The systematic review «Al-Rubaie Z, Askie LM, Ray JG ym. The performance ...»1 of Al-Rubaie et al (2016) aimed to compare simple models including only clinical risk factors to methods combining clinical risk factors, biomarkers and uterine artery Doppler measurements before 17th week of gestation in singleton pregnant women. 29 studies reporting 70 screening models predicting early-onset (< 34 weeks) or late-onset (≥ 34 weeks) pre-eclampsia (PE) published until 2014 were included. 22 simple models included most commonly parity, race, history of hypertension, body mass index and blood pressure and 48 combined models additionally included biomarkers (pregnancy-associated plasma protein A) (PAPP-A), human chorionic gonadotrophin (hCG), placental growth factor (PlGF), placental protein (PP) -13) and/or uterine artery Doppler measurements. 14 study populations were used for development of the models, two of these had a low risk of bias, both of which reported simple models. Twelve studies were classified as high risk of bias due to lack of external or internal validation, not reporting on model calibration or accounting for overfitting. Six studies were externally validated.
Four externally validated studies showed better discrimination for early-onset (AUC 0.74-0.76, sensitivity 29-40 % at 90 % specificity) than late-onset PE (AUC 0.65-0.72, 22-23 % sensitivity at 90 % specificity) when using simple models. Six validation studies including combined model based on the same UK development population showed no discriminatory power to predict any type of PE. Models including clinical risk factors, uterine artery pulsatility index (PI) and/or PAPP-A, PlGF or PP-13 discriminated women at risk to preterm PE moderate to good (AUC 0.70-0.93). To predict late-onset PE with combined models, two externally validated models showed good discrimination (AUC 0.75-0.93), but to be noticed, the incidences of late-onset PE were notably high (9 %) in these populations. Two models showed only moderate discrimination (AUC 0.64-0.70) for late-onset PE.
Sixteen of 17 models in 9 studies directly comparing simple models to combined models reported better discrimination when using combined models. Improvement in AUC values ranged from -0.005 to 0.240 when using combined models and the discrimination was better for early-onset PE than late-onset PE. Median difference in sensitivity was 18 % (range 0-56 %) at 90-95 % specificity when comparing combined models to simple models.
US Preventive Services Task Force Recommendation (2017) included a systematic review «Henderson JT, Thompson JH, Burda BU, Cantor A, Bei...»2 of four validation studies based on three development studies determining accuracy of first trimester combined screening models to predict early-onset or preterm PE. Validation populations included 541–2 833 women in UK, Australia, US and Norway. Early-onset PE (< 34 wk) was the primary outcome in 3 studies and preterm PE (< 37 wk) in one study. Incidences were 0.4-1.2 % for PE. Sensitivities to predict early-onset or preterm PE were 52, 80, 80 and 92 % at 90 % specificity. PPVs were 3.6, 4.2, 6.8 and 11.3; and NPVs were 99.6-99.9. Screening models included several clinical characteristics (race, chronic hypertension, parity, blood pressure, conception mode in 3 studies; PAPP-A in all; PlGF in one; PP-13 in one and Doppler ultrasound uterine artery PI in all studies).
In a prospective multicenter ASPRE validation study «O'Gorman N, Wright D, Poon LC ym. Multicenter scre...»3 O'Gorman et al (2018) compared clinical risk factors determined using National Institute for Health and Care Excellence (NICE) and American College of Obstetricians and Gynecologists (ACOG) guidelines to combined screening using maternal clinical risk factors, mean arterial pressure, biomarkers (PlGF and PAPP-A) and mean uterine artery PI defined by Fetal Medicine Foundation (FMF) in the screening of PE (Table «Clinical risk factors according to NICE guidelines, ACOG recommendations, FMF algorithm and simple risk model from the POP study....»1). 8 775 singleton pregnant women were screened at 11+0 - 13+6 weeks of gestation for PE (n = 239, 2.7 %) and preterm PE (< 37 weeks, n = 59, 0.7 %). The highest sensitivity in the prediction of preterm PE was 75 % at 90 % specificity by using FMF model. Sensitivity was 39 % when using the NICE criteria (at 90 % specificity) and sensitivity was 90 % when using ACOG criteria with 36 % specificity. Screen-positive rates were 10 % with FMF and NICE criteria and 64 % with ACOG criteria (Table «Performance of the models predicting preterm PE (< 37 weeks). Comparison between the prospective studies by O'Gorman, Tan, Sovio and Sandstrom....»2 and «O'Gorman N, Wright D, Poon LC ym. Multicenter scre...»3: Table «Clinical risk factors according to NICE guidelines, ACOG recommendations, FMF algorithm and simple risk model from the POP study....»1). Positive predictive values in predicting preterm PE were 4.8 for FMF, 2.5 for NICE and 0.9 for ACOG; and negative predictive values were 99.8 %, 99.5 % and 99.8 %, respectively (Table «Performance of the models predicting preterm PE (< 37 weeks). Comparison between the prospective studies by O'Gorman, Tan, Sovio and Sandstrom....»2).
In the prospective UK multicenter SPREE study «Tan MY, Wright D, Syngelaki A ym. Comparison of di...»4 of Tan et al (2018) 16 747 singleton pregnant women were screened for preterm PE (< 37 weeks) using NICE criteria or combined methods. NICE criteria were compared to mini-combined model including NICE criteria, mean arterial blood pressure (MAP) and/or PAPP-A and FMF screening method (Table «Clinical risk factors according to NICE guidelines, ACOG recommendations, FMF algorithm and simple risk model from the POP study....»1). Screen-positive rate of 10 % was observed for NICE criteria and was used for each model. Sensitivity to predict preterm PE was 41 % when using NICE criteria, 53 % when using NICE + MAP + PAPP-A and 82 % when using FMF method (at fixed 90 % specificity). Positive and negative predictive values were 3.4 and 99.4; 4.4 and 99.6; and 6.8 and 99.8, respectively (Table «Performance of the models predicting preterm PE (< 37 weeks). Comparison between the prospective studies by O'Gorman, Tan, Sovio and Sandstrom....»2). Improvement in sensitivities were observed between the models: Differences in sensitivities were 41.6 % comparing FMF to NICE criteria and 12.7 % when comparing NICE + MAP + PAPP-A to NICE criteria. To be noticed, 749 (4.5 %) women used acetylsalicylic acid 75-150 mg daily between < 14 and 36th week of gestation according to local routine clinical standards. After adjustment of ASA usage the results did not change significantly.
The prospective Pregnancy Outcome Prediction (POP) study «Sovio U, Smith G. Evaluation of a simple risk scor...»5 by Sovio et al (2019) screened 4 184 singleton nulliparous pregnant women for preterm PE (< 37 weeks) using three clinical risk factor based methods. Binary NICE criteria that categorizes risk as "at risk" or "no risk", clinical risk score based on FMF algorithm and clinical risk score adapted for nulliparous women (see «Sovio U, Smith G. Evaluation of a simple risk scor...»5: Table «Clinical risk factors according to NICE guidelines, ACOG recommendations, FMF algorithm and simple risk model from the POP study....»1). Women using ASA before 20th week of gestation were excluded from the study.
Only 28 (0.67 %) women developed preterm PE requiring delivery before 37 weeks of gestation. Sensitivity to predict preterm PE was 53.6 % when using NICE criteria, 57.1 % when using clinical risk score for nulliparas and 60.7 % when using clinical risk score based on FMF method (at 90 % specificity). Positive and negative predictive values were 3.3 and 99.7; 4.2 and 99.7; and 4.1 and 99.7, respectively (Table «Performance of the models predicting preterm PE (< 37 weeks). Comparison between the prospective studies by O'Gorman, Tan, Sovio and Sandstrom....»2). Fixed screen-positive rate closest to 10 % were used. Adding either first trimester (12 week) biomarkers (PAPP-A, PlGF) to risk models did not improve the discrimination for preterm PE. AUC values were improved when adding MAP (p = 0.034 for FMF; p = 0.033 for Sovio's model) or uterine artery PI measured at 20th week of gestation (p < 0.0001) to the models.
A Swedish population based cohort study «Sandström A, Snowden JM, Höijer J ym. Clinical ris...»6 of 62 562 nulliparous pregnancies aimed to predict PE in early pregnancy by three different models based on the clinical risk factors. The predictive models in the study by Sandstrom et al (2019) were created by three methods; logistic regression models using 1) FMF algorithm 2) backward selection and 3) a Random forest model. The models 2) and 3) used 36 candidate predictors consisting maternal demografic and clinical factors (social, reproductive and medical history consisting cardiovascular, endocrinological, neurological, infectious and psychiatric conditions as well as blood type). In addition, the performance of these three methods were compared to NICE guidelines (Table «Clinical risk factors according to NICE guidelines, ACOG recommendations, FMF algorithm and simple risk model from the POP study....»1) in prediction for PE. The outcome measures were the diagnosis of PE and delivery < 34, < 37 and ≥ 37 weeks' gestation. A restricted population of 58 276 pregnancies without major anomalies and without use of aspirin during pregnancy was evaluated separately.
In the total study population, 2 773 (4.4 %) developed PE during pregnancy and of these 216 (0.3 %) developed PE with delivery < 34 weeks, 497 (0.8 %) < 37 weeks and 2 276 (3.6 %) ≥ 37 weeks, respectively. The receiver operating characteristic (ROC) curves of the variables ability to predict PE at < 34 weeks with the three different multivariable methods were 0.68 (95 % confidence interval 0.64-0.72); 0.66 (95 % confidence interval 0.62-0.70); 0.58 (95 % confidence interval 0.54-0.62), < 37 weeks 0.68 (95 % confidence interval 0.65-0.70); 0.66 (95 % confidence interval 0.63-0.68); 0.60 (95 % confidence interval 0.57-0.63) and ≥ 37 weeks 0.67 (95 % confidence interval 0.66-0.69); 0.67 (95 % confidence interval 0.66-0.68); 0.61 (95 % confidence interval 0.60-0.62), respectively. The results of restricted population did not differ statistically compared with total population.
When using the binary NICE-guidelines risk classification system for identifying women at risk of PE in this population, 5.8 % of all nulliparous women would have been classified as high risk (screen positive). The detection rate for PE with delivery < 34 weeks would have been 22.2 % (95 % confidence interval 16.8-28.4), < 37 weeks 19.5 % (95 % confidence interval 16.1-23.3) and ≥ 37 weeks 12.2 % (95 % confidence interval 10.9-13.7), all with a fixed false-positive rate (FPR) of about 5.6 %. The detection rate using FMF algorithm were 25.8 % at 10 % FPR rate for the prediction of PE < 37 weeks. PPV and NPV for the prediction of PE < 37 weeks were 2.8 and 99.3 by using NICE criteria and 2.0 and 99.3 by using FMF algorithm (Table «Performance of the models predicting preterm PE (< 37 weeks). Comparison between the prospective studies by O'Gorman, Tan, Sovio and Sandstrom....»2).
NICE a | ACOG 2015 b | FMF algorithm c | POP study d |
---|---|---|---|
a NICE (National Institute for Health and Care Excellence) allows dichotomic estimation "at risk" or "no risk" and recommend low-dose aspirin
treatment for "at risk" patients for PE b ACOG (American College of Obstetricians and Gynecologists): According to ACOG, use of aspirin should be reserved for women with history of PE in two or more previous pregnancies or PE requiring delivery < 34weeks of gestation. Risk factors only helps to identify the patients who are in increased risk to develop PE. c FMF (Fetal Medicine Foundation) algorithm: allows estimation of individual patient-specific risks for PE combining maternal demographic characteristics with biophysical and biochemical factors. d POP (Pregnancy Outcome Prediction) ( «Sovio U, Smith G. Evaluation of a simple risk scor...»5): study allows simple risk score estimation derived from ASPRE algorithm modified by certain coefficients (see also «Sovio U, Smith G. Evaluation of a simple risk scor...»5: Table «Clinical risk factors according to NICE guidelines, ACOG recommendations, FMF algorithm and simple risk model from the POP study....»1). BMI, body-mass index; IVF, in-vitro fertilization; SLE, systemic lupus erythematosus; APS, antiphospholipid syndrome; MAP, mean arterial blood pressure; PAPP-A, pregnancy-associated plasma protein A; PlGF, placental growth factors; UtAPI, uterine artery pulsatility index |
|||
One of the following: | Nulliparity | Parity | Age |
History of hypertensive disease in previous pregnancy | Age > 40 years | Age | Height |
Chronic hypertension | Interpregnancy interval > 10 years | Smoking | Ethnicity |
Chronic kidney disease | BMI ≥ 35 kg/m2 | BMI | Chronic hypertension (CH) |
Diabetes type I or II | Conception by IVF | IVF | SLE or APS |
Autoimmune disease | History of previous pregnancy with PE | History of previous pregnancy with PE | IVF |
Or two of the following: | Family history of PE | Mother's PE | If no CH, add |
Nulliparity | Chronic hypertension | Chronic hypertension | Weight |
Age ≥ 40 years | Chronic kidney disease | Ethnicity | Family history of PE |
BMI ≥ 35 kg/m2 | Diabetes type I or II | Diabetes type I or II | Diabetes type 1 or 2 |
Family history of PE | SLE | SLE / APS | |
Interpregnancy interval > 10 years |
Thrombophilia | MAP | |
PAPP-A, PlGF, mean UtAPI |
Sens. | Spes. | PPV | NPV | Screen-pos. | Accuracy | LR+ | LR- | |
---|---|---|---|---|---|---|---|---|
PPV, positive predictive value; NPV, negative predictive value; LR +, positive likelihood ratio; LR -, negative likelihood ratio; NICE, National Institute for Health and Care Excellence; ACOG, American College of Obstetricians and Gynecologists; MAP, mean arterial blood pressure; PlGF, placental growth factor; UtAPI, uterine artery pulsatility index; FMF, Fetal Medicine Foundation | ||||||||
% | % | % | % | % | % | |||
O'Gorman 2017 «O'Gorman N, Wright D, Poon LC ym. Multicenter scre...»3 | ||||||||
Clinical: NICE | 39.0 | 89.8 | 2.5 | 99.5 | 10.4 | 89.5 | 3.8 | 0.68 |
Clinical: ACOG 2015 | 90 | 35.8 | 0.9 | 99.8 | 64.4 | 36.2 | 1.4 | 0.28 |
Clinical + MAP + PIGF + UtAPI (FMF) | 74.6 | 90.0 | 4.8 | 99.8 | 10.4 | 89.9 | 7.5 | 0.28 |
Tan 2018 «Tan MY, Wright D, Syngelaki A ym. Comparison of di...»4 | ||||||||
Clinical: NICE | 40.8 | 89.9 | 3.4 | 99.4 | 10.3 | 89.5 | 4.1 | 0.66 |
Clinical + MAP | 49.3 | 90.3 | 4.2 | 99.5 | 10.0 | 90.0 | 5.1 | 0.56 |
Clinical + MAP + PAPP-A | 53.5 | 90.1 | 4.4 | 99.6 | 10.3 | 89.7 | 5.4 | 0.52 |
Clinical + MAP + PIGF | 69.0 | 90.5 | 5.9 | 99.7 | 10.0 | 90.3 | 7.3 | 0.34 |
Clinical + MAP + UtAPI | 73.9 | 90.5 | 6.3 | 99.8 | 10.0 | 90.4 | 7.8 | 0.29 |
Clinical + MAP + PIGF + UtAPI (FMF) | 82.4 | 90.3 | 6.8 | 99.8 | 10.3 | 90.2 | 8.5 | 0.19 |
Sovio 2019 «Sovio U, Smith G. Evaluation of a simple risk scor...»5 | ||||||||
Clinical: NICE | 53.6 | 89.4 | 3.3 | 99.7 | 10.9 | 89.1 | 5.1 | 0.52 |
Clinical score for nullparas (Sovio) | 57.1 | 91.2 | 4.2 | 99.7 | 9.1 | 91.0 | 6.5 | 0.47 |
Clinical risk factors of FMF | 60.7 | 90.4 | 4.1 | 99.7 | 9.9 | 90.2 | 6.3 | 0.43 |
Sandstrom 2019 «Sandström A, Snowden JM, Höijer J ym. Clinical ris...»6 | ||||||||
Clinical: NICE | 19.5 | 94.5 | 2.8 | 99.3 | 5.6 | 93.9 | 3.5 | 0.85 |
Clinical: FMF | 25.8 | 90.0 | 2.0 | 99.3 | 10.1 | 89.5 | 2.6 | 0.82 |