Equine Veterinary Journal ISSN 0425-1644

DOI: 10.1111/evj.12279

Analytical Clinical Studies

Re-evaluation of the sepsis score in equine neonates

E. J. WEBER, L. C. SANCHEZ* and S. GIGUÈRE†
Hofmann Neonatal Intensive Care Unit, College of Veterinary Medicine, University of Florida, Gainesville, USA
†
Department of Large Animal Medicine, College of Veterinary Medicine, University of Georgia, Athens, USA.

*Correspondence email: sanchezl@ufl.edu; Received: 31.12.13; Accepted: 05.04.14

Portions of this work were presented in abstract form at the University of Florida College of Veterinary Medicine Phi Zeta Day, Gainesville, Florida in March 2013
and the Dorothy Russell Havemeyer Foundation Neonatal Sepsis Workshop, Saugherties, New York in October 2013.

Summary
Reasons for performing study: The modified sepsis scoring system provides a method to identify sepsis in foals early in the disease process, but
inconsistent results have been obtained from its clinical application in previous studies.
Objectives: To evaluate the sensitivity and specificity of the modified sepsis score in a larger population of foals. A secondary objective was to identify
factors associated with sepsis.
Study design: Retrospective cohort study.
Methods: Records were retrospectively analysed for neonatal admissions to the University of Florida Large Animal Hospital from 1982 to 2008. Backwards
stepwise multivariable logistic regression was used to evaluate objective clinical factors associated with sepsis. Receiver operating chararacteristic ROC
curve analysis was performed on the modified sepsis score and used to determine an optimal cut point.
Results: A total of 1065 foals were included in the study. The modified sepsis score had a sensitivity of 56.4% and specificity of 73.4% for the originally
suggested cut point (>11). Receiver operating chararacteristic analysis revealed an optimal cut point of >7 (sensitivity 84.4%, specificity 41.8%). The modified
sepsis score performed better than the generated regression model which included age, creatinine, lymphocytes, potassium and IgG>8 g/l (P = 0.026).
Diagnostic performance of the modified sepsis score did not change significantly over time.
Conclusions: Sensitivity and specificity of the modified sepsis score were not as high as previously documented, but the score still outperformed a
regression model derived from objective clinical data.
The Summary is available in Chinese – see Supporting information.

Keywords: horse; foal; sepsis; neonatal illness; diagnosis; scoring system

Introduction Materials and methods

Neonatal sepsis is one of the most challenging disease processes for
equine practitioners and has been documented as a major cause of
neonatal morbidity and mortality. Although survival rates have improved
over time [1], sepsis can still carry a guarded prognosis [2]. Infection may
occur pre- or post partum and is influenced by a number of factors,
including maternal disease processes, complications of parturition,
transfer of passive immunity, management practices and environmental
exposure [3]. Early identification of the disease process and initiation of
treatment are critical in case outcome, as therapy becomes more
challenging once clinical signs have become severe [4]. Although blood
culture is an objective measure of bacteraemia, the necessity of beginning
treatment at an early stage requires a rapid method of sepsis identification.
A sepsis scoring system was initially developed for this purpose in the
early 1980s. The ‘sepsis score’ included subjective clinical criteria as well
as objective clinicopathological parameters [5]. Additionally, a modified
sepsis score was developed to make the sepsis score more readily usable
by practitioners, eliminating the variables ‘metabolic acidosis’ and ‘PaO2’
[6]. Initial retrospective analysis in a population of 209 foals from the same
hospital as that described in the current report yielded a sensitivity and
specificity of 94.0 and 85.9%, respectively, for the original sepsis score and
92.8 and 87.5% for the modified sepsis score; both scoring systems
outperformed a computer-generated score [6]. Later retrospective and
prospective analysis of the modified scoring system in a population of 168
foals from a different geographic region yielded a sensitivity of 67% and a
specificity of 76% [7]. This discrepancy suggests that the performance of
the scoring system probably varies among populations of foals or over
time in a given population.
The objectives of the present study were to evaluate the sensitivity and
specificity of the modified sepsis score in a large population of hospitalised
foals from the same hospital where the sepsis score was developed, to
determine whether the performance of the sepsis scoring system has
changed over time, and to identify objective factors associated with sepsis
in the same population.
Equine Veterinary Journal 47 (2015) 275–278 © 2014 EVJ Ltd
Inclusion criteria
Cases were selected retrospectively from equine neonatal (<14 days of
age) admissions to the Large Animal Hospital at the University of Florida
College of Veterinary Medicine between 1982 and 2008. Selection criteria
included recorded results of a blood culture performed at any point during
hospitalisation, documentation of either discharge from the hospital or
nonsurvival (death or euthanasia), and a recorded modified sepsis score or
the ability to generate one from available data.
Classification
Foals were classified as septic if they fulfilled any or all of the following
criteria: 1) positive blood culture, 2) clinical course consistent with sepsis
(i.e. multifocal sites of infection) or 3) post mortem evidence of
disseminated septic processes [6]. Foals that did not meet any of the above
criteria were classified as nonseptic [6]. Foals alive at the time of discharge
from the hospital were considered survivors. Nonsurvivors included foals
that died or were subjected to euthanasia.
Factors included in analysis
Information recovered from the record of each foal included the following:
age, duration of gestation (d), sex, breed, year of presentation, duration of
hospitalisation, outcome (survived/died), vital signs (temperature, pulse
rate, respiratory rate), CBC (total WBC, total RBC, PCV, MCV, MCH, MCHC,
myelocytes, metamyelocytes, band neutrophils, segmented neutrophils,
lymphocytes, monocytes, plasma protein, fibrinogen, neutrophil toxicity,
platelets), biochemical profile (sodium, potassium, chloride, magnesium,
creatinine, blood urea nitrogen, total calcium, glucose, phosphorus, total
bilirubin, alkaline phosphatase, gamma glutamyl transferase, aspartate
aminotransferase, creatine kinase), arterial blood gas (pH, PCO2, PO2,
HCO3−, base excess), IgG and sepsis score [6]. All variables were taken at or
near the time of admission, unless otherwise specified.
275
Sepsis scoring in foals E. J. Weber et al.

TABLE 1: Sensitivity, specificity and predictive values of various cut TABLE 2: Comparison of the sepsis score over time
points of the modified sepsis score for the diagnosis of sepsis in 1065
Decade n ROC AUC ± s.e. 95% CI for ROC AUC
foals
1980s 185 0.745 ± 0.0423 0.675–0.806
Postive Negative
1990s 290 0.685 ± 0.0329 0.628–0.738
predictive predictive
Criterion Sensitivity Specificity value* value* 2000s 459 0.683 ± 0.0248 0.638–0.725

>0 99.83 1.94 61.8 87.5 Objective variables associated with sepsis by univariable logistic
>1 99.30 3.05 61.9 73.3 regression are presented in Table 3. Variables retained in the multivariable
>2 98.60 5.82 62.4 72.4 model (overall significance P<0.0001) included age, creatinine, lymphocyte
>3 97.03 10.53 63.3 69.1 count, potassium and IgG <8 g/l (Table 3). The model correctly classified
>4 95.81 17.73 64.9 72.7 62.3% of cases. The modified sepsis score (AUC 0.705 ± 0.0174) was
significantly (P = 0.026) better at predicting sepsis than the multivariable
>5 92.32 26.04 66.5 68.1
model (AUC 0.649 ± 0.0203) (Fig 1).
>6 88.48 34.63 68.2 65.4
>7 84.29 41.83 69.7 62.7
>8 78.01 50.69 71.5 59.2 Discussion
>9 71.90 59.83 74.0 57.3 Previous studies have shown variable diagnostic performance of the
>10 64.57 67.59 76.0 54.6 modified sepsis score in a clinical setting [6,7]. In the current study,
>11 56.37 73.41 77.1 51.5 although sensitivity and specificity values were lower than originally
>12 48.52 77.29 77.2 48.6 reported [6], they were similar to those reported more recently [7].
>13 41.71 81.99 78.6 47.0 Suggested reasons for the discrepancy between earlier reports included
>14 33.16 86.98 80.2 45.1 geographical location, hospital population and temporal associations.
Because the results of the current report more closely approximate those
>15 25.83 90.30 80.9 43.4
of a different [7], rather than similar [6], geographical location, hospital
>16 19.55 94.18 84.2 42.4
population alone cannot explain the results. Also, our data indicate that
>17 15.01 96.68 87.8 41.7 the performance of the modified sepsis score has not changed significantly
>18 11.52 97.51 88.0 41.0 over the 20 year period evaluated. In the current study, the modified sepsis
score also outperformed the constructed model. This supports prior data
*Predictive values at the observed prevalence of sepsis (60.2%) in the current
supporting use of the modified sepsis score over a previously generated
study.
computer model [6]. Because the modified sepsis score includes findings
that are also used to define sepsis, namely localising signs of sepsis such as
Data analysis swollen joints or respiratory distress, this system will undoubtedly have an
Potential associations between each variable and sepsis were first advantage over any system not including such factors. Our goal for the
screened using univariable logistic regression. Variables for which the multivariable analysis was to include only objective data, excluding
screening P value was <0.10 were considered for inclusion in the localising signs of infection that would obviate the need for scoring.
multivariable model. Variables with tolerance <0.40 and a variance inflation
TABLE 3: Factors associated with sepsis in a population of 1065 foals
factor >2.50 were deleted to avoid multicollinearity. The multivariable
model was a backward stepwise model, whereby variables were removed Variable n OR (95% CI) P
sequentially starting with that having the largest P value until only those
variables with P<0.05 remained. Goodness of fit of the final model was Univariable analysis
evaluated using the Hosmer and Lemeshow test. Odds ratios (OR) and 95% Age (days) 1065 1.07 (1.03–1.12) 0.001
confidence intervals (CI) were calculated. An OR >1 corresponds to a Base excess (mmol/l) 761 0.97 (0.95–0.99) 0.0112
positive association with sepsis, whereas a ratio <1 corresponds to a BUN (mg/dl) 773 1.01 (1.00–1.01) 0.006
negative association. Calcium (mg/dl) 976 0.92 (0.86–0.99) 0.0492
The overall performance of the multivariable regression models and
Creatinine (mg/dl) 1021 1.04 (0.99–1.09) 0.0524
sepsis score in predicting sepsis were assessed by receiver operating
characteristic (ROC) curve analysis [8]. The significance of the difference HCO3- (mmol/l) 789 0.97 (0.95–0.99) 0.0044
between the AUC of different ROC curves was assessed using the method IgG (mg/dl) 824 0.99 (0.99–0.99) 0.0166
described by Hanley and McNeil [9]. For all analyses, P<0.05 was IgG <400 (mg/dl) 824 1.48 (1.06–2.05) 0.0188
considered statistically significant. IgG <800 (mg/dl) 824 1.47 (1.11–1.96) 0.0074
Lymphocyte count (/µl) 993 0.99 (0.99–1.00) 0.0346
Metamyelocytes (/µl) 1041 1.00 (1.00–1.00) 0.0139
Results Monocyte count (/µl) 978 1.00 (1.00–1.00) 0.0016
A total of 1065 foals were included in the study; 557 (52%) were colts, 494 PCV (%) 1015 1.01 (0.99–1.02) 0.0679
(46%) were fillies and 14 (1%) foals did not have their sex recorded; 574 foals pH 806 0.19 (0.05–0.71) 0.0133
were excluded from analyses. A total of 641 foals (60.2%) were classified as Platelet count (K/µl) 463 0.99 (0.99–0.99) 0.0037
septic and 453 foals (42.5%) were bacteraemic. Overall survival to Potassium (mEq/l) 1008 1.35 (1.16–1.57) 0.0001
discharge for the study population was 72.8% (773 foals).
The modified sepsis score predicted sepsis in the study population with OR 95%CI P
a sensitivity of 56.4% and specificity of 73.4% with the originally suggested
cut point of >11; a cut point of >7 was deemed optimal via ROC analysis, Multivariable analysis (n = 748)
with a sensitivity of 84.3% and specificity of 41.8% (Table 1). A sepsis score Age (days) 1.17 1.09 –1.26 <0.0001
>7 correctly classified 67.8% of cases. To determine whether the diagnostic Creatinine (mg/dl) 1.05 1.00–1.11 0.0422
performance of the modified sepsis score changed over time, foals were Lymphocyte count (/µl) 0.99 0.99–0.99 0.0022
classified based on the decade of admission. The AUC of sepsis score for Potassium (mEq/l) 1.37 1.13–1.65 0.0010
the prediction of sepsis was not significantly different between decades IgG <800 (mg/dl) 1.46 1.07–1.99 0.0161
(Table 2).
276 Equine Veterinary Journal 47 (2015) 275–278 © 2014 EVJ Ltd
E. J. Weber et al.
100
80
Sensitivity (%)
60
P = 0.026
40
20
Sepsis score
Multivariate model
0 Authors’ declaration of interests
0 20 40 60 80 100
100 specificity (%)
Fig 1: Receiver operator characteristic curves for sensitivity and 100 specificity for the
modified sepsis score (solid line) and the calculated multivariable model (dotted line)
for classification of sepsis in 1065 neonatal foals.
One of the clinically relevant findings from the current study is the
suggestion of a cut point of >7, rather than the previously recommended
cut point of 11 [6], for a determination of sepsis. We chose to use
ROC analysis as an overall measure of diagnostic performance. The
area under the curve (AUC) of a ROC curve is a summary statistic of
overall diagnostic performance of a test. Diagnostic tests with perfect
discrimination between negative and positive reference samples
have an AUC equal to 1.00, whereas completely noninformative tests
have an AUC equal to 0.5. Given the severity of the consequences to
untreated sepsis in a neonate and the decrease in efficacy of therapy once
clinical signs have progressed, it may be appropriate to consider
sensitivity as a more important parameter than specificity in the clinical
setting [3]. Although using a cut point with lower specificity may result
in higher incidence of unnecessary treatment, this is less likely to be
detrimental to patients than a delay in treatment for neonates that
are truly septic. It is important to consider that the prevalence of sepsis
in the current population was relatively high (60.2%), thus positive
predictive values may be lower in populations with a lower prevalence
of sepsis.
Diagnostic performance of the modified sepsis score could possibly
be further improved by refining the variables and the value of points
assigned. In the current scoring system, hyperthermia (rectal temperature
>38.9°C/102°F) receives more points than hypothermia (rectal temperature
<36.7°C /98°F). However, fever is not a consistent finding in foals with
sepsis and clinical experience suggests that foals with severe sepsis are in
fact more likely to be hypothermic [4]. Additionally, physical indicators of
fetal maturity, rather than numerical gestational length, may improve
the diagnostic performance of this variable as it is included in the modified
sepsis score. Loss of the variables derived from history and physical
examination appeared to reduce diagnostic performance, as only age
and clinicopathological values were retained in the final multivariable
model.
One of the limitations of the modified sepsis score is the availability of
information for scoring. While physical examination data are typically
readily obtained or available, historical and clinicopathological data may
be unknown and many veterinarians rely upon machine-generated vs.
manual white blood cell differential analyses. Thus, the degree of
neutrophil toxicity may not be known, even if the remaining white blood
cell parameters are recorded.
Although the goal of a sepsis scoring system is to standardise the initial
evaluation and classification of septic neonates, clinical acumen is still a
critical component of case evaluation. Overall diagnostic performance of
this and all prior work evaluating sepsis scoring systems in foals remains
poor to moderate at best and is hindered by the lack of a true gold
standard ante mortem definition. For consistency, and to allow for
comparisons with the original study, the definition of sepsis previously
used by Brewer [5] was used in the current study. In man, sepsis is defined
Equine Veterinary Journal 47 (2015) 275–278 © 2014 EVJ Ltd
Sepsis scoring in foals
as an infection accompanied by a systemic inflammatory response
syndrome [10]. While the systemic inflammatory response syndrome
(SIRS) has clear, objective criteria based upon body temperature,
heart rate, hyperventilation and white blood cell count, the definition
of sepsis depends upon a documented or suspected infection and
‘some’ of a variety of other variables [10]. The consensus group
based that recommendation upon the fact that reality still relies
upon clinicians’ determination that patients ‘look septic’ and did not
want research entry criteria for sepsis to take precedence over the
clinical reality [10]. Equine neonatal medicine is currently faced with a
similar dilemma. Thus, clinicians may decide to use SIRS-based criteria
in addition to a suspected infectious focus as a sepsis definition for
future trials.
No competing interests have been declared.
Ethical animal research
Ethical review not currently required by this journal: retrospective study of
clinical records. Explicit owner informed consent for participation in this
study was not stated.
Sources of funding
The study was not funded.
Acknowledgements
The authors would like to thank J. Lyle-Dugas for assistance with data entry
and all UF medicine clinicians, technicians and students for their care of the
foals.
Authorship
Study design was undertaken by L.C. Sanchez and S. Giguère, study
execution by E.J. Weber and data analysis by S. Giguère. All authors
contributed to manuscript preparation and approved the final version of
the manuscript.
References
1. Sanchez, L.C., Giguère, S. and Lester, G.D. (2008) Factors associated
with survival of neonatal foals with bacteremia and racing performance of
surviving Thoroughbreds: 423 cases (1982-2007). J. Am. Vet. Med. Ass. 233,
1446-1452.
2. Roy, M.F. (2004) Sepsis in adults and foals. Vet. Clin. N. Am.: Equine Pract. 20,
41-61.
3. Paradis, M.R. (1994) Update on neonatal septicemia. Vet. Clin. N. Am.: Equine
Pract. 10, 109-135.
4. Koterba, A.M., Brewer, B.D. and Tarplee, F.A. (1984) Clinical and
clinicopathological characteristics of the septicaemic neonatal foal: review of
38 cases. Equine Vet. J. 16, 376-382.
5. Brewer, B.D. and Koterba, A.M. (1988) Development of a scoring system for
the early diagnosis of equine neonatal sepsis. Equine Vet. J. 20, 18-22.
6. Brewer, B.D., Koterba, A.M., Carter, R.L. and Rowe, E.D. (1988) Comparison of
empirically developed sepsis score with a computer generated and weighted
scoring system for the identification of sepsis in the equine neonate. Equine
Vet. J. 20, 23-24.
7. Corley, K.T.T. and Furr, M.O. (2003) Evaluation of a score designed to predict
sepsis in foals. J. Vet. Emerg. Crit. Care 13, 149-155.
8. Delong, E.R., Delong, D.M. and Clarkepearson, D.I. (1988) Comparing the areas
under 2 or more correlated receiver operating characteristic curves – a
nonparametric approach. Biometrics 44, 837-845.
277
Sepsis scoring in foals E. J. Weber et al..

9. Hanley, J.A. and McNeil, B.J. (1982) The meaning and use of the area under a Supporting information
receiver operating characteristic (roc) curve. Radiol. 143, 29-36.
10. Levy, M.M., Fink, M.P., Marshall, J.C., Abraham, E., Angus, D., Cook, D., Cohen, Additional Supporting Information may be found in the online version of
J., Opal, S.M., Vincent, J.L. and Ramsay, G. (2003) 2001 this article at the publisher’s web-site:
SCCM/ESICM/ACCP/ATS/SIS international sepsis definitions conference. Crit.
Care Med. 31, 1250-1256. Summary in Chinese.

EVJ Bookshop
Manual of Equine Neonatal Medicine 4th Edition
Editor: John E Madigan
Publisher: Live Oak Publishing, April 2013
Paperback, 451 pages
The new fourth edition from renowned equine veterinarian, Dr. John Madigan, is now
available. This manual has been considered the single most authoritative, easy to use,
source of information for the veterinarian working on foals in the field, clinic or critical
care unit.
Veterinary students, animal technicians and horse owners around the world have used
this handy guide to find quick information to aid in the management, diagnosis and
treatment of young foals. This fourth edition is completely updated, has several new
sections added, all continuing with the tradition of a format for rapid access to needed
information. 
New topics include: Neonatal Ultrasound - Gastrointestinal, Musculoskeletal, Cardiac • EVJ price:
Fetal Ultrasound • New Drug Formulary with specific doses for foals • New Method of £55.50 plus p&p
Foal  Restraint  -  non-chemical  -  Madigan  Squeeze  Method  •  Limb  Deformity BEVA member price:
Management by Dr. Hans Castelijns • New Sedation & Anesthesia Guidelines. £49.95 plus p&p

EVJ Bookshop, Mulberry House, 31 Market Street, Fordham, Ely, Cambs. CB7 5LQ, UK
Tel: 01638 723555 ✦ Fax: 01638 724043 ✦ Email: bookshop@evj.co.uk ✦ www.beva.org.uk

278 Equine Veterinary Journal 47 (2015) 275–278 © 2014 EVJ Ltd