Objective
To derive a prediction rule for preeclampsia and early onset preeclampsia requiring delivery <34 weeks using first trimester maternal, ultrasound, and serum markers.
Study Design
Prospective cohort study of women enrolled at first trimester screening. Maternal history, demographics, anthropometry, ultrasound parameters, and serum analytes were compared between women with preeclampsia and normal outcome. The prediction rule was derived by Lasso logistic regression analysis.
Results
In 2441 women, 108 (4.4%) women developed preeclampsia, and 18 (0.7%) early preeclampsia. Nulliparity, prior hypertension, diabetes, prior preeclampsia, mean arterial pressure, and the log pregnancy-associate pregnancy protein-A multiples of the median were primary risk factors. Prediction rules for preeclampsia/early preeclampsia had an area under the curve of 0.82/0.83 respectively. Preeclampsia was predicted with 49% sensitivity and early preeclampsia with 55% sensitivity for a 10% false positive rate.
Conclusion
First trimester prediction rules using parameters currently available at first trimester screening identify a significant proportion of women with subsequent preeclampsia.
Preeclampsia (PET) remains a significant contributor to maternal and fetal morbidity and mortality. Improved and early prediction of women at risk for PET would make it possible to institute preventative measures and offer appropriate surveillance. Several studies have adopted the approach that has been established in first-trimester aneuploidy screening and developed multimarker algorithms for prediction of PET. Although these studies agree that a multifactorial approach is the most promising one, they diverge in the reported predictive accuracies. This could be because of variations in population risk profiles, the serum analytes used as well as the statistical approach or use of low-dose aspirin prophylaxis.
For clinical implementation, a predictive method that can be incorporated into existing obstetric care would be the most practical one. Pregnancy-associated plasma protein-A (PAPP-A) and free beta human chorionic gonadotropin (βhCG) concentrations are routinely determined by first-trimester testing. Yet, most first trimester screening studies use serum analytes that are either not yet commercially available, or that have not been validated for clinical use. It was the aim of this study to develop a risk algorithm for PET in prospectively enrolled women using parameters that are available in the context of first-trimester screening.
Materials and Methods
Pregnant women presenting for first trimester screening to any of 4 centers in the Baltimore, MD, metropolitan area for first-trimester screening from 2007-2010 were offered enrollment in this prospective observational study. The ultrasound examination and all study procedures were carried out after informed written consent. The study was approved by the institutional review boards of the University of Maryland School of Medicine, Mercy Medical Center, and the MedStar Research Institute. The standardized ultrasound examination at these centers confirms pregnancy dating, and includes crown-rump length measurements and transabdominal uterine artery Doppler; the specific ultrasound techniques have been previously described. Uterine artery notching was defined as early diastolic blood flow acceleration after the end-systolic nadir, producing a notched appearance of the waveform. Standardized written questionnaires on medical history, current and prior pregnancy histories, and social and demographic information were then obtained.
Following completion of the questionnaire, maternal height, weight, and blood pressure were measured. After 5 minutes of rest, qualified staff performed a single blood pressure measurement with the woman in a seated position and the arm at the level of the heart. The Dinamap Pro 1000 V3 (GE Medical Systems, Milwaukee, WI) automated sphygmomanometer was used, with a cuff size appropriate for maternal arm circumference. Sphygmomanometer calibration occurred every 6 months in accordance with the Association for the Advancement of Medical Instrumentation guidelines. Blood samples were obtained by trained nursing staff and submitted on standardized filter paper forms to NTD Labs (Perkin Elmer, Melville, NY) for analysis. NTD Labs reported the results as multiples of the median after absolute measurements of pregnancy-associated plasma protein-A (PAPP-A) and free βhCG were compared with reference ranges of a normal population.
Women were followed through pregnancy by the respective study enrollment center. Pregnancy outcomes, including the development of PET in the current pregnancy, were recorded at delivery. Proteinuric PET was defined as new-onset or worsening proteinuria (comparing results with the first-trimester urinalysis) and maternal systolic blood pressure ≥140 mm Hg or diastolic blood pressure ≥90 mm Hg on 2 separate occasions, 6 or more hours apart, after 20 weeks’ gestation.
All information from the collaborating clinical sites was reported directly to the primary investigator at the University of Maryland School of Medicine. Study data were collected, validated by source documentation, and entered into a dedicated study database. It is our local practice to offer women at risk for PET low dose aspirin if bilateral uterine artery notching is noted on the ultrasound examination. For logistic purposes these women were enrolled but subsequently excluded from this analysis if aspirin was recommended before 16 weeks.
Normality of continuous data was tested using the Kolmogorov-Smirnov test. We first performed univariate analysis to identify factors that were associated with an increased risk of PET. The statistical significance of observed associations was assessed using Fisher exact test as implemented for multiple-row tables in SAS 9.2 (SAS Institute, Cary, NC).
To derive a multivariable prediction rule we used lasso logistic regression as implemented by the R package GLMnet. This approach fits a logistic regression model, however, relative to the standard logistic regression fitting approach, the parameter estimates from the lasso approach are shrunk towards 0. This avoids extreme values that are likely to perform poorly in data not used to fit the model and has generally been shown to result in better prediction. The lasso approach also automatically chooses a subset of variables to include in the model (ie, those that are not shrunk all the way to zero). A tuning parameter to determine the degree of shrinkage was chosen by cross validation of the model prediction in 10 equal sized subsets of the patient population. Based on univariate analysis in our population and previously screening studies candidate variables for the prediction model included: age, race, history of diabetes, hypertension, nephropathy, thrombophilia, autoimmune disease, or prior PET, parity, body mass index (BMI), systolic, diastolic, and mean arterial pressure, left and right uterine artery pulsatility index as well as the mean pulsatility index in both uterine arteries multiples of the median (MoM), left and right uterine artery notching, number of early diastolic notches counting both uterine arteries, PAPP-A MoM, and free βhCG MoM. We then used receiver operator curve statistics to determine sensitivity and specificity and the 95% confidence interval (CI) for the area under the curve for the predictive probabilities derived from each prediction rule.
Results
Of 3422 screened and enrolled women, 323 (9.4%) were lost to follow-up, 634 (18.5%) received the recommendation for first-trimester aspirin, and 24 (0.7%) were excluded for other reasons leaving 2441 meeting inclusion criteria. OF these 108 (4.4%) developed PET; 18 (0.7%) required delivery before 34 weeks. African-American and whites constituted the majority of ethnicities among enrolled women (n = 1196, 49% and 1068, 43.8%, respectively). The median maternal age was 30 years (range, 18–55) and 1058 (43.3%) were nulliparous. A prior history of chronic hypertension (n = 174, 7.1%), diabetes mellitus (n = 91, 3.7%) were the most common preexisting medical conditions, and 91 women (3.7%) reported a history of prior PET ( Table 1 ). Delivery was at a median gestational age of 39.1 weeks (22.4-43 weeks’ gestation) and the median birthweight was 3264 g (100-5360 g). Of those, 171 (7%) infants were large and 217 (8.9%) were small for gestational age with a birthweight below the 10th percentile.
| Patient characteristic | Number (%) with early PET | P value | Number (%) with any PET | P value |
|---|---|---|---|---|
| Age | .99 | .31 | ||
| 18-24 (n = 605) | 4 (0.7%) | 35 (5.8%) | ||
| 25-29 (n = 561) | 4 (0.7%) | 26 (4.6%) | ||
| 30-34 (n = 578) | 5 (0.9%) | 19 (3.3%) | ||
| 35-39 (n = 570) | 4 (0.7%) | 24 (4.2%) | ||
| 40+ (n = 127) | 1 (0.8%) | 4 (3.2%) | ||
| Race | .0030 | .025 | ||
| Caucasian (n = 1068) | 2 (0.2%) | 35 (3.3%) | ||
| African-American (n = 1196) | 12 (1.0%) | 65 (5.4%) | ||
| Asian (n = 133) | 2 (1.5%) | 4 (3.0%) | ||
| Hispanic (n = 34) | 2 (5.9%) | 4 (11.8%) | ||
| Other (n = 10) | 0 (0.0%) | 0 (0.0%) | ||
| Prior deliveries | .63 | .0027 | ||
| None (n = 1058) | 11 (1.0%) | 65 (6.1%) | ||
| 1 (n = 766) | 4 (0.5%) | 20 (2.6%) | ||
| 2 (n = 395) | 2 (0.5%) | 14 (3.5%) | ||
| ≥3 (n = 222) | 1 (0.5%) | 9 (4.0%) | ||
| History of diabetes | .0004 | .0001 | ||
| No (n = 2350) | 13 (0.6%) | 95 (4.0%) | ||
| Yes (n = 91) | 5 (5.5%) | 13 (14.3%) | ||
| History of hypertension | < .0001 | < .0001 | ||
| No (n = 2267) | 10 (0.4%) | 82 (3.6%) | ||
| Yes (n = 174) | 8 (4.6%) | 26 (14.9%) | ||
| History of nephropathy | > .99 | .17 | ||
| No (n = 2437) | 18 (0.7%) | 107 (4.4%) | ||
| Yes (n = 4) | 0 (0.0%) | 1 (25.0%) | ||
| History of autoimmune disease | .11 | .52 | ||
| No (n = 2425) | 17 (0.7%) | 107 (4.4%) | ||
| Yes (n = 16) | 1 (6.3%) | 1 (6.3%) | ||
| History of thrombophilia | > .99 | .61 | ||
| No (n = 2420) | 18 (0.7%) | 107 (4.4%) | ||
| Yes (n = 21) | 0 (0.0%) | 1 (4.8%) | ||
| History of PET | .14 | < .0001 | ||
| No (n = 2350) | 16 (0.7%) | 91 (3.9%) | ||
| Yes (n = 91) | 2 (2.2%) | 17 (18.7%) |
Table 1 shows the rates of preeclampsia by patient demographics and clinical history. The rates of early or any PET differed significantly by maternal ethnicity and were significantly higher among those with a history of diabetes and a history of hypertension. Nulliparous women were significantly more likely to develop PET and parous women with a history of PET were almost 5 times more likely to develop recurrence.
Table 2 shows the relationship between variables measured at the first-trimester screening and rates of PET. Those with higher blood pressure (as measured by either systolic, diastolic, or mean arterial pressure) had significantly higher rates of PET. High BMI and low PAPP-A-MoM were also associated with higher rates of PET.
| Characteristic | Distribution | Number (%) with early PET | P value | Number (%) with any PET | P value |
|---|---|---|---|---|---|
| BMI | <20 (n = 145) | 0 (0.0%) | .23 | 6 (4.1%) | .0024 |
| 20-25 (n = 798) | 6 (0.8%) | 29 (3.6%) | |||
| 25-30 (n = 685) | 2 (0.3%) | 19 (2.8%) | |||
| 30-35 (n = 383) | 5 (1.3%) | 21 (5.5%) | |||
| 35+ (n = 430) | 5 (1.2%) | 33 (7.7%) | |||
| SBP | <120 (n = 1635) | 6 (0.4%) | < .0001 | 33 (2.0%) | < .0001 |
| 120-129 (n = 514) | 3 (0.6%) | 32 (6.2%) | |||
| 130-139 (n = 202) | 4 (2.0%) | 27 (13.4%) | |||
| 140+ (n = 90) | 5 (5.6%) | 16 (17.8%) | |||
| DBP | <70 (n = 1584) | 4 (0.3%) | < .0001 | 31 (2.0%) | < .0001 |
| 70-79 (n = 659) | 5 (0.8%) | 50 (7.6%) | |||
| 80-89 (n = 169) | 7 (4.1%) | 18 (10.7%) | |||
| 90+ (n = 29) | 2 (6.9%) | 9 (31.0%) | |||
| MAP | <70 (n = 78) | 0 (0.0%) | < .0001 | 2 (2.6%) | < .0001 |
| 70-80 (n = 757) | 0 (0.0%) | 6 (0.8%) | |||
| 80-90 (n = 1103) | 6 (0.5%) | 35 (0.3%) | |||
| 90+ (n = 503) | 12 (2.4%) | 65 (12.9%) | |||
| PAPP-A MoM | 1st quartile (n = 613) | 7 (1.1%) | .50 | 50 (8.2%) | < .0001 |
| 2nd quartile (n = 595) | 5 (0.8%) | 21 (3.5%) | |||
| 3rd quartile (n = 611) | 3 (0.5%) | 17 (2.8%) | |||
| 4th quartile (n = 622) | 3 (0.6%) | 20 (3.2%) | |||
| ßhCG MoM | 1st quartile (n = 625) | 7 (1.1%) | .45 | 27 (4.3%) | .87 |
| 2nd quartile (n = 601) | 2 (0.3%) | 30 (5.0%) | |||
| 3rd quartile (n = 614) | 5 (0.8%) | 27 (4.4%) | |||
| 4th quartile (n = 601) | 4 (0.7%) | 24 (4.0%) | |||
| LUA PI z-score | 1st quartile (n = 682) | 1 (0.2%) | .079 | 24 (3.5%) | .024 |
| 2nd quartile (n = 635) | 5 (0.8%) | 19 (3.0%) | |||
| 3rd quartile (n = 605) | 5 (0.8%) | 36 (6.0%) | |||
| 4th quartile (n = 519) | 7 (1.4%) | 29 (5.6%) | |||
| RUA PI z-score | 1st quartile (n = 685) | 5 (0.7%) | .37 | 35 (5.1%) | 0.41 |
| 2nd quartile (n = 649) | 2 (0.3%) | 25 (3.9%) | |||
| 3rd quartile (n = 592) | 5 (0.8%) | 30 (5.1%) | |||
| 4th quartile (n = 515) | 6 (1.1%) | 18 (3.5%) | |||
| MUA PI z-score | 1st quartile (n = 687) | 3 (0.4%) | .13 | 30 (4.4%) | 0.72 |
| 2nd quartile (n = 640) | 4 (0.6%) | 24 (3.8%) | |||
| 3rd quartile (n = 623) | 3 (0.5%) | 29 (4.7%) | |||
| 4th quartile (n = 491) | 8 (1.6%) | 25 (5.1%) | |||
| Uterine artery notching | None (n = 1504) | 8 (0.5%) | .17 | 60 (4.0%) | .32 |
| 1 (n = 488) | 4 (0.8%) | 23 (4.7%) | |||
| 2 (n = 449) | 6 (1.3%) | 25 (5.6%) |
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
