PELOD-2: An Update of the PEdiatric Logistic

Organ Dysfunction Score

Stéphane Leteurtre, MD, PhD1,2; Alain Duhamel, PhD2,3; Julia Salleron2,3; Bruno Grandbastien, MD2,4;
Jacques Lacroix, MD5; Francis Leclerc, MD, PhD1,2 on behalf of the Groupe Francophone de
Réanimation et d’Urgences Pédiatriques (GFRUP)

Objective: Multiple organ dysfunction syndrome is the main cause Setting: Nine multidisciplinary, tertiary-care PICUs of university-
of death in adult ICUs and in PICUs. The PEdiatric Logistic Organ affiliated hospitals in France and Belgium.
Dysfunction score developed in 1999 was primarily designed Patients: All consecutive children admitted to these PICUs (June
to describe the severity of organ dysfunction. This study was 2006–October 2007).
undertaken to update and improve the PEdiatric Logistic Organ Intervention: None.
Dysfunction score, using a larger and more recent dataset.
Measurements and Main Results: We collected data on variables
Design: Prospective multicenter cohort study.
considered for the PEdiatric Logistic Organ Dysfunction-2 score
during PICU stay up to eight time points: days 1, 2, 5, 8, 12,
1
Pediatric Intensive Care Unit, Jeanne de Flandre University Hospital, Lille, 16, and 18, plus PICU discharge. For each variable considered
France. for the PEdiatric Logistic Organ Dysfunction-2 score, the most
2
UDSL, Univ Lille Nord de France, Lille, France.
abnormal value observed during time points was collected. The
3
Department of Biostatistics, CHU Lille, Lille, France.
outcome was vital status at PICU discharge. Identification of the
4
Department of Epidemiology and Public Health, Calmette University
Hospital, Lille, France. best variable cutoffs was performed using bivariate analyses. The
5
Pediatric Intensive Care Unit, Sainte-Justine Hospital, Université de PEdiatric Logistic Organ Dysfunction-2 score was developed
Montréal, Montréal, Canada. by multivariable logistic regressions and bootstrap process. We
Authors belonging to Groupe VALIDscore of the GFRUP: D. Biarent
(Bruxelles, Belgium), R. Cremer (Lille, France); S. Dauger (Robert
used areas under the receiver-operating characteristic curve to
Debré-Paris, France), M. Dobrzynski (Brest, France), G. Emériaud evaluate discrimination and Hosmer-Lemeshow goodness-of-
(Grenoble, France), S. Renolleau (Trousseau-Paris, France), M. Roque-
fit tests to evaluate calibration. We enrolled 3,671 consecutive
Gineste (Toulouse, France), D. Stamm, N. Richard (Lyon, France), and I.
Wroblewski (Besançon, France). patients (median age, 15.5 mo; interquartile range, 2.2–70.7).
All authors contributed to the study design drafted by Dr. Leteurtre. Mortality rate was 6.0% (222 deaths). The PEdiatric Logistic
Drs. Leteurtre, Grandbastien, and Leclerc contributed to the clinical
implementation of the study and supervision of the patients. Dr. Duhamel
Organ Dysfunction-2 score includes ten variables corresponding
and Mr. Salleron designed and did the statistical analysis and verified to five organ dysfunctions. Discrimination (areas under the
its accuracy. Dr. Leclerc obtained funding and supervised the study. Dr.
receiver-operating characteristic curve = 0.934) and calibration
Leteurtre, Dr. Duhamel, Ms. Salleron, Dr. Grandbastien, and Dr. Leclerc
had full access to all data. All authors helped draft this report or critically (chi-square test for goodness-of-fit = 9.31, p = 0.317) of the
revised the draft. All authors reviewed and approved the final version of the PEdiatric Logistic Organ Dysfunction-2 score were good.
report and had final responsibility for the decision to submit for publication.
Supplemental digital content is available for this article. Direct URL Conclusion: We developed and validated the PEdiatric Logistic
citations appear in the printed text and are provided in the HTML and PDF Organ Dysfunction-2 score, which allows assessment of the severity
versions of this article on the journal’s website (http://journals.lww.com/
of cases of multiple organ dysfunction syndrome in the PICU with
ccmjournal).
Supported, in part, by grant from the French Ministry of Health. a continuous scale. The PEdiatric Logistic Organ Dysfunction-2
The sponsors of the study and the funding source had no role in the study score now includes mean arterial pressure and lactatemia in the
design, data collection, data analysis, data interpretation, writing of the cardiovascular dysfunction and does not include hepatic dysfunction.
manuscript, or in the decision to submit to publication.
The score will be in the public domain, which means that it can be
The authors have disclosed that they do not have any potential conflicts
of interest. freely used in clinical trials. (Crit Care Med 2013; 41:1761–1773)
For information regarding this article, E-mail: stephane.leteurtre@chru-lille.fr Key Words: intensive care units; multiple organ failure; outcome
Copyright © 2013 by the Society of Critical Care Medicine and Lippincott assessment; pediatric; scoring methods; severity of illness index
Williams & Wilkins
DOI: 10.1097/CCM.0b013e31828a2bbd

Critical Care Medicine www.ccmjournal.org 1761

Leteurtre et al

D
escribing the severity of illness of critically ill patients February 7, 2007. The study design was approved by the ethics
while they are in an ICU is very important: Reliable committee of the Société de Réanimation de Langue Française
quality assurance and quality assessment cannot be on April 27, 2007.
done without such data. Furthermore, an accurate marker of
severity of illness can be used as an outcome measure in clinical Item Selection
studies. Multiple organ dysfunction syndrome (MODS), defined The variables used to create and validate the PELOD-2 were
as the presence of two or more organ dysfunctions, is a good abstracted from the PELOD (Glasgow Coma Score, pupillary
candidate marker of severity of illness because MODS is the reactions, heart rate, systolic blood pressure, creatinine,
main cause of death in adult ICU (1) and in PICU patients (2, 3). Pao2/Fio2, Paco2, mechanical ventilation, WBCs, platelets,
Several definitions of organ dysfunctions and several sets aspartate transaminase, prothrombin time, and international
of diagnostic criteria of MODS have been published (4–6). normalized ratio) and the P-MODS (lactate, Pao2/Fio2,
In the PICU, the relationship between the number of organ bilirubin, fibrinogen, and blood urea nitrogen) scores (2, 8).
dysfunction, which is somewhat quantitative, and mortality is Furthermore, the mean arterial pressure, which is an item of
better than it is between the presence or absence of MODS, the SOFA score for adults, was added because it is considered a
which is dichotomous, and mortality (2, 6). Furthermore, good marker of organ perfusion (9).
it is important to develop scores that consider the higher
and the lower risk of death associated with the different Data Management
organ dysfunctions (2). Adult MODS scores were developed Patients were monitored until they died or were discharged from
using mortality as a dependent variable (7, 8). It is with the PICU, whichever happened first. Our team of investigators
these considerations in mind, the Pediatric-Multiple Organ showed in 2010 that one does not need to collect data on the
Dysfunction Score (P-MODS) (9), the PEdiatric Logistic items of the PELOD everyday: Thus, we collected data during
Organ Dysfunction (PELOD) score (2, 10, 11), and a modified PICU stay according to the set of 8 days (days 1, 2, 5, 8, 12,
Sequential Organ Failure Assessment (SOFA) score for children 16, and 18, plus the discharge day) that were identified as the
were created (12). The PELOD was developed in 1999, more optimal time points to estimate the daily PELOD (dPELOD)
than a decade ago (10). It is by far the most frequently used score scores (16). Days were counted by 24-hour interval, from the
aiming to describe the severity of cases of P-MODS. Because time of admission to 24 hours after admission and so on.
of changes over time in case mix and clinical practice, the Variables were measured only if the attending physician thought
it appropriate (i.e., if justified by clinical status of patient). If a
performance of prognostic and descriptive models deteriorate,
variable was not measured, we assumed that it was identical to
and there is a need to re-calibrate them (13). Furthermore, the
the previous measurement (i.e., the physician thought the value
PELOD was not free of some limitations; for example, a Brazilian
of the variable had not changed) or normal (i.e., the physician
study reported that the PELOD kept a very good discriminative
thought the value of the variable was normal). Physiologic data
capacity in Brazilian PICUs, but its calibration was poor (14).
from the preterminal period (the last 4 hr of life) were discarded
Also, even though PELOD is quantitative, it is discontinuous,
(17). As for previously published severity and MODS scores, the
which may cause problems when doing some statistical analyses
most abnormal value of each variable observed during each of
(15). This study was undertaken to update and to improve the
these time points was considered to build the PELOD-2 score.
PELOD score, using a larger and more recent dataset.
Clinical data were prospectively recorded on a standardized
case-report form. Previously trained physicians (one per
METHODS center) entered data into a web-based database respecting
The 33 university-affiliated PICUs that were members of confidentiality requirements (Epiconcept, Paris, France). A
the Groupe Francophone de Réanimation et Urgences research assistant screened the database weekly for quality
Pédiatriques (GFRUP) were invited to participate: nine PICUs control and, if needed, sent a report to investigators. One
provided, on a voluntary basis, the information requested quality control visit was done in each center during the study.
(eight in France and one in Belgium). All consecutive children Patients’ data were collected anonymously, but investigators
admitted to these PICUs between June 21, 2006, and October held a nominal list for quality control.
31, 2007, were included. Children with a history of prematurity
and hospitalized after birth were enrolled. Patients over 18 Statistical Analysis
years and newborns who were premature (< 37-wk gestation) Identification of Covariates and Their Cutoff.  The association
and admitted at birth were excluded. Although the period of between the outcome (death/survival at PICU discharge) and
data collection varied between units, for each unit, all patients each variable was first investigated using bivariate logistic
admitted consecutively during the study period were included. regression. As the log-linearity assumption for continuous
variables was not verified, they were transformed in categorical
Ethical Considerations variables. This transformation process took into account two
The PELOD score does not require that more measurements groups of variables according to literature data: those for which
be done than in standard practice. The study and its database the normal values depend on the patient’s age (namely heart
were declared safe and were approved by the French authorities rate, systolic and mean arterial pressure, and creatinine) and
(Commission Nationale de l’Informatique et des Libertés) on the others. The cutoffs were identified by using a decision tree

1762 www.ccmjournal.org July 2013 • Volume 41 • Number 7

 Pediatric Critical Care

procedure with the Chaid method (14). The final cutoff values Table 1. Baseline and Clinical Characteristics,
were validated on the basis of their clinical relevance, the results of Reason for Admission and Primary Disease
the bivariate logistic regression, and the existence of a monotone
at Entry, and Outcomes of Children Admitted
relation between the death rates and the levels of the categorized
variables (supplemental data, Supplemental Digital Content 1,
to Nine PICUs (June 2006–October 2007)
http://links.lww.com/CCM/A639). Characteristics and Outcomes Value
Identification of the Predictive Model. A multivariable
logistic regression was performed with all variables (full Baseline characteristics
model). The simplification of this full model was done using   Gender (male), n (%) 2,097 (57.1)
another multivariable logistic regression with backward   Age (mo), median (IQR) 15.5 (2.2; 70.7)
selection at the level p = 0.05. The stability of the selected model
   0 to < 1, n (%) 627 (17.1)
was investigated using the bootstrap resampling method (16)
(supplemental data, Supplemental Digital Content 1, http://   1–11, n (%) 1,068 (29.1)
links.lww.com/CCM/A639).   12–23, n (%) 399 (10.9)
Creation of the PELOD-2 Score. For some categorized   24–59, n (%) 559 (15.2)
variables, some levels were pooled according to the odds ratio
  60–143, n (%) 562 (15.3)
estimates. All cutoff values were rounded to the nearest integer
to have a user-friendly score. The model was rebuilt considering   ≥ 144, n (%) 456 (12.4)
these simplifications (supplemental data, Supplemental Digital   Recovery post procedure,a n (%) 955 (26.0)
Content 1, http://links.lww.com/CCM/A639). The PELOD-2   Pediatric Index of Mortality 2 score 1.42 (0.78; 4.34)
was obtained from the coefficients of this final multivariable   (predicted death rate in %) median (IQR)
logistic regression. The coefficients were multiplied by two and
  Primary reason for PICU admission, n (%)
rounded to the nearest integer to have a user-friendly score.
Validation.  The comparison of the PELOD-2 between the   Respiratory 1,664 (46.6)
two groups (death/survival) was performed by a Student t test.   Neurologic 662 (18.5)
The discriminant power of the PELOD-2 score was estimated   Cardiovascular 673 (18.8)
using the area under the receiver-operating curve (AUC)   Hepatic 40 (1.1)
with 95% CI, and calibration was assessed using the Hosmer-
Lemeshow chi-square test. We addressed the optimism bias   Genitourinary 96 (2.7)
using a bootstrap resampling method. The stability of the score   Gastrointestinal 205 (5.7)
was estimated by cross validation (17) (supplemental data,   Endocrine 57 (1.6)
Supplemental Digital Content 1, http://links.lww.com/CCM/
  Musculoskeletal 45 (1.3)
A639). We used a logistic regression to investigate the relation
between each organ dysfunction and mortality. We used a   Hematologic 45 (1.3)
stepwise multiple regression to evaluate the relative weight of   Miscellaneous/undetermined 99 (2.7)
each organ dysfunction on the PELOD-2 scoring system. For   Mixed 85 (2.4)
these two analyses, the independent variables were ordinal
  Cause of primary diseases at entry, n (%)
variables of each organ dysfunction.
  Infection 863 (23.5)
Comparison Between the PELOD and PELOD-2 Scores. The
PELOD score was computed and calibrated using a logistic   Trauma 325 (8.9)
regression to compare distribution and AUC of both scores.   Congenital disease 1,123 (31.0)
The results are expressed by medians and interquartile   Drug poisoning 72 (2.0)
ranges (IQR) or means and sd for continuous variables and by
  Cancer 120 (3.3)
frequencies and percentages for categorical variables. Statistical
analyses were performed using SAS software version 9.2 (SAS   Diabetes 41 (1.1)
Institute, Cary, NC). A p value of less than 0.05 was considered   Allergic/immunologic diseases 55 (1.5)
statistically significant.   Miscellaneous/undetermined 1,072 (29.2)
  Elective PICU admission, n (%)a
970 (26.4)
RESULTS Outcomes
The nine participating PICUs were devoted to medical,
trauma, and postoperative care (including cardiac surgery)   Mechanical ventilation, n (%) 1,926 (52.5)
and were representative of the 33 university-affiliated PICUs   Length of ICU stay (d), median (IQR) 2 (1; 5)
of the GFRUP in terms of recruitment (medical/surgical,  Mortality, n (%) 222 (6.0)
neonatal/pediatric) (supplemental data, Supplemental Digital a
According to Pediatric Index of Mortality 2 instructions.
Content 1, http://links.lww.com/CCM/A639 for details on the Total number of patients included in the study are 3,671.
nine participating and the 24 nonparticipating sites). Patients IQR = interquartile range.

Critical Care Medicine www.ccmjournal.org 1763

Leteurtre et al

were enrolled for a median period of 15 months (IQR, 7–16). in Table 5. Finally, the PELOD-2 includes 10 variables involving
The median number of admissions per PICU during the study five organ dysfunctions. For each variable, the severity level is
period was 442 (IQR, 132–581). ranging from 0 (normal) to a maximum of 6 (Table 6).
Among 3,675 consecutively screened children, two older than The AUC of the PELOD-2 was 0·942 (95% CI, 0.925–
18 years and two with incomplete data were excluded. Thus, 0.960). The calibration assessed by the Hosmer-Lemeshow chi-
3,671 patients were retained for analysis, including 222 who square test was equal to 6.74 (p = 0.565) (supplemental data,
died in the PICU (mortality rate, 6.0%). Baseline characteristics Supplemental Digital Content 1, http://links.lww.com/CCM/
and outcomes of the enrolled population are detailed in A639). After correction for the optimism bias, the AUC and the
Table 1. The proportion of medical and surgical cases was 66.5% calibration of the PELOD-2 were 0.934 and 9.31 (p = 0.317),
to 33.5%. The median age was 15.5 months (IQR, 2.2–70.7). respectively. The cross validation covariate was associated with
The bivariate analyses demonstrated that all the candidate a β value of 0.89 close to 1.
variables were predictive of death except bilirubin (p = 0.53); Median PELOD-2 score was 4 (IQR, 2–7); mean PELOD-2
these 17 significant variables are listed with their cutoffs in score was 4.8 (sd, 4.3). PELOD-2 was significantly (p < 0.0001)
Tables 2 and 3. Mean arterial pressure and systolic blood pressure higher in nonsurvivors than in survivors (mean, 14.9 [sd, 6.1]
provided similar information; we selected mean arterial pressure vs mean, 4.2 [sd, 3.2], respectively).
to describe cardiovascular dysfunction. The results of the All organ dysfunctions retained in the PELOD-2 score were
multivariable logistic regression performed on these 16 variables closely related to the risk of mortality. The maximum points
are reported in the supplemental data (Supplemental Digital for each organ ranged between 2 and 10. Neurologic and
Content 1, http://links.lww.com/CCM/A639) (full model). respiratory dysfunctions were the most important markers,
After backward selection with bootstrap validation, 10 variables explaining, respectively, 48% and 29% of the variance with
were retained (Table 4). From the results of this multivariable respect to the risk of mortality (Table 7). Figure 1 shows the
analysis, the following levels were pooled with the reference level: distribution of patients for each organ dysfunction.
lactatemia (mmol/L) between 3.97 and 5.37, Pao2 (mm Hg)/ Five hundred fifty-six (15%) patients had no organ
Fio2 ratio between 60.5 and 136.3, noninvasive ventilation, and dysfunction, 1,016 (28%) patients had one, 994 (27%)
WBC count (× 109/L) between 2.15 and 4.09 (significant level patients had two, and 1,105 (30%) patients had three or more.
> 0.2). For creatinine, the two levels of risk had nearly equal Table 8 shows mean PELOD-2 scores and outcome stratified
odds ratio values (2.42 and 2.90, respectively); they were pooled by number of organ dysfunctions.
in a unique class (≥ cutoff 1). All cutoff values were rounded The distribution of the PELOD-2 is different from that of
to be more user friendly, when appropriate. The final logistic the PELOD: The PELOD-2 is a continuous score that can take
regression, which considered these simplifications, is detailed all integer values from 0 to 33 (Fig. 2). The value of the AUC of

Table 2. Identification of Cutoff Values for Age-Dependent Variables

Age (mo)

Variable and Cutoff <1 1–11 12–23 24–59 60–143 ≥ 144

Creatinine (µmol/L)
  Cutoff 1 70 22 34 50 58 93
  Cutoff 2 94 47 59 75 83 117
Heart rate (beats/min)
 Cutoff 207 215 203 191 176 167
Mean arterial pressure (mm Hg)
  Cutoff 1 16 25 30 32 35 37
  Cutoff 2 30 39 44 46 49 51
  Cutoff 3 46 55 60 62 65 67
Systolic arterial pressure (mm Hg)
  Cutoff 1 22 38 47 49 54 58
  Cutoff 2 34 49 58 60 65 69
  Cutoff 3 43 58 67 69 74 78
  Cutoff 4 52 68 76 79 84 87
  Cutoff 5 63 79 87 90 95 98

1764 www.ccmjournal.org July 2013 • Volume 41 • Number 7

 Pediatric Critical Care

Table 3. Candidate Variables for the Pediatric Logistic Organ Dysfunction-2 Score
Variable Survivors (n = 3,449) (%) Nonsurvivors (n = 222) (%) p
Non–age-dependent variables
  Glasgow Coma Score
  3–4 53 (1.54) 103 (46.61)
  5–10 337 (9.77) 28 (12.67) < 0.001
  ≥ 11 3,059 (88.69) 91 (40.99)
  Pupillary reaction
  Both fixed 37 (1.07) 108 (48.65)
  Both reactive 108 (98.93) 114 (51.35) < 0.001
  Lactatemia (mmol/L)
  < 3.97 3,121 (90.49) 106 (47.75)
  3.97–5.36 158 (4.58) 20 (9.01)
  5.37–11.06 136 (3.94) 43 (19.37) < 0.0001
  ≥ 11.07 34 (0.99) 53 (23.87)
  Uremia (mg/dL)
  < 27 2,205 (63.93) 71 (31.98)
  27–36 595 (17.25) 29 (13.06) < 0.0001
  ≥ 37 649 (18.82) 122 (54.95)
 Pao2 (mm Hg)/Fio2 ratio
  < 60.5 81 (2.35) 47 (21.27)
  60.5–136.2 250 (7.25) 23 (10.41) < 0.001
  ≥ 136.3 3,118 (90.40) 152 (68.47)
 Paco2 (mm Hg)
  < 58.5 3,099 (89.85) 155 (69.82)
  58.5–94 315 (9.13) 49 (22.07) < 0.0001
  ≥ 94.5 35 (1.01) 18 (8.11)
 Ventilation
  No 1,470 (42.62) 10 (4.50)
  Noninvasive 261 (7.57) 4 (1.80) < 0.001
  Invasive 1,718 (49.81) 208 (93.69)
  WBC count (× 109/L)
  < 2.15 85 (2.46) 34 (15.32)
  2.15–4.0 170 (4.93) 21 (9.46) < 0.0001
  ≥ 4.1 3,194 (92.61) 167 (75.23)
  Platelets (× 10 /L)
9

  < 76.5 237 (6.87) 72 (32.43)

  76.5–141 287 (8.32) 36 (16.22) < 0.0001
  ≥ 141.5 2,925 (84.81) 114 (51.35)

(Continued)

Critical Care Medicine www.ccmjournal.org 1765

Leteurtre et al

Table 3. (Continued). Candidate Variables for the Pediatric Logistic Organ

Dysfunction-2 Score
Variable Survivors (n = 3,449) (%) Nonsurvivors (n = 222) (%) p
  Fibrinogen (mg/dL)
  < 81 92 (2.67) 31 (13.96)
  81–146 156 (4.52) 32 (14.41) < 0.0001
  ≥ 147 3,201 (92.81) 159 (71.62)
  Aspartate transaminase (IU/L)
  < 111.5 3,102 (89.94) 122 (54.95)
   111.5 to < 339.5 235 (6.81) 48 (21.62) < 0.0001
  ≥ 340 112 (3.25) 52 (23.42)
  Prothrombin time (s)
  < 34.5 110 (3.19) 61 (27.60)
  34.5–55 335 (9.71) 56 (25.34)
  55.5–69 418 (12.12) 29 (13.12) < 0.001
  ≥ 69.5 2,586 (74.98) 76 (34.23)
  International normalized ratio
  < 3 3,429 (99.42) 217 (97.75)
  ≥ 3 20 (0.58) 5 (2.25) 0.0149
Age-dependent variables a

  Mean arterial pressure (mm Hg)

   < Cutoff 1 21 (0.61) 39 (17.57)
   Cutoff 1–cutoff 2 276 (8.0) 69 (31.08) < 0.0001
   Cutoff 2–cutoff 3[ 1,654 (47.96) 95 (42.79)
  ≥ Cutoff 3 1,498 (43.43) 19 (8.56)
  Systolic arterial pressure (mm Hg)
   < Cutoff 1 15 (0.43) 42 (18.92)
   Cutoff 1–cutoff 2 58 (1.68) 33 (14.86)
   Cutoff 2–cutoff 3 193 (5.60) 38 (17.12) < 0.0001
   Cutoff 3–cutoff 4 519 (15.05) 47 (21.17)
   Cutoff 4–cutoff 5 1,057 (30.65) 45 (20.27)
  ≥ Cutoff 5 1,607 (46.59) 17 (7.66)
  Heart rate (beats/min)
   < Cutoff 1 3,286 (95.27) 180 (81.08) < 0.0001
  ≥ Cutoff 1 163 (4.73) 42 (18.92)
  Creatinine (µmol/L)
  < Cutoff 1 2,255 (65.38) 49 (22.07)
   Cutoff 1–cutoff 2 785 (22.76) 73 (32.88) < 0.0001
  ≥ Cutoff 2 409 (11.86) 100 (45.05)
Cutoff of age-dependent variables is described in Table 2.
a

Reference levels are in bold font.

1766 www.ccmjournal.org July 2013 • Volume 41 • Number 7

 Pediatric Critical Care

Table 4. Multivariable Logistic Regression After Backward Selection With Bootstrap Validation
Odds Ratio
Variable and Cutoff Coefficient (95% CI) p
Glasgow Coma Score
  ≥ 11 1
 5–10 0.635 1.89 (1.12–3.18) 0.017
 3–4 1.904 6.71 (3.58–12.59) < 0.0001
Pupillary reaction
  Both reactive 1
  Both fixed 2.481 11.95 (6.48–22.03) < 0.0001
Lactatemia (mmol/L)
  < 3.97 1
 3.97–5.37 0.326 1.39 (0.70–2.74) 0.347
 5.37–11.07 0.642 1.90 (1.05–3.43) 0.033
  > 11.07 1.814 6.13 (2.93–12.83) < 0.0001
Mean arterial pressure (mm Hg)
  ≥ Cutoff 3a 1
  Cutoff 2–cutoff 3 a
0.841 2.32 (1.28–4.18) 0.005
  Cutoff 1–cutoff 2 a
1.372 3.94 (2.00–7.79) < 0.0001
  < Cutoff 1a 2.961 19.31 (6.88–54.21) < 0.0001
Creatinine (µmol/L)
  < Cutoff 1a 1
  Cutoff 1–cutoff 2 a
0.883 2.42 (1.49–3.94) < 0.0001
  ≥ Cutoff 2 a
1.065 2.90 (1.69–4.99) < 0.0001
Pao2 (mm Hg)/Fio2 ratio
  > 136.3 1
 60.5–136.3 –0.197 0.82 (0.44–1.53) 0.535
  < 60.5 0.889 2.43 (1.27–4.67) 0.008
Paco2 (mm Hg)
  < 58.5 1
 58.5–94.4 0.529 1.70 (1.00–2.88) 0.050
  ≥ 94.5 1.604 4.97 (1.94–12.76) 0.001
Ventilation
  No ventilation 1
  Noninvasive ventilation, yes 0.166 1.18 (0.28–5.05) 0.823
  Invasive ventilation, yes 1.446 4.25 (2.02–8.93) < 0.0001
WBC count (× 109/L)
  ≥ 4.10 1
 2.15–4.09 –0.276 0.76 (0.36–1.60) 0.468
  < 2.15 0.693 2.00 (0.95–4.19) 0.067
Platelets (× 10 /L)9

  ≥ 141.5 1
 76.5–141.4 0.399 1.49 (0.84–2.64) 0.172
  < 76.5 0.782 2.19 (1.27–3.77) 0.005
Cutoffs of age-dependent variables are defined in Table 2.
a

Critical Care Medicine www.ccmjournal.org 1767

Leteurtre et al

Table 5. Final Logistic Regression Defining the Pediatric Logistic Organ

Dysfunction-2 Score
Type of Variables Odds Ratio Pediatric Logistic Organ
and Cutoffs Coefficient (95% CI) p Dysfunction-2 Points
Glasgow Coma Score
  ≥ 11 1
 5–10 0.650 1.92 (1.14–3.22) 0.014 1
 3–4 1.942 6.97 (3.74–12.99) < 0.0001 4
Pupillary reaction
  Both reactive 1
  Both fixed 2.510 12.30 (6.68–22.65) < 0.0001 5
Lactatemia (mmol/L)
  < 5.0 1
 5.0–10.9 0.508 1.66 (0.98–2.83) 0.062 1
  ≥ 11.0 1.804 6.07 (3.08–11.96) < 0.0001 4
Mean arterial pressure (mm Hg)
  ≥ Cutoff 3a 1
  Cutoff 2–cutoff 3 a
0.806 2.24 (1.25–4.02) 0.007 2
  Cutoff 1–cutoff 2 a
1.332 3.79 (1.93–7.43) < 0.0001 3
  < Cutoff 1 a
2.891 18.01 (6.52–49.77) < 0.0001 6
Creatinine (µmol/L)
  < Cutoff 1a 1
  ≥ Cutoff 1 a
0.959 2.61 (1.68–4.06) < 0.0001 2
Pao2 (mm Hg)/Fio2 ratio
  ≥ 61 1
  ≤ 60 0.964 2.62 (1.39–4.94) 0.003 2
Paco2 (mm Hg)
  ≤ 58 1
 59–94 0.484 1.62 (0.97–2.73) 0.068 1
  ≥ 95 1.514 4.55 (1.79–11.52) 0.001 3
Ventilation
  No or noninvasive 1
 Invasive 1.384 3.99 (2.07–7.70) < 0.0001 3
WBC count (× 10 /L) 9

  >2 1
  ≤2 0.761 2.14 (1.04–4.40) 0.039 2
Platelets (× 10 /L)9

  ≥ 142 1
 77–141 0·373 1.45 (0.82–2.57) 0.200 1
  ≤ 76 0·782 2.19 (1.30–3.67) 0.003 2
Cutoffs of age-dependent variables are defined in Table 2.
a

1768 www.ccmjournal.org July 2013 • Volume 41 • Number 7

 Pediatric Critical Care

Table 6. Scoring the Pediatric Logistic Organ Dysfunction-2 Score

Points by Severity Levels
Organ Dysfunctions
and Variablesa 0 1 2 3 4 5 6
Neurologicb
  Glasgow Coma Score ≥ 11 5–10 3–4
  Pupillary reaction Both reactive Both fixed
Cardiovascular c

  Lactatemia (mmol/L) < 5.0 5.0–10.9 ≥ 11.0

  Mean arterial pressure (mm Hg)
   0 to < 1 mo ≥ 46 31–45 17–30 ≤ 16
  1–11 mo ≥ 55 39–54 25–38 ≤ 24
  12–23 mo ≥ 60 44–59 31–43 ≤ 30
  24–59 mo ≥ 62 46–61 32–44 ≤ 31
  60–143 mo ≥ 65 49–64 36–48 ≤ 35
  ≥ 144 mo ≥ 67 52–66 38–51 ≤ 37
Renal
  Creatinine (µmoL/L)
   0 to < 1 mo ≤ 69 ≥ 70
  1–11 mo ≤ 22 ≥ 23
  12–23 mo ≤ 34 ≥ 35
  24–59 mo ≤ 50 ≥ 51
  60–143 mo ≤ 58 ≥ 59
  ≥ 144 mo ≤ 92 ≥ 93
Respiratory d

 Pao2 (mm Hg)/Fio2 ≥ 61 ≤ 60

 Paco2 (mm Hg) ≤ 58 59–94 ≥ 95
  Invasive ventilation No Yes
Hematologic
  WBC count (× 109/L) >2 ≤2
  Platelets (× 109/L) ≥ 142 77–141 ≤ 76
a
All variables must be collected, but measurements can be done only if justified by the patient’s clinical status. If a variable is not measured, it should be
considered normal. If a variable is measured more than once in 24 hr, the worst value is used in calculating the score. Fio2: fraction of inspired oxygen.
b
Neurologic dysfunction: Glasgow Coma Score: use the lowest value. If the patient is sedated, record the estimated Glasgow Coma Score before sedation.
Assess only patients with known or suspected acute central nervous system disease. Pupillary reactions: nonreactive pupils must be > 3 mm. Do not assess
after iatrogenic pupillary dilatation.
c
Cardiovascular dysfunction: Heart rate and mean arterial pressure: do not assess during crying or iatrogenic agitation.
d
Respiratory dysfunction: Pao2: use arterial measurement only. Pao2/Fio2 ratio is considered normal in children with cyanotic heart disease. Paco2 can be
measured from arterial, capillary, or venous samples. Invasive ventilation: the use of mask ventilation is not considered invasive ventilation.
Logit (mortality) = –6.61 + 0.47 × PELOD-2 score.
Probability of death = 1/(1 + exp [–logit(mortality)]).

PELOD-2 was near to the value obtained for the PELOD on the DISCUSSION
training set (0.98 [95% CI, 0.960–0.999], published in 1999) We developed and validated the PELOD-2, a continuous scale
(10). After recalibration on the present data, the AUC of the that allows assessment of the severity of cases of MODS in
PELOD-2 was significantly higher than the AUC of the PELOD the PICU. This updated version of the PELOD includes 10
0.857 (95% CI, 0.834–0.879), p < 0.0001. variables involving five organ dysfunctions. Compared with the
Critical Care Medicine www.ccmjournal.org 1769
Leteurtre et al

Table 7. Relative Contribution of Each Organ Dysfunction to Mortality (Logistic Regression)

and to the Pediatric Logistic Organ Dysfunction-2 score (Multiple Regression)
Logistic Regression Multiple Regression
Maximum Value in Each
Dysfunction Dysfunction p Partial R²
Neurologic 9 < 0.0001 0.48
Respiratory 8 < 0.0001 0.29
Cardiovascular 10 < 0.0001 0.12
Renal 2 < 0.0001 0.08
Hematologic 4 < 0.0001 0.03

previous version, the PELOD-2

100 Neurological 100 Cardiovascular
now includes mean arterial
80 80 pressure and lactatemia in the
Patients (%)

cardiovascular dys­function,
60 60
which are present in the SOFA
40 40 and the P-MODS scores. On
the other hand, the PELOD-2
20 20
does not include hepatic
0 0 dysfunction, which was part of
0 1 4 5 6 9 0 1 2 3 4 6 7 10 the first version of the PELOD.
PELOD-2 score PELOD-2 score Hepatic dysfunction was linked
n 3111 355 60 39 10 96 n 1457 43 1562 413 63 82 42 9 with mortality in PELOD but
accounted for in only 0.1% of
its variance (2), and it did not
100 Respiratory predict death in the P-MODS
100 Renal
80 (9) score.
80
To be useful, a descriptive
Patients (%)

60 60 index, such as the PELOD-2,

40
should be clinically credible,
40
accurate, user friendly, and
20 20 generalizable and provide
useful additional information
0 0
0 2 0 1 2 3 4 5 6 8
to clinicians and scientists (18).
PELOD-2 score PELOD-2 score What remains to be determined
n 2304 1367 n 1641 88 4 1558 227 63 78 12
is if the responsiveness of the
PELOD-2 to preventive and/
or therapeutic interventions is
100 Haematological
good, if so, this will support the
80 idea that the PELOD-2 can be
Patients (%)

60
used as an outcome measure in
randomized clinical trials. The
40 responsiveness of a test is good
20 if the test reacts significantly to
active therapeutic approach,
0 when the response of the test
0 1 2 3 4
is in the right direction and
PELOD-2 score
when the response of the test is
n 3005 311 271 12 72
proportional to the stimulus,
Figure 1. Patients’ distribution and mortality rate for each organ dysfunction score. Black diamond = mortality as demonstrated for the
rate, PELOD-2 = Pediatric Logistic Organ Dysfunction-2. PELOD (19). Thus, it makes

1770 www.ccmjournal.org July 2013 • Volume 41 • Number 7

 Pediatric Critical Care

Table 8. Relationship Between the Number of Organ Dysfunctions, the Pediatric Logistic
Organ Dysfunction-2 Score, and Mortality
Pediatric Logistic Organ Deaths:
Number of Organ Dysfunction-2 score No. of ­
Dysfunctions No. of ­Patients (%) Mean (sd) Patients (%)
0 556 (15.2) 0 (0.0) 2 (0.4)
1 1,016 (27.7) 2.3 (0.8) 3 (0.3)
2 994 (27.1) 4.9 (1.3) 12 (1.2)
3 687 (18.7) 7.5 (2.0) 49 (7.1)
4 318 (8.7) 11.5 (4.4) 97 (30.5)
5 100 (2.7) 16.8 (5.2) 59 (59.0)

sense to believe that the responsiveness of the PELOD-2 should Physiology and Chronic Health Evaluation system (20), need
be good, but it remains to be proven. to be updated regularly because patients’ demographics, disease
Descriptive scores, like the PELOD, and predictive scores, prevalence, monitoring, treatment, and mortality rates change
like the Pediatric Risk of Mortality score (17) and the Acute over time. In the future, scoring systems are likely to become
more complex and dependent on new information technology.
They may require additional variables, adjustment for treatment
A 900 1
0.9
limitations, and diagnostic precision (21). For example, a
800
noninvasive variable, such as Spo2/Fio2 ratio instead of Pao2/Fio2
700 0.8
Probability of mortality

ratio (22), might be considered more appropriate and/or more

0.7
Number of patients

600 informative.
0.6 MODS scores seem to satisfy most criteria proposed
500
0.5 by Segers et al to assess the validity of a surrogate outcome,
400 even though some authors underline that surrogate outcome
0.4
300
0.3
endpoints have not yet been rigorously validated (23, 24).
200 There is also evidence that MODS scores (PELOD, SOFA,
0.2
MOD) can be a useful tool for clinical investigation: MODS
100 0.1 scores are largely used as secondary endpoint and primary
0 0 outcome measure when the mortality cannot be used as a
0 10 20 30 40 50 60 primary outcome measure because its incidence rate is too low
PELOD score (25). This is particularly important in PICUs where mortality
rate is much lower than in adult ICUs. In fact, the first version
B 900 1 of the PELOD was used in many large multicenter studies (26–
800 0.9 28). Also, MODS scores may be used to describe the effects of
therapy, and most trials now include repeated measures of a
700 0.8
MODS score as part of routine patient assessment (29).
Probability of mortality

0.7 The equation of probability of death that we used while doing

Number of patients

600
0.6 the goodness-of-fit test (calibration) is given in Table 6. However,
500
0.5 if the risk of death is to be predicted in a population different
400 from that which PELOD-2 was developed and validated, this
0.4
300 should be done using customization steps (first or second level
0.3 of customization), as recommended in the literature (21). Also,
200 0.2 we would emphasize that the aim of organ dysfuction scores,
100 0.1 such as the P-MODS (9), PELOD (2), SOFA (8), MODS (7), and
Logistic Organ Dysfunction (30) scores, is to describe the severity
0 0
of illness of critically ill patients and not to predict mortality rate.
0 2 4 6 8 10 12 14 16 18 20 22 24 26 29
This study has some limitations. First, the data in this study were
PELOD-2 score
collected using the set of 8 days (days 1, 2, 5, 8, 12, 16, and 18, plus
Figure 2. Descriptive characteristics of the Pediatric Logistic Organ the PICU discharge) in PICU that were previously identified as
Dysfunction (PELOD) (A) and PELOD-2 (B) scores in the study the optimal time points for measurement of dPELOD (16). Thus,
population. Dotted lines = probability of mortality calculated from the
score according to the values of the score; bars = number of patients an abnormal value of a variable measured on a day outside this set
according to the values of the score. could be missed; in theory, this could cause some underestimation

Critical Care Medicine www.ccmjournal.org 1771

Leteurtre et al

of the score. However, data of the day of discharge or death were 5. Proulx F, Fayon M, Farrell CA, et al: Epidemiology of sepsis and
multiple organ dysfunction syndrome in children. Chest 1996;
recorded, and thus, worst values of variables of patients who 109:1033–1037
died were taken into account up to 4 hours before death. Even 6. Wilkinson JD, Pollack MM, Ruttimann UE, et al: Outcome of pediatric
though we considered the worst value of each variable over the patients with multiple organ system failure. Crit Care Med 1986;
PICU stay to build the PELOD-2, as performed with other organ 14:271–274
dysfunction scores for adults (such as the MOD score of Marshall 7. Marshall JC, Cook DJ, Christou NV, et al: Multiple organ dysfunction
score: A reliable descriptor of a complex clinical outcome. Crit Care
et al [7] and the SOFA score of Vincent et al [8]), this may Med 1995; 23:1638–1652
have exaggerated the discrimination (AUC). Consequently, an 8. Graciano AL, Balko JA, Rahn DS, et al: The Pediatric Multiple Organ
external validation of the whole PICU stay and dPELOD-2 scores Dysfunction Score (P-MODS): Development and validation of an
objective scale to measure the severity of multiple organ dysfunction
is needed. Second, the PELOD-2 was developed and validated in critically ill children. Crit Care Med 2005; 33:1484–1491
with a dataset that originated from only two countries (France 9. Vincent JL, Moreno R, Takala J, et al: The SOFA (Sepsis-related
and Belgium). The population of our study was different from Organ Failure Assessment) score to describe organ dysfunction/
U.S. and U.K. populations (31, 32). Thus, the extensibility of our failure. On behalf of the Working Group on Sepsis-Related Problems
of the European Society of Intensive Care Medicine. Intensive Care
findings to other countries, particularly to PICUs that receive a Med 1996; 22:707–710
higher percentage of cases from the operating room (32), has 10. Leteurtre S, Martinot A, Duhamel A, et al: Development of a pediatric
to be verified. Also, because of changes in case mix and clinical multiple organ dysfunction score: Use of two strategies. Med Decis
practice, the performances of prognostic models deteriorate Making 1999; 19:399–410
11. Leteurtre S, Duhamel A, Grandbastien B, et al: Paediatric logistic
over time. To counterbalance this deterioration, models often organ dysfunction (PELOD) score. Lancet 2006; 367:897; author
need to be customized (33). Third, interobserver variability reply 900–902
was not studied for the PELOD-2 and should be evaluated in 12. Shime N, Kageyama K, Ashida H, et al: Application of modified
future studies on new populations. However, using an electronic sequential organ failure assessment score in children after cardiac
surgery. J Cardiothorac Vasc Anesth 2001; 15:463–468
clinical information system with an automated archiving
13. Tilford JM, Roberson PK, Lensing S, et al: Differences in PICU
method intelligently excluding unreliable variables values should mortality risk over time. Crit Care Med 1998; 26:1737–1743
decrease the risk of inaccurate data collection (34). 14. Garcia PC, Eulmesekian P, Branco RG, et al: External validation of
This study has several strengths. First, the data were collected the paediatric logistic organ dysfunction score. Intensive Care Med
2010; 36:116–122
from nine typical European multidisciplinary university-
15. Tibby SM: Does PELOD measure organ dysfunction and is organ
affiliated PICUs. Second, the score development used a large function a valid surrogate for death? Intensive Care Med 2010;
dataset of 3,671 consecutive patients. Third, it is continuous. 36:4–7
Fourth, its adjusted discriminative value is 0.934, which is 16. Leteurtre S, Duhamel A, Grandbastien B, et al: Daily estimation of the
considered “excellent” in the scale advocated by Hanley et al severity of multiple organ dysfunction syndrome in critically ill children.
CMAJ 2010; 182:1181–1187
(35) to estimate the descriptive value of a test. 17. Pollack MM, Patel KM, Ruttimann UE: PRISM III: An updated
In summary, we created and validated the PELOD-2, Pediatric Risk of Mortality score. Crit Care Med 1996; 24:743–752
which has an excellent discriminative power. The PELOD-2 18. Altman DG, Royston P: What do we mean by validating a prognostic
score now includes mean arterial pressure and lactatemia in model? Stat Med 2000; 19:453–473
the cardiovascular dysfunction and does not include hepatic 19. Nguyen TC, Han YY, Kiss JE, et al: Intensive plasma exchange
increases a disintegrin and metalloprotease with thrombospondin
dysfunction. We believe that the PELOD-2 allows assessment of motifs-13 activity and reverses organ dysfunction in children with
the severity of cases of MODS in the PICU and can be useful as thrombocytopenia-associated multiple organ failure. Crit Care Med
an outcome measure in clinical trials (36). The PELOD-2 will be 2008; 36:2878–2887
20. Zimmerman JE, Kramer AA, McNair DS, et al: Intensive care unit length
available in the public domain, which means that it can be freely of stay: Benchmarking based on Acute Physiology and Chronic Health
used in clinical trials, as it was the case with the PELOD (37). Evaluation (APACHE) IV. Crit Care Med 2006; 34:2517–2529
21. Zimmerman JE, Kramer AA, McNair DS, et al: Acute Physiology
and Chronic Health Evaluation (APACHE) IV: Hospital mortality
ACKNOWLEDGMENTS assessment for today’s critically ill patients. Crit Care Med 2006;
We thank the patients who volunteered for the study and the 34:1297–1310
clinicians responsible for their day-to-day clinical care. 22. Leteurtre S, Dupré M, Dorkenoo A, et al: Assessment of the Pediatric
Index of Mortality 2 with the Pao₂/Fio₂ ratio derived from the Spo₂/Fio₂
ratio: A prospective pilot study in a French pediatric intensive care
REFERENCES unit. Pediatr Crit Care Med 2011; 12:e184–e186
1. Ferreira AM, Sakr Y: Organ dysfunction: General approach, 23. Ioannou N, Terblanche M: Surrogate end points in critical illness
epidemiology, and organ failure scores. Semin Respir Crit Care Med research: Some way to go yet. Crit Care Med 2011; 39:2561–2562
2011; 32:543–551 24. Segers AE, Prins MH, Lensing AW, et al: Is contrast venography
2. Leteurtre S, Martinot A, Duhamel A, et al: Validation of the paediatric a valid surrogate outcome measure in venous thromboembolism
logistic organ dysfunction (PELOD) score: Prospective, observational, prevention studies? J Thromb Haemost 2005; 3:1099–1102
multicentre study. Lancet 2003; 362:192–197 25. Marshall JC, Vincent JL, Guyatt G, et al: Outcome measures for clinical
3. Proulx F, Joyal JS, Mariscalco MM, et al: The pediatric multiple organ research in sepsis: A report of the 2nd Cambridge Colloquium of the
dysfunction syndrome. Pediatr Crit Care Med 2009; 10:12–22 International Sepsis Forum. Crit Care Med 2005; 33:1708–1716
4. Goldstein B, Giroir B, Randolph A; International Consensus 26. Bateman ST, Lacroix J, Boven K, et al; Pediatric Acute Lung Injury
Conference on Pediatric Sepsis: International pediatric sepsis and Sepsis Investigators Network: Anemia, blood loss, and blood
consensus conference: Definitions for sepsis and organ dysfunction transfusions in North American children in the intensive care unit. Am
in pediatrics. Pediatr Crit Care Med 2005; 6:2–8 J Respir Crit Care Med 2008; 178:26–33

1772 www.ccmjournal.org July 2013 • Volume 41 • Number 7

 Pediatric Critical Care

27. Lacroix J, Hébert PC, Hutchison JS, et al; TRIPICU Investigators; 32. Marcin JP, Song J, Leigh JP: The impact of pediatric intensive care
Canadian Critical Care Trials Group; Pediatric Acute Lung Injury and unit volume on mortality: A hierarchical instrumental variable analysis.
Sepsis Investigators Network: Transfusion strategies for patients in Pediatr Crit Care Med 2005; 6:136–141
pediatric intensive care units. N Engl J Med 2007; 356:1609–1619 33. Keegan MT, Gajic O, Afessa B: Severity of illness scoring systems in
28. Vlasselaers D, Milants I, Desmet L, et al: Intensive insulin therapy the intensive care unit. Crit Care Med 2011; 39:163–169
for patients in paediatric intensive care: A prospective, randomised 34. Hug CW, Clifford GD, Reisner AT: Clinician blood pressure
controlled study. Lancet 2009; 373:547–556 documentation of stable intensive care patients: An intelligent
29. Vincent JL, Bruzzi de Carvalho F: Severity of illness. Semin Respir archiving agent has a higher association with future hypotension. Crit
Crit Care Med 2010; 31:31–38 Care Med 2011; 39:1006–1014
30. Le Gall JR, Klar J, Lemeshow S, et al: The Logistic Organ Dysfunction 35. Hanley JA, McNeil BJ: The meaning and use of the area under a receiver
system. A new way to assess organ dysfunction in the intensive care operating characteristic (ROC) curve. Radiology 1982; 143:29–36
unit. ICU Scoring Group. JAMA 1996; 276:802–810 36. Petros AJ, Marshall JC, van Saene HK: Should morbidity replace
31. Leteurtre S, Grandbastien B, Leclerc F, et al; Groupe Francophone mortality as an endpoint for clinical trials in intensive care? Lancet
de Réanimation et Urgences Pédiatriques; Paediatric Intensive Care 1995; 345:369–371
Audit Network: International comparison of the performance of the 37. Société Française d’Anesthésie et de Réanimation (SFAR): PELOD
paediatric index of mortality (PIM) 2 score in two national data sets. score (Pediatric Logistic Organ Dysfunction). Available at: http://www.
Intensive Care Med 2012; 38:1372–1380 sfar.org/scores/pelod.php. Accessed October 23, 2012

Critical Care Medicine www.ccmjournal.org 1773