FullPIERS Paper-Lancet
FullPIERS Paper-Lancet
FullPIERS Paper-Lancet
Authors
1
Alexandra L Millman, BSc 1
James A Russell, MD 3
Departments of Obstetrics and Gynaecology1, School of Population and Public Health2, Medicine3,
Anesthesiology, Pharmacology and Therapeutics4, and the CFRI Reproduction and Healthy
Obstetrics and Gynaecology11 and Medicine12, University of Otago, Christchurch, New Zealand;
2
Departments of Medicine and Obstetrics and Gynaecology, King Edward Memorial Hospital, WA,
Australia15; and Department of Reproductive Health Research, World Health Organization, Geneva,
Switzerland16.
Dr Peter von Dadelszen, 2H30-4500 Oak Street, Vancouver, BC V6H 3N1, Canada
3
ABSTRACT
BACKGROUND
Pre-eclampsia is a leading cause of maternal deaths. These deaths primarily result from eclampsia,
Estimate of RiSk) model was developed and validated for women with pre-eclampsia to identify
their risk of life-ending, -altering, or -threatening complications within 48h of hospital admission
with pre-eclampsia.
METHODS
We assessed the vulnerable organ systems of 2023 women with pre-eclampsia admitted to tertiary
centres. The outcome of interest was maternal mortality or other serious complications of pre-
eclampsia. Routinely reported and informative variables were included in stepwise backward
elimination regression model to predict the adverse maternal outcome. Performance was assessed
using area under the curve (AUC) statistics. Standard bootstrapping techniques were used to assess
potential overfitting and performance was also assessed in 3 other relevant populations of women
FINDINGS
Predictors of adverse maternal outcome included gestational age, chest pain/dyspnoea, oxygen
saturation (SpO2), platelet count, creatinine, and aspartate transaminase. The fullPIERS model AUC
was 0.88 [95%CI: 0.84, 0.92]. There was no significant overfitting. fullPIERS performed well (AUC
>0.7) up to 7d after eligibility, and for the other HDP cohorts admitted to various levels and places
of care.
INTERPRETATION
4
The fullPIERS model identifies women at increased risk of adverse outcomes up to 7d before
complications arise and can thereby modify direct patient care (e.g., timing of delivery, place of care),
improve the design of clinical trials, and inform biomedical investigations related to pre-eclampsia.
5
INTRODUCTION
systemic inflammation, and remains a leading direct cause of maternal morbidity and mortality
worldwide (1). Reducing the burden of illness associated with pre-eclampsia (2) will address, in part,
the aims of Millennium Development Goal 5 (3;4). In high income countries, this excess maternal
morbidity and mortality relates to both uncontrolled hypertension and the pulmonary and hepatic
The only cure for pre-eclampsia is delivery. For pre-eclampsia arising remote from term, supportive
and temporising measures (“expectant management”) are used to improve perinatal outcomes.
However, the magnitude of the maternal risks associated with expectant management is unclear (1).
The perinatal benefits of expectant management near term are even less clear (1). Concerns around
maternal risk have caused experts to hesitate in recommending expectant management either remote
from, or near to, term (1). At term, maternal benefits derive from a policy of effecting delivery (7).
The best method of risk assessment in pre-eclampsia pregnancies being managed expectantly or
during induction of labour remains unclear (8). Currently, assessment is directed by expert opinion-
based guidelines that perform poorly when operationalised (9). A validated tool that allows real-time
maternal risk stratification is needed to guide care (e.g., expectant management both remote from
term or during an induction of labour). Previous modelling was unsuccessful in predicting adverse
outcomes occurring at any time after admission with pre-eclampsia (10). However, being able to
predict adverse maternal outcomes within a time frame that would inform and guide clinical care
(e.g., 48 hours - 7 days) would optimise both the management of women admitted with pre-
6
Standardising antenatal and postnatal assessment and surveillance of pre-eclampsia with protocols
that recognise the systemic inflammatory model of pre-eclampsia (1) has been associated with
reduced maternal morbidity (11). Using this standardised approach, we have developed and validated
METHODS
fullPIERS was developed and internally validated in a prospective, multicentre study of women who
fulfilled a research definition of pre-eclampsia, and who were admitted to participating academic
tertiary obstetric centres in Canada, New Zealand, Australia, and the United Kingdom (Appendix).
All centres had a general policy of expectant management remote from term to maximise temporal
PIERS was conducted as either a continuous quality improvement (n=4 sites) or a consented
research (n=4 sites initially, eventually only one) project depending on local ethics committee
requirements. Inpatient women with either suspected or confirmed pre-eclampsia received care that
included predetermined guidelines for initial assessment and ongoing surveillance (for details see
(11;12)).
Women were included if admitted with pre-eclampsia, or having developed pre-eclampsia following
admission. Pre-eclampsia was defined as: i) blood pressure (BP) ≥140/90mmHg (at least one
component, twice, ≥4 hours apart, after 20 weeks) and either proteinuria (of ≥2+ by dipstick,
hyperuricaemia (greater than local upper limit of local non-pregnancy normal range) (5), ii) HELLP
syndrome even in the absence of hypertension or proteinuria (13), or iii) superimposed pre-
7
eclampsia (rapidly increasing requirements for antihypertensives, systolic BP (sBP) >170mmHg or
diastolic BP (dBP) >120mmHg, new proteinuria, or new hyperuricaemia). This definition, although
differing from many international definitions (14) reflects both the variable and multisystem nature
of pre-eclampsia at presentation and the spectrum of women seen in clinical practice (15). Women
were excluded if either admitted in spontaneous labour or having achieved any component of the
maternal outcome prior to either fulfilling eligibility criteria or collection of predictor data.
The candidate maternal and fetal predictor variables chosen were those that were predictive,
available, measurable, frequent, and reliable (Table 1) (20). Symptoms, although difficult to quantify,
were included for face validity due to their use to classify severe disease (12;14), and potential
predictive performance in pre-eclampsia (16). While some of the candidate predictors were
associated with components of the outcome (e.g. the predictor of creatinine and the outcome
component of renal insufficiency or failure) they were retained for consideration in the model
because we were interested in predicting the development of adverse events in the future based on
information available at the time of admission. As our study criteria specifically excluded women who
had achieved any component of the outcome, all women included in the modelling had the potential
The components of the combined adverse maternal outcome (Table 3) were: maternal mortality or
morbidities. This outcome was developed by iterative Delphi consensus (17;18). A single case of
Bell’s palsy and two cases of severe ascites were included as the onset and resolution were
8
Customised case report forms and database were utilised by all participating sites. Data were
collected from the patient medical record(s) and predictor variables were collected within 48 hours
of eligibility. If absent, the ‘last observation carried forward’ method was used by which any
preceding observation performed within two weeks of admission was considered current unless
replaced by a more recent value. While not universally supported (19), this is consistent with clinical
practice as clinicians do not re-evaluate what they believe has not changed, and is conservative in
underestimating the effect of any given variable in modelling. For example, 24 hour urine
proteinuria of 0.6g/d measured 4 days previously could be carried forward to the day of delivery for
Lead-time bias: We selected either the date/time of admission with pre-eclampsia or the post-
admission development of pre-eclampsia (which ever was later) to standardise for the level of
Missing values and misclassification: We undertook abstractor training, checked the data collection
methods, monitored data logic, and performed random re-abstraction of charts (randomly in 102
(5%) cases and for all adverse maternal or perinatal outcomes were suspected or confirmed). Cases
of uncertainty (n=13 [1.0%]) were resolved by iterative discussion between PvD, LAM, BP, and the
One highly informative variable, oxygen saturation by pulse oximetry (SpO2), was prone to missing
data before all participating centres achieved regular pulse oximetry. Missing pulse oximetry data
points were assigned a value, 97%, to lie within the normal range (95-100%), assuming that non-use
of oximetry was associated with better clinical state, and biasing analyses to underestimate the
Study size
9
In response to a falling incidence of adverse outcomes observed in all centres, and previously
reported in one site (11), an early decision was made to assess the model iteratively once 200 women
were entered into the database, and monthly thereafter, so that non-informative variables (p>0.2)
in the model (= 15), and I = incidence of the combined adverse outcome (0.13 at any time after
eligibility).
Quantitative variables
For lactate dehydrogenase (LDH), values were corrected to the midpoint of the relevant laboratory’s
normal range to standardise across sites. For women who developed de novo postpartum pre-
Statistical methods
Only candidate predictor variables available for ≥80% of the women were included in modelling, as,
independent variables collected over the first 48 hours to predict the combined adverse maternal
outcome occurring within the first 48 hours after eligibility (Table 1). The ‘worst value’ (e.g., highest
sBP or lowest platelet count) measured prior to outcome occurrence or completion of the 48 hour
time period, whichever was first, was used. A 48 hour time period was chosen because it would
improve perinatal outcomes by giving time for steroid administration remote from term and it
would inform decisions about the place of delivery/in utero transfer from level 1 and 2 units.
10
The relationship between each predictor variable and the combined adverse maternal outcome was
assessed by univariable logistic regression. Continuous variables were modelled using quadratic
terms, and categorised based on risk thresholds to evaluate the potential for non-linearity. Variables
associated with the outcome (p<0.1) were included in the initial multivariable regression model
along with variables deemed important, a priori, on clinical grounds. To avoid colinearity, the
correlation between variables was assessed and the more clinically relevant variable of a pair of
highly correlated variables included. Clinical expectations regarding possible interactions were
specifically examined.
Stepwise backward elimination was used to build the parsimonious final model. The AUC of the
receiver-operating characteristics curve (ROC) was calculated using standard methods (24). The final
model was internally validated using Efron’s enhanced bootstrap method (details available
validation approaches (e.g., splitting data into training and test datasets) because it maximises
Performance was assessed using calibration ability, stratification capacity, and classification accuracy
(27).
women with pregnancy hypertension. First, women admitted with pre-eclampsia to five level I/II
obstetric centres in British Columbia (n=4) and Western Australia (n=1). Second, women admitted
to BC Women’s with either pre-existing or gestational hypertension (17). Third, women with pre-
eclampsia admitted to three academic centres in low and middle income countries (Fiji, South
11
RESULTS
From 1 September 2003 - 31 January 2010, data for 2023 women (2221 fetuses) were entered into
the fullPIERS database from eight international sites (Table 2). There were 261 (12.9 %) combined
adverse maternal outcomes at any time following eligibility. Compared with the women who did not
develop adverse outcomes, the women who developed adverse outcomes were of lower gestational
age at eligibility, and less likely to be either parous, to smoke during the pregnancy, or to be eligible
on the basis of hyperuricaemia. They were more likely to develop HELLP syndrome, and to receive
both antihypertensives and/or antenatal corticosteroids (for either fetal lung maturation or HELLP).
Maternal blood pressure indices, dipstick proteinuria, and AST were higher in women who
developed adverse outcomes, while platelet counts were lower. The eligibility-to-delivery interval did
not vary between groups, except among women eligible at <34wk. Such women who developed
outcomes had briefer eligibility-to-delivery intervals. Women who developed adverse outcomes were
more likely to receive MgSO4 during their clinical course (62% vs 30%) and to deliver babies earlier
and of lower birth weight. Perinatal and infant mortality did not differ significantly between groups.
The median eligibility-to-outcome interval was 4 days (Table 3). These adverse outcomes occurred
antenatally in 6.0% of women, intrapartum in 3.4%, and postnatally in 3.5%. The most common
outcomes reached were pulmonary oedema (63 (3%)) or blood product transfusion (85 (4%)).
Having excluded some historically important variables after univariable modelling, we modelled
using variables with possible explanatory power (Table 4; full list of tested variables and univariable
Developed with data from 1935 women during the first 48 hours after eligibility, fullPIERS, predicts
adverse maternal outcomes within 48 hours of eligibility (AUC ROC 0.88 [95% CI 0.84, 0.92])
12
(Figure 1). The final fullPIERS equation was: logit(pi) = 2.68 + -5.41 x 10-2 (gestational age
(eligibility)) + 1.23 (chest pain/dyspnoea) + -2.71 x 10-2 (creatinine) + 2.07 x 10-1 (platelets) + 4.00
x 10-5 (platelets2) + 1.01 x 10-2 (AST) + -3.05 x 10-6 (AST2) + 2.50 x 10-4 (creatinine x platelet) + -
6.99 x 10-5 (platelet x AST) + -2.56 x 10-3 (platelet x SpO2). On-line fullPIERS propability
calculator available (www.piers.cfri.ca). After 200 cycles of bootstrapping the average optimism was
fullPIERS successfully stratified the population into clinically relevant risk categories (Table 5), with
a large percentage (65%) of women classified into a low risk group (predicted probability of <0.025),
and 4% of women into the highest risk group (predicted probability ≥0.30). The majority (60%) of
women with a predicted probability ≥0.30 had an adverse outcome. Conversely, the adverse
<0.025, and in only 0.4% of women with a predicted probability <0.01 (negative predictive value:,
99.6%).
The classification accuracy of fullPIERS was good. For example, using a predicted probability of
0.05 as a threshold, fullPIERS identified >75% of women who subsequently had events as being
‘high risk,’ while only 16% of the population was incorrectly identified as being ‘high risk.’ In
practice, the predicted probability would best be used as a continuous value, “probability of an
fullPIERS also performed well predicting adverse maternal outcome from 2 to 7 days following
These AUC and risk stratification findings were replicated for women admitted with pre-eclampsia
prior to 34+0 weeks (AUC ROC 0.85 [95% CI 0.79, 0.92]) and for primigravid women admitted with
13
pre-eclampsia defined solely as proteinuric gestational hypertension (AUC ROC 0.87 [95% CI 0.82,
confirmed its performance (i.e., AUC ROC >0.7). The AUC ROC for fullPIERS was 0.77 [95% CI
0.45, 1.00], 0.85 [95% CI 0.65, 1.00], and 0.80 [95% CI 0.66, 0.94] for women admitted to level I/II
centres with pre-eclampsia (n=6 outcomes/139 women), one tertiary centre (level III) with a non-
pre-eclampsia HDP (n=4/224), and LMIC centres with pre-eclampsia (n=17/145), respectively.
DISCUSSION
Key results
We carried out a prospective, international study to develop and validate a maternal outcome
prediction model for women admitted to tertiary units with pre-eclampsia. Among women admitted
to hospital with pre-eclampsia, fullPIERS predicted adverse maternal outcomes occurring within the
first 48 hour following eligibility [AUC ROC 0.88]. The model included the following predictors:
gestational age at eligibility, chest pain/dyspnoea, SpO2, platelet count, serum creatinine, and AST.
PIERS modelling identified SpO2, a clinical variable that has not been included traditionally in lists
of adverse features. All components of the model fulfilled the requirement for clinical face validity,
in view of the particular risks of pre-eclampsia (5), especially remote from term (1). fullPIERS
attained similar stratification capacity, calibration ability, and classification accuracy as established
cardiovascular, adult critical care, and neonatal critical care scores (28-30). fullPIERS should assist
decisions around delivery, especially at gestational ages when expectant management has important
Limitations
14
There are several limitations to this study.
First, to attain generalizability, our population included women who fulfilled a broad definition of
pre-eclampsia, including women without significant proteinuria. Restricting the analysis to the
tightest possible research definition (primigravid women with proteinuric hypertension) did not
Second, while components of our combined adverse maternal outcome are not of equal value, all
components were assessed and validated by iterative Delphi consensus (17;18) and are
Third, the study was performed solely in high income country tertiary obstetric units and in women
fulfilling our research definition of pre-eclampsia. We have begun to address these limitations
through initial assessments of the predictive ability of fullPIERS across the HDP spectrum and are
developing and validating a specific, symptom- and sign-based, version of PIERS (miniPIERS) for
use in rural and remote settings in high, middle, and low income countries..
A fourth limitation was the relatively small sample size, especially when considering the low rate of
adverse maternal outcomes. This may be particularly important with uncommon outcomes such as
eclampsia, as headache and/or visual symptoms did not contribute independently to fullPIERS.
Therefore, our bootstrap validation was only able to confirm the predictive ability of fullPIERS for
the occurrence of the combined maternal outcome. Since internal validation methods such as the
bootstrap have limitations (31), we have commenced a process of external validation of fullPIERS
The fifth limitation is that fullPIERS is limited to maternal surveillance and does not address the
Interpretation
15
fullPIERS accurately predicted adverse maternal outcomes for up to 48 hours, a clinically useful time
period that permits steroid administration, transfer, or induction. Also, fullPIERS maintained good
performance (AUC ROC >0.8), beyond 3 days post-eligibility, and maintained reasonable
performance (AUC ROC >0.7) up to 7 days post-eligibility. Remote from term, measurable perinatal
gains accrue at weekly intervals (32). However, like Ganzevoort et al (10), we were unable to predict
adverse maternal outcomes at any time following admission to hospital with pre-eclampsia. This was
anticipated, as deteriorating maternal and/or fetal status directs clinical decision making, especially
In the PIERS cohort, gestational age on admission for pre-eclampsia was significantly lower, and
independently predictive, in women destined to develop complications. Disease onset <32 weeks is
Many traditional clinical variables of importance were not included in the final model either because
they were collected in <80% of cases (e.g., 24 hour urine), they lacked univariable association with
the combined adverse outcome (9), or they were displaced within the multivariable modelling (e.g.,
blood pressure, ‘heavy’ proteinuria, uric acid, ALT, and LDH) by variables with greater independent
explanatory power. Our findings support the view that once significant proteinuria has been
identified, serum creatinine can be used for monitoring renal function and risk in women with pre-
eclampsia (33).
For face validity, we did examine whether or not blood pressure could be forced into fullPIERS.
Blood pressure did not independently predict adverse maternal outcomes in the multivariable model,
perhaps as it is the sole element of the maternal syndrome amenable to intervention. Effective
antihypertensive agents exist for severe and non-severe pregnancy hypertension (1). During the first
48 hours after eligibility, women who proceeded to develop adverse outcomes had blood pressure
16
indices 3-10mmHg higher than those women with uncomplicated courses. We do not advocate that
Severe systolic (≥160 mmHg) and diastolic hypertension (≥ 110mmHg) convey significant maternal
risks and should be brought into the non-severe or normotensive range (1).
Our results suggest that only one of AST or ALT need to be measured, and that the measurement of
LDH is redundant in these women. Other tests that could reasonably be abandoned in light of these
Why were 24 hour collections performed in fewer than 50% of these women? Pragmatically, we
believe that clinicians faced with a hypertensive woman with proteinuria on dipstick analysis at term
will decide to advise delivery rather than accept the delay inherent in a 24 hour collection; a decision
supported by both the HYPITAT trial (7), and the inaccuracy of 24 hour urine collections for
proteinuria estimation in pregnancy (33). We suggest that dipstick proteinuria, despite its inherent
The low rate of MgSO4 administration to women who developed adverse outcomes (62%) in these
academic tertiary centres was surprising; these women all developed significant personal
complications of pre-eclampsia. While the results of the randomised controlled trials of MgSO4 as
eclampsia prophylaxis are compelling (34), for women with ‘mild’ pre-eclampsia there remains
apparent uncertainty about when, and with whom, to start MgSO4 (12).
Generalisability
First, we believe that these data will help clinicians gain a fuller sense of disease evolution. This may
be what underlay the reduced incidence of adverse maternal outcomes associated with the single site
17
Second, we propose that gestational age, maternal symptoms, pulse oximetry, serum creatinine,
platelet count, and AST be used to stratify maternal risk during the assessment and surveillance of
women admitted with pre-eclampsia using the fullPIERS equation (available on-line at
Third, it appears reasonable to abandon redundant tests. For example, the testing of AST, ALT, and
LDH might be replaced by AST alone without losing important information and with reduced
laboratory costs.
An important impact of fullPIERS may be to identify women at lowest risk of adverse outcomes,
who can be offered expectant management either remote from term for perinatal benefit or at or
By grouping women according to the risk of adverse maternal outcomes, fullPIERS should also
POP-Q in pelvic floor prolapse (35), fullPIERS may, over time, aid in describing the heterogeneous
populations in the pre-eclampsia literature, and enhance the development of new treatments and
interventions.
Although the model-making process is not finished (36), we hope that the planned external
validation (through prospective data collection and using extant international databases) and
implementation of fullPIERS will help to reduce the risk of the life-ending, life-altering (e.g., stroke),
18
ACKNOWLEDGEMENTS
We acknowledge the funding support of the Canadian Institutes for Health Research (CIHR;
International Federation of Obstetricians and Gynecologists (FIGO), Michael Smith Foundation for
Health Research (salary: JMA, LAM, PvD, KRW), and Child and Family Research Institute (salary
The fullPIERS study centres were: in Canada, British Columbia's Women's Hospital/University of
British Columbia, Vancouver, BC; Kingston General Hospital/Queen's University, Kingston, ON,
the Ottawa Hospital (General Campus)/University of Ottawa, Ottawa, ON; and centre hospitalier
Memorial Hospital for Women/University of Western Australia, Subiaco, WA; and in the UK,
Hospital/University of Leeds, Leeds, Yorks. The level I/II units were: St Paul’s Hospital,
Vancouver, BC; the Richmond Hospital, Richmond, BC; Surrey Memorial Hospital, Surrey, BC;
Kootenay Regional Hospital, Cranbrook, BC; and Osborne Park Hospital, Osborne Park, WA,
Australia. The LMIC sites were: the Colonial War Memorial Hospital/University of the South
Pacific, Suva, Fiji; Tygerberg Hospital/Stellenbosch University, Cape Town, South Africa; and
The members of the Delphi consensus were: Canada: P von Dadelszen (maternal-fetal medicine
Walley (critical care medicine (CCM)), JA Russell (CCM) (Vancouver), SK Lee (neonatology)
19
(Toronto), A Gruslin (MFM) (Ottawa), GN Smith (MFM) (Kingston), AM Côté (OIM), J-M
Walters (OIM) (Perth); Brazil: N Sass (MFM) (São Paulo); China: T Duan (MFM), J Zhou (MFM)
(Shanghai); Fiji: S Mahajan (MFM), A Noovao (MFM) (Suva); New Zealand: LA McCowan (MFM)
(Auckland), P Kyle (MFM; now London, UK), MP Moore (OIM) (Christchurch); Pakistan: SZ
Bhutta (MFM), ZA Bhutta (neonatology) (Karachi); South Africa: DR Hall (MFM), DW Steyn
(MFM) (Cape Town); UK: F Broughton Pipkin (PhD), P Loughna (MFM) (Nottingham), S Robson
(MFM) (Newcastle-on-Tyne), M de Swiet (OIM) (London), JJ Walker (MFM and OIM) (Leeds);
The other member of the PIERS research group are: in Canada, Geoffrey Cundiff, Paula Lott,
Brenda Wagner, Lynne Palmer, Sayrin Lalji, D Keith Still, Dany Hugo, Dorothy Shaw (for FIGO)
and George Tawagi; in Fiji: Swati Mahajan, Amanda Noovao; in South Africa: David Hall, D
Wilhelm Steyn; in Uganda: Christine Biryabarema, Florence Mirembe, Annettee Nakimuli; and in the
United States: William A Grobman, Eleni Tsigas (for the Preeclampsia Foundation).
We acknowledge the efforts of the PIERS site co-ordinators and research assistants (Daniel
Johnstone, Ziguang Qu, Kalie Kissoon, Samantha Benton, Jane Hoffer and Aleksandra Kuzmanovic
(St Paul’s) Vancouver, Richmond and Cranbrook; Winnie Lau, Surrey; Lynn Bissonnette,
Sherbrooke; Svetlana Shachkina, Ottawa; Heather Ramshaw, Kingston; Jane Hayes, Leeds; Amanda
Green, Nottingham; Barbra Pullar, Christchurch; Claire Parker, Subiaco; Erika van Papendorp, Cape
Town; and the charge midwives and medical librarian, Mulago Hospital, Kampala) for their
assistance. Thanks to Pamela Lutley, Terry Viczko, Jonathan Lam, and Kelly Richardson for their
roles during the start of this project, and to Yi Lin for preliminary statistical analyses.
20
References
(1) Steegers EA, von Dadelszen P, Duvekot JJ, Pijnenborg R. Pre-eclampsia. Lancet 2010 June
30.
(2) Langer A, Villar J, Tell K, Kim T, Kennedy S. Reducing eclampsia-related deaths--a call to
(3) Hill K, Thomas K, AbouZahr C, Walker N, Say L, Inoue M et al. Estimates of maternal
mortality worldwide between 1990 and 2005: an assessment of available data. Lancet 2007
October 13;370(9595):1311-9.
(4) Horton R. What will it take to stop maternal deaths? Lancet 2009 October
24;374(9699):1400-2.
(5) Lewis, G. and Drife, J. O. Why women die; 2000-2002. Report on confidential enquiries into
(6) Martin JN, Jr., Thigpen BD, Moore RC, Rose CH, Cushman J, May W. Stroke and severe
preeclampsia and eclampsia: a paradigm shift focusing on systolic blood pressure. Obstet
(7) Koopmans C, van den Berg P, Mol BW, Groen H, Willekes C, Kwee A et al. Pregnancy-
induced hypertension and preeclampsia after 36 weeks: induction of labour versus expectant
21
(9) Menzies J, Magee LA, Li J, Lam J, Richardson K, Douglas JM et al. The Canadian
Hypertension Society and National High Blood Pressure Education Program criteria for
2007;26(4):447-62.
(10) Ganzevoort W, Rep A, de Vries JI, Bonsel GJ, Wolf H. Prediction of maternal
August;195(2):495-503.
(11) Menzies J, Magee LA, Li J, Yin R, Lam E, Stuart H et al. Instituting surveillance guidelines
and adverse outcomes in preeclampsia: a single site study. Obstet Gynecol 2006;110(1):121-
7.
(12) Magee LA, Helewa ME, Moutquin JM, von Dadelszen P, Cardew S, Cote AM et al. SOGC
(13) Audibert F, Friedman SA, Frangieh AY, Sibai BM. Clinical utility of strict diagnostic criteria
for the HELLP (hemolysis, elevated liver enzymes, and low platelets) syndrome. Am J
(14) Brown MA, Hague WM, Higgins J, Lowe S, McCowan L, Oats J et al. The detection,
22
(15) Homer CS, Brown MA, Mangos G, Davis GK. Non-proteinuric pre-eclampsia: a novel risk
(17) Milholland AV, Wheeler SG, Heieck JJ. Medical assessment by a Delphi group opinion
(18) Pill J. The Delphi method: substance, context, a critique and an annotated bibliography.
(19) Molenberghs G. Editorial: what to do with missing data? J Roy Stat Soc Ser A 2007;170:861-
3.
(20) Williams AJ. ABC of oxygen: assessing and interpreting arterial blood gases and acid-base
(21) Altman DG, Lyman GH. Methodological challenges in the evaluation of prognostic factors
(22) Harrell FE, Jr., Lee KL, Mark DB. Multivariable prognostic models: issues in developing
models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med
(23) Wasson JH, Sox HC, Neff RK, Goldman L. Clinical prediction rules. Applications and
23
(24) Hanley JA, McNeil BJ. The meaning and use of the area under a receiver operating
(25) Efron B, Tibsherani R. An introduction to the bootstrap. New York: Chapman and Hall;
1993.
(26) Harrell FE, Jr. Regression modeling strategies. New York: Springer Science & Business, Inc;
2001.
(27) Janes H, Pepe MS, Gu W. Assessing the value of risk predictions by using risk stratification
(28) Knuiman MW, Vu HT. Prediction of coronary heart disease mortality in Busselton, Western
and WHO ERICA risk scores. J Epidemiol Community Health 1997 October;51(5):515-9.
(29) Richardson DK, Corcoran JD, Escobar GJ, Lee SK. SNAP-II and SNAPPE-II: Simplified
newborn illness severity and mortality risk scores. J Pediatr 2001 January;138(1):92-100.
(30) Zimmerman JE, Kramer AA, McNair DS, Malila FM. Acute Physiology and Chronic Health
Evaluation (APACHE) IV: hospital mortality assessment for today's critically ill patients. Crit
(32) von Dadelszen P, Magee LA, Lee SK, Stewart SD, Simone C, Koren G et al. Activated
24
(33) Lindheimer MD, Kanter D. Interpreting abnormal proteinuria in pregnancy: the need for a
(34) Duley L, Gulmezoglu AM, Henderson-Smart DJ. Magnesium sulphate and other
2003;CD000025.
(35) Novara G, Artibani W. Surgery for pelvic organ prolapse: current status and future
(36) Leushuis E, van der Steeg JW, Steures P, Bossuyt PM, Eijkemans MJ, van der Veen F et al.
Prediction models in reproductive medicine: a critical appraisal. Hum Reprod Update 2009
September;15(5):537-52.
(37) Lessoway VA, Schulzer M, Wittmann BK, Gagnon FA, Wilson RD. Ultrasound fetal
biometry charts for a North American Caucasian population. J Clin Ultrasound 1998
November;26(9):433-53.
(38) Hadlock FP, Harrist RB, Martinez-Poyer J. In utero analysis of fetal growth: a sonographic
(39) Kramer MS, Platt RW, Wen SW, Joseph KS, Allen A, Abrahamowicz M et al. A new and
improved population-based Canadian reference for birth weight for gestational age.
(40) Bellomo R, Kellum JA, Ronco C. Defining and classifying acute renal failure: from advocacy
to consensus and validation of the RIFLE criteria. Intensive Care Med 2007
March;33(3):409-13.
25
26
Table 1 Variables considered in the PIERS modelling
Variable
≥1 symptom (y/n)
27
MPV/plt ratio Pr:Cr ratio (mg/mM)
Fibrinogen (μmol/L)
midpoint (U/L)
Albumin (g/L)
* missing data filled assuming 97% (described in the Methods); † classified as 0, trace, 1+, 2+, 3+,
(http://www.rcog.org.uk/resources/public/pdf/efm_guideline_final_2may2001.pdf).
AC abdominal circumference; AFI amniotic fluid index; ALT alanine transaminase; APTT activated
partial thromboplastin time; AST aspartate transamninase; BPP biophysical profile; DAP deepest
amniotic fluid pocket; dBP diastolic blood pressure; EDD expected date of delivery; EFW estimated
fetal weight (Hadlock (38)); FHR fetal heart rate; GDM gestational diabetes mellitus; INR
international normalised ratio; LDH lactate dehydrongenase; MAP mean arterial pressure; MPV
28
mean platelet volume; preg pregnancy; RUQ rught upper quadrant; SpO2 oxygen saturation (pulse
oximetry); sBP systolic blood pressure; UA EDF umbilical artery Doppler end diastolic flow
29
Table 2 Characteristics of women in the PIERS study (median [interquartile range] and number (%))
adverse adverse
outcomes outcomes
(N=261 ) (N=1762 )
Gestational age at eligibility (weeks) 33.9 [30.0, 36.6] 36.3 [33.4, 38.3] 8.2E-20
Gestational age at eligibility <34 weeks 133 (51.0%) 503 (28.5%) 3.3E-12
Pre-eclampsia description
Mean arterial pressure 123 [116, 133] 120 [113, 129] 5.2E-05
30
Lowest platelets (x 109/L) 170 [121, 230] 194 [153, 243] 2.8E-06
Interventions
Pregnancy outcomes
(d)
Gestational age at delivery (wk) 34.7 [30.7, 37.0] 37.0 [34.6, 38.7] 8.2E-20
Birth weight (g) 1938 [1189, 2750] 2685 [1935, 3300] 4.5E-18
≥500g)
dBP diastolic blood pressure; EDD expected date of delivery; HELLP haemolysis, elevated liver
enzymes, low platelets; sBP systolic blood pressure. * Data from Kramer et al (39).
31
Table 3 Adverse maternal outcomes (definitions available [www.piers.cfri.ca])
One or more of maternal morbidity or mortality: within 48h within 7d any time
Maternal death 0 0 0
Eclampsia (≥1) 6 10 11
Cardiorespiratory
Myocardial ischaemia/infarction 1 1 1
SpO2 <90% 11 30 41
Pulmonary oedema 22 52 63
Haematological
Hepatic
32
Dysfunction 9 11 12
Haematoma/rupture 0 0 0
Renal
Dialysis 0 0 1
Placental outcomes
Placental abruption 15 24 34
Severe ascites 1 2 2
Bell’s palsy 0 1 1
33
Table 4 Univariable analyses of candidate predictor variables with p<0.1 and collected in >80% of
cases
CI]
Demographics
Maternal age at EDD (yr) 2020 (99.9) 0.99 [0.96, 1.02] 0.57 0.51 [0.46, 0.57]
Number of fetuses (n) 2020 (99.9) 0.86 [0.43, 1.70] 0.66 0.51 [0.45, 0.57]
Gravidity (n) 2020 (99.9) 0.894 [0.81, 0.99] 0.03 0.56 [0.50, 0.61]
Weight at eligibility (kg) 1784 (88.2) 0.99 [0.97, 0.998] 0.03 0.59 [0.53, 0.65]
BMI (kg/m2) 1647 (81.4) 0.99 [0.95, 1.02] 0.45 0.55 [0.49, 0.61]
Height (cm) 1763 (87.2) 0.99 [0.97, 1.02] 0.62 0.52 [0.46, 0.58]
Hypertension (y/n) 2015 (99.6) 0.57 [0.29, 1.11] 0.10 0.53 [0.48, 0.58]
Symptoms
Severe nausea and vomiting 2020 (99.9) 2.14 [1.22, 3.73] 0.008 0.54 [0.48, 0.60]
(y/n)
RUQ/epigastric pain (y/n) 2020 (99.9) 2.92 [1.94, 4.39] 2.7E-07 0.61 [0.55, 0.66]
Headache (y/n) 2020 (99.9) 1.23 [0.83, 1.83] 0.30 0.53 [0.47, 0.58]
Visual disturbance (y/n) 2020 (99.9) 0.99 [0.60, 1.63] 0.96 0.50 [0.45, 0.56]
Chest pain/dyspnoea (y/n) 2020 (99.9) 6.13 [3.56, 10.54] 5.6E-11 0.58 [0.52, 0.66]
Number of symptoms (n) 2020 (99.9) 1.49 [1.26, 1.76] 3.2E-06 0.62 [0.57, 0.68]
Cardiovascular signs
dBP on eligibility (mmHg) 2020 (99.9) 1.04 [1.02, 1.05] 8.6E-05 0.63 [0.57, 0.68]
34
sBP on eligibility mmHg) 2020 (99.9) 1.03 [1.02, 1.04] 1.1E-08 0.65 [0.59, 0.70]
MAP on eligibility (mmHg) 2019 (99.8) 1.04 [1.03, 1.06] 4.6E-08 0.65 [0. 60, 0.71]
Respiratory
SpO2 (filled) (%) * 2020 (99.9) 0.63 [0.58, 0.70] 4.8E-22 0.72 [0.67, 0.78]
Renal
Dipstick (continuous) † 1949 (96.3) 1.43 [1.24, 1.66] 1.6E-06 0.65 [0.59, 0.71]
Creatinine (μmol/L) 2000 (98.9) 1.02 [1.02, 1.03] 4.2E-09 0.63 [0.57, 0.69]
Uric acid (mmol/L) 2008 (99.3) 1.004 [1.00, 1.01] 1.1E-04 0.59 [0.53, 0.65]
Haematological
Platelet count (x 109/L) 2015 (99.6) 0.99 [0.98, 0.99] 4.9E-17 0.69 [0.63, 0.75]
Mean platelet volume (fL) 1953 (96.5) 1.00 [0.88, 1.13] 0.939 0.51 [0.46, 0.57]
MPV x 106/platelet count 1952 (96.5) 45.46 [1.63, 1269] 2.9E-25 0.66 [0.59, 0.72]
ratio
International normalised ratio 1758 (86.9) 2710 [143, 51381] 1.3E-07 0.64 [0.58, 0.70]
(INR)
Activated partial 1759 (87.0) 1.04 [1.02, 1.07] 1.7E-04 0.64 [0.58, 0.70]
Hepatic
Aspartate transaminase (U/L) 1947 (96.2) 1.005 [1.00, 1.01] 1.6E-14 0.73 [0.67, 0.79]
Alanine transaminase (U/L) 2011 (99.4) 1.005 [1.00, 1.01] 7.9E-16 0.72 [0.66, 0.78]
Lactate dehydrogenase (U/L) 1623 (80.2) 1.63 [1.43, 1.86] 2.0E-13 0.75 [0.68, 0.81]
Bilirubin (μmol/L) 1911 (94.5) 1.15 [1.11, 1.18] 8.7E-17 0.67 [0.60, 0.73]
Albumin (g/L) 1749 (86.5) 0.92 [0.88, 0.96] 7.8E-05 0.62 [0.56, 0.68]
35
Fetal assessment tests
FHR 1829 (90.4) 2.15 [1.47, 3.14] 6.9E-05 0.58 [0.51, 0.64]
EFW (%ile category) 1665 (82.3) 0.99 [0.98, 0.99] 5.9E-05 0.61 [0.55, 0.68]
AC (%ile category) 1665 (82.3) 0.99[0.98, 0.99] 5.8E-05 0.61 [0.55, 0.68]
* missing data filled assuming 97% (described in the Methods); † classified as 0, 0.5, 1, 2, 3, 4.
%ile percentile; AC abdominal circumference; BMI body mass index; dBP diastolic blood pressure;
MAP mean arterial pressure; MPV mean platelet volume; RUQ right upper quadrant; SpO2 oxygen
36
Table 5 Risk stratification table assessing the value of the fullPIERS model in risk prediction
outcome
within 48
hours
37
Sens: 0.755
Spec: 0.869
PV+: 0.236
PV‐: 0.985
AUC ROC 0.88 [95% CI 0.84, 0.92]
Figure 1 Performance of the fullPIERS model developed with data from first 48h after eligibility.
Combined adverse maternal outcome predicted within 48h of eligibility using only data collected
prior to the outcome (an on-line tool to calculate fullPIERS probabilities is available at
www.piers.cfri.ca).
AUC ROC area under the curve of the receiver operator characteristic; PV- negative predictive
38
1.0
AUC ROC fullPIERS
0.9
(mean & 95% CI)
0.8
0.7
0.1
0.0
0 1 2 3 4 5 6 7
days after eligibility
Figure 2 fullPIERS areas under the receiver-operator curves (AUCs; error bars: 95% confidence
39