Content uploaded by Fabrice MAMBU Mbika
Author content
All content in this area was uploaded by Fabrice MAMBU Mbika on Oct 27, 2022
Content may be subject to copyright.
Available via license: CC BY-NC-ND 4.0
Content may be subject to copyright.
Development of the PREDS score to predict in-hospital
mortality of patients with Ebola virus disease under
advanced supportive care: Results from the EVISTA
cohort in the Democratic Republic of the Congo
Marie Jaspard,
a,b,1
Sabue Mulangu,
c,1
Sylvain Juchet,
a,b,1
Beatrice Serra,
a,b
Ibrahim Dicko,
a
Hans-Joeg Lang,
a
Baweye Mayoum Baka,
a
Gaston Musemakweli Komanda,
d
Jeremie Muhindo Katsavara,
d
Patricia Kabuni,
e
Fabrice Mbika Mambu,
c
Margaux Isnard,
f
Christophe Vanhecke,
g
Alexia Letord,
h
Ibrahima Dieye,
i
Oscar Patterson-Lomba,
i
Olivier Tshiani Mbaya,
c
Fiston Isekusu,
e
Donatien Mangala,
e
Jean Luc Biampata,
c
Richard Kitenge,
j
Moumouni Kinda,
a
Xavier Anglaret,
b
Jean Jacques Muyembe,
c
Richard Kojan,
a,1
Khaled Ezzedine,
b,k,1
and Denis Malvy
b,l,1
*
a
Alliance for International Medical Action (ALIMA), Dakar, Senegal
b
University of Bordeaux, National Institute for Health and Medical Research (Inserm), Research Institute for Sustainable
Development (IRD), Bordeaux Population Health Center, UMR 1219, Bordeaux, France
c
National Biomedical Research Institute (INRB), Kinshasa, Democratic Republic of the Congo
d
Beni General Hospital, North Kivu province, Democratic Republic of the Congo
e
Kinshasa University Hospital, Democratic Republic of the Congo
f
Intensive Care Unit, Savoie Hospital, France
g
Internal Medicine Department, West R
eunion Hospital, R
eunion, France
h
Surgical Intensive Care Unit, Henri Mondor University Hospital, Cr
eteil, France
i
Analysis Group Inc., Boston, MA 02199, USA
j
Ministry of Health, National Emergency and Humanitarian Action Program, Democratic Republic of the Congo
k
Department of Dermatology, AP-HP, Henri Mondor University Hospital, Cr
eteil, France and Universit
e Paris Est (UPEC), Epi-
DermE research unit, Paris, France
l
Department of Infectious Diseases and Tropical Medicine, Tropical Medicine and Clinical International Health Unit, H^
opital
Pellegrin Bordeaux University Hospital, Bordeaux, France
Summary
Background As mortality remains high for patients with Ebola virus disease (EVD) despite new treatment options,
the ability to level up the provided supportive care and to predict the risk of death is of major importance. This analy-
sis of the EVISTA cohort aims to describe advanced supportive care provided to EVD patients in the Democratic
Republic of the Congo (DRC) and to develop a simple risk score for predicting in-hospital death, called PREDS.
Methods In this prospective cohort (NCT04815175), patients were recruited during the 10
th
EVD outbreak in the
DRC across three Ebola Treatment Centers (ETCs). Demographic, clinical, biological, virological and treatment data
were collected. We evaluated factors known to affect the risk of in-hospital death and applied univariate and multivar-
iate Cox proportional-hazards analyses to derive the risk score in a training dataset. We validated the score in an
internal-validation dataset, applying C-statistics as a measure of discrimination.
*Corresponding author at: Department of Infectious Diseases and Tropical Medicine, Tropical Medicine and Clinical International
Health Unit, H^
opital Pellegrin Bordeaux University Hospital, Bordeaux, France.
E-mail addresses: marie.jaspard@coral.alima.ngo (M. Jaspard), sabuemulo@yahoo.fr (S. Mulangu), sylvain.juchet@coral.alima.
ngo (S. Juchet), beatrice.serra@coral.alima.ngo (B. Serra), barahimadicko@gmail.com (I. Dicko), hansjoerg.lang@alima.ngo
(H.-J. Lang), bmayoum@yahoo.ca (B.M. Baka), drkomanda.gas@gmail.com (G.M. Komanda), jeremiemkats@gmail.com
(J.M. Katsavara), pkabuni@gmail.com (P. Kabuni), mambumbika2@gmail.com (F.M. Mambu), margauxisnard@gmail.com
(M. Isnard), christophevanhecke@yahoo.fr (C. Vanhecke), alexia.letord@gmail.com (A. Letord), ibrahimadieye@g.harvard.edu
(I. Dieye), oscar.Patterson-Lomba@analysisgroup.com (O. Patterson-Lomba), pipombaya@hotmail.com (O.T. Mbaya), mpinda.
isekusu@gmail.com (F. Isekusu), donatienmangala@gmail.com (D. Mangala), jlbiampata@gmail.com (J.L. Biampata),
richardkitenge2@gmail.com (R. Kitenge), moumouni.kinda@alima.ngo (M. Kinda), xavier.anglaret@u-bordeaux.fr (X. Anglaret),
jjmuyembet@gmail.com (J.J. Muyembe), richard.kojan@alima.ngo (R. Kojan), khaled.ezzedine@aphp.fr (K. Ezzedine), denis.
malvy@chu-bordeaux.fr (D. Malvy).
1
These authors contributed equally to the work.
eClinicalMedicine
2022;54: 101699
Published online xxx
https://doi.org/10.1016/j.
eclinm.2022.101699
www.thelancet.com Vol 54 December, 2022 1
Articles
Findings Between August 1
st
2018 and December 31
th
2019, 711 patients were enrolled in the study. Regarding supportive
care, patients received vasopressive drug (n= 111), blood transfusion (n= 101), oxygen therapy (n= 250) and cardio-pulmo-
nary ultrasound (n= 15). Overall, 323 (45%) patients died before day 28. Six independent prognostic factors were identified
(ALT, creatinine, modified NEWS2 score, viral load, age and symptom duration). The final score range from 0 to 13 points,
with a good concordance (C = 86.24%) and calibration with the Hosmer-Lemeshow test (p=0.12).
Interpretation The implementation of advanced supportive care is possible for EVD patients in emergency settings.
PREDS is a simple, accurate tool that could help in orienting early advanced care for at-risk patients after external
validation.
Funding This study was funded by ALIMA.
Copyright Ó2022 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND
license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Keywords: Ebola virus; Sub-Saharan Africa; Supportive care; Outcome; Predictive score; In-hospital mortality
Research in context
Evidence before this study
Mortality from Ebola virus disease (EVD) remains a
major issue, despite the development of new specific
treatments. Advanced supportive care and early detec-
tion of severe cases are levers that could further
reduce mortality. We search Pubmed from January 1
st
2014 to July 1
st
2022 for cohorts related to intensive
care in EVD patients on low-resource settings. The
search was done with no language restriction and
using the search terms “Ebola virus”,“West Africa”,
“Guinea”,“Sierra Leone”,“Liberia”,“Democratic Repub-
lic of the Congo”and “intensive care”in titles and
abstracts. Advance supportive care were already pro-
vided to EVD patients in Sierra Leone in an Ebola treat-
ment center (ETC) equipped with intensive care unit
on a small number of patients in comparison with “reg-
ular ETC”. Access to intensive care was then associated
with lower mortality. In addition, several article high-
light the urgent need to scale up the level of support-
ive care to improve mortality.
We did a similar search for articles related to the
development or validation of prognostic models for
in-hospital death (search terms “prognostic”,“death”,
“outcome”,“score”). Two scores were developed
using clinical, socio-demographics, or virological data
from the Sierra Leone 2014−2015 outbreak data but
were not internally or externally validated and did
not include biological variables. Furthermore, the clin-
ical relevance and feasibility of these two scores is
debatable. One score includes a lot of clinical varia-
bles, including socio-demographic variables whose
association with mortality is not obvious, with no bio-
logical variables, while the other includes very few
variables, with no biological considerations either.
The scoring of EVD severity remains then a key ques-
tion in order to orientate rapidly patients in in
advanced care unit.
Added value of this study
We collected clinical, biological, and virological data on
all patients attending to the three ETCs providing
advanced supportive care and investigational treat-
ments during the 10
th
outbreak in the DRC. With these
data, we were able to develop a prognostic score model
that predicts the day 28 in-hospital risk of death of EVD
patients.
Implication of all the available evidence
This cohort has demonstrated that access to optimized
supportive care can be offered during EVD outbreaks
even in low-resource settings. After being externally val-
idated, early triage using our prognostic score may help
rapid identification of at-risk patients in order to provide
EVD patients with adapted critical care.
Introduction
The Ebola virus disease (EVD) outbreak in the Demo-
cratic Republique of the Congo (DRC) that occurred
between August 2018 and June 2020 involved 3,470
cases and resulted in 2287 deaths. It affected three prov-
inces in the east of the country (Ituri, North Kivu and
South Kivu).
1
After being validated in the 2014−2016
West African outbreak, the specific Ebola Virus (EBOV)
vaccine (r-VSV-ZEBOV, ERVEBO, Merck) was used for
ring vaccination of contacts and contacts of contacts.
2,3
In addition, during this DRC 10
th
outbreak, a specific
treatment was identified. The MEURI (Monitored
Emergency Used for Unregistered Intervention) and
the PALM trials evaluated 3 investigational drugs, two
of them showed efficacy in term of mortality improve-
ment: REGN-EB3and MAb114 (monoclonal antibody
products).
4
Even using these treatment strategies, the mortality
remains high for severe cases.
4
Thus, any strategy that
Articles
2 www.thelancet.com Vol 54 December, 2022
can help reduce mortality should be explored. We pro-
vide in this paper two leads that could help in future
outbreaks: advanced supportive care and early triage of
at-risk patients.
Supportive care is an important therapeutic strat-
egy.
5−10
However, the precise spectrum of supportive
care is wide and there is no clear indication as to its
application, as it depends on a combination of several
complicated actions.
11
The spectrum of supportive care
therefore requires definition in order to optimize its fea-
sibility in the field.
Early triage of patients also remains of prime impor-
tance in reducing mortality rates. Understanding which
patients are more at risk of death and have the highest
chance of being saved is a crucial issue to address.
The present study set out to describe the supportive
care provided to patients during the 10
th
Ebola outbreak
in the DRC, and to develop an individual death risk
score model, herein referred to as PREDS (PRedicting
Ebola Death risk Score).
Methods
Participants and setting
Within the context of the operational response to Ebola
outbreaks in the DRC, the national Ministry of Health,
with the support of the French NGO ALIMA (Alliance
for International Medical Action), has established three
Ebola Treatment Centers (ETCs) in the North Kivu and
Ituri provinces, at the Beni, Katwa and Mambassa urban
centers.
EVISTA (Ebola Virus STAndard of care) is an obser-
vational cohort set up for the duration of the outbreak,
with the aim of describing the standard of care provided
to Ebola patients, and the clinical and biological course
of the disease. All patients with a positive EBOV RT-
PCR test admitted to an ETC during the inclusion
period were included. Patients were followed from
admission until discharge or death in ETC.
Care, treatment and follow-up
All EVD patients received a standard of care in accor-
dance with the DRC Ministry of Health recommenda-
tions and WHO guidelines.
11
Clinical evaluation was
carried out using a modified version of the NEWS2
(mNEWS2) score without the respiratory rate, for practi-
cal reasons (Figure S4).
12
Advanced supportive care
tools were used to improve clinical management. Four
investigational drugs were available through PALM or
MEURI (Remdesivir, REGN-EB3, Zmapp and MAb11).
Follow-up was similar during PALM and MEURI. An
innovative measure, the Biosecure Emergency Room
(BER), was developed to treat cases in individual spaces
within the units.
5,13,14
Haemotological, biochemical, and
virological analysis were done in each ETC (more detail
on the devices available in annex 4).
Data collection
Data regarding the demographic and clinical situations,
biological parameters and medication received were col-
lected for all patients at admission. The outcome and a
summary of the hospitalization were collected at dis-
charge or death. Data regarding vaccination status were
based on patient declaration.
Statistical analysis
Baseline characteristics, follow-up supportive care, treat-
ment, and observed outcomes were summarized using
median and interquartile range (IQR) for continuous
variables, and count and percentage for categorical vari-
ables. Statistical comparisons were carried out using the
Wilcoxon rank-sum test for continuous variables and
chi-squared test for categorical variables. Time-to-event
was estimated as the period from date of admission to
date of death for the patients who died. Patients who
survived the whole study period ceased to be monitored
at the end of their follow-up period, i.e. day 28. Univari-
ate Cox proportional hazard regression models were
used to assess the association of key baseline character-
istics with risk of death for all patients.
Prognostic model development
The selection of candidate predictors included in the
model was based on the literature, the results from the
univariate Cox models, and input from clinical experts.
Three multivariable Cox proportional hazard model
were explored: one full model (i.e. with all candidate
predictors), and two least absolute shrinkage and selec-
tion operator (LASSO) models. The outcome was death
before day 28 after admission. To develop and validate
the PREDS score, patients with non-missing data on all
candidate predictors and the outcome were randomly
allocated to either a training sample (2/3 of the data) or
a validation sample (1/3 of the data). The training sam-
ple was used to develop and parametrize the prediction
model, and the validation sample was used to test out-
of-sample prediction model performance. Predictive
performance was assessed in the validation set by 1)
model concordance, measured by the C-statistic, and 2)
model calibration for risk of death at 28 days, quantified
by the Hosmer-Lemeshow goodness-of-fit test. The Bre-
slow estimator of the baseline hazard was combined
with the HRs to obtain the predicted risk of death for
each patient at 28 days from admission date.
Prognostic score development
PREDS was developed as a simple points-based system
to characterize patient risk scores using the methods of
Sullivan and D’Agostino.
15
Creatinine was categorized
according to a clinically relevant cut-off, and a reference
value operationalized as the midpoint was calculated
using the 1
st
and 99
th
percentile values to minimize the
Articles
www.thelancet.com Vol 54 December, 2022 3
influence of extreme values. A base risk profile was set
to correspond to the lowest risk category for each vari-
able. The constant for the points system (number of
regression units corresponding to one point) was
defined as the increase in risk associated with a 2-unit
(mg/dL) increase in creatinine. Factor states associated
with a higher risk of death were assigned more points, so
a higher points total represented greater risk. The points
for each predictor were totaled, and a table indicating risk
(defined as the probability of dying within 28 days of
admission) for each points total was established.
Prognostic score internal validation
To evaluate the robustness of the risk-scoring algorithm,
the points system based on the training sample was
used to generate risk scores for patients in the validation
sample, and the correlation between the points-based
risk scores and the Cox model-based risk scores was
assessed. To evaluate the impact of missing data, a sen-
sitivity analysis was carried out to assess the model’s
predictive performance in multiple imputed held-out (i.
e., validation) datasets. This approach allowed the
robustness of the prediction model to be evaluated
across a range of imputed datasets to assess how well
the model would perform in other datasets with no
missing data. Multiple imputation was carried out using
the multivariate imputation by chained equations
(MICE) approach,
16
which has been the preferred
approach for handling missing data based on published
simulation studies.
17
In this analysis, 200 validation
imputations were generated. Then, model concordance
C-statistic, p-value for the Hosmer−Lemeshow good-
ness-of-fit test, and correlation coefficient between
points-predicted risk and model-predicted risk were
obtained for each imputed validation set. The mean val-
ues across all 200 validation sets were then calculated
and compared with the respective values obtained in the
non-imputed (empirical) validation set.
All statistical analyses were performed with the R sta-
tistical package version 4.0.3 or later (R Foundation for
Statistical Computing). The results were reported follow-
ing the TRIPOD statement (supplementary material).
Ethical consideration
The consent of patients included in the cohort was (i)
extrapolated from their signed consent to participation
in the RCT or MEURI studies (the data collected were
part of the medical support we had provided for these
patients) or (ii) collected through specific consent forms
for EVISTA. For patients unable to sign the consent
form, relatives were contacted to sign for them. The
EVISTA protocol was approved by the DRC National
Ethics Committee (151/CNES/BN/PMMFI2019) and
registered with clinicaltrial.gov (NCT04815175).
Role of funding sources
The funding sources took no part in designing the
study, collecting, analyzing or interpreting the data,
writing the report or making the decision to submit the
article for publication.
Results
Description of the cohort (Table 1 and S1)
Between August 1
st
2018 and December 31
th
2019, 711
patients were eligible for analysis (Figure 1). The mean
(§SD) age was 28 (§17) years. Patients were admitted
at a median of 4 days from onset of symptoms. In terms
of vaccination status, 175 (33%) declared they had been
vaccinated with the rVSV-ZEBOV-GP vaccine. Of the
134 patients who reported their vaccination date, 93 had
been vaccinated 10 days or less before admission
(median 7 days, IQR 5-8), and 41 more than 10 days
before admission (median 15 days, IQR 12-41). Biologi-
cal parameters at baseline (Table S2) showed a low
platelet count in 158 patients (54%), a high white blood
cell count in 138 (46%), and an increased level of creati-
nine in 271 (49%). The CRP value was above 5 mg/l for
512 patients (99%). Serum sodium was low (<130mmol/
l) for 183 patients (33%) and 166 (35%) had an abnormal
level of potassium. Values for AST and ALT were more
than 5 times the upper limit for normal rang in 224
(59%) and 262 patients (48%) respectively. The albumin
level was low (≤35 g/l) in 490 patients (88%). In terms
of viral load measurement, the EBOV RT-PCR NP Ct
value was ≤22 for 356 patients (54%), while the GP Ct
value was ≤22 for only 83 patients (13%).
Follow-up (Table S1)
Almost all patients (94%) experienced viral-illness-like
symptoms such as fatigue (91%), anorexia (70%) and
headache (57%). Half of them showed respiratory symp-
toms and a critical oxygen saturation below 92%. In
addition, 86% of patients had digestive symptoms, of
which 75% diarrhea and 34% dysphagia. A total of 238
patients (35%) experienced an impaired level of con-
sciousness (CVPU on the ACVPU scale) and 264 (37%)
had no neurological symptoms. The most common neu-
rological symptoms were agitation and coma (in 25%
and 29% of patients respectively). 103 patients (15%)
experienced seizure during follow-up. Overall, 275
patients (39%) had bleeding symptoms of some type,
the most common being venous puncture point bleed-
ing, melena, gingival bleeding and hematemesis (20%,
14%, 12% and 11% respectively).
Investigational treatment and patient care
Overall, 623 patients (88%) were given an investiga-
tional treatment, 453 (73%) in the context of PALM and
170 (27%) in MEURI. Most patients were managed in
Articles
4 www.thelancet.com Vol 54 December, 2022
the BER from the time of admission, mainly for bio-
safety reasons. All the supportive interventions provided
to the patients are detailed in Table 2.
Outcomes
Of the 711 patients involved, 323 (45%) died after a
median of 2 days (IQR 1-5) from admission and 9 days
(IQR 6-12) from onset of symptoms (Table 2). In addi-
tion, lethality dropped to 36% and 29% in patients hos-
pitalized for more than 24 h and 48 h respectively.
The case fatality rate (CFR) was 39% (243/623)
among the patients who received an investigational
treatment, and 91% (79/87) among those who did not
(64 of them could not receive it because they died before
having the possibility to receive the treatment, within
the first 24 h of admission). The CFR of the 175 patients
who declared being vaccinated was 30% (vs. 49% for
unvaccinated). A total of 32 pregnant women and 85
children aged 5 or under were included in the study.
Details of the outcomes for these specific groups are
given in Figure S1 and table S2 respectively.
Prognostic model development
Six candidate predictors were evaluated: four from the
univariate analysis and the literature (ALT, creatinine,
mNEWS2 and EBOV RT-PCR), and two from clinical
expertise (age and time from 1
st
symptom to admission,
widely used by clinicians on the field to evaluate severity
at admission). Some variables were significant at the
univariate analysis (table S2) but were not kept to pro-
pose a score that is not field dependant (e.g. socio-eco-
nomic variables), that can be used on resource-limited
terrains (e.g. biological variables), and with reliable data
(e.g. vaccination). The training sample to develop the
prediction model comprised 279 patients randomly
selected (Figure 2). Compared to the two LASSO mod-
els, prediction performance was similar and calibration
at 28 days was better for the full model (i.e. including
all candidate predictors). Proportional Hazard was
checked graphically and with scaled Schoenfeld resid-
uals (annex 3). The score was developed using this
model.
Prognostic score development
The PREDS score was developed by attaching adding
points according to the HR of the multivariable Cox
model: ALT >5N(aHR=2.48[95%CI1.46-4.23],3
points), creatinine >1.3 mg/dl (aHR = 1.19 [95%CI 1.12
−1.26],3points),mNEWS2score>4(aHR=2.311
[95%CI 1.54−3.48], 2 points), EBOV RT-PCR NP Ct value
≤22(aHR=2.56[95%CI1.56−4.19], 3 points), age ≥
50 years old (aHR = 1.50 [95%CI 0.89−2.54], 1 point)
and time from 1
st
symptom to admission ≥4days
(aHR=1.44[95%CI0.97−2.16], 1 point) (Table 3). The
final score ranged from 0 to 13 points, with lower score
indicatingalowerriskoftheoutcome.Intheinternally
validated dataset, the estimated risk of death at day 28
ranged from 1.9% (0 point) to 80.8% (13 points) (annex
3).Threeriskgroupsweredefined:low(score0−5, sur-
vival probability at D28 92.8% (95%CI 86.3−99.9)),
medium (score 6−9, survival probability at D28 53.7%
(95%CI 40.4−71.3)), high (score 10−13, survival probabil-
ity at D28 11.9% (95%CI 5.2−27.1)) (Figure 3).
Figure 1. Inclusion flowchart.
Articles
www.thelancet.com Vol 54 December, 2022 5
Figure 2. Kaplan Meier probability of death for PREDS individual parameters in the training sample (N= 279).
Articles
6 www.thelancet.com Vol 54 December, 2022
Value N
Age, n (%) 676
≤5 yr 85 (13%)
≤28 days 6 (1%)
6 to 18 yr 101 (15%)
19 to 49 yr 406 (60%)
≥50 yr 84 (12%)
Sex, n (%) 711
Male 302 (42%)
Female 409 (58%)
Positive result on pregnancy test, n (%) 32 (8%)
Vaccination
Patient-reported vaccination with rVSV,
n (%)
175 (33%) 533
Time from vaccination to admission
(days),median [IQR]
8 [5;11] 134
>10 days before admission, n (%) 41 (31%)
≤10 days before admission, n (%) 93 (69%)
Time from 1st symptom to admission
(days), median [IQR]
4 [2;7] 697
Vital signs, n (%)
Respiratory rate ≥24 breaths per
minute
328 (51%) 641
Heart rate ≥110 per minute 207 (31%) 666
Diastolic Arterial Pressure <60 mmHg 134 (23%) 579
Systolic Arterial Pressure <90 mmHg 89 (15%) 579
SpO2 <92% 63 (10%) 602
Mean Arterial Pressure <65 mmHg
a
49 (8%) 579
Symptoms, n (%)
Any type of bleeding symptom
b
123 (17%) 711
Any type of neurologic symptom
c
99 (14%) 711
Level of consciousness, n (%)
ACPVU classification:
d
638
Alert 577 (90%)
Impaired 61 (10%)
mNEWS2 score, n (%)
e
496
Total score ≤4 357 (72%)
Total score >4 139 (28%)
Hematology, median [IQR]
Hemoglobin —g/dl 14 [12;15] 301
Hematocrit —% 40 [36;45] 298
Platelets —G/l 145 [97;222] 294
Leucocytes —G/l 10 [5;21] 300
Lymphocytes —G/l 3 [1;6] 232
Neutrophils —G/l 5 [3;12] 224
Monocytes —G/l 0.4 [0.2;1.0] 222
Renal function, median [IQR]
Urea —mg/dl 18 [9;48] 553
Creatinine —mg/dl 1.1 [0.7;3.6] 548
Electrolytes and Biochemistry, median [IQR]
CRP —mg/l 39 [11;109] 517
CPK —U/liter 674 [267;2116] 519
Sodium —mmol/l 131 [128;134] 551
Potassium —mmol/l 4.2 [3.7;4.8] 469
Table 1 (Continued)
Value N
Amylase —U/liter 106 [72;177] 534
Corrected calcemia—mmol/l
f
2.3 [2.2;2.5] 554
Albumine —g/l 28 [23;33] 556
Glycemia —mg/dl 97 [79;121] 556
Liver function, median [IQR]
Bilirubin —mg/dl 0.6 [0.5;1.1] 511
ALT —U/liter 212 [60;589] 544
AST —U/liter 292 [93;1018] 378
Virology, n (%)
EBOV RT-PCR NP Ct value ≤22 356 (54%) 655
EBOV RT-PCR GP Ct value ≤22 83 (13%) 644
Table 1: Baseline characteristics (N=711).
a
Mean Arterial pressure = 2/3 Diastolic Arterial Pressure + 1/3 Systolic
Arterial Pressure.
b
Epistaxis, hematemesis, hematuria, hemoptysis, melena, purpura,
conjunctival bleeding, gingival bleeding, venous puncture point bleeding.
c
Headache, focal neurological deficit, disorientation/confusion, agita-
tion, convulsions, coma/consciousness disorder, photophobia.
d
ACVPU classification: A: Alert; C: new confusion; V: responsive to
voice; P: responsive to pain, U: Unresponsive.
e
NEWS2 score: National Early Warning Score 2nd version; mNEWS2:
modified NEWS2 score without respiratory rate.
f
Corrected calcemia = measured calcemia −0.025*(albuminemia −
40); measured calcemia in mmol/L, albuminemia in g/L.
Value N
Supportive care
Antibiotics, n (%)
a
636 (100%) 638
Duration (days), median [IQR] 8 [4;14]
Inotropes, n (%)
b
111 (16%) 711
Received 1 dose 69 (62%)
Received 2 doses or more 42 (38%)
Oxygen therapy, n (%) 250 (38%) 661
Duration (days), median [IQR] 2 [1;3] 188
Highest flow, n (%) 218
<10 l/min 154 (71%)
≥10 l/min 64 (29%)
Transfusion, n (%) 101 (14%) 711
Volume (ml), median [IQR] 450 [250;525] 99
Urinary catheter, n (%) 199 (32%) 628
Feeding tubes, n (%) 45 (7%) 654
UltraSound, n (%) 15 (2%) 711
Intraosseus catheter, n (%) 7 (1%) 659
Hospitalized in cube, n (%) 560 (85%) 656
Investigational treatment
Investigational treatment received,
n (%)
623 (88%) 710
Program, n (%)
MEURI
c
170 (27%)
PALM RCT
d
453 (73%)
Drugs, n (%)
MAb114 197 (32%)
Table 2 (Continued)
Articles
www.thelancet.com Vol 54 December, 2022 7
Prognostic score validation
The predictive performance of the multivariable Cox model
showed good concordance (C = 86.24%) and calibration
with the Hosmer- Lemeshow test (p=0.12)forriskof
death at 28 days in the validation sample (Figure S3).
The correlation between points-based risk scores and
the Cox model-based risk scores in the validation sam-
ple (N= 140) was 93.8%. A plot of the observed survival
rates at different time points stratified by predicted risk
quartiles (Figure S2) shows that the model-predicted
risk scores discriminate well between patient groups
with different survival profiles. The sensitivity analyses
indicated that the model had a similar predictive value
across the 200 multiple imputed held-out datasets. The
mean C-statistic (SD) was 82.2% (0.96), compared to
86.2% in the empirical validation set. The mean p-value
for the Hosmer−Lemeshow goodness-of-fit test p-value
(SD) was 0.09 (0.11), while the corresponding value in
the empirical dataset was 0.12. Finally, the mean (SD)
correlation coefficient between points-predicted risk and
model-predicted risk was 92.7% (0.67), compared to
93.8% in the empirical dataset.
Discussion
This study, characterizing more than 700 patients, rep-
resents to our knowledge the largest prospective obser-
vational cohort for EVD.
The initial clinical presentation of these patients was
similar to that described during the 2014−2016 Ebola
outbreak in West Africa.
14,18−22
Likewise,
Value N
Regeneron 191 (31%)
Remdesivir 122 (20%)
Zmapp 107 (17%)
Outcome
Dead, n (%) 323 (45%) 711
Time from admission to death (days),
median [IQR]
2 [1;5]
Alive, n (%) 388 (55%) 711
Time from admission to discharge
(days), median [IQR]
17 [14;22]
Table 2: Care, treatment and outcome (N=711).
a
Ceftriaxone, cefixim, metronidazole, gentamycin, amoxicillin, cipro-
floxacin, cloxacillin
b
Adrenaline, noradrenaline.
c
Consultation on Monitored Emergency Use of. Unregistered and
Investigational Interventions.
d
Mulangu et al, NEJM, 2019.
Risk factor Prevalence,
n%
Hazard Ratio
(95% CI)
Pvalue bregression
coefficient
Points
b
ALT >5N (U/L) 124 (44) 2.48 (1.46, 4.23) 0.0008 * 0.91 3
Creatinine (mg/dL) - 1.19 (1.12, 1.26) per 1mg/dl increase <0.0001 * 0.17 3
c
mNEWS2 score >4 107 (38) 2.31 (1.54, 3.48) 0.0001 * 0.84 2
EBOV RT-PCR NP Ct value ≤22 117 (42) 2.56 (1.56, 4.19) 0.0002 * 0.94 3
Age ≥50 years 38 (14) 1.5 (0.89, 2.54) 0.1281 0.41 1
Time from 1st symptom to admission (days) ≥4 147 (52) 1.44 (0.97, 2.16) 0.0733 0.37 1
Table 3: Multivariate cox proportional-hazards analysis of the training sample and PREDS scoring system (N=279)
a
.
a
The sample size of 279 patients correspond to patients randomly selected from the 711 patients of the cohort with no missing data on all candidate risk fac-
tors (i.e. training sample). CI denotes confidence interval, Hazard Ratios for each variable is adjusted on the other variables shown in the table. Crude Hazard
Ratios corresponding to the univariate analysis are shown in Table S4.
b
Assignment of points to risk factors was based on a linear transformation of the corresponding bregression coefficient. The constant for the points system
(number of regression units corresponding to one point) was defined as the increase in risk associated with a 2-unit (mg/dL) increase in creatinine (B=0.34). (cf
annex 3).
c
In the Multivariate Cox Proportional-Hazards Analysis creatinine is used as a continuous variable, but to assign points for the PREDS score it was catego-
rized according to a clinically relevant cut-off (cfTable 4).
Risk factors Categories Points
ALT >5N (U/L) No 0
Yes 3
Age (years) <50 0
≥50 1
Creatinine (mg/dL) ≤1.3 0
>1.3 3
Time from 1st symptom to
admission (days)
<40
≥41
mNEWS2 score >4No0
Yes 2
EBOV RT-PCR NP Ct value ≤22 No 0
Yes 3
Score category
a
Low 0−5 Points
Medium 6−9 Points
High 10−13 Points
Table 4: Points associated with each of the categories of the
PREDS score and risk categorization.
a
The score is obtained by adding the points obtained by individual risk
factors.
Articles
8 www.thelancet.com Vol 54 December, 2022
hyponatremia, renal dysfunction, thrombocytopenia,
dyskaliemia and hepatic failure are factors previously
described in EVD.
21,23,24
We also recognized that acute
renal failure, high viral load, and a high ALT level were
indicators of a fatal outcome.
14,23,24
The overall mortal-
ity rate for our cohort was 45% - higher than that
reported in the PALM trial in which patients with severe
symptoms on admission were not included.
4
Figure 3. Estimated risk of death and survival probability for a low (Score 0−5), Medium (Score 6−9), or High (Score 10−13)
PREDS score.
Articles
www.thelancet.com Vol 54 December, 2022 9
Because there have been strong calls for improve-
ments in care standard of care by caregivers,
6−10
symp-
tomatic treatment has improved since the West African
epidemic.
9,25
Innovative options such as admission and
management in BER, cardiopulmonary ultrasound to
guide vascular filling, monitoring of biological parame-
ters allowing the adaptation of electrolyte compensation
and oxygen therapy are now available in the field.
5,13
Some patients have even benefited from the placement
of a central venous or bone catheter, or treatment with a
vasopressive drug to manage cardio-circulatory shock.
Specific treatments only have a partial impact on fatality
rates in severe patients on admission,
4,26
which is
largely supported by the experience of EVD manage-
ment in northern countries with high levels of support-
ive care and showing mortality rates of less than 20%,
even without specific treatments.
27
The optimization of
supportive care should be mandatory in future outbreak
management and therefore always involve three ele-
ments: (i) very close monitoring of the patient’s hydra-
tion level (vital constant, electrolytes and possibly
cardiopulmonary ultrasound), (ii) early access to oxygen
therapy, antibiotics and blood transfusion and (iii) the
possibility of advanced resuscitation care such as central
catheter placement, dialysis or mechanical ventilation,
which could lower fatality rates. The availability of inno-
vative measures such as BERs for severe patients could
also help with close monitoring while maintaining bio-
medical safety for caregivers.
5,13
However, recourse to these advanced supportive care
measures, even where they are available, remains lim-
ited in poor-resource settings. In this context, patient tri-
age is of prime importance. The NEWS2 score appears
to be a useful marker of severity that is compatible with
operational constraints; however, it omits important
risk factor variables that have been included in the
PREDS scoring model developed here.
Two previously-published study attempted to gener-
ate a scoring system for evaluation of death. Our col-
leagues in Sierra Leone developed a score including
age, level of education, occupation and symptom
description.
28
Similarly, Hartley et al proposed a score
based on age, symptom at triage and day since first
symptom).
29
Nonetheless, none of those scores include
biological and virological parameters, and the use of
some socio-demographical data are not easily replicable
across different field. Finally, the score developed by
Kangbai et al is too complicated (5 groups and up to 9
sub groups of variables) to be used during outbreaks.
PREDS, on the other hand, does include those biologi-
cal and clinical data, which might help early distinction
of patients requiring immediate advanced supportive
care from those at a lower risk of negative outcome.
Moreover, in view of the data currently available in case
of epidemics, its operational feasibility is high. The
NEWS2 has demonstrated its simplicity in large cohorts
of Lassa fever patients,
30
standard biological devices are
now widely deployed in ETCs and viral load is systemati-
cally performed at patient admission. We have also
developed a form to be filled in for the calculation of the
PREDS score and the determination of the risk level at
admission (annex 5). In the framework of our study we
propose, with the aim to improve survival rates, that
patients who score 6 or more (medium or high PREDS
score) should benefit immediately from innovative
advanced care. This aggressive strategy should be tested
with external cohorts to confirm its general applicability
and accuracy.
This study has some limitations. First, due to its
observational nature, it did not allow the effects of
advanced supportive care to be distinguished from those
of specific investigational treatments on mortality rates.
Second, PREDS was only internally validated and
requires external validation in future epidemics in the
field. Third, data collection was a major challenge in the
field context (war in the region, extremely isolated sites),
leading to issue in data quality and missingness. To
address missing data, we conducted a sensitivity analy-
sis using multiple imputed validation datasets to test
the external validity of the prediction model and subse-
quent risk-scoring system, which indicated that the
model performed well on these multiple imputed held-
out datasets. Finally, our study population was heteroge-
neous, because some patients were vaccinated and some
received specific treatment or benefit from advanced
supportive care which might alter the interpretation of
case fatality rate. However, it should be underlined that
this study was set up in real life condition, and reflects
the daily life of field teams.
In conclusion, despite the development of specific
treatments for EVD, mortality remains high, especially
in patients who present with critical symptoms on admis-
sion to the ETC. Improving the availability of advanced
supportive care is therefore essential as a first step in stra-
tegic management of the disease. In addition, early triage
after admission at the ETC, aimed at identifying patients
at risk of death, is key to managing strategies. The
PREDS score allows this early referral of patients to
appropriate medical care for them. Considering the
trends in EVD cases in recent outbreaks, it is likely that
interventions will be carried out at multiple small ETCs
quickly set up near to where cases of the disease develop,
and that PREDS will be useful in this regard for identify-
ing severe cases at these small intervention units.
Contributors
Marie Jaspard (MJ), Sabue Mulangu (SM), Sylvain
Juchet (SJ), Beatrice Serra (BS), Richard Kojan (RK),
Xavier Anglaret (XA) and Denis Malvy (DM) designed
the study.
MJ, SM, SJ, BS, RK, Hans-Joerg Lang (HJL), Ibrahim
Dicko (ID), Baweye Mayoum Baka (BMB), Gaston
Musemakweli Komanda (GMK), Jeremie Muhindo Kat-
savara (JMK), Jean Louis Muanza Nyengele (JLM),
Articles
10 www.thelancet.com Vol 54 December, 2022
Patricia Kabuni (PK), Fabrice Mbika Mambu (FMM),
Margaux Isnard (MI), Christophe Vanhecke (CV),
Alexia Letord (AL), Olivier Tshiani (OT, Fiston Isekusu
(FE, Jean Luc Biampata (JB) and Jean Jacques Muyembe
(JM) set up the study in DRC, enrolled and monitored
the patients and recorded clinical data.
MJ, SM, SJ, BS, Ibrahima Dieye (ID), Oscar Peterrson-
Lomba (OPS) had access to the raw data.
MJ, SM, SJ, BS, ID, OPL, Khaled Ezzedine (KE) carried
out the analyses. MJ, SM, SJ, BS, RK, Moumouni Kinda
(MK), KE and DM drafted the manuscript.
All authors revised the manuscript critically for impor-
tant intellectual content and approved the final version
before submission.
Data sharing statement
The anonymized individual data and data dictionary for
the study will be made available to other researchers by
Professor Denis Malvy (denis.malvy@chu-bordeaux.fr)
after a methodologically sound proposal has been
approved and a data access agreement signed.
Declaration of interests
No conflict of interest is declared by any of the authors.
SM is listed as the inventor on the patent application
for mAb 114, US Application No.62/087, 087 (PCT
Application No.PCT/US2015/060733) related to anti-
Ebola virus antibodies and their use.
Acknowledgments
Our thanks to all the patients involved in the current
study as well as their caregivers, the investigators and
the research staff at the participating care centers.
Our thanks to the World Health Organization for tech-
nical advice.
Supplementary materials
Supplementary material associated with this article can
be found in the online version at doi:10.1016/j.
eclinm.2022.101699.
References
1 10th Ebola outbreak in the Democratic Republic of the Congo
declared over; vigilance against flare-ups and support for survivors
must continue. https://www.who.int/news/item/25-06-2020-10th-
ebola-outbreak-in-the-democratic-republic-of-the-congo-declared-
over-vigilance-against-flare-ups-and-support-for-survivors-must-con
tinue. Accessed 21 July 2022.
2Henao-Restrepo AM, Camacho A, Longini IM, et al. Efficacy and
effectiveness of an rVSV-vectored vaccine in preventing Ebola virus
disease: final results from the Guinea ring vaccination, open-label,
cluster-randomised trial (Ebola ¸C a Suffit!). Lancet Lond Engl.
2017;389:505–518.
3Coller B-AG, Blue J, Das R, et al. Clinical development of a recom-
binant Ebola vaccine in the midst of an unprecedented epidemic.
Vaccine. 2017;35:4465–4469.
4Mulangu S, Dodd LE, Davey RT, et al. A randomized, controlled
trial of Ebola virus disease therapeutics. N Engl J Med.
2019;381:2293–2303.
5Fischer WA, Crozier I, Bausch DG, et al. Shifting the paradigm -
applying universal standards of care to ebola virus disease. N Engl J
Med. 2019;380:1389–1391.
6Lamontagne F, Fowler RA, Adhikari NK, et al. Evidence-based
guidelines for supportive care of patients with Ebola virus disease.
Lancet Lond Engl. 2018;391:700–708.
7Leligdowicz A, Fischer WA, Uyeki TM, et al. Ebola virus disease
and critical illness. Crit Care Lond Engl. 2016;20:217.
8Langer M, Portella G, Finazzi S, et al. Intensive care support and
clinical outcomes of patients with Ebola virus disease (EVD) in
West Africa. Intensive Care Med. 2018;44:1266–1275.
9Fowler RA, Fletcher T, Fischer WA, et al. Caring for critically ill
patients with ebola virus disease. Perspectives from West Africa.
Am J Respir Crit Care Med. 2014;190:733–737.
10 Lamontagne F, Cl
ement C, Fletcher T, Jacob ST, Fischer WA,
Fowler RA. Doing today’s work superbly well−treating Ebola with
current tools. N Engl J Med. 2014;371:1565–1566.
11 Optimized supportive care for Ebola virus disease. https://www.
who.int/publications-detail/optimized-supportive-care-for-ebola-
virus-disease. Accessed 6 January 2020.
12 National Early Warning Score (NEWS) 2. RCP Lond. 2017; published
online Dec 19. https://www.rcplondon.ac.uk/projects/outputs/
national-early-warning-score-news-2. Accessed 17 September 2021.
13 Lamontagne F, Cl
ement C, Kojan R, Godin M, Kabuni P, Fowler
RA. The evolution of supportive care for Ebola virus disease. Lancet
Lond Engl. 2019;393:620–621.
14 Malvy D, McElroy AK, de Clerck H, G€
unther S, van Griensven J.
Ebola virus disease. Lancet Lond Engl. 2019;393:936–948.
15 Sullivan LM, Massaro JM, D’Agostino RB. Presentation of multi-
variate data for clinical use: the Framingham study risk score func-
tions. Stat Med. 2004;23:1631–1660.
16 Buuren S, Groothuis-Oudshoorn C. MICE: multivariate imputa-
tion by chained equations in R. J Stat Softw. 2011;45. https://doi.
org/10.18637/jss.v045.i03.
17 Marshall A, Altman DG, Holder RL. Comparison of imputation
methods for handling missing covariate data when fitting a Cox
proportional hazards model: a resampling study. BMC Med Res
Methodol. 2010;10:112.
18 Jacob ST, Crozier I, Fischer WA, et al. Ebola virus disease. Nat Rev
Dis Primer. 2020;6:13.
19 Bah EI, Lamah M-C, Fletcher T, et al. Clinical presentation of
patients with Ebola virus disease in Conakry, Guinea. N Engl J
Med. 2015;372:40–47.
20 Mobula LM, MacDermott null N, Hoggart null C, et al. Clinical
manifestations and modes of death among patients with Ebola
virus disease, Monrovia, Liberia, 2014. Am J Trop Med Hyg.
2018;98:1186–1193.
21 Schieffelin JS, Shaffer JG, Goba A, et al. Clinical illness and out-
comes in patients with Ebola in Sierra Leone. N Engl J Med.
2014;371:2092–2100.
22 Lado M, Walker NF, Baker P, et al. Clinical features of patients iso-
lated for suspected Ebola virus disease at Connaught Hospital,
Freetown, Sierra Leone: a retrospective cohort study. Lancet Infect
Dis. 2015;15:1024–1033.
23 van Griensven J, Bah EI, Haba N, et al. Electrolyte and metabolic
disturbances in Ebola patients during a clinical trial, Guinea, 2015.
Emerg Infect Dis. 2016;22. https://doi.org/10.3201/eid2212.161136.
24 Hunt L, Gupta-Wright A, Simms V, et al. Clinical presentation, bio-
chemical, and haematological parameters and their association
with outcome in patients with Ebola virus disease: an observational
cohort study. Lancet Infect Dis. 2015;15:1292–1299.
25 Loignon C, Nouvet E, Couturier F, et al. Barriers to supportive care
during the Ebola virus disease outbreak in West Africa: results of a
qualitative study. PloS One. 2018;13:e0201091.
26 Iversen PL, Kane CD, Zeng X, et al. Recent successes in therapeu-
tics for Ebola virus disease: no time for complacency. Lancet Infect
Dis. 2020;20:e231–e237.
27 Uyeki TM, Mehta AK, Davey RT, et al. Clinical Management of
Ebola virus disease in the United States and Europe. N Engl J Med.
2016;374:636–646.
28 Kangbai JB, Heumann C, Hoelscher M, Sahr F, Froeschl G. Sever-
ity score for predicting in-facility Ebola treatment outcome. Ann
Epidemiol. 2020;49:68–74.
29 Hartley M-A, Young A, Tran A-M, et al. Predicting Ebola severity: a
clinical prioritization score for Ebola virus disease. PLoS Negl Trop
Dis. 2017;11:e0005265.
30 Duvignaud A, Jaspard M, Etafo IC, et al. Lassa fever outcomes and
prognostic factors in Nigeria (LASCOPE): a prospective cohort
study. Lancet Glob Health. 2021;9:e469–e478.
Articles
www.thelancet.com Vol 54 December, 2022 11
