RESULTS
During the study period 86 women were admitted for early-onset severe PE. 68 of them had no maternal complications at admission and no fetal indication for imminent delivery. Blood samples for angiogenic factors measurement were not collected from 4 of these patients due to a breach of the study protocol. In one additional case, blood sample was not obtained at delivery. Thus, a total of 63 women were available for analysis with both at-admission and at-delivery measurements. Out of them, 23 patients (36.5%) had a complication. Table 1 shows the baseline characteristics by the presence of complications and supplementary table 1 details the maternal complications. The median time interval between admission and delivery was 9 days (ranging from 2 to 24). It was shorter in complicated pregnancies [7 (2-22)] vs [10 (2-24)] in those without complications.
Supplementary table 2 shows the levels of the angiogenic factors at admission and at delivery, with the delta values (expressing average daily change). Of note, all the evaluated angiogenic factors significantly differed between admission and delivery. Table 2 and supplementary figure 1 depict the angiogenic factors levels in uncomplicated and complicated pregnancies. Only the sFlt-1 delta values were significantly different between those without and with complication (median values: 343.7 vs. 1079.5 pg/mL/day; p=0.04). Expressed as relative changes from the admission values, this corresponds to 2.6% vs. 8.2% daily increase. Supplementary figure 2 shows the individual and averaged changes of sFlt-1 by the occurrence of complications.
On the multivariate analysis, the baseline model (PREP-L risk score and sFlt-1 at admission) explained 6.6% of the uncertainty of complications occurrence (R2 Naegelkerke). The addition of sFlt-1 delta values significantly improved this performance (R2 Naegelkerke of 23.2%; p=0.004). Table 3 shows the logistic regression model which includes all three predictors. While the area under the curve of the baseline model was 0.6 (95%CI 0.46-0.75), it increased to 0.75 (95% CI 0.62-0.87) when the sFlt-1 delta was added. For a 20%, 30% and 40% of false-positive rates, detection rates were better for the model including sFlt-1 delta: 57.1% vs. 33.3%; 66.7% vs. 38.1%; and 76.2% vs. 47.6%, respectively. Figure 1 shows both ROC curves.
Figure 2 shows the Kaplan-Meier graph of the time interval to delivery: the median time was 3 days (95%CI 1.96-4.05) in those in the highest quartile of delta sFlt-1 values vs. 10 (95%CI 8.1-11.9) in those in the 1st-3rd quartiles [Log Rank (Mantel-Cox) p<0.001].
Table 3 shows the Cox regression modelling of the time interval from admission to delivery. Of note, to meet the proportional hazards assumption, log-transformed sFlt-1 delta values had to be categorized in quartiles and entered in the model as binary (highest quartile vs. those the 1st-3rd quartile): hazard ratio of 3.15 (95%CI 1.67-5.86; p<0.001).