Interpretation
In the SCOPE study, a predictive model for preeclampsia in nulliparous women based on maternal clinical predictors including MAP at 15 weeks’ gestation, AUC under internal validation was 0.71.36 This was similar to our findings in nulliparous women with an AUC of 0.69 in gestational week 24. In our previous study of preeclampsia prediction in early pregnancy, we found an AUC of 0.68 and 0.67 for preterm and term preeclampsia, respectively.44
In the latest United States Preventive Services Task Force (USPSTF) recommendations, screening for preeclampsia with blood pressure measurements throughout pregnancy is emphasised.2 Previous knowledge suggests that prehypertension, and blood pressure patterns throughout pregnancy, may be important for prediction of preeclampsia.21–24 We have not found any longitudinal predictive studies of preeclampsia restricted to nulliparous women. A previous study using maternal characteristics and serial blood pressure measurements in a multivariable model (all parities), demonstrated improved prediction of preeclampsia from gestational week 28 and onwards (AUC 0.79 [95% CI 0.77–0.82] in gestational week 24, to 0.88 [95% CI 0.86–0.90] in week 36).18 Our findings in nulliparous women displayed a similar improvement of preeclampsia prediction, although including additional longitudinal predictors to blood pressure measurements. Prediction of both preterm and term preeclampsia can be improved by using maternal factors and serial MAP, compared to MAP from only one trimester.45
The USPSTF’s preeclampsia screening recommendations further emphasize the need for high quality studies and models using parameters available in routine care.2 Early pregnancy clinical risk factors are well established.46 However, there is also a body of evidence supporting the importance of incorporating pregnancy trajectories of clinical examinations. Examples of factors are: weight gain during pregnancy,27–29 especially for the risk of term preeclampsia,27 symphysis-fundal height, since fetal growth restriction is strongly associated with preeclampsia,1 hyperglycemia and gestational diabetes.30,31 Isolated gestational proteinuria is associated with preeclampsia,32,33 and so is increased hemoglobin concentration in second trimester (proxy for plasma volume expansion).26,47 To our knowledge, no previous study has addressed the predictive capacity of these variables in a multivariable model.
Late-onset preeclampsia (delivery ≥ 34 weeks’ gestation), comprise the majority of preeclampsia cases, and is 3–7 times more common than the generally more severe early-onset preeclampsia.48 Late-onset preeclampsia is nevertheless associated with fetal growth restriction, perinatal morbidities and deaths, and maternal eclamptic seizures.48,49 Irrespective of using MAP in first, second, or third trimester, the predictive capacity for term preeclampsia is consistently lower compared to preterm preeclampsia.15,16,50,51 This is accordance with our results, but we found a major improvement for prediction of term preeclampsia by adding serial information. Close monitoring of the high risk group in third trimester enables diagnosis of hypertensive disorders at an early stage, and improve perinatal outcomes by both optimized treatment, and selection of appropriate time, place and method of delivery.7 Compared to expectant management, planned delivery from 34 to 37 weeks’ gestation is associated with reduced maternal morbidity in women with mild hypertensive disease without adverse neonatal outcomes among term pregnancies.7,52
Biophysical examinations and numerous biomarkers have been proposed in second and third trimester prediction of preeclampsia. Improved prediction can be reached by adding second and/or third trimester uterine artery ultrasound examinations, and some biomarkers have displayed fairly good risk discrimination when used in second or third trimester, especially in combination with maternal characteristics.34,53−57
Strengths and limitations
The strengths of this study include our linkage of multiple population registers including electronic medical records during antenatal, delivery and postpartum care and inpatient/outpatient visit records. The comprehensive range of prospectively measures in a standardized way, including serial medical examinations throughout pregnancy, compose a distinct strength. There is generally a minimal level of missing values and the data is population-based on a large population increasing the likelihood of accurate prediction and allowed us to study preeclampsia subtypes.
The analytical approach efficiently exploits the information jointly contained in the baseline and longitudinal predictors. The shared-effects joint longitudinal model utilizes all of the longitudinal measurements on each woman, irrespective of timing and number of observations, without using imputed data. These updated predictions are potentially more precise than those from traditional prognostic methods where routinely information is not evaluated in a multivariable or longitudinal fashion.
Several limitations should be noted. Analysis of secondary data did not make it possible to assess the potential misclassification of the mainly self-reported maternal characteristics. This would however probably reflect the outcomes of the model in the clinical setting. The medical examinations are performed according to guidelines by trained midwives in antenatal care. Blood pressure measurements may though have been rounded to closest 5 or 10 when registered in the medical record, influencing the specificity of the prediction.
The use of ICD-10 codes for preeclampsia instead of data from medical records can introduce misclassification bias. In order to improve the accuracy of the diagnosis, one diagnosis in in-patient or two diagnoses in outpatient care was required. The Swedish version of ICD-10 diagnoses defined preeclampsia with mandatory proteinuria during the study period, which is less sensitive but more specific compared to current international recommendations of the diagnosis.11 Overall rates of preeclampsia in nulliparous women in our study were consistent with previous populations from western countries.36,37,53