here is our experience that for two studies the institutional review board of the center invited to participate did not agree with anonymized data sharing. Moreover, data from cohort studies are apparently not stored as well as are data from randomized trials, although the advantagesthe

possibilities for answering new questions with the use of available good-quality data rather than having to include women in a new studyare clear and have been highlighted before30,31 . Within the shared datasets, we could only use the available patient characteristics. As all studies were primarily designed to investigate uterine artery Doppler, the number and type of patient characteristics differed between studies and hampered our ability to study all possible combinations in all available datasets. Although we could only investigate a limited number of patient characteristics, we believe that those available include the most relevant factors for the prediction of pre-eclampsia thus far identied. Out of all shared datasets with second-trimester Doppler ndings we chose a common and preferably unselected group of women, as test performance is best investigated in a population with an average risk for the chosen outcome, a population that resembles the one that will be tested once the test is introduced into daily practice. As most Doppler studies identied in our search had been performed in women selected in various ways for their risk of pre-eclampsia (for example based on pre-eclampsia in a previous pregnancy or pre-existent medical conditions), we chose the relatively unselected group of nulliparous women. The pre-eclampsia risk was similar for all included studies, as reected in the comparable incidences of pre-eclampsia in the studies. The data shared by Diab and Macleod consisted of women selected on the basis of abnormal Doppler results out of all the women included in their studies. Assuming that the higher incidence of pre-eclampsia was related to the abnormal Doppler ndings but not to the patient

Copyright 2013 ISUOG. Published by John Wiley & Sons Ltd.

Ultrasound Obstet Gynecol 2013; 42: 257267.

266

Kleinrouweler et al. screen-positive women would be worried, monitored or even treated unnecessarily, using Doppler as an exclusion test only failed to identify 2% of women who developed pre-eclampsia. It is possible that the majority of the low-risk women among these 2% have mild, late-onset pre-eclampsia. Until further research shows more effective screening or therapeutic options for preeclampsia, Doppler ultrasound could be used to support clinicians judgment of low-risk pregnancy.

characteristics (which were comparable with those in the other included studies), these women were only included in the estimation of predictive performance of patient characteristics alone. One could speculate that since the data shared by Thilaganathan formed the largest part of our dataset, our results may partly reect the associations present in their data. However, in a random effects model the weighted larger inuence of larger studies is unweighted by the extent of variability of the effect sizes over all included studies32 . Moreover, because not all variables were present in the dataset of Thilaganathan, it was not part of all prediction models. The SCOPE study investigated the predictive performance of patient characteristics and Doppler ultrasound in a large prospective cohort of nulliparous women with a similar 5% incidence of pre-eclampsia33 . In that study the AUC of the ROC curve of the best prediction model was around 0.71 on internal validation, but Doppler ultrasound did not add to the predictive performance of patient characteristics (the AUC remained 0.71). In addition to the patient characteristics blood pressure and BMI that were included in our prediction models, the SCOPE model also contained information on age, family history, smoking and alcohol use, fruit intake, bleeding, previous miscarriages, duration of sexual relationship, time to conception and the womans own birth weight. It seems that Doppler ultrasound either does not contain predictive information beyond that provided by these patient characteristics, or that abnormal Doppler ndings are a sign of impaired placentation and, as such, early disease that can be predicted by patient characteristics34,35 . In contrast, we did not nd any improvement in our models by the addition of data on smoking and/or age, and we have shown that, in the absence of more predictive patient information, Doppler ultrasonography does have added value in identifying women at risk. So should we use this test in daily practice? Offering Doppler ultrasonography (alone or in combination with other factors in a prediction model) as a screening test for pre-eclampsia would be justied if there were a safe, effective and affordable intervention to prevent pre-eclampsia in high-risk women. So far, there is not enough evidence that giving calcium or vitamins to highrisk women (or all women) can prevent this disease well enough to be cost-effective, neither has the benet of intensive monitoring been proven36,37 . Although aspirin treatment is thought to have a relative risk of 0.90 for almost all subgroups, there are indications that starting in early pregnancy is more effective than starting later in pregnancy4 . This raises questions regarding the potential effectiveness of a screen-and-treat strategy in the second trimester. Doppler measurements, as incorporated in a multivariable prediction model, could also be suggested as a low-risk identication tool. Our calibration plots show that women at higher risk (> 1015%) can be well differentiated from women with lower risks (2%). Whereas using the model as a screening test with a positivity threshold of 10 or 15% risk would imply that 8590% of

ACKNOWLEDGMENTS
Emily Kleinrouweler is supported by a PhD Scholarship from the AMC Graduate School. The AMC Graduate School had no involvement in the design and conduct of the study; collection, management, analysis and interpretation of the data; in the preparation, review and approval of the manuscript or in the decision to submit the manuscript for publication.

REFERENCES
1. Sibai B, Dekker G, Kupferminc M. Pre-eclampsia. Lancet 2005; 365: 785799. 2. Askie LM, Duley L, Henderson-Smart DJ, Stewart LA. Antiplatelet agents for prevention of pre-eclampsia: a metaanalysis of individual patient data. Lancet 2007; 369: 17911798. 3. Duley L, Henderson-Smart DJ, Meher S, King JF. Antiplatelet agents for preventing pre-eclampsia and its complications. Cochrane Database Syst Rev 2007: CD004659. 4. Bujold E, Roberge S, Lacasse Y, Bureau M, Audibert F, Marcoux S, Forest JC, Gigu` ere Y. Prevention of preeclampsia and intrauterine growth restriction with aspirin started in early pregnancy: a meta-analysis. Obstet Gynecol 2010; 116: 402414. 5. Cnossen JS, ter Riet G, Mol BW, van der Post JA, Leeang MM, Meads CA, Hyde C, Khan KS. Are tests for predicting pre-eclampsia good enough to make screening viable? A review of reviews and critical appraisal. Acta Obstet Gynecol Scand 2009; 88: 758765. 6. Conde-Agudelo A, Villar J, Lindheimer M. World Health Organization systematic review of screening tests for preeclampsia. Obstet Gynecol 2004; 104: 13671391. 7. Audibert F, Boucoiran I, An N, Aleksandrov N, Delvin E, Bujold E, Rey E. Screening for preeclampsia using rst-trimester serum markers and uterine artery Doppler in nulliparous women. Am J Obstet Gynecol 2010; 203: 383.e18. 8. Poon LC, Karagiannis G, Leal A, Romero XC, Nicolaides KH. Hypertensive disorders in pregnancy: screening by uterine artery Doppler imaging and blood pressure at 1113 weeks. Ultrasound Obstet Gynecol 2009; 34: 497502. 9. Thilaganathan B, Wormald B, Zanardini C, Sheldon J, Ralph E, Papageorghiou AT. Early-pregnancy multiple serum markers and second-trimester uterine artery Doppler in predicting preeclampsia. Obstet Gynecol 2010; 115: 12331238. 10. Tuuli MG, Odibo AO. First- and second-trimester screening for preeclampsia and intrauterine growth restriction. Clin Lab Med 2010; 30: 727746. 11. Zhong Y, Tuuli M, Odibo AO. First-trimester assessment of placenta function and the prediction of preeclampsia and intrauterine growth restriction. Prenat Diagn 2010; 30: 293308. 12. Broeze KA, Opmeer BC, van der Veen F, Bossuyt PM, Bhattacharya S, Mol BW. Individual patient data meta-analysis:

Copyright 2013 ISUOG. Published by John Wiley & Sons Ltd.

Ultrasound Obstet Gynecol 2013; 42: 257267.

IPD meta-analysis of UtA for PE

a promising approach for evidence synthesis in reproductive medicine. Hum Reprod Update 2010; 16: 561567. Stroup DF, Berlin JA, Morton SC, Olkin I, Williamson GD, Rennie D, Moher D, Becker BJ, Sipe TA, Thacker SB. Metaanalysis of observational studies in epidemiology: a proposal for reporting. Meta-analysis Of Observational Studies in Epidemiology (MOOSE) group. JAMA 2000; 283: 20082012. Brown MA, Lindheimer MD, de Swiet M, Van Assche A, Moutquin JM. The classication and diagnosis of the hypertensive disorders of pregnancy: statement from the International Society for the Study of Hypertension in Pregnancy (ISSHP). Hypertens Pregnancy 2001; 20: IXXIV. Whiting PF, Weswood ME, Rutjes AW, Reitsma JB, Bossuyt PN, Kleijnen J. Evaluation of QUADAS, a tool for the quality assessment of diagnostic accuracy studies. BMC Med Res Methodol 2006; 6: 9. Steyerberg EW. Clinical Prediction Models. A practical approach to development, validation and updating. Springer: New York, 2009. Harrell FE Jr. Regression Modeling Strategies. Springer-Verlag: New York, 2001. Twisk JWR. Software for multilevel analysis. In Applied Multilevel Analysis (Twisk JWR). Cambridge University Press : Cambridge, 2007; 146150. Arenas J, Fernandez-Inarrea J, Rodriguez-Mon C, Dupla B, Diez E, Gonzalez-Garcia A. Doppler screening of the uterine arteries to predict complications during pregnancy. [Spanish]. Clinica e Investigacion en Ginecologia y Obstetricia 2003; 30: 178184. Deurloo KL, Spreeuwenberg MD, Bolte AC, van Vugt JM. Color Doppler ultrasound of spiral artery blood ow for prediction of hypertensive disorders and intra uterine growth restriction: a longitudinal study. Prenat Diagn 2007; 27: 10111016. Diab AE, El-Behery MM, Ebrahiem MA, Shehata AE. Angiogenic factors for the prediction of pre-eclampsia in women with abnormal midtrimester uterine artery Doppler velocimetry. Int J Gynaecol Obstet 2008; 102: 146151. Khan F, Belch JJ, MacLeod M, Mires G. Changes in endothelial function precede the clinical disease in women in whom preeclampsia develops. Hypertension 2005; 46: 11231128. Ohkuchi A, Minakami H, Sato I, Mori H, Nakano T, Tateno M. Predicting the risk of pre-eclampsia and a small-for-gestationalage infant by quantitative assessment of the diastolic notch in uterine artery ow velocity waveforms in unselected women. Ultrasound Obstet Gynecol 2000; 16: 171178. Napolitano R, Santo S, DSouza R, Bhide A, Thilaganathan B. Sensitivity of higher, lower and mean second-trimester uterine artery Doppler resistance indices in screening for pre-eclampsia. Ultrasound Obstet Gynecol 2010; 36: 573576.

267
25. Vollebregt KC, Gisolf J, Guelen I, Boer K, van Montfrans G, Wolf H. Limited accuracy of the hyperbaric index, ambulatory blood pressure and sphygmomanometry measurements in predicting gestational hypertension and preeclampsia. J Hypertens 2010; 28: 127134. 26. Rang S, van Montfrans GA, Wolf H. Serial hemodynamic measurement in normal pregnancy, preeclampsia, and intrauterine growth restriction. Am J Obstet Gynecol 2008; 198: 519.e19. 27. Audibert F, Benchimol Y, Benattar C, Champagne C, Frydman R. Prediction of preeclampsia or intrauterine growth restriction by second trimester serum screening and uterine Doppler velocimetry. Fetal Diagn Ther 2005; 20: 4853. 28. Singh T, Leslie K, Bhide A, DAntonio F, Thilaganathan B. Role of second-trimester uterine artery Doppler in assessing stillbirth risk. Obstet Gynecol 2012; 119: 256261. 29. Khan KS, Bachmann LM, ter Riet G. Systematic reviews with individual patient data meta-analysis to evaluate diagnostic tests. Eur J Obstet Gynecol Reprod Biol 2003; 108: 121125. 30. Broeze KA, Opmeer BC, Bachmann LM, Broekmans FJ, Bossuyt PM, Coppus SF, Johnson NP, Khan KS, ter Riet G, van der Veen F, van Wely M, Mol BW. Individual patient data meta-analysis of diagnostic and prognostic studies in obstetrics, gynaecology and reproductive medicine. BMC Med Res Methodol 2009; 9: 22. 31. Individual Patient Data meta-analysis of diagnostic and prognostic studies in obstetrics, gynaecology and reproductive medicine. http://www.studies-obsgyn.nl/IPD [Accessed 10 May 2012]. 32. Senn S. Trying to be precise about vagueness. Stat Med 2007; 26: 14171430. 33. North RA, McCowan LM, Dekker GA, Poston L, Chan EH, Stewart AW, Black MA, Taylor RS, Walker JJ, Baker PN, Kenny LC. Clinical risk prediction for pre-eclampsia in nulliparous women: development of model in international prospective cohort. BMJ 2011; 342: d1875. 34. Steegers EA, von Dadelszen P, Duvekot JJ, Pijnenborg R. Preeclampsia. Lancet 2010; 376: 631644. 35. James JL, Whitley GS, Cartwright JE. Pre-eclampsia: tting together the placental, immune and cardiovascular pieces. J Pathol 2010; 221: 363378. 36. Meads CA, Cnossen JS, Meher S, Juarez-Garcia A, ter Riet G, Duley L, Roberts TE, Mol BW, van der Post JA, Leeang MM, Barton PM, Hyde CJ, Gupta JK, Khan KS. Methods of prediction and prevention of pre-eclampsia: systematic reviews of accuracy and effectiveness literature with economic modelling. Health Technol Assess 2008; 12: iiiiv, 1270. 37. Shmueli A, Meiri H, Gonen R. Economic assessment of screening for pre-eclampsia. Prenat Diagn 2012; 32: 2938.

13.

14.

15.

16.

17. 18.

19.

20.

21.

22.

23.

24.

SUPPORTING INFORMATION ON THE INTERNET

The following supporting information may be found in the online version of this article: Appendix S1 This appendix contains additional information on the search strategy, quality assessment and analysis methods that are described concisely in the paper. In addition, the investigation of the association of the available patient characteristics with pre-eclampsia and the selection of the most predictive patient characteristics are described.

Copyright 2013 ISUOG. Published by John Wiley & Sons Ltd.

Ultrasound Obstet Gynecol 2013; 42: 257267.