Prediction of preeclampsia before 11th week of gestation: a secondary analysis of the ASPIRIN trial

Abstract

Background: Early screening for preeclampsia is crucial for preventing adverse maternal and fetal events. Current first-trimester algorithms for predicting preeclampsia are designed to evaluate individual risk between 11.0 and 13.6 weeks of gestation based on various maternal characteristics while integrating biophysical and biochemical features. However, there is limited information regarding risk assessment during earlier stages of pregnancy (i.e., <11.0 weeks gestation).

Objective: To develop a prediction model for preeclampsia/eclampsia before 11.0 weeks of gestation as a proof-of-concept in a secondary analysis of the ASPIRIN trial.

Study design: This study is a secondary analysis of the ASPIRIN trial, a multinational, randomized, double-blind, placebo-controlled trial. The ASPIRIN trial database, obtained from NICHD DASH, included 11,976 nulliparous pregnant women aged 18-40 with gestational ages of 6.0-13.6 weeks at randomization. Participants were assigned to receive either aspirin (81 mg/day) or placebo until 36.0 weeks or delivery. This secondary analysis included pregnancies delivered at ≥20.0 weeks, excluding those in the aspirin group or with gestational ages ≥11.0 weeks at enrollment. The composite outcome was preeclampsia/eclampsia, as reported in the ASPIRIN trial. Predictor variables available in the dataset included maternal age, education (4 levels), body mass index (BMI kg/m²), gravidity, baseline hemoglobin, baseline systolic blood pressure, and baseline diastolic blood pressure. Logistic regression, with logarithmic transformation for continuous variables, was used to develop the model. The area under the ROC curve with a 95% confidence interval (CI) estimated via bootstrap resampling (1,000 iterations) and the P-value of the Hosmer-Lemeshow statistical test are reported as discrimination and calibration measures. This study used the entire available sample using a complete case approach.

Results: A total of 3421 participants met the inclusion criteria, with a cumulative incidence of preeclampsia/eclampsia of 2.9% (99/3,421). Maternal age (21.96 ± 4.13 vs 20.86 ± 3.21, P<.001) and BMI (22.49 ± 4.77 vs 20.79 ± 3.55, P<.001) were significantly higher in the preeclampsia/eclampsia group. Gravidity was lower (P=.023), and hemoglobin levels were slightly elevated (11.88 ± 1.52 g/dL vs 11.50 ± 1.61 g/dL, P=.019) in the preeclampsia/eclampsia group. Educational level (P=.070), systolic blood pressure (P=.720), and diastolic blood pressure (P=.390) showed no significant differences between groups. The logistic regression model yielded an AUC of 0.69 (95% CI 0.63-0.74), and the Hosmer-Lemeshow test P-value was 0.094, indicating acceptable discrimination and calibration.

Conclusions: This proof-of-concept logistic regression model using first-trimester maternal characteristics demonstrated acceptable predictive performance for preeclampsia/eclampsia before 11.0 weeks of gestation. During this critical period, several interventions could be proposed to reduce preeclampsia risk, including medication adjustments, lifestyle changes, and appropriate referral if needed. Further studies are required to validate these findings and assess their clinical utility in different settings.

Keywords: aspirin; aspirin trial; first trimester of pregnancy; logistic regression; prediction model; preeclampsia; pregnancy.

Figure 1

Area under the ROC curve and corresponding 95% confidence interval for the proposed logistic regression predictive model

Figure 2

Area under the ROC curve for the proposed LASSO predictive model