Study identification and individual participant data acquisition
One hundred and twenty-five researchers from 73 teams in 25 countries had joined the IPPIC network by October 2017 and provided access to pseudonymised individual data for 3,674,684 pregnancies.42,103–178 The most common reason for not obtaining the IPD was not receiving a response from the author to the request to share data (28/180) ().
Flow diagram of studies included in the IPD meta-analysis, showing reasons why IPD were not obtained.
Our search up to September 2014 of reviews that evaluated the performance of single or combined tests for predicting pre-eclampsia identified 73 citations. After evaluation of the abstracts, we included 62 published reviews evaluating one or more tests for predicting pre-eclampsia (). Clinical characteristics were studied in 32.3% (20/62) of published reviews, biochemical markers were studied in 59.7% (37/62) and ultrasound markers were studied in 8.1% (5/62).
Review of reviews on predictors of pre-eclampsia
Characteristics of data sets in the IPPIC data repository
Seventy-eight data sets contributed data to the IPPIC data repository.42,103–178 More than half of the data sets received (58%, 45/78) were prospective cohort studies; 15% (12/78) were randomised controlled trials and 17% (13/78) were large prospective registry data sets or birth cohorts. One data set was IPD made up of 31 RCTs. Most of the data sets were from participants in Europe (60%, 47/78), 18% (14/78) were from North America, 6% (5/78) were from South America, 5% (4/78) were from Asia and Australia, and one (1%) was from Africa. Three of the data sets provided included participants from multiple countries, such as Argentina, Colombia, Kenya, India, Peru, Thailand and New Zealand. Ninety-seven per cent (3,570,993) of the 3,674,684 pregnancies in the IPPIC repository were singleton pregnancies. Individual data set size ranged from 42 to 1,663,167 pregnancies, and the total number of reported pre-eclampsia outcomes in each data set ranged from 0 to 4252 for early-onset pre-eclampsia, from 0 to 38,305 for late-onset pre-eclampsia and from 3 to 42,608 for any-onset pre-eclampsia (see Appendix 4). About one-third of the data sets received were on women with high-risk pregnancies only (29%, 23/78), 14% (11/78) of the data sets were on women with low-risk pregnancies and more than half (55%, 43/78) of the data sets included women with pregnancies of any risk. Detailed study characteristics of all IPPIC data sets are provided in Appendix 5 and a summary of the missing data for prioritised predictors and each pre-eclampsia outcome is provided in Appendix 6.
Prioritisation of predictors of pre-eclampsia
In April 2017, the online survey was designed and run using smartsurvey.co.uk (see Appendix 7). Ninety-eight members of the IPPIC collaborative network who had agreed to share data by this date were sent an e-mail introducing the survey and explaining the participation requirements and survey objectives. Collaborators had 7 days within which to complete the online survey.
Fifty-four candidate predictor variables were identified (37 clinical characteristics, nine biochemical markers and eight ultrasound markers) and ranked by 33 (34%) IPPIC collaborators. Seventy per cent (23/33) of responders were from Europe, 12% were from both the American (4/33) and Asian (4/33) continents, and 6% (2/33) were from Africa. A consensus group made up of five clinical academics reviewed 13 candidate predictor variables ranked by the online survey participants as being ‘moderately important’. This included eight clinical characteristic variables, two biochemical markers and three ultrasound markers. Two each of the clinical characteristic variables and ultrasound markers reviewed by the consensus group (mode of conception, substance misuse in current pregnancy, umbilical artery pulsatility index and estimated fetal weight centile) were included following assessment by the group.
Overall, fewer than half (48%, 26/54) of all assessed predictors were ranked as being important, with 54% (20/37) of clinical characteristics, 33% (3/9) of biochemical markers and 38% (3/8) of ultrasound markers being prioritised as important in predicting pre-eclampsia ().
Predictors of pre-eclampsia prioritised by online survey and consensus meeting
Quality of the IPPIC data sets
Risk-of-bias assessment using the PROBAST resulted in 77% (60/78) of the included IPD data sets being classified as having an overall low risk of bias, while 22% (17/78) were classified as having an unclear risk of bias. Only one data set (1%, 1/78) received an overall high risk of bias assessment. All of the included data sets had a low risk of bias in the domain of participant selection. For the domain of predictors, 94% (73/78) had a low risk of bias, while 1% (1/78) had a high and 5% (4/78) an unclear risk of bias assessment. The risk of bias in the outcome domain was unclear for 22% (17/78) of the included data sets and low in the rest (78%, 61/78). Detailed assessment of the risk of bias for the IPPIC data sets is presented in Appendix 8.
Characteristics of identified prediction models
From our updated literature search (up to December 2017), we identified 131 models developed to predict pre-eclampsia. About half of these (53%, 70/131) reported the model equation in the publication, and only one-fifth of all models (18%, 24/131) from 12 publications met the inclusion criteria for the external validation of their predictive performance in the IPPIC-UK data sets.115,128,147,229–237 The primary reasons for not including a model for external validation were the full prediction formula not being reported in the publication (47%, 61/131) and the absence of the predictor information in the IPPIC-UK data sets (27%, 35/131). Other reasons for not validating the models include pre-eclampsia being poorly defined in the study (8%, 10/131) and not enough events in the IPPIC-UK data set to validate the model (1%, 1/131). is the flow chart of prediction model selection for external validation, and Appendix 9 shows published pre-eclampsia prediction models reporting a model equation.
Flow chart of pre-eclampsia prediction model selection for external validation in IPPIC-UK data set using IPD meta-analysis. Reproduced from Snell et al. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, (more...)
