A two-stage landmark competing risks approach: A prediction model for predicting preterm preeclampsia in a retrospective pregnancy cohort.

OBJECTIVES

This study develops a prediction model for preterm birth with preeclampsia, addressing challenges such as different gestational age at first presentation, time-dependency, and births without preeclampsia.

STUDY

DESIGN

Data were obtained from a retrospective real-world pregnancy cohort.

MAIN OUTCOME MEASURES: A two-stage landmark prediction model was constructed. First, longitudinal trajectories of biomarkers (mean arterial pressure, mean pulsatility index of uterine arteries, soluble fms-like tyrosine kinase-1 to placental growth factor ratio) were modeled at predefined landmark time points using a linear mixed-effects model. Second, based on Cox proportional hazards models the probability of preterm preeclampsia was predicted.

RESULTS

Of 2,930 pregnancies, 311 resulted in a preterm delivery with current diagnosis of preeclampsia. Unadjusted estimates yielded a probability of preterm delivery with preeclampsia of 25% (95% confidence interval 21-28%). Across all landmark time points, the model demonstrated good sensitivity (76-89%) and specificity (65-77%) for predicting preterm preeclampsia by gestational week 37 and within the next two weeks (sensitivity 61-89% and specificity 65-95%). However, the low positive predictive value (11-23%) indicates a high false-discovery rate. Compared to a simple strategy based solely on the soluble fms-like tyrosine kinase-1 to placental growth factor ratio, the new prediction model showed disadvantages for detection of preterm preeclampsia by gestational week 37 but advantages for preterm preeclampsia within the next two weeks.

CONCLUSION

The two-stage landmark prediction model may support early identification of pregnancies at elevated risk for preeclampsia in high-risk cohorts. However, the high false-positive rate might result in an over-surveillance.