Placental growth factor (alone or in combination with soluble fms-like tyrosine kinase 1) as an aid to the assessment of women with suspected pre-eclampsia: systematic review and economic analysis

Epidemiology

Pre-eclampsia affects up to 5% of pregnancies, and severe PE occurs in about 1–2% of pregnancies. 7 In 2012–13, 12,356 pregnant women were admitted to hospital in England for PE, and 294 for eclampsia. 9 Maternal deaths attributable to PE have fallen,10 and in the UK and Ireland , in 2010–12, only nine deaths directly attributable to PE or eclampsia were recorded (0.38 per 100,000), although deaths from related conditions also occurred, including two caused by placental abruption (0.49 per 100,000). 11 According to Action on Pre-eclampsia, fetal mortality is much higher, and around 1000 babies die each year as a result of PE, mostly because of complications associated with early delivery. 12

Definitions of pre-eclampsia and related conditions

There is no international consensus on the criteria by which to diagnose PE and related conditions, although criteria provided by different organisations overlap, as shown in Table 1. The criteria for diagnosing PE that are relevant to the current diagnostic assessment are those provided by the National Institute for Health and Care Excellence (NICE),13 the American Congress of Obstetricians and Gynecologists (ACOG)14 and the International Society for the Study of Hypertension in Pregnancy (ISSHP),15 as these are the criteria referred to in studies of the diagnostic accuracy of placental growth factor (PlGF) tests and soluble fms-like tyrosine kinase 1 (sFlt-1) to PlGF ratio tests.

Pre-eclampsia is defined by NICE as gestational hypertension accompanied by gestational proteinuria (i.e. hypertension and proteinuria occurring during pregnancy) after week 20 of pregnancy. 13 The presence of either hypertension or proteinuria alone during pregnancy can also indicate a risk of developing PE. 13 PE is classified as early onset if it occurs before week 34 of pregnancy or as late onset if it occurs after week 34. 4

Hypertension in pregnancy is defined by NICE, ACOG and ISSHP as a systolic blood pressure of at least 140 mmHg and/or a diastolic blood pressure of at least 90 mmHg, and they all define proteinuria in pregnancy as 300 mg or more of protein in a 24-hour urine collection (which equates to 30 mg/dl or 30 mg/mmol)15 (see Table 1).

Some researchers have developed amended versions of the ACOG definition of PE. For example, the PETRA study6 (included in this diagnostic assessment, see Chapter 4) used an ‘expanded’ definition of PE, to address the limitations of the 2002 ACOG definition, in that the diagnostic criteria were considered too rigid and give insufficient recognition to the progressive nature and varied rate of development of PE. 16

Another condition associated with PE is haemolysis, elevated liver enzymes, low platelet count (HELLP) syndrome. Experts disagree about whether or not this is a variant of PE or a separate syndrome17 (see Table 1), but HELLP syndrome is diagnosed in pregnant women with signs of PE or even eclampsia, combined with laboratory findings of haemolysis (breakdown of red blood cells), elevated liver function test results and low platelet count, hence the acronym HELLP.

Women with gestational hypertension without proteinuria occasionally develop severe PE, eclampsia, HELLP syndrome, disseminated intravascular coagulation, acute renal or hepatic failure, or placental abruption. 18 Some women with isolated gestational proteinuria may also later develop hypertension and PE,19 and in these cases the PE may be more severe than in women who present with both hypertension and proteinuria. 20

Impact on pregnant women and babies

Hypertension in pregnancy carries risks for mother and baby and increases a woman’s lifetime risk of hypertension, PE in subsequent pregnancies,21 ischaemic heart disease, stroke, type 2 diabetes mellitus and venous thromboembolism. 10,22 Negative consequences of PE for the baby include fetal growth restriction and preterm birth,3 which can lead to complications including intracranial haemorrhage, nutritional compromise, necrotising enterocolitis and breathing difficulties (neonatal respiratory distress syndrome)7 and necessitate a stay in a neonatal intensive care unit (NICU). Decisions about when to deliver the baby when mothers have PE involve a balance between the best outcomes for both the mother and the baby. 23 Before 34 weeks of gestation, clinicians would aim to prolong the pregnancy long enough for the fetus to develop as much as possible before birth. Some babies die because of complications related to early delivery, and a few are stillborn. 7 Babies born early, or small for gestational age, may experience preschool developmental delays24 or be at increased risk of adult disease. 25 However, the baby may be delivered early if there is a risk that the mother may develop severe PE, HELLP syndrome, disseminated intravascular coagulation, acute renal failure, hepatic failure, placental abruption or eclampsia.

Suspected PE may have a negative impact on pregnant women if it involves hospitalisation, loss of work days and/or anxiety, and previously pre-eclamptic women, particularly those with severe PE, have reported poorer quality of life than women with normotensive pregnancies. 26,27 PE can be stressful for both parents, owing to worry about the condition of the unborn baby and the risk of morbidity and mortality as a result of preterm birth. 28 Having a condition that can deteriorate rapidly, being kept in hospital for monitoring, uncertainty about what will happen and undergoing emergency caesarean section can also cause women to experience fear, anxiety, loss of control over their situation and anxiety about future pregnancies. 29 Women’s partners and friends can be affected because of fear of losing the mother or the baby. 29 Evidence is mixed but, generally, PE or HELLP syndrome is associated with increased prevalence or severity of depression and post-traumatic stress disorder. 30 In one study31 post-partum depression was identified in 23% of women with mild PE and in 44% of women with severe PE, although the higher incidence among women who had suffered severe PE was attributed to infants’ admission to a neonatal unit or perinatal death, rather than to the severity of disease alone.

Significance for the NHS

Pregnant women need to be monitored during routine antenatal care, and should be given advice about the action that they need to take if they experience symptoms that may indicate PE. 13,32 If proteinuria is identified on a dipstick (‘qualitative’) test, a spot urinary protein-to-creatinine ratio or 24-hour urine collection is needed to quantify the level of proteinuria. Twenty-four-hour urine collection may necessitate an overnight stay in hospital, refrigeration of the urine during collection and laboratory-based analysis. Women suspected of having PE should be referred to a specialist and admitted to hospital for both maternal and fetal monitoring; if not admitted to hospital, women would need ongoing regular monitoring in case signs of PE develop. The uncertainty around PE prediction increases the economic burden on the NHS by increasing false-positive diagnoses, antenatal admissions, fetal monitoring and preterm delivery. 33 A 2011 study of women who had PE in a previous pregnancy found that there were longer maternal inpatient stays (12.97 days for women with PE in their current pregnancy compared with 5.06 days for women without PE in their current pregnancy) and more inpatient days in a neonatal or special care baby unit (14.7 days for babies whose mothers had PE in their current pregnancy compared with 1.35 days for babies of women without PE). 34

Triage PlGF test (Alere)

The Triage PlGF test is a Conformité Européenne (CE)-marked single-use fluorescence immunoassay device that is used in conjunction with the Alere Triage MeterPro (Alere Inc., San Diego, CA, USA) point-of-care analyser for the quantitative determination of PlGF in blood plasma samples. 40 The test is intended for use in conjunction with clinical judgement and other existing diagnostic tests, to aid the diagnosis of PE and to assess the level of risk for delivery arising from PE within 14 days of testing. 40 Each Triage PlGF test device contains mouse monoclonal antibodies against PlGF, fluorescent dye and stabilisers. Prior to use of the test device, a blood sample is centrifuged for around 3 minutes to obtain an ethylenediaminetetraacetic acid (used to prevent clotting)-anticoagulated plasma specimen. A 250-µl sample of plasma is then added to the Triage PlGF test device’s sample port where it reacts with fluorescent antibody conjugates and flows through the test device, via capillary action, to a measurement zone in which complexes of the fluorescent antibody conjugates are captured. 41 The test device is inserted into the Triage MeterPro analyser, which measures levels of fluorescence from the antibody–conjugate complexes. It has been reported that the test has a limit of detection of 9 pg/ml and a measurable range of 12–3000 pg/ml. 42 The test turnaround time is reported as approximately 15 minutes. 41

The Triage PlGF test is recommended for use in pregnant women with a gestational age of between 20 weeks and 34 weeks⁺⁶ days. The test cut-off points, derived from a population with suspected PE5 and recommended by the company, are shown in Table 4.

DELFIA Xpress PlGF test (PerkinElmer)

The DELFIA Xpress PlGF test is a CE-marked, solid-phase, two-site fluoroimmunometric sandwich assay for the quantitative determination of PlGF in serum samples. The test is intended as an aid to the diagnosis of PE during the second and third trimesters of pregnancy, is used in conjunction with clinical assessment and is a laboratory-based rather than a near-patient test. 43 The assay includes both immobilised and europium-labelled monoclonal antibodies, which bind to PlGF molecules present in the sample to form PlGF–monoclonal antibody complexes. The resulting europium fluorescence from each sample is proportional to the concentration of PlGF. The assay has a limit of detection of 1.9 pg/ml (measuring range 1.9–4000 pg/ml) and a limit of quantitation of 3.3 pg/ml. The assay is compatible with the PerkinElmer 6000 DELFIA Xpress random access analyser (PerkinElmer, Wallac Oy, Turku, Finland). The company advises that cut-off values for PlGF measurements obtained during the second trimester are highly dependent on gestational day and should be established by individual laboratories. 43 In the third trimester the company advises that, in addition to laboratory calculated cut-off values based on the gestational day, a fixed cut-off point of 184 pg/ml can be used. Levels of PlGF lower than 184 pg/ml indicate an elevated probability of PE developing. 43 Cut-off values were calculated in a case–control study that included samples from women with PE but without chronic hypertension. 44

Elecsys sFlt-1 to PlGF ratio test (Roche Diagnostics)

The Elecsys immunoassay measures the relative amounts of sFlt-1 to PlGF [also known as vascular endothelial growth factor receptor 1 (VEGFR1)] in serum samples from women with suspected PE. 45 The ratio is formed by combining the results from two CE-marked sandwich electrochemiluminescence immunoassays (the Elecsys PlGF and Elecsys sFlt-1 assays), which are compatible with both the Elecsys and the Cobas^® e automated clinical chemistry analysers (Roche Diagnostics Limited, Burgess Hill, UK). 46,47 The sFlt-1 to PlGF ratio is calculated and reported to the user alongside the individual assay values by the laboratory information system. The Elecsys immunoassay sFlt-1 to PlGF ratio is intended for use in aiding the diagnosis of PE in conjunction with clinical judgement and other diagnostic tests. In addition, the ratio may be used as an aid to predict PE, eclampsia and HELLP syndrome in the short term. 45 The Elecsys sFlt-1 assay has a limit of detection of 10 pg/ml (measuring range 10–85,000 pg/ml) and a limit of quantitation of 15 pg/ml. The Elecsys PlGF assay has a limit of detection of 3 pg/ml (measuring range 3–10,000 pg/ml) and a limit of quantitation of 10 pg/ml. 42 The time required to measure the sFlt-1 to PlGF ratio is 18 minutes. 48

The Elecsys immunoassay sFlt-1 to PlGF ratio may be used for testing pregnant women with suspected PE from a gestational age of 20 weeks up until the time of delivery. The test cut-off points previously recommended by the company, derived from case–control studies of patients with PE or normal pregnancy outcome, are shown in Table 5. 49–51

After further research and clinical consensus, these recommendations have been updated to include a more confident rule-in and rule-out of PE with a sFlt-1 to PlGF ratio cut-off point of 38, so that clinicians have greater certainty when making decisions about patient management (Table 6).

BRAHMS sFlt-1 Kryptor to BRAHMS PlGF plus Kryptor ratio test (Thermo Fisher Scientific)

The BRAHMS Kryptor sFlt-1 to PlGF ratio test is formed by combining the results from two automated immunofluorescent sandwich assays, the BRAHMS sFlt-1 Kryptor and BRAHMS PlGF plus Kryptor assays. The assays are indicated for the quantitative determination of sFlt-1 and PlGF in serum samples and are compatible with the BRAHMS Kryptor compact plus analyser (Thermo Fisher Scientific GmbH, Hennigsdorf, Germany). The assays are intended to be run simultaneously, with the analyser reporting both the concentrations for each assay and the sFlt-1 to PlGF ratio to the user. The BRAHMS Kryptor sFlt-1 to PlGF ratio test is intended to be used in conjunction with clinical assessment to aid the diagnosis of PE. 53 The total durations of the assays are 9 minutes (sFlt-1) and 29 minutes (PlGF). 48

The BRAHMS sFlt-1 Kryptor assay has a limit of detection of 22 pg/ml (measuring range 22–90,000 pg/ml) and a limit of quantification of 34 pg/ml. 48 The BRAHMS PlGF plus Kryptor assay has a limit of detection of 3.6 pg/ml (measuring range 3.6–7000 pg/ml) and a limit of quantification of 6.9 pg/ml. 48 Reference ranges for each of the assays for singleton pregnancies with a normal outcome are provided in the product inserts,54,55 and the company recommends that individual laboratories should validate these ranges or establish their own reference ranges prior to use. The company suggests a cut-off point of 85, based on a study of PE cases and controls that included women with singleton pregnancies and normal pregnancy outcome. 56

Population

The population of relevance to the decision problem is pregnant women, between gestation week 20 and gestation week 36⁺⁶ who, on the basis of screening tests and clinical symptoms (hypertension plus other signs or symptoms that may include proteinuria, haematological abnormalities, frontal headache, severe pain just below the ribs, vision problems, vomiting and/or severe swelling of the face or hands), are suspected of having PE.

There are four potential subgroups of women for this decision problem: those with chronic hypertension; those with pre-existing or gestational diabetes mellitus; those with renal conditions; and those with an autoimmune condition.

Index tests

According to the paradigm of diagnostic test accuracy assessment, the index test is the new test that is to be evaluated for its diagnostic accuracy when compared against an existing gold standard test (the reference standard). Four index tests are eligible for inclusion in the current diagnostic assessment. These are:

1. Triage PlGF test

2. Elecsys immunoassay sFlt-1 to PlGF ratio

3. DELFIA Xpress PlGF test

4. BRAHMS Kryptor sFlt-1 to PlGF ratio test.

As stated in the NICE scope, each of these tests could be employed:

• in conjunction with standard clinical assessment of suspected PE (i.e. as an add-on test to assessments of hypertension, proteinuria, and other clinical criteria for suspecting PE)

• in conjunction with standard clinical assessment excluding quantitative determination of proteinuria.

For assessing diagnostic accuracy these index tests should be compared against the reference standard, that is, standard clinical assessment of suspected PE. Assessment of the relative diagnostic accuracy of any pairwise comparisons among the four index tests is permissible, subject to the availability of relevant evidence (i.e. any direct head-to-head comparisons among pairs of index tests and/or indirect comparisons).

Reference standard

The gold standard reference for diagnosis of PE is clinical assessment, guided by a combination of the following clinical information:

• maternal hypertension (categorised as mild, moderate or severe)

• quantitative proteinuria test

• clinical symptoms suggestive of PE (e.g. headache, oedema or visual disturbances)

• fetal growth restriction.

Maternal hypertension and/or proteinuria with or without clinical symptoms may be sufficient to diagnose PE, or they may also occur in combination with fetal growth restriction and/or signs of biochemical or haematological impairment.

Outcomes

Inclusion and exclusion criteria

The eligibility criteria for the systematic review of test performance are:

• Study design: this review included primary diagnostic research, but was not limited to particular study designs. Instead, issues of methodological rigour relating to study design (specifically, risk of bias and applicability of the study findings) were evaluated during formal quality assessment (see Critical appraisal).

• Population: women with suspected PE between 20 weeks and 36⁺⁶ weeks of pregnancy who have received blood pressure assessment and qualitative assessment of proteinuria. In the present review, ‘suspected PE’ means that blood has been taken for one of the index tests, but a formal diagnosis of PE has not yet been made.

• Index tests: Triage PlGF, Elecsys immunoassay measuring the sFlt-1 to PlGF ratio, DELFIA Xpress PlGF and BRAHMS Kryptor sFlt-1 to PlGF ratio test, in conjunction with standard clinical assessment, or in conjunction with standard clinical assessment excluding quantitative determination of proteinuria.

• Reference standard: clinical assessment guided by maternal hypertension, proteinuria, symptoms suggestive of PE, and ultrasound fetal growth measurements. A combination of maternal hypertension and/or proteinuria, with or without clinical symptoms, may be sufficient to diagnose PE, or they may also occur in combination with fetal growth restriction and/or signs of biochemical or haematological impairment.

• Test performance outcomes: diagnostic and prognostic test accuracy (sensitivity, specificity, prevalence and related outcome measures) for PE.

• Intermediate measures: time to test result, test failure rate, time to diagnosis, proportion of women diagnosed with PE, time to onset of PE and/or eclampsia, proportion of women returned to less intensive follow-up, length of inpatient hospital stay and time to delivery.

Characteristics of the included studies

The 12 full-text documents included in the systematic review of test accuracy reported on four unique primary research studies. Two of these studies used the Triage PlGF test and two studies used the Elecsys sFlt-1 to PlGF ratio for predicting PE. None of the studies that met the systematic review inclusion criteria had employed the PerkinElmer DELFIA Xpress PlGF test or the BRAHMS Kryptor sFlt-1 to PlGF ratio for assessment of women with suspected PE.

Studies on the Triage PlGF test included in the review are:

• The PETRA study, reported in an unpublished, confidential company submission to NICE provided by Alere (Sibai6). Confidential data from PETRA are not presented in the current report; however, they were available for consideration by the NICE Diagnostics Assessment Committee and the EAG.

• The PELICAN study, reported by Chappell et al. in an academic journal paper,5 three meeting abstracts,70–72 and the Triage PlGF test product insert. 40

Studies on the Elecsys sFlt-1 to PlGF ratio test included in the review are:

• the PROGNOSIS study, reported in a company submission to NICE provided by Roche Diagnostics,61 an unpublished academic manuscript73 that was subsequently published by Roche Diagnostics (Zeisler et al. 51), a meeting abstract (Zeisler et al. 74) and the Elecsys sFlt-1 to PlGF ratio test product insert46

• a study by Álvarez-Fernández et al. reported in an academic journal paper75 and a meeting abstract76 (the EAG contacted the authors of this study to confirm that it met the eligibility criteria).

For brevity, studies are cited in this report by their primary reference only, unless the reference is intentionally to a different document.

General characteristics of the study participants

The three published studies varied in their designs and locations (Table 8). 5,51,75 Both the PELICAN study on the Triage PlGF test5 and the PROGNOSIS study on the Elecsys sFlt-1 to PlGF ratio test51 included women in the UK, although in the PROGNOSIS study only 1 of the 30 study centres was in the UK, with 76 UK participants represented among 1050 total participants. The remaining study by Álvarez-Fernández et al. 75 was not conducted in the UK. Two of the studies were of a single-cohort design, whereas the PROGNOSIS study51 had two cohorts: a ‘development’ cohort to derive a cut-off value-based prediction model for the sFlt-1 to PlGF ratio (n = 500) and a ‘validation’ cohort to test the model (n = 550). Two of the studies recruited patients prospectively5,51 and one retrospectively. 75 The major studies, PELICAN5 and PROGNOSIS,51 had relatively large sample sizes for those gestational age groups that are relevant to the current diagnostic assessment, analysing 424–1050 patients. All the studies covered the majority of the period of pregnancy specified in the NICE scope (20⁺⁰ to 36⁺⁶ weeks, i.e. the second and third trimesters), although only the PELICAN study5 covered the whole of this period. However, only the PROGNOSIS study (Zeisler et al. 51) reported how gestational age was determined (calculated from the last menstrual period or first sonography date and recorded at visit 1). Groups of women who presented with fetuses at gestational ages exceeding those specified in the NICE scope (i.e. after 36⁺⁶ weeks) were reported in two studies;5,75 these groups are not considered in the current review.

Key characteristics of the participants in the three published studies are shown in Table 9. Women were aged in their early thirties. Median BMI ranged from 24.9 kg/m² in the PROGNOSIS study51 to 31.2 kg/m² in the study by Álvarez-Fernández et al. 75 Two studies reported the proportion of women who were nulliparous, and this ranged from 43% in the PELICAN study5 to 84% in the study by Álvarez-Fernández et al. 75 One study analysed only singleton pregnancies (PROGNOSIS51), while the proportion of singleton pregnancies in the remaining studies ranged from 74% to 93% in the study by Álvarez-Fernández et al. 75 and up to 90% to 96% in the PELICAN study. 5 Ethnicity of the participants was reported only in the PELICAN and PROGNOSIS studies, and all three studies were dominated by white or Caucasian participants. None of the published studies reported the socioeconomic status of their participants (i.e. their educational achievement, employment status, income or dependency on any community services). When reported, in only two studies, the majority of participants (> 70%) were not smokers.

Prognostic characteristics of the study participants

Population characteristics relevant to the prognosis of PE were reported inconsistently in the published studies and, therefore, it is difficult to present a comparison of the prognostic characteristics across the studies. Many characteristics of potential prognostic relevance were only reported by one or two studies (Table 10).

New-onset proteinuria was reported in two studies. The proportion of women with new-onset proteinuria was highest in the PELICAN study5 (56–62%) and lowest in the PROGNOSIS study51 (30–44%). Álvarez-Fernández et al. 75 reported that 73% of the PE group and 3% of those without PE diagnosis had proteinuria, but did not state whether or not this was new onset. Although all three studies reported hypertension, only Álvarez-Fernández et al. 75 reported the proportion of patients who had chronic hypertension (8–9%). Cases of new-onset hypertension were reported only in the PELICAN study5 (54–67%), although the PROGNOSIS study51 did report cases of elevated blood pressure, without specifying whether or not this meant hypertension; elevated blood pressure was found to be more common in women who developed PE (42–53%) than in those without PE (18–40%). Worsening of existing hypertension was also reported only in the PELICAN5 (15–20%) and PROGNOSIS51 (11–24% in women who developed PE; 12–14% in those who did not) studies. The same two studies reported the proportions of women who had epigastric or upper right quadrant pain or headache, although they differed in whether or not the epigastric pain was stated as being persistent. Overall, the proportion of women with any type of epigastric or upper right quadrant pain in these studies was in the range of 6–10% and the proportion of patients with headache was in the range 28–35%. All three published studies reported the proportion of women who had (suspected) abnormal fetal growth, which was described as suspected fetal growth restriction (small for gestational age, < 10th customised birthweight percentile) in the PELICAN study5 (3–9%) and as IUGR in the PROGNOSIS study51 (6–23%) and the study by Álvarez-Fernández et al. 75 (3–14%). However, the PROGNOSIS study51 and Álvarez-Fernández et al. 75 defined IUGR differently. The PROGNOSIS study51 defined IUGR as estimated fetal weight or abdominal circumference < 5th percentile, presence of pathological process that inhibits expression of normal intrinsic growth potential on at least one occasion after gestational week 22, amniotic fluid index < 10th percentile or pulsatility index > 95th percentile, serial ultrasonography growth curve anomalies or serial growth curve anomalies based on local manual measurement. Álvarez-Fernández et al. 75 defined IUGR as fetal weight < 3rd percentile for gestational age or between the 3rd and 10th percentiles with abnormal uterine arteries or altered blood flow on ultrasound.

Four potential subgroups of women were identified by NICE for this decision problem: those with chronic hypertension, gestational diabetes mellitus, renal impairment and autoimmune conditions. Although some of the included primary studies collected data on these baseline characteristics, there were no relevant subgroup analyses that could contribute to the present diagnostic assessment.

The published studies included in the review differed in the amount of detail they provided about why PE was suspected (Table 11). The most detailed information was presented for the PROGNOSIS study. 51

Risks of bias

The results of the EAG’s critical appraisal of the included studies using QUADAS criteria are shown in Table 12. At the time of preparing the present report, the PETRA study (Sibai6) was confidential. The EAG critically appraised the PETRA study and the results of the appraisal were made available to the NICE Diagnostics Assessment Committee. However, owing to the data confidentiality, the EAG’s critical appraisal of the PETRA study is not included in Table 12 or discussed below. The impact of the PETRA study on the overall conclusions of the current diagnostic assessment is considered in Chapter 6, Discussion.

‘Yes’ answers to QUADAS questions 1 to 9 imply low risk of bias for each of the types of bias being assessed (for explanations of the bias types addressed by each question see Table 7). ‘Yes’ answers to QUADAS questions 10 and 11 reflect adequacy of reporting and require further information (see the paragraphs following) in order for the risks of bias to be assessed. Explanations for the judgements reached by the EAG are provided in the data extraction form for each study (see Appendix 4 for an example data extraction form for review of test accuracy).

As shown in Table 12, all three of the published studies were rated favourably on QUADAS questions 1 to 7, suggesting that the studies would have low risk of spectrum bias, verification bias, disease progression bias, differential verification bias, incorporation bias or diagnostic review bias.

QUADAS question 8 assesses whether or not studies could be at risk of test review bias, that is, when the reference standard result influences interpretation of the index (biomarker) test result. The EAG judged all three studies to be at low risk of this type of bias.

QUADAS question 9 assesses whether or not studies could be at risk of clinical review bias, that is, when the information used when interpreting the index (biomarker) test does not reflect that which would be available in clinical practice. As the PlGF and sFlt-1 to PlGF ratio tests are intended (in this diagnostic assessment) to be add-on tests to standard clinical assessment, it is implicit that in order to make a diagnosis they would be interpreted in conjunction with other clinical information. The EAG judged that in all three studies the interpretation of the biomarker test would not reflect clinical practice, as diagnosis of PE was based solely on whether test values were above or below the cut-off point; in clinical practice such test values would probably be interpreted in conjunction with information about hypertension, proteinuria and/or other clinical signs or symptoms.

QUADAS question 10 assesses whether or not studies could be at risk of test classification bias, that is, when inclusion or exclusion of results classified as intermediate, indeterminate or uninterpretable may systematically influence test accuracy. Only two of the three studies mentioned intermediate, indeterminate or uninterpretable test results, but these were reported differently in each study. In the PELICAN study,5 11.1% of women changed PlGF status from low to very low and 3.5% changed from low to normal when a duplicate run of the test was performed. In the PROGNOSIS study,51 it was stated only that the visit 1 sample was not available or not usable from 34 participants.

QUADAS question 11 assesses whether or not studies could be at risk of attrition bias. In studies of test accuracy, attrition might lead to bias if it differentially impacts on the numbers of true-positive, true-negative, false-positive or false-negative test results. The numbers of exclusions were reported in all three studies. In the PELICAN study,5 24 participants were excluded out of 649 enrolled (3.7%) because of a lack of enrolment samples, missing sample barcodes or no outcome data being available (reasons were not stated). The risk of attrition bias in the PELICAN study is unclear because it is not known if the 3.7% of women with missing data would have been equally distributed among the true-positive, false-positive, true-negative and false-negative groups. In the PROGNOSIS study,51 a relatively large number of women [223 out of 1273 enrolled (17.5%)] were excluded. Reasons given were withdrawal of informed consent (0.9%), failure to meet the inclusion criteria (3.8%), multiple gestation pregnancy (6.1%), lack of an available/usable visit 1 sample (2.7%) and loss to follow-up as they had no PE during the first four visits but no visit after 28 days to confirm lack of PE (4.0%). The risk of attrition bias in the PROGNOSIS study51 is unclear because it is not known if the 6.7% of women who either lacked a usable visit 1 sample or lacked follow-up after 28 days would have been equally distributed among the true-positive, false-positive, true-negative and false-negative groups. In the study by Álvarez-Fernández et al. ,75 19 out of 281 enrolled women (6.8%) were excluded because of lack of delivery data (4.3%), suspicion of PE without urinalysis (1.4%), or because PE was diagnosed before presentation at triage (1.1%). As with the other studies, the risk of attrition bias is unclear for Álvarez-Fernández et al. ,75 as it is not known if the 4.3% of women with missing data would have been equally distributed among the true-positive, false-positive, true-negative and false-negative groups.

Generalisability

The assessment of study generalisability (often referred to as applicability) draws together information on the studies’ designs and their participants’ characteristics (reported in Tables 8–11) and how PE was defined in each study. Some aspects of generalisability are also captured in the QUADAS criteria reported, for example ‘patient spectrum bias’, which is assessed by QUADAS question 1, and refers to whether or not the study participants would be representative of those likely to be seen in clinical practice.

The studies varied in the amount of information they provided about their participants, and in some cases generalisability is unclear. For example, Álvarez-Fernández et al. 75 did not report ethnicity, and none of the published studies reported socioeconomic status.

Only two of the studies included any centres in the UK (PELICAN5 and PROGNOSIS51). A factor that might influence the generalisability of study findings is whether or not management of PE is expectant or based mainly on PE diagnostic criteria. According to the Alere company submission,6 a more expectant management approach appears to be used in the UK than in the USA (although no source of evidence for this was cited). However, the EAG is not aware of any data that would enable different countries or health systems to be classified according to the degree of expectant management of PE that takes place. As such, it is unclear whether or not the geographical location of a study alone would influence generalisability of its test accuracy outcomes to clinical practice in England and Wales. Most likely to be more important is how PE is defined in each study (see Definition of pre-eclampsia).

Relevance of the evidence to the NICE decision problem

The decision problem (see Chapter 2) specifies that the angiogenic biomarker tests should be assessed (1) in addition to routine clinical assessment (which includes monitoring of blood pressure and proteinuria) and (2) as an alternative to quantitative proteinuria testing. The four included studies employed the biomarker tests in addition to routine clinical assessment. Therefore, the available evidence only informs the first part of the decision problem; no evidence is available for assessing the accuracy of these biomarker tests as alternatives to proteinuria testing.

Each of the four included studies assessed only one of the tests specified in the NICE scope. Therefore, there are no direct head-to-head comparisons available in relevant populations between any of the four diagnostic tests specified in the NICE scope. Furthermore, evidence from the included studies is limited to only two of the four biomarker tests specified in the NICE scope, that is the Triage PlGF test and the Elecsys sFlt-1 to PlGF ratio. No studies were identified that compared either the PerkinElmer DELFIA Xpress PlGF test or the BRAHMS Kryptor sFlt-1 to PlGF ratio to usual clinical practice.

The three published studies reporting the Alere and Roche Diagnostics tests differ in their participant characteristics and the test accuracy outcomes they report, for example providing outcomes for different test cut-off points and different gestational timings. For this reason, the EAG considered it inappropriate to conduct meta-analyses to quantitatively combine test accuracy outcomes across studies to improve their precision. Instead, the results of the studies are presented narratively, with tabulation of the test accuracy outcomes [sensitivity and specificity, with confidence intervals (CIs)] together with their source data (true and false positives and true and false negatives), prevalence of PE, and supporting statistics (PPV, NPV, positive and negative likelihood ratios and areas under ROC curves). These outcomes capture the best-available diagnostic test accuracy values that could be employed in our economic model (see Chapter 5).

The diagnostic test accuracy outcomes are reported below for each diagnostic test and are then discussed in the context of any key issues of bias or generalisability that were identified at the critical appraisal step above (see Critical appraisal of the included studies).

Triage PlGF test (Alere)

Results for this test were available to the NICE Diagnostics Assessment Committee and the EAG from the PELICAN study5 (n = 287 for gestation weeks 20⁺⁰ to 34⁺⁶; n = 137 for weeks 35⁺⁰ to 36⁺⁶) and from the PETRA study. 6 As mentioned, at the time of preparing the present report, the PETRA study (Sibai6) was confidential and results from this study are therefore not reported here. The impact of the PETRA study on the overall conclusions of the current diagnostic assessment is considered in Chapter 6.

The most relevant outcome to the decision problem reported in the PELICAN study is prediction of PE requiring delivery within 14 days of testing (Tables 13 and 14 and Figure 3). Some of these data are used to inform the EAG economic model. Other related test accuracy outcomes (which do not directly inform the economic model but were considered by the NICE Diagnostics Assessment Committee and the EAG as supporting information) were prediction of preterm PE and prediction of delivery independent of the PE diagnosis (Table 15). Summary AUC estimates for ROC plots for these outcomes are given in Table 16; the AUC data are not used directly in the EAG economic model, but provide an overview of how well the PlGF test performed for different outcomes.

FIGURE 3.

Accuracy of the Triage PlGF test at three cut-off points for predicting PE requiring delivery within 14 days. a, Data from the PETRA study (Sibai6) (100 pg/ml cut-off point, < 35⁺⁰ weeks) were also considered by the EAG and the NICE Diagnostics Assessment Committee, but are confidential and are not reproduced here. FN, false negative; FP, false positive; TN, true negative; TP, true positive.

Elecsys sFlt-1 to PlGF ratio test (Roche Diagnostics)

Diagnostic and prognostic accuracy outcomes for the Elecsys sFlt-1 to PlGF ratio test are available in two studies that assessed three test cut-off points: 23, 38, and 85. The studies are not directly comparable because they differed in the cut-off points employed and the time periods during which the rule-in or rule-out of PE applied. The test accuracy outcomes are therefore grouped within each study in the tables presented here. The most relevant outcomes to the NICE decision problem reported in the included studies are:

• rule-out of PE (broadly defined – including HELLP syndrome) within 1 week of testing using a cut-off point of 38 (PROGNOSIS study51) (Table 17 and Figure 4)

• rule-in of PE (broadly defined – including HELLP syndrome) within 4 weeks of testing using a cut-off point of 38 (PROGNOSIS study51) (see Table 17 and Figure 4).

• rule-out of PE within an unspecified time period (assumed by the EAG to be 3 weeks) using cut-off points of 23 and 85 (Álvarez-Fernández et al. 75) (Table 18 and Figure 5).

FIGURE 4.

Accuracy of the Elecsys sFlt-1 to PlGF ratio test at cut-off point 38 for rule-in and rule-out of PE/eclampsia/HELLP syndrome. Composite diagnostic outcome of PE/eclampsia/HELLP syndrome. FN, false negative; FP, false positive; NR, not reported; TN, true negative; TP, true positive.

FIGURE 5.

Accuracy of the Elecsys sFlt-1 to PlGF ratio test at cut-off points 23 and 85 for predicting PE. FN, false negative; FP, false positive; NR, not reported; TN, true negative; TP, true positive.

Comparison of the Triage PlGF test (Alere) and the Elecsys sFlt-1 to PlGF ratio test (Roche Diagnostics)

The Triage PlGF test and Elecsys sFlt-1 to PlGF ratio tests cannot be compared directly in terms of their performance, as the evidence available for each test relates to different test accuracy outcomes. Gencay et al. 85 compared diagnostic accuracy of these tests, but this was a case–control study without a suspected PE population, so was excluded from our systematic review of test accuracy. Although AUC estimates from ROC curve analyses can help to compare how well different tests perform, the studies are unbalanced in the extent to which AUC estimates are presented for different outcomes, precluding comparisons (different gestational ages, different primary outcomes; compare Tables 16 and 19). For example, in the PETRA study (Triage PlGF test) the outcome was preterm PE necessitating delivery within ≤ 14 days or ≤ 7 days among women presenting with suspected PE between 20 and 35 weeks of gestation, while in the PROGNOSIS study (Elecsys sFlt-1 to PlGF ratio test) outcomes were rule out PE within 1 week or rule in PE within 4 weeks among women presenting between weeks 24 and 36⁺⁶.

Comparison of the PerkinElmer and Thermo Fisher Scientific tests

In support of the current diagnostic assessment, PerkinElmer provided evidence showing a good correlation between PlGF concentrations measured by the DELFIA Xpress PlGF test and Elecsys PlGF assays,86 while Thermo Fisher Scientific provided evidence showing good correlations between PlGF concentrations, sFlt-1 concentrations, and the sFlt-1 to PlGF ratios measured by the BRAHMS Kryptor and Elecsys assays. 48,87 However, only one of these documents, a study by Andersen et al. ,48 compared diagnostic accuracy of the tests. None of these documents meets the scope for the current diagnostic assessment (and hence they were excluded from our systematic review of test accuracy), as the biomarker samples were not obtained from women with suspected PE. The EAG notes that, although the biomarker test measurements correlated well across tests, slopes of the correlations generally differed from unity: sFlt-1 to PlGF ratios measured with the BRAHMS Kryptor assay were systematically higher than those measured with the Elecsys assay,48,87 while PlGF concentrations measured by the DELFIA Xpress PlGF test were systematically lower than those measured with the Elecsys PlGF assay. 86

Given that no studies on the DELFIA Express PlGF test and the BRAHMS Kryptor sFlt-1 to PlGF ratio met the inclusion criteria for the review of test accuracy, the EAG is unable to determine the diagnostic accuracy of these two tests. A study by Anderson et al. 48 has compared the accuracy of the Elecsys and BRAHMS Kryptor sFlt-1 to PlGF ratio tests for predicting PE, in a case–control study that included women with PE (n = 39, i.e. a relatively small sample size) and women with healthy pregnancies (n = 76). While it may be tempting to use the results from Andersen et al. 48 as illustrative of how the Elecsys and BRAHMS Kryptor sFlt-1 to PlGF ratio tests might perform in women with suspected PE, the EAG cautions against this, as both sensitivity and specificity would be overestimated to an unknown extent, which might differ between the tests. In addition, the study was conducted in Denmark, included only singleton pregnancies, and was retrospective (thereby at possible risk of selection bias).

As the study by Anderson et al. 48 is the only diagnostic accuracy comparison that has been made between the BRAHMS Kryptor and Elecsys sFlt-1 to PlGF ratio assays, a brief summary of the comparison is provided here with the proviso that this should be interpreted as illustrative only.

Andersen et al. 48 made a number of comparisons between the BRAHMS Kryptor and the Elecsys sFlt-1 to PlGF ratio assays, with post hoc subgroup analyses for early-onset and late-onset PE and for non-obese (BMI of < 30 kg/m²) and obese (BMI of ≥ 30 kg/m²) women. Sensitivity, specificity, PPVs and NPVs, and areas under the ROC curve for the sFlt-1 to PlGF ratio are concisely tabulated by Andersen et al. 48 in their paper, for test cut-off points of > 85, < 33 and > 110, although the EAG has not extracted these data. At all cut-off points, the sensitivity of the BRAHMS Kryptor sFlt-1/PlGF assay was either higher than or similar to that of the Elecsys sFlt-1 to PlGF ratio. At all cut-off points, the specificity of the BRAHMS Kryptor sFlt-1/PlGF assay was lower than that of the Elecsys sFlt-1 to PlGF ratio for the overall gestational period or late gestation, but both tests had similar specificity during early gestation (i.e. diagnosis up to 34 weeks). For both companies’ tests, both sensitivity and specificity were highest during early gestation.

Receiver operator characteristic analyses conducted by Andersen et al. 48 indicated that the BRAHMS Kryptor PlGF assay performed marginally better than the Elecsys PlGF assay, while the BRAHMS Kryptor sFlt-1 to PlGF ratio had similar accuracy to the Elecsys sFlt-1 to PlGF ratio. Both sFlt-1 to PlGF ratio tests showed best performance (AUC > 0.92) for early-onset PE (i.e. before week 34 of gestation). Both of the sFlt-1 to PlGF ratio tests performed better (AUC > 0.8) in non-obese women than in obese women (AUC < 0.77).

Repeating the caveats above, the EAG emphasises that these findings may not necessarily reflect the relative performance of these tests if applied in a population of women with suspected PE.

Systematic review of economic studies

The methods detailed in Chapter 3 were used to systematically review the cost and cost-effectiveness literature. The populations, interventions and comparators included are the same as for the systematic review of test accuracy (as described in Inclusion and exclusion criteria), with the exception of study design and outcomes. As the systematic review of test accuracy searches applied no study type filters (see Chapter 3), the results of these searches were suitable for identifying cost-effectiveness evidence by applying the relevant economic evaluation inclusion criteria.

Studies were included if they were full economic evaluations, assessing both costs and consequences, or cost studies for the specified index tests. Outcomes included are those consistent with full economic evaluations and cost studies, including intermediate outcomes (budget impact, cost per patient or cost per case of PE correctly managed), or final outcomes (life-years or QALYs gained). Each step of the review was conducted by one health economist and checked by a second health economist, with any disagreements resolved by discussion.

Review of HRQoL studies

The EAG undertook a series of sequential searches designed to identify HRQoL data in gestational hypertension, PE and general pregnancy. The goal of these searches was to identify European Quality of Life-5 Dimensions (EQ-5D) utility values for use in the economic model or to identify utility values that could be mapped to EQ-5D using published algorithms, in line with the NICE reference case. 88 The eligible instruments were identified using Oxford University’s Health Economics Research Centre database of mapping studies. 89 The following instruments were eligible for inclusion: EQ-5D, Short Form questionnaire-36 items (SF-36) (using all scales), Short Form questionnaire-12 items (SF-12), Short Form questionnaire-6 items (SF-6D), Nottingham Health Profiles and the Health Assessment Questionnaire. The relevant population is women who are pregnant and/or who have gestational hypertension or PE and their neonates. Non-intervention study designs were preferred, unless the intervention directly addressed hypertensive disorders of pregnancy (including PE). Studies assessing specific symptoms of pregnancy, rather than general pregnancy HRQoL, that were not potentially related to gestational hypertension or PE (such as urinary incontinence or emesis) were excluded. Studies in subpopulations of pregnant women that are not directly related to gestational hypertension or PE were also excluded (such as human immunodeficiency virus, thyroid conditions and cancer). Clinical outcomes relevant to the model were defined as delivery before 35 weeks of gestation, delivery after 35 weeks of gestation, NICU stay, hospital stay, birth by induction, birth by caesarean section, standard vaginal birth, severe complications of birth, mild PE and severe PE.

The sequential approach for identifying HRQoL studies was followed and, unless stated otherwise, all steps were conducted by one health economist and checked by a second health economist, with any disagreements resolved through discussion:

1. References already identified through the systematic review of test accuracy (see Chapter 4, Quantity and quality of research available) were searched for HRQoL studies. Two health economists independently marked any titles and abstracts that appeared to be relevant. Following this process, only one study, reported by Shmueli et al. ,90 was identified (described in detail in Results of the review of health-related quality-of-life studies).

2. Information on HRQoL was sought from the company submissions provided for this diagnostic assessment.

3. The NICE Guideline on Hypertensive Disorders in Pregnancy (CG107)13 was scrutinised to identify any further HRQoL data.

4. Finally (as the preceding searches yielded limited relevant information), additional systematic searches of bibliographic databases were conducted for HRQoL data in pregnant women or women with hypertensive disorders of pregnancy.

The systematic searches were conducted in MEDLINE, MEDLINE In-Process & Other Non-Indexed Citations, EMBASE, and DelphiS. The search strategy (see Appendix 6) was designed in MEDLINE and adapted for the other databases. The inclusion and exclusion criteria for eligibility screening of titles and abstracts are given in Box 1. The same eligibility criteria were also used for screening full-text records, with an exception that more stringent criteria for HRQoL outcomes were applied at full-text screening. Studies could be excluded at full-text screening if the HRQoL scales were incomplete (e.g. if six scales of the SF-36 instead of eight were reported, or if only the mental component scale of the SF-12 was reported), as the mapping algorithm to obtain EQ-5D scores requires full reporting of the scale to be mapped. In addition, at full-text screening, studies that reported only summary QALYs (i.e. cost-effectiveness results) without utility scores were excluded.

Hadker et al.39,96

Critical appraisal of the studies reported by Hadker et al. in 201039 and 201396 has been combined as they used the same model, populated with identical clinical inputs, to address UK39 and German96 health-care payer perspectives, respectively. The discussion of the validity and generalisability of the results focuses on the analysis that is more relevant to the UK setting. 39 The UK analysis is described as a budget impact model, although both studies might be better described as cost analyses, which are concerned with determining potential changes in resource use (hence cost) associated with improved diagnostic performance achieved using the sFlt-1 to PlGF ratio (primarily through increased specificity, given the relatively low prevalence of PE in the population being assessed). The population studied was all women receiving obstetric care, assessed for PE beyond 20 weeks of gestation.

Sensitivity and specificity for usual clinical assessment were based on values reported in a systematic review by Meads et al. 95 The parameter values used in 2010 by Hadker et al. 39 are reported as pooled averages (the paper does not state whether these are means or medians) of the values reported by Meads et al. No detail was provided on the method of pooling and there was no indication of which diagnostic methods reported by Meads et al. were included in the pooled estimate. Sensitivity for diagnostic tests in Meads et al. ranged from 9% to 66% and specificity ranged from 74% to 96% across a range of tests. Sensitivity and specificity for the sFlt-1 to PlGF ratio were taken from a single publication by Verlohren et al. ,84 which was excluded from our review of test accuracy (see Chapter 3) because the study population had already been diagnosed with PE (i.e. not suspected PE). Neither study reports how this paper was identified or selected for use in the model, and no quality assessment of these or other inputs into the model was provided. The model assumes that all patients who initially test negative on the sFlt-1 to PlGF ratio test receive two repeat tests, and it includes these in resource-use estimates, but does not adjust diagnostic outcomes for repeated testing. Identification or diagnosis of PE in the model appears to be based entirely on the modelled outcome of the sFlt-1 to PlGF ratio test and does not take account of the other maternal symptoms or risk factors (i.e. the presence or absence of clinical signs of PE).

Hadker et al. 39 provided little information on resource use assumptions. The study reports that these were based on published literature and expert opinion, but does not explain how sources were identified. The text of the paper states that ‘PE management costs include physician office visits, physical exams, regular blood pressure checks, blood and urine tests, and cardiotocography as well as hospital stays for day assessments, intensive care, inpatient monitoring and delivery or termination of pregnancy’, but provides no further information on the assumed frequency of these events or the proportions of women affected. In a table summarising cost assumptions, two published studies are referenced (Meads et al. 95 and Murphy and Stirrat100). However, no detail was provided on which resource use data were extracted from these sources. As a result it is difficult to judge the validity of the resource use or cost assumptions included in the model (Table 21). The cost year for the UK analysis was 2009.

In addition to the base-case analysis reported in Table 20, Hadker et al. 39 present deterministic sensitivity and scenario analyses conducted on disease prevalence, sensitivity and specificity of usual clinical assessment, the proportion of high-risk patients and the cost of the new test. Separate scenario analyses that involved increasing the prevalence of PE, increasing the sensitivity of usual clinical assessment and an increased cost for the sFlt-1 to PlGF ratio test had minimal impact on the estimated saving associated with the addition of the diagnostic test to current clinical assessment (£969, £955 and £928, respectively, compared with £945 per patient in the base case). Increased specificity of usual clinical assessment resulted in a large decrease in the estimated saving (to £208 per patient), while increasing or decreasing the proportion of high-risk patients increased costs identically for usual clinical assessment and the new test, but did not affect the estimated saving. In each of these analyses uncertainty was assessed by varying parameters by a fixed (arbitrary) proportion rather than using any statistically derived measure of variation (such as 95% confidence limits for PE prevalence), using data from Bhattacharya and Campbell,101 and sensitivity and specificity of usual clinical assessment derived from Meads et al. 95

Overall, this study indicates that improved diagnostic performance, using the Elecsys sFlt-1 to PlGF ratio test for identifying PE in women after 20 weeks of gestation, may be associated with reduced resource use and that the majority of this benefit would be realised by reducing false-positive results. This study was conducted for a general population of pregnant women, not for those with suspected PE, and is therefore not directly relevant to the current decision problem.

Schnettler et al.33

Schnettler et al. 33 reported a study comparing observed management of women presenting to hospital with suspected PE at less than 34 weeks of gestation and modelled management using the Elecsys sFlt-1 to PlGF ratio diagnostic test. Current clinical assessment was based on a combination of blood pressure, urinary protein excretion, levels of alanine aminotransferase (ALT) and platelet counts with the modelled management based on these same measures combined with the sFlt-1 to PlGF ratio, and treating clinicians were unaware of the PlGF and sFlt1 values. The paper does not indicate specific threshold values for hypertension or proteinuria that would lead to diagnosis of PE, but does provide threshold values for elevated transaminases (ALT or aspartate aminotransferase ≥ 80 units/l) and low platelet count (≤ 100 × 10⁹/l). The sFlt-1 to PlGF ratio threshold value adopted in the study was ≥ 85 (referenced to Rana et al. 102). Resource-use information was collected over a 2-week period following enrolment to the study and characterised according to test outcome (true/false positive and true/false negative). Resource use was valued using institutional charges (converted to US$ using the 2012 charge-to-cost conversion factor) for triage evaluations, maternal and fetal radiological studies, laboratory tests, admissions, consultations, deliveries and miscellaneous items (such as intravenous tubing and fluids). The cost of the diagnostic test was US$101.14 (sourced from Hadker et al. 39).

Sensitivity and specificity of usual assessment for predicting PE were based on diagnoses observed in normal clinical practice for the 149 women enrolled into the study. These are reported in Table 20 and indicate high sensitivity, but poor specificity, for usual clinical assessment. Diagnostic outcome, using the sFlt-1 to PlGF ratio, is reported (in terms of true/false-positive and true/false-negative status), but it is not clear from the study whether this classification was based entirely on the diagnostic test results or on a combination of the diagnostic test and usual clinical assessment (and if so, what combination algorithm was used). Sensitivity and specificity of the diagnostic test (estimated by the EAG from values reported by Schnettler et al. 33) are reported in Table 20, showing poorer sensitivity than usual clinical assessment but substantially improved specificity. The number of true negatives increased from 35 (24% of the total population) with usual clinical assessment to 92 (62% of the total population) using the diagnostic test at the expense of an increase in false-negative results from 3 (2% of the total) to 12 (8% of the total). Table 22 presents the prevalence of PE in the population studied by Schnettler et al. 33 and key outcomes reported by diagnostic test outcome.

As the study that informed the model was a retrospective cohort, patients who had false-positive results with the Elecsys sFlt-1 to PlGF ratio test might not have been treated with the same intensity as patients who had a true-positive result. Likewise, treatment intensity for false-negative results did not closely match that for true-positive results. This is likely to be a result of different lengths of admission to hospital. A woman with a false-positive test result may be hospitalised for a short period of time with suspicion of PE and a woman with a false-negative result may spend some time out of hospital before her symptoms force admission.

The study reported very limited sensitivity analyses. These were undertaken with respect to the sFlt-1 to PlGF ratio cut-off value (which varied between 5 and 200; the results of this analysis are not reported in the study) and costs by test outcomes (i.e. patients’ true-positive, false-positive, true-negative or false-negative status, which were varied between 50% and 200%). Schnettler et al. 33 do not appear to have considered uncertainty in the estimation of test outcomes, from individual events used in the resource-use estimation (including neonatal outcomes) or in the unit costs applied (other than indirectly through percentage variation applied to costs by test outcomes).

Overall, this study indicates that improved diagnostic performance for identifying PE in women presenting with suspected PE up to 34 weeks of gestation may be associated with reduced resource use, and that the majority of this benefit would be realised by reducing the number of false positives. In this study this potential financial saving was associated with a diagnostic strategy that was inferior in terms of correctly identifying false-negative results (12 compared with three using standard clinical assessment). Schnettler et al. 33 reported that women in the false-negative group were predominantly obese (mean BMI 36.8 kg/m², compared with 29.9, 27.8 and 33.9 kg/m² for true-positive, false-positive and true-negative cases, respectively). This study is relevant to the decision problem as stated by NICE, as it was conducted in pregnant women being evaluated for suspected PE, but it does not cover the full range of gestation indicated in the scope and has a small sample size. Moreover, the study was conducted in a single hospital in the USA and may lack generalisability to the UK NHS context.

Summary of economic studies

A summary critical appraisal checklist for quality assessment of the included studies is shown in Table 23.

Overall, the three published economic evaluations appear applicable to the UK NHS in terms of the patient population and comparator, although Schnettler et al. 33 had a different health-care system and setting than is usual in the UK, and Hadker et al. 93 was a study in Germany that has unclear relevance to the UK NHS setting. In all three studies, the modelling methodology, model structure and assumptions were appropriate, and data inputs were described, although these were not fully justified in two studies. 39,96 Of the three published studies, none was based on a systematic review, none measured health benefits in QALYs, none used standardised and validated generic instruments, none described and justified the resource costs used, none assessed uncertainty through appropriate sensitivity analyses and none reported whether or not their model had been validated.

HRQoL evidence from searches of test accuracy

Of the 1972 references originally identified in the searches of test accuracy studies (see Chapter 4, Quantity and quality of research available), only one study, by Shmueli et al. ,90 reported relevant information about HRQoL. Shmueli et al. 90 constructed a model of a screening programme that allowed some low-impact prevention of PE through management with aspirin, calcium, folate and vitamins in an Israeli health system context. The HRQoL in the model is tied to neonatal mortality, mortality in the first year of an infant’s life, and an assumption of higher rates of diabetes mellitus in children born to mothers with PE based on studies cited from the literature. 103–105 These three studies cited by Shmueli et al. 90 were all from Israel. One study was based on maternal mortality data from 1964 to 1976,103 and the other two studies were reported only in conference abstracts,104,105 only one of which was available to the EAG. 105 The available abstract was for long-term outcomes of maternal gestational diabetes mellitus, not PE, on offspring. 105 None of the data used in Shmueli et al. ’s90 model to determine HRQoL differences appear generalisable to the UK, and none appears relevant to the current decision problem.

HRQoL evidence from company submissions

As mentioned, Alere and Roche Diagnostics each produced economic models to simulate the addition of their diagnostic test to current clinical practice. These were critically appraised by the EAG, but were specified as confidential by the companies and are not presented here. Both models were cost models and did not include HRQoL. PerkinElmer and Thermo Fisher Scientific did not provide any economic models or any documents that contained relevant data on HRQoL.

HRQoL evidence in the NICE Guideline on Hypertensive Disorders in Pregnancy

The Guideline Development Group (GDG) for the NICE Guideline on hypertensive disorders in pregnancy13 created several models of gestational hypertension and PE treatment: a model of prophylactic aspirin use for women suspected of having PE (see appendix H in the guideline), a model comparing expectant monitoring or immediate induction for the management of gestational hypertension (see appendix I), a model comparing expectant monitoring to immediate induction for the management of women with PE (see appendix J) and further models for proteinuria measured via dipstick testing (see appendix K) and via automated urinalysis (see appendix L).

In the prophylactic aspirin model of women with suspected PE, the GDG searched the Harvard Cost-Effectiveness Registry for HRQoL data for normotensive women. Only one study was found, by Sonnenberg et al. ,106 reporting costs and health effects of contraception usage. The utility values in the study were derived using time trade-off (TTO) methods (rather than the EQ-5D, which is the preferred method of utility measurement in NICE appraisals) from a convenience sample of unreported size. 106 The EAG believes that these utility values are inadequate. The GDG aspirin model assumed that women who developed PE have the same quality of life as normotensive women and assumed that all children discharged alive would live a normal healthy life up to 80 years and have 27.7 discounted QALYs. No utility decrements were assigned for hospitalisations or adverse outcomes related to birth, although it is unlikely that these events would have a significant effect on cost-effectiveness because of the transitory nature of hospitalisation and shared characteristics with regard to hospitalisations within the comparator arms.

The model for hypertension in pregnancy assumed a lower utility score for women with severe disease based on utility for intensive care stay derived from a study on the treatment of severe hospital infections (not pregnancy or hypertensive disorders of pregnancy). 107 Only patients who developed severe disease had lower HRQoL. Further examination of the referenced study shows that the use of utility data from the study is inconsistent in CG107. 13 CG107 reports that severe complications of PE had a utility of 0.019, derived from an Edwards et al. 107 utility value divided by 52 and multiplied by two to represent the amount of time spent in intensive care for severe complications of PE. However, the utility decrement presented in Edwards et al. 107 is 0.402. Performing the operations referenced by the GDG on this value does not produce 0.019. It is unclear how the GDG developed the utility measure used in appendix I for severe complications of PE. The PE model in appendix J of the NICE guideline13 used the same assumptions.

Models in appendices K and L of NICE guideline CG10713 added maternal death with an assumption of 24.8 discounted QALYs lost for a maternal death. Clinicians consulted by the EAG indicated that maternal death occurs in a vanishingly small proportion of pregnancies and probably did not warrant inclusion in modelling.

All of the models in the NICE guideline13 assumed full health for mothers and infants after successful birth and loss of full health in the case of mortality. As acknowledged by the GDG, the assumption that women who develop PE have perfect health for the rest of their lives is likely to be an overestimate. Many women who develop gestational hypertension or PE are overweight or already suffering from hypertension before pregnancy. Likewise, the incidence of both developmental disability and cerebral palsy is higher among children born prematurely. 108–110 As acknowledged by the GDG, the guideline models probably overestimate QALYs for these children. In addition, the models used utility scores derived by two different methods, TTO and EQ-5D. It is well known that utility scores from different instruments or methods of measurement produce different results that may not be comparable. 88 The EAG considers that the utility scores used in NICE CG10713 are inadequate for the purposes of the current assessment. Table 24 presents the utility values directly from Sonnenberg et al. ,106 and Edwards et al. ,107 without any modifications; the values do not perfectly match those reported in the NICE guideline13 and do not perfectly match after performing operations referenced by the NICE guideline.

HRQoL evidence from additional systematic searches

The systematic searches identified nine potentially relevant studies (Figure 7). Seven of these were identified directly from database searches. Among the excluded references there were two studies by Bijlenga et al. 111,112 that did not report sufficient data to map SF-36 to EQ-5D, and did not report raw scores for EQ-5D. The lead author of these studies was contacted and provided unpublished data for each of the subscales of SF-36 and summary scores for EQ-5D for each trial arm. These data have been included in the review of HRQoL, bringing the total number of included studies to nine (Table 25).

FIGURE 7.

Flow chart for the identification of studies on HRQoL.

Only three studies reported EQ-5D, while six reported HRQoL outcomes that could be mapped to EQ-5D. As more than one source may be necessary to capture HRQoL for women during pregnancy, delivery and the post-partum period, the use of these values in the EAG economic model is assessed further below when considering relevant model parameters (see Derivation of utility estimates from health-related quality of life).

Of the nine studies, only one was in the UK,119 but five were of European populations. 111,112,115,116,119 Four studies had relatively small patient numbers (fewer than 250 total participants). 116–118,120 Only the Petrou et al. 119 and the Bijlenga et al. 111,112 studies utilised the EQ-5D questionnaire. Petrou et al. used the EQ-5D in a general pregnancy population, with two groups of patients classified by spontaneous or non-spontaneous birth.

Decision problem

The scope developed by NICE states the decision problem as follows:

• What is the cost-effectiveness of diagnostic testing, in addition to standard clinical assessment, for the diagnosis of PE in the second and third trimesters of pregnancy in women presenting with suspected PE between 20 weeks and 36⁺⁶ weeks of pregnancy?

• What is the cost-effectiveness of PlGF-based diagnostic testing as a replacement for quantitative proteinuria tests in the diagnosis of PE in the second and third trimesters of pregnancy?

The first part of the decision problem is assessed in the base-case analysis, while the second part of the decision problem is assessed in sensitivity analysis only because of lack of any relevant diagnostic accuracy data according to the systematic review of test accuracy.

Strategies and comparators

The reference standard defined in the NICE scope is standard clinical assessment for PE, which is guided by a combination of maternal blood pressure measurement, proteinuria tests, assessment of clinical symptoms suggestive of the condition and ultrasound fetal growth measurements. The index tests included in the scope for the assessment are diagnostic tests for PE (Triage PlGF test, Elecsys sFlt-1 to PlGF ratio, DELFIA Xpress PlGF test and BRAHMS sFlt-1 Kryptor/PlGF plus Kryptor sFlt-1 to PlGF ratio) in addition to standard clinical assessment. However, as indicated in Chapter 4, Assessment of test accuracy, evidence of diagnostic test accuracy was only identified for two of the tests: Triage PlGF test and Elecsys sFlt-1 to PlGF ratio. The remaining tests, DELFIA Xpress PlGF test and BRAHMS Kryptor sFlt-1 to PlGF ratio, are not included in the current economic analysis. In addition, no clinical test accuracy evidence to inform the assessment of biomarker tests as alternatives to proteinuria testing was identified (see Chapter 4, Assessment of test accuracy). Thus, the second part of the decision problem cannot be addressed reliably in the economic analysis.

Description of the decision analytical model

The EAG developed a decision analytical model to assess the cost-effectiveness of diagnostic tests based on PlGF or sFlt-1 to PlGF ratio test results when used in addition to standard clinical assessment compared with standard clinical assessment alone, in accordance with the scope of the appraisal issued by NICE.

The model was structured to include outcomes identified in the scope issued by NICE for this diagnostic assessment. Suitable data on the sequelae of alternative approaches to the management of suspected PE in pregnancy (in terms of maternal and neonatal morbidity and/or mortality as well as associated resource use) were identified in our systematic review of test accuracy evidence, from company submissions and through targeted searches. The model evaluates costs (GBP using a 2014 price base) from the perspective of the NHS and Personal Social Services. Outcomes in the model are expressed as QALYs. The time horizon of the analysis was < 1 year, corresponding with the length of prebirth monitoring and immediate post-partum monitoring, so no discounting was applied to costs or benefits, in line with current guidance. 69,88,121

Modelling approach and model structure

The model developed for this assessment was a decision tree, incorporating the management of clinical symptoms of suspected PE, timing and mode of delivery, and maternal and neonatal outcomes. A decision tree was considered more appropriate than a Markov model because events in the model deal with only one pregnancy, as repeat pregnancies are highly uncertain (may not occur, unknown timing and difficult to control for risk factors), removing the need for modelling recurrent events (an advantage of a Markov model) and because there was a paucity of data on long-term outcomes in PE and preterm birth, eliminating the Markov advantage of covering complex future events. With the limitations of the data listed above, a decision tree model structure is sufficient to represent the decision problem.

The decision tree can be considered to have four main structural components:

• Risk stratification (high, intermediate or low risk of PE) of women with suspected PE, determined on the basis of clinical signs, symptoms or findings (Box 2) in the case of standard clinical assessment or the same signs, symptoms and clinical findings with the addition of a PlGF-based diagnostic test.

• Pre-eclampsia management (identified as expectant management or immediate delivery based on key symptoms of PE or emergent eclampsia). Expectant management includes monitoring of clinical signs, symptoms and findings (assumed to follow the NICE CG for management of suspected PE122), active management of conditions such as hypertension (assumed to follow the NICE CG for management of hypertension in pregnancy13) and planned delivery at 37 weeks of gestation. Immediate delivery relates to a requirement to deliver within 24 hours irrespective of gestational age because of clinical findings indicating severe risk to a pregnant woman or fetus, and is the assumed treatment for PE detected after 35 weeks of gestation.

• Maternal outcomes (in terms of admission to intensive care, extended hospital stay and morbidity associated with PE).

• Fetal and neonatal outcomes (in terms of admission to intensive care, extended hospital stay, and morbidity and mortality associated with fetal conditions that may be associated with the underlying cause of maternal PE and/or with early delivery).

BOX 2 - Clinical signs, symptoms or findings indicating suspected PE

New-onset elevated blood pressure.

Aggravation of pre-existing hypertension.

New-onset protein in urine.

Aggravation of pre-existing proteinuria.

Abnormal uterine perfusion.

Suspected IUGR.

Headache.

Oedema.

Epigastric pain.

Visual disturbance.

Sudden weight gain.

Low platelets.

Elevated liver transaminases.

Figure 8 shows a simplified schematic outline of the model, indicating the main structural components. The model distinguishes between suspected PE presenting up to 35 weeks and that presenting from 35 to 37 weeks of gestation. This distinction has been adopted to take account of differing accuracy of biomarker tests according to gestational age. (This relates specifically to sensitivity and specificity of the Triage PlGF test which are reported for pregnancies presenting up to 35 weeks of gestation, for pregnancies between 35 and 37 weeks of gestation and those beyond 37 weeks of gestation. Women beyond 37 weeks of gestation are outside the scope of this assessment; sensitivity and specificity of the Elecsys sFlt-1 to PlGF ratio test are reported for gestation ages between 24 and 37 weeks.)

FIGURE 8.

Schematic outline of the EAG economic model.

Risk stratification

The probability of patients with suspected PE being identified as being at high, intermediate or low risk of PE is based on disease prevalence (i.e. the proportion with PE within the suspected PE population), as well as the reported sensitivity and specificity of each diagnostic strategy. These outcomes were modelled based on sensitivity and specificity for standard clinical assessment (using signs/symptoms and clinical findings, i.e. the reference standard) and for assessments including a diagnostic test (signs/symptoms and clinical findings plus a relevant biomarker test, i.e. the index test) reported in studies included in the systematic review of test accuracy (see Chapter 4).

In the simple case, when patients are categorised as either positive or negative on the basis of a test with a single cut-off point (assuming a positive test result indicates high risk, a negative test result indicates low risk and, therefore, there is no intermediate risk group) the probability of positive and negative results can be calculated using the formulae below:

⁽¹⁾ pPositive = prevalence × sensitivity + ( 1 − prevalence ) × ( 1 − specificity )

⁽²⁾ pNegative = prevalence × ( 1 − sensitivity ) + ( 1 − prevalence ) × specificity .

These calculations apply for standard clinical assessment where women are identified either as positive or negative for PE based on a combination of signs/symptoms and clinical findings.

In a situation where a high-risk cut-off point is used to rule in PE and a low-risk cut-off point is used to rule out PE, as is done for the Triage PlGF test, the calculation is modified as shown below:

⁽³⁾ pHighRisk = prevalence × sensitivity RuleIn + ( 1 − prevalence ) × ( 1 − specificity RuleIn )

⁽⁴⁾ pLowRisk = prevalence × ( 1 − sensitivity RuleOut ) + ( 1 − prevalence ) × specificity RuleOut

⁽⁵⁾ pIntermediate = 1 − pHighRisk − pLowRisk .

A similar calculation is applied for the Elecsys sFlt-1 to PlGF ratio test, which has a specific rule-in cut-off point (for PE within 4 weeks) and a specific rule-out cut-off point (for PE within 1 week).

Expectant monitoring/immediate delivery

Clinical guidelines indicate that expectant monitoring is the preferred strategy for pregnancies presenting with suspected PE (particularly those presenting prior to 34 weeks of gestation), with the principal objective of reducing the risk of neonatal respiratory distress syndrome or other complications due to prematurity, while maintaining close monitoring to minimise maternal risk. 123 The ACOG clinical guidance16 indicates that immediate delivery should be offered for PE between 34 and 36 completed weeks when there is evidence of specific abnormal maternal or fetal findings. Abnormal maternal findings include severe hypertension that is refractory to treatment, persistently abnormal haematological or biochemical findings, severe persistent right upper quadrant or epigastric pain (that is unresponsive to treatment and not accounted for by other diagnosis), pulmonary oedema and cerebral or visual disturbance. The fetal findings include abnormal cardiotocography or evidence of fetal compromise on ultrasound.

The model assumes that women presenting with PE before 35 weeks will be managed using expectant monitoring when there are none of the above signs of increased risk for the mother or neonate. From 35 weeks of gestational age onwards, the model assumes that the pregnancy will be managed by immediate delivery. The time to delivery for both strategies is determined by the PELICAN study. 5 These assumptions are in line with NICE guidance. 13

Disease status

Maternal and fetal outcomes in the model are assumed to be primarily related to the underlying condition (i.e. the presence or absence of PE). As a result, the outcome components [3: maternal outcome and 4: fetal outcome (see Figure 8)] of the model are preceded by an evaluation of true disease status. This is determined on the basis of the proportion of each risk group (high/intermediate/low) identified as having PE as shown:

⁽⁶⁾ pDisease HighRisk = prevalence × sensitivity RuleIn pHighRisk pDisease LowRisk = prevalence × ( 1 − sensitivity RuleOut ) pLowRisk pDisease Intermediate = prevalence × ( 1 − sensitivity RuleIn − ( 1 − sensitivity RuleOut ) ) pIntermediate ,

where pDisease_RiskLevel is the probability of PE (disease) among women identified as being at the given level of risk during the risk stratification stage of the model.

Maternal outcome

The NICE guideline CG107 (Box 3) provides advice on the timing of birth for women with PE.

BOX 3 - Advice on the timing of birth for women with PE from NICE CG107

The NICE guideline states the following (see Timing of birth in NICE CG107):13

Manage pregnancy in women with PE conservatively (that is, do not plan same-day delivery of the baby) until 34 weeks.

Consultant obstetric staff should document in the woman’s notes the maternal (clinical, biochemical and haematological) and fetal thresholds for elective birth before 34 weeks in women with PE.

Consultant obstetric staff should write a plan for antenatal fetal monitoring during birth.

Offer birth to women with PE before 34 weeks, after discussion with neonatal and anaesthetic teams and a course of corticosteroids has been given if:

severe hypertension develops refractory to treatment

maternal or fetal indications develop as specified in the consultant plan.

Recommend birth for women who have PE with severe hypertension after 34 weeks when their blood pressure has been controlled and a course of corticosteroids has been completed (if appropriate).

Offer birth to women who have PE with mild or moderate hypertension at 34⁺⁰ to 36⁺⁶ weeks depending on maternal and fetal condition, risk factors and availability of neonatal intensive care.

Recommend birth within 24–48 hours for women who have PE with mild or moderate hypertension after 37⁺⁰ weeks.

National Institute of Health and Clinical Excellence (2010). Adapted from CG107: Hypertension in Pregnancy: Diagnosis and Management. 13 Available from www.nice.org.uk/Guidance/cg107. Reproduced with permission from NICE. The material was accurate at the time of going to press

Point 6 above recommends that women presenting with PE and mild or moderate hypertension from 34 to 37 weeks be offered immediate delivery. This recommendation was based on the findings of the Dutch Hypertension and Pre-eclampsia Intervention Trial at Term (HYPITAT) I study. 124 The HYPITAT I study was criticised by a member of the NICE CG107 guideline group for including hypertension as one of the outcomes contributing to a composite outcome of maternal morbidity and for not including neonatal outcomes. 125 Awareness of the limitations of the HYPITAT I study may have influenced the formulation of the HYPITAT II study. 124 Clinical experts consulted by the EAG made clear that neonatal outcomes were at least as important as maternal outcomes, and a recently published editorial in the British Medical Journal also supports the importance of neonatal outcomes. 123 Consequently, for women presenting between 34 and 37 weeks, we have replaced the values in NICE CG107 (see appendix I)13 that were from the HYPITAT I study124 to reflect maternal and neonatal outcomes measured in the HYPITAT II study. 126

The HYPITAT II study compared expectant monitoring for hypertensive disorders of pregnancy (including PE) in women between 34 and 37 weeks of gestation against induction within 2 days of presentation. In one arm of the trial, women were assigned to induction within 2 days. In the other trial arm, women were expectantly monitored until 37 weeks, at which point these women were induced. Induction at 37 weeks is broadly in line with NICE CG107 on management of women with PE. 13

Figure 9 shows a simplified schematic for the delivery and maternal outcome component of the model. This stage of the model has been developed following model outlines presented in the NICE Hypertension in Pregnancy: The Management of Hypertensive Disorders During Pregnancy CG10713 (relating to aspirin for prevention of PE, gestational hypertension and PE management; these last two models were primarily used to evaluate cost-effectiveness of immediate delivery compared with expectant management).

FIGURE 9.

Delivery and maternal outcome subtree.

The maternal outcome component begins with delivery whether as a result of spontaneous labour (which is expected to be the minority of deliveries in women presenting with suspected PE before 37 weeks of gestation), induced labour or planned caesarean section. This branch corresponds to section 3 of the model in Figure 8. This branch has been included in the model on the expectation that delivery costs and outcomes are likely to be highly influenced by mode of delivery, and that the balance of the modes of delivery is likely to differ for women presenting before 35 weeks of gestation and those presenting between 35 and 37 weeks of gestation. We have adopted a slightly different structure to the NICE gestational hypertension and PE management models, by modelling spontaneous and induced deliveries separately. Induction of delivery will be associated with a higher cost than spontaneous delivery because of the need to administer medication to induce labour and a requirement for maternal monitoring during induction. Each of these modes of delivery may be associated with a risk of conversion to assisted/instrumental vaginal delivery or to emergency caesarean section, and the probability of these outcomes may differ according to whether or not labour was initially spontaneous or induced.

Each mode of delivery is associated with a risk of a severe adverse event associated with the progression of severity of PE during the delivery, which results in convulsions. These adverse events confer higher maternal risk, higher risk of admission to intensive or high-dependency care and a requirement for administration of anticonvulsive therapy. The model assumes that women who do not experience convulsions are transferred to the ward following delivery and those who do not experience any further adverse events have a normal length of stay for the given mode of delivery.

Fetal and neonatal outcomes

Figure 10 shows a simplified schematic for the fetal outcome component of the model. This branch corresponds to section 4 of the model in Figure 8. As noted for maternal outcomes, this stage of the model has been developed with reference to CG107. 13 The model takes a simplified approach to assessing fetal and neonatal outcomes, in which we do not directly model morbidity in terms of clinical manifestations (such as respiratory distress syndrome or sepsis) but instead we model groups of outcomes that may be associated with increased resource use, developmental deficits or differences in HRQoL, and fetal or neonatal mortality. We regard this as a reasonable approach to ensure tractability of the modelling task, but inevitably it involves some simplification of the clinical picture.

FIGURE 10.

Fetal and neonatal outcome subtree. ICU, intensive care unit; HDU, high-dependency unit.

The first branch in the neonatal outcome component of the model establishes whether or not the labour results in a live birth or a stillbirth. This is included in the expectation that the probability of stillbirth is likely to be higher in early deliveries, whether or not these occur because of PE. The following branch in the model relates to admission to neonatal high-dependency or intensive care. The probability of admission to neonatal intensive or high-dependency care is expected to be related to gestational age, presence or absence of PE, principal cause of early delivery (maternal condition vs. fetal distress), mode of delivery and the presence or absence of complication(s) during delivery. As a result, this subtree is applied for all risk groups in the model, with or without PE. However, the probability of experiencing neonatal adverse outcomes of delivery may vary according to the factors mentioned above.

The economic model developed for this assessment has been subjected to a number of validation procedures, including assessment of the clinical validity and credibility of the model structure and data used to populate it. The structure of the model has been presented to, and discussed in detail with, a range of clinical experts (see Acknowledgements). The model has also been exposed to a range of technical validation exercises. Each component of the model has been tested against published sources (using their data) to establish technical and internal validity of model calculations, as reported in the next section.

Model parameters

The following sections report parameters included in the model. For ease of reference, a list of all model parameters and their sources is provided in Appendix 8, and a list of the key model assumptions and their justification is provided in Appendix 9.

The model parameters include diagnostic test accuracy, clinical inputs (such as onset of labour, mode of delivery and birth outcomes) and health sector costs (including costs of biomarker tests, antenatal management, delivery and costs of complications). When possible, clinical data were sourced from the PELICAN study,5 which was conducted in the UK. Targeted searches were conducted to find alternative data sources for model parameters where these were not reported in the PELICAN study. 5 Resource-use assumptions for costing diagnostic and management strategies are presented in full below (based on the companies’ suggested approaches to diagnostic testing, NICE CG10713 and expert opinion). Unit costs were taken from National Schedule of Reference Costs 2013–14 for NHS Trusts and NHS Foundation Trusts,127 NHS Payment by Results Tariff 2013/14,128 and the British National Formulary. 120 Targeted searches were conducted to identify unit costs in cases where these sources were inadequate.

Model validation

The initial technical validation of the model was conducted by replication of published analyses reviewed in Results of the review of economic studies and of analyses developed for (and reported in) the NICE CG on Hypertension in Pregnancy: The Management of Hypertensive Disorders during Pregnancy. 122

For technical validation against the study by Hadker et al. 39 the simplified model structure illustrated in Figure 8 was populated with the reported data and evaluated for the cost analysis reported in the paper. This validation yielded identical results to the published paper. The technical validation against the model developed for the NICE CG on Hypertension in Pregnancy: The Management of Hypertensive Disorders during Pregnancy did not yield identical results to those reported in the appendices to the guideline,122 although the results were similar. Identification of reasons for the differences in results was made difficult by a lack of clarity in the reporting of the model structures used in the analyses and some inconsistencies in the reporting of data used to populate the model.

A second stage of technical validation of the model used data from key clinical studies included in our systematic review of test accuracy (see Chapter 4). Based on the company-recommended cut-off points, these studies provided data on the diagnostic accuracy for ruling in and ruling out PE using the Triage test (PELICAN study5) and the Elecsys sFlt-1 to PlGF ratio test (PROGNOSIS study;133 Álvarez-Fernández et al. 75). The simplified model structure was populated with sensitivity and specificity data for women presenting before 34 weeks and between 34 and 37 weeks of gestation. The model predictions for the number of women with PE (among the suspected PE population) classified into the high-, intermediate- and low-risk categories were compared with those reported in 2 × 2 contingency tables extracted from the studies. The model predictions were identical to those presented in the clinical studies, indicating technical validity of the calculations in the model.

Base-case cost-effectiveness results

This section reports the cost-effectiveness results for women presenting for assessment of suspected PE, prior to 35 weeks and between 35 and 37 weeks of gestation, using the PlGF-based tests in addition to standard clinical assessment as compared against standard clinical assessment alone. The results for costs and QALYs are presented for each diagnostic strategy with incremental costs and QALYs calculated compared with the next best alternative (based on dominance and extended dominance).

Sensitivity analyses

Scenario analyses

This section reports two scenario analyses that were conducted. The first examined the impact of processing and analysing the PlGF-based test results in a near-patient setting instead of in a central laboratory. This is based on an assumption that the Triage test could be employed in a midwifery day unit. The second examined the potential impact of using a PlGF-based biomarker test as a replacement for quantitative proteinuria testing as part of standard clinical assessment.

Clinical effectiveness (test accuracy)

Four studies met the inclusion criteria for the systematic review of test accuracy. These assessed the Triage PlGF test (PETRA study6 and PELICAN study5) and the Elecsys sFlt-1 to PlGF ratio (PROGNOSIS study133 and Álvarez-Fernández et al. study75). As noted above, the PETRA study on the Triage PlGF test was confidential at the time of preparing the current report and has been excluded, although it was available to the EAG and NICE Diagnostics Assessment Committee. Critical appraisal of the three published studies on these tests suggested that the studies were probably at low risk of bias. An exception is a high risk of clinical review bias in all three studies, as the studies diagnosed PE solely according to biomarker test results, whereas in clinical practice the biomarker test results would be interpreted alongside hypertension, proteinuria and/or other signs or symptoms. However, it is unclear whether or not this difference would have led to systematic under- or overestimation of test accuracy outcomes.

The decision problem for this diagnostic assessment was not fully met by the available evidence. No test accuracy studies were found for the PerkinElmer DELFIA Xpress PlGF test or BRAHMS Kryptor sFlt-1 to PlGF ratio test. In addition, no test accuracy studies were identified that assessed index tests as an alternative to quantitative proteinuria testing (part 2 of the NICE decision problem). All the identified evidence from primary studies relates to use of the biomarker tests where the test assays were performed in a laboratory; none of the studies reported assays that could be done in an antenatal clinic (near-patient) setting.

Based on the available published evidence, the Triage PlGF test has high prognostic sensitivity for predicting PE requiring delivery within 14 days of testing, while the Elecsys sFlt-1 to PlGF ratio has high diagnostic sensitivity for rule-out of PE within 1 week of testing and good specificity for rule-in of PE within 4 weeks, although it has a high false-positive rate. However, the primary studies included in the test accuracy review reported different outcomes for each test (prognostic accuracy for the Triage PlGF test; diagnostic accuracy for the Elecsys sFlt-1 to PlGF ratio test), which makes direct comparisons difficult.

For the Triage PlGF test cut-off points of < 100 pg/ml and < 5th percentile of PlGF concentration (test positive), the PELICAN study gave high sensitivity (96%) with good precision (i.e. narrow 95% CIs) for identifying women likely to develop PE requiring delivery within 14 days when presenting with suspected PE at up to 35 weeks of gestation. Diagnostic accuracy outcomes for the Elecsys sFlt-1 to PlGF ratio are for three test cut-off points: 23, 38 and 85; however, the majority of data are from the PROGNOSIS study, which employed the 38 cut-off point. The PROGNOSIS study outcomes suggest that the Elecsys sFlt-1 to PlGF ratio can rule out PE within 1 week of testing in approximately 99% of patients (based on the NPVs for two study cohorts) and has reasonable specificity for ruling in PE within 4 weeks of testing (specificity 83% for two study cohorts, although with a likelihood of false positives: PPVs were approximately 40% for both study cohorts).

The three published studies used definitions of PE that were different from the NICE definition, appearing to include a wider range of women than the NICE definition would permit; notably, the main Elecsys test study (PROGNOSIS) included HELLP syndrome in the definition of PE.

None of the included studies evaluated more than one test, meaning that head-to-head comparisons of tests are not available. Meta-analysis was not possible because of heterogeneity of outcomes, with different test cut-off points reported in the studies.

Cost-effectiveness

The systematic review of economic studies identified a small number of cost models. No models assessed health benefits to mothers or neonates. Targeted systematic searches for HRQoL studies identified a similarly sparse evidence base for HRQoL in gestational hypertension and PE. Most studies were in general pregnancy and post-partum populations and very few used the EQ-5D or any other preference-based utility instrument. Utility values used in the cost-effectiveness model were primarily derived from SF-36 data mapped to EQ-5D. The mapping equation appeared to overestimate utility when compared with directly measured EQ-5D scores. However, it is unclear whether this reflects limitations of the EQ-5D or of the mapping process (see Uncertainties).

The cost-effectiveness model found that both the Triage PlGF test and the Elecsys sFlt-1 to PlGF ratio test were cost-saving compared with standard clinical assessment. The differences in QALYs were very small, requiring four decimal places to show a difference in the base-case analyses for prior to 35 weeks of gestational age and finding no difference between diagnostic assessments for women with suspected PE presenting between 35 and 37 weeks of gestation. The cost differences between the Triage PlGF test and the Elecsys sFlt-1 to PlGF ratio test were slightly in favour of the Triage PlGF test, for women presenting both before and after 35 weeks of gestation.

Sensitivity analyses required up to five decimal points (four decimal places minimum) to show a difference in QALYs for women presenting with suspected PE before 35 weeks of gestation. The most influential parameters in the model were associated with the probability and cost of stay for neonates in the NICU. Most other parameters had very small effects on the model results.

Owing to a lack of available evidence, the accuracy and cost-effectiveness of the DELFIA Xpress PlGF test and of the BRAHMS Kryptor sFlt-1 to PlGF ratio could not be assessed.

Strengths of the assessment

The systematic reviews and economic analysis presented in this report have been carried out independent of competing interests, and were based on methods specified a priori in a peer-reviewed protocol, consistent with the NICE scope and decision problem for this diagnostic assessment. All studies included in the systematic review of test accuracy were critically appraised using a standard approach to identify possible threats to validity and generalisability. A multidisciplinary advisory group commented on the research protocol and on a draft of the final report. Additional clinical experts (see Acknowledgements) also commented on a draft version of the economic model.

The de novo economic model developed by the EAG is based on recognised guidelines. The model structure and data inputs are presented in the current report with explanatory rationale. The economic model is based on data identified from systematic searches for test accuracy, economic studies and HRQoL evidence, and other best available information. The model structure has been subjected to comment and external validation by experts. Additional validation checks were undertaken for model input parameters using published models. The model has been subjected to deterministic and scenario sensitivity analyses to test the robustness of this model compared with alternative data inputs.

Limitations of the assessment

The current assessment is dependent on a relatively limited evidence base, meaning that only four primary studies of test accuracy met the inclusion criteria, and these report on only two of the biomarker tests specified in the decision problem. No evidence to address the second part of the decision problem (exploring the effectiveness of the biomarker tests as a replacement for quantitative proteinuria testing) was identified. Meta-analysis of the primary evidence was not feasible, meaning that the majority of test accuracy information included in this report comes primarily from three studies, namely PETRA and PELICAN for the Triage PlGF test and PROGNOSIS for the Elecsys sFlt-1 to PlGF ratio test.

Data from the PETRA study are confidential and, therefore, have not been presented in detail in this report. However, this does not influence our conclusions, as the test accuracy data for the Triage PlGF test in the economic model were obtained from the PELICAN study, which we considered to be more relevant to a UK population.

The EAG cost-effectiveness model and subsequent analyses based on the model outputs have several limitations. The EAG was forced to make a number of assumptions, with clear justifications, due to the lack of data. The model was designed to perform a probabilistic sensitivity analysis, as indicated in the protocol. However, the absence of evidence directly comparing tests and standard care, or an evidence structure that could support robust, bivariate meta-analysis of diagnostic accuracy data precluded the possibility of conducting an appropriate probabilistic sensitivity analysis. While variation in the individual components of sensitivity and specificity for standard care or the diagnostic tests could be introduced into the model, no data exist to inform on the correlation between these parameters for each test and standard clinical assessment or between the tests and standard assessment. The results of such a probabilistic sensitivity analysis would be of little benefit in informing decision-making, and indeed may be detrimental by mischaracterising uncertainty. The model drew data from a wide range of data sets, which may not all be relevant or generalisable to the UK. Owing to lack of adequate diagnostic effectiveness data, only the Triage PlGF test and the Elecsys sFlt-1 to PlGF ratio test were included in the base-case analysis.

There was a lack of data on utility scores in PE. By necessity, this required expanding the searches to include pregnant and post-partum women who did not have PE. The model requires some assumptions with regard to whether or not these data are applicable, but it is better to include estimates and acknowledge some uncertainty than to exclude measurements and in essence assume no uncertainty in utility differences. In this model, the differences in utility scores are very small, and it is unknown whether or not utility scores measured from women with PE would cause different conclusions. The short time horizon of the model, and rarity of maternal and neonatal mortality in the studies used to populate the models make utility scores unlikely to be a driving factor in the model. No data were available on long-term outcomes of birth for our model population; consequently, this model has not overcome this specific limitation of other models in this treatment area.

Searches were limited to studies published in the English language. This was a pragmatic decision made when developing the review protocol because the current diagnostic assessment is specifically focused on clinical practice in England and Wales. Non-English-language studies would be unlikely to be generalisable to the UK clinical setting because the management of women suspected of having PE varies by country.