- To determine the incidence of three specific forms of maternal morbidity (e.g. Eclampsia; Cardiac arrest and Amniotic Fluid Embolism (AFE)) in a consecutive two-year timeframe in the
- To compare management strategies in cases of eclampsia compared to a control group of women with severe preeclampsia without eclampsia.
- To design a prediction model for eclampsia.
After identification of a case, anonymised photocopies of all relevant case-files will be mailed to the principal investigator, showing only the study-ID obtained after contacting the principal investigator. The Principal investigator will distil the necessary data from the anonymous case-files and enter the parameters in a dedicated ProMISe database from the Leiden University Medical Centre.
A control group will be constructed using the case as a reference. We will identify two controls before and two controls after the case of eclampsia. These controls are subdivided in a group of women who develop severe preeclampsia but no eclampsia and a group representing the general population.
Prof. Dr J. van Roosmalen. LUMC, Leiden
Prof A Franx, UMCU, Utrecht
Dr J.J. Zwart, Deventer ziekenhuis, Deventer
prof. Dr. J.M.M. van Lith, LUMC, Leiden
Health Technology Assessment
All interventions in cases of eclampsia and severe pre-eclampsia will be registered in a database. Analysis of time-dependent variables will be performed to compare management strategies between both group.
Using the cases and controls we will perform a logistic regression analysis to predict the primary endpoint (eclampsia) from clinical characteristics. For individual dichotomous and continuous variables, univariate pooled odds ratios and 95% confidence intervals (CI), as well as P-values will be calculated. Using this, it is possible to perform a multivariable logistic regression analysis with a stepwise backward selection of predictors to construct a prediction model. To evaluate the discriminative performance of the logistic model, the area under the receiver ľoperating characteristics (ROC) curve will be calculated, comparing actual outcomes to the outcome predicted by the model. Subsequently, we can evaluate the calibration of the prediction model by plotting observed and predicted event rates for 10 subgroups of patients on the basis of deciles of the predicted probability. Furthermore, we can assess the reliability of the model with the Hosmer and Lemeshow test for goodness of fit. Internal validation and extent of over fitting of the model can be assessed by bootstrapping.
Drs. Timme Schaap, AIOS gynaecologie & promovendus NethOSS