uva-dare (digital academic repository) who is (not) at ... · z íì | ^µuu Ç bw_kleinrouweler...

UvA-DARE is a service provided by the library of the University of Amsterdam (http://dare.uva.nl)

UvA-DARE (Digital Academic Repository)

Who is (not) at risk? Prognostic tests and models in obstetrics with a focus on pre-eclampsiaand preterm birth

Kleinrouweler, C.E.

Link to publication

Citation for published version (APA):Kleinrouweler, C. E. (2013). Who is (not) at risk? Prognostic tests and models in obstetrics with a focus on pre-eclampsia and preterm birth.

General rightsIt is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s),other than for strictly personal, individual use, unless the work is under an open content license (like Creative Commons).

Disclaimer/Complaints regulationsIf you believe that digital publication of certain material infringes any of your rights or (privacy) interests, please let the Library know, statingyour reasons. In case of a legitimate complaint, the Library will make the material inaccessible and/or remove it from the website. Please Askthe Library: https://uba.uva.nl/en/contact, or a letter to: Library of the University of Amsterdam, Secretariat, Singel 425, 1012 WP Amsterdam,The Netherlands. You will be contacted as soon as possible.

Download date: 08 Aug 2019

https://dare.uva.nl/personal/pure/en/publications/who-is-not-at-risk-prognostic-tests-and-models-in-obstetrics-with-a-focus-on-preeclampsia-and-preterm-birth(7448a8c3-284d-433d-b11f-c1e627426df2).html

Chapter 10|Summary

BW_Kleinrouweler 2.indd 169 05-11-13 16:43

Kleinrouweler_PROEFSCHRIFT (all).ps Front - 85 T1 - Black CyanMagentaYellow


Kleinrouweler_PROEFSCHRIFT (all).ps Back - 85 T1 - Black CyanMagentaYellow

Summary

171

Chapter

10

Effective and efficient care relies on accurate and individualized estimates of risk, or the

probability, that a certain health outcome (favourable or unfavourable) occurs. Thereby only

those patients who are at high risk for an outcome can be selected to undergo a procedure (e.g.

only women with a high probability of successful external cephalic version or vaginal birth after

a caeserean section) or to receive (preventive) treatment or intervention (only women with a

high risk for preterm birth receive progesterone or cerclage, and only women with a high risk

of pre-eclampsia are monitored more frequently). Prognostic factors, or more than one factor

combined in a multivariable model, can be used to estimate such risks.

Indeed many scientific efforts have been performed to develop prediction models for various

obstetrical outcomes, which forms the scope of part 1 of this thesis.

In chapter 2 a systematic review is presented that gives an overview of available prediction models

in obstetrics. A total of 263 prediction models for 40 different outcomes were identified with widely

varying estimates of model discrimination (area under the receiver operating characteristics

curve (AUC) ranged between 0.56 and 0.99), although discrimination and calibration were only

reported for 63% and 17% of models, respectively. The main focus in this field seems to be on

model development, as for many outcomes there was more than one available model (up to 69

models that were developed for pre-eclampsia and 63 models for preterm birth) but externaI

validation was assessed for only 23 (8.7%) of the prediction models and we did not identify any

studies investigating model impact. Further investigation of validity and impact of a model may

be firstly hampered by poor reporting of a prediction rule or score that allows others to use the

model (missing in 38%) and secondly by insufficiently described recommendations on how the

model should be used in clinical practice (i.e. management of women with risks below and above

a certain risk threshold, missing in 89%). We conclude that since crucial information, namely the

external validity and impact of prediction models on clinical outcomes, is lacking, the conclusion

as to whether their use would be beneficial to clinical practice cannot be drawn.

Chapter 3 presents the use of the ‘fullPIERS model’ to estimate risk for adverse maternal

outcomes in women admitted with pre-eclampsia and to investigate whether changes in risk

during the initial 48 hours of admission can predict adverse outcomes occurring within 2-7 days.

We used data of 846 of the 1935 women from the original fullPIERS study cohort, i.e. those that

had two assessments in the first 48 hours of admission for pre-eclampsia. Adverse outcomes

occurred in 32 women (3.8%) within 48 hours of admission and in 84 women (9.9%) within 7

days of admission. Risk changes over time in the first 48 hours (from the first to the second

assessment after admission) had poor prognostic accuracy (all AUC ≤0.57). However, accuracy

improved from the first assessment as a single entity (AUC 0.54 for adverse outcomes within

48 hours) to the second assessment (AUC 0.64) as more recent information became available,

indicating the added value of repeated assessments. Further research will have to determine the

optimal between-test interval for women in whom delivery is not immediately indicated.



Chapter 10

172

Part 2 of this thesis focuses on prediction of pre-eclampsia. Many, often small, studies have been

performed to investigate the accuracy of patient characteristics, uterine Doppler ultrasound or

circulating proteins in maternal blood, or their combinations, to predict pre-eclampsia. Smaller

studies increase the probability of overoptimistic or inconsistent findings, and a current synthesis

of available evidence was lacking for several questions. We performed three meta-analyses to

get a clearer view of the current available evidence.

Chapter 4 describes an individual patient data meta-analysis of eight datasets, in total reporting

on 6708 nulliparous women, of whom 302 (4.5%) developed pre-eclampsia. Adding the results

of second trimester uterine artery Doppler measurements (pulsatility index or resistance index

combined with bilateral notching) to patient characteristics (systolic blood pressure and/or

body mass index) improved the identification of women at increased risk for pre-eclampsia.

Discrimination with models including both Doppler and patient characteristics was significantly

higher (AUCs up to 0.85) than with models based on patient characteristics (AUCs up to 0.67) or

Doppler characteristics alone (AUCs up to 0.78). Based on calibration plots, women with risks over

10-15% could be well differentiated from women with lower risks, but using this risk threshold of

10-15% would also imply that the far majority of women (85-90%) considered to be at high risk

would be worried or treated unnecessarily. Considering that there is no definite treatment other

than delivery and that antiplatelet agents or aspirin started in the second trimester of pregnancy

can probably prevent not more than 10% of cases of pre-eclampsia, we conclude that at present

we cannot recommend routine Doppler ultrasound (in combination with patient characteristics)

as a risk assessment strategy for pre-eclampsia.

Chapter 5 presents a systematic review and meta-analysis of 34 studies to investigate the accuracy

of circulating placental growth factor (PlGF), vascular endothelial growth factor (VEGF), soluble

fms-like tyrosine kinase 1 (sFLT1) and soluble endoglin (sENG) in the prediction of pre-eclampsia.

Measured in serum or plasma of pregnant women before 30 weeks of gestation and before (any)

clinical symptoms or signs of pre-eclampsia became apparent, mean concentrations of PlGF (27

studies) were significantly lower, those of VEGF (3 studies) were lower but not significantly so,

and those of sFLT1 (19 studies) and sENG (10 studies) were significantly higher in women that

subsequently developed pre-eclampsia. However, test accuracies of all four markers were poor,

with summary diagnostic odds ratios of 9.0 for PlGF (95% CI 5.6 – 14.5), 6.6 for sFLT1 (95% CI

3.1 – 13.7) and 4.2 for sENG (95% CI 2.4 – 7.2), which correspond to sensitivities of 32%, 26% and

18%, respectively, for a 5% false-positive rate. Based on these findings, accurate prediction of

pre-eclampsia in clinical practice based on one of these markers alone is not possible.

Chapter 6 covers a systematic review and meta-analysis of 33 studies that have identified

differentially expressed genes in the pre-eclamptic placenta, compared to placental tissue of

normotensive women. Using a vote-counting method based on a comparative meta-profiling

algorithm, we determined a meta-signature that characterizes the significant intersection of

differentially expressed genes from 26 independent mRNA gene signatures and 4 microRNA



Summary

173

Chapter

10

(miRNA) signatures. Covering a total of 254 pre-eclampsia samples and 278 control samples, the

mRNA signature consisted of 40 genes of which 14 transcripts were not reported previously in

relation to pre-eclampsia. Ten of these 14 are also expressed in the first trimester placenta and

three encode a secreted protein, which makes them interesting targets for further investigation

for use as a biomarker. Covering 39 pre-eclampsia samples and 38 control samples, the miRNA

meta-signature consisted of 17 miRNAs. As miRNAs have a regulatory function over mRNA gene

expression, further research into the combined differential expression of the identified mRNA

and miRNA transcripts might also provide further insights in the pathogenesis of pre-eclampsia

and the role of (gene expression in) the placenta.

Part 3 covers two studies that aimed to identify women at high risk and low risk, respectively, for

preterm birth.

In chapter 7 a case-control study is presented of 129 women with spontaneous preterm births

before 32 weeks of gestation (cases) and 129 matched controls, identified from the obstetric

database of the Academic Medical Center. All women had a singleton pregnancy and underwent

an ultrasound measurement of fetal crown-rump length (CRL) between 8+0 and 13+6 weeks

of gestation, had a regular menstrual cycle of 28±4 days and a certain first day of their last

menstrual period. In contrast to previous studies that found an association with a short CRL in

the first trimester (i.e. shorter than expected based on pregnancy dating using the last menstrual

period) and preterm birth (spontaneous or indicated), we could not find this association in our

group of only spontaneous preterm births. CRL (in multiples of the median (MoM) of expected

CRL and adjusted for factors known to influence CRL, included those matched for) had an odds

ratio for preterm birth of 1.10 (95% CI 0.89–1.36) per 0.10 point increase in CRL-MoM. The

association was not influenced by timing of the ultrasound. We conclude that the finding of

short CRL in early pregnancy does not indicate intensive monitoring or treatment for increased

preterm birth risk.

Chapter 8 describes a pooled analysis of individual patient data from seven studies. Included were

1316 asymptomatic women with one or more prior spontaneous preterm births who underwent

transvaginal cervical length measurements between 18 and 24 weeks of gestation. Recurrence

rates of preterm birth before 32, 34 and 37 weeks were 9.1%, 14% and 31%, respectively. Cervical

length and details of the obstetric history (gestational age of the earliest and most recent prior

spontaneous preterm birth, gestational age of the most recent birth (term or preterm), and

number of prior spontaneous preterm births) were associated with recurrent preterm birth and

could stratify women into distinct risk groups with risks of preterm birth before 37 weeks ranging

between 15-20% and 100%. Still, even in the group at lowest risk, the risks are considerably

higher than those of women with no prior spontaneous preterm birth (4-8%). Thus, we were not

able to identify a subgroup of women in whom withholding treatment (progestagens, cerclage

or pessary) can be advised. Future research should investigate if longitudinal changes in cervical



Chapter 10

174

length, or sequential cervical length measurements, can differentiate between impending

preterm birth and uneventful continuing pregnancy.

Implications for clinical practice and recommendations for future research

Overlooking the results of the studies presented in this thesis it is our impression that so far,

efforts to predict outcomes in obstetrics have not moved far beyond the development phase. We

did not find evidence that any test or prediction model described here (in obstetrics in general,

or for pre-eclampsia or preterm birth in particular) would improve clinical outcomes. Thus, at this

stage we cannot advise a new model or test that should be used in clinical practice. However,

individualized care would be difficult without patient-specific risks and prediction models and

they will likely always be a part of medicine; but first further research into their benefit for

improving clinical outcomes is needed.

We would like to give several recommendations for further research. We encourage researchers

to perform external validation studies and impact studies (comparing care strategies with

and without the use of the model) for existing models. In addition, we advise to first critically

question the added value of a new model before undertaking developing efforts and adhere to

recommended methods for model development. Moreover, reporting should be transparent,

proper and comprehensive to enable assessment of methodological study quality and use of the

results in the future. We argue in favour of a registry of current and planned observational studies

that allows for coordination of efforts and an overview of all available evidence and data (even

from unfinished or unpublished studies) for inclusion in new analyses or meta-analyses. Apart

from registering the study itself, its data should be stored to enable analysis of individual patient

data (IPD), or the original, ‘raw’ data from the study. Thereby final conclusions can be drawn or

new hypotheses can be tested (for example: multivariable models or subgroup-analyses) within

relatively short time as there is no need to collect data from new patients. We will promote

awareness and encourage collaboration for this approach. Apart from prediction of an outcome,

prediction models could be valuable to predict the effect of treatment; thus potential predictive

variables should be measured in all patients in a cohort or trial investigating (new) treatments.



uva-dare (digital academic repository) who is (not) at ... · z íì | ^µuu Ç bw_kleinrouweler...

Documents