ORIGINAL ARTICLE

Risk factors and diagnostic markers of

bacteremia in Stevens-Johnson
syndrome and toxic epidermal
necrolysis: A cohort study of
176 patients
Hui Kai Koh, MBBS,a Zi Teng Chai, MD, MRCP,b Hui Wen Tay, BS,a Stephanie Fook-Chong, MS, CStat,c
Karen J. L. Choo, MBChB, MRCP,b Choon Chiat Oh, MBBS, MRCP, FAMS (Dermatology),b
Yi Wei Yeo, MBBS, MRCP, MMed (Int Med), FAMS (Dermatology),b
Hong Yi Koh, MBBS, MRCP, MMed (Int Med), FAMS (Dermatology),b
Shiu Ming Pang, MBBS, FRCP, Dip Dermatology, FAMS (Dermatology),b and
Haur Yueh Lee, MBBS, MRCP, MMed (Int Med), FAMS (Dermatology)b
Singapore

Background: Sepsis is the main cause of death in Stevens-Johnson syndrome (SJS) and toxic epidermal
necrolysis (TEN).

Objectives: Our aim was to identify admission risk factors predictive of bacteremia and the accompanying
clinical or biochemical markers associated with positive blood cultures.

Methods: A retrospective cohort study over a 14-year period (2003-2016) was performed.

Results: The study included 176 patients with SJS (n = 59), SJS-TEN overlap (n = 51), and TEN (n = 66).
During hospitalization, bacteremia developed in 52 patients (29.5%), who experienced poorer outcomes,
including higher intensive care unit admission (P \ .0005), longer length of stay (P \ .0005), and higher
mortality (P \ .0005). There were 112 episodes of bacteremia, and isolates included Acinetobacter
baumannii (27.7%, n = 31) and Staphylococcus aureus (21.4%, n = 24). On multivariate analysis, clinical
factors present at admission that were predictive of bacteremia included hemoglobin #10 g/dL (odds ratio
[OR] 2.4, confidence interval [CI] 2.2-2.6), existing cardiovascular disease (OR 2.10, CI 2.0-2.3), and body
surface area involvement $10% (OR 14.3, CI 13.4-15.2). The Bacteremia Risk Score was constructed with
good calibration. Hypothermia (P = .03) and procalcitonin $1 g/L (P = .02) concurrent with blood culture
sampling were predictive of blood culture positivity.

Limitations: This is a retrospective study performed in a reference center.

Conclusion: Hemoglobin #10 g/dL, cardiovascular disease, and body surface area involvement $10% on
admission were risk factors for bacteremia. Hypothermia and elevated procalcitonin are useful markers for
the timely detection of bacteremia. ( J Am Acad Dermatol https://doi.org/10.1016/j.jaad.2019.05.096.)

Key words: adverse drug reactions; bacteremia; diagnostic markers; microbiology; risk factors; sepsis;
Stevens-Johnson syndrome; toxic epidermal necrolysis.

From the Yong Loo Lin School of Medicine, National University of Reprint requests: Haur Yueh Lee, MBBS, MRCP, MMed (Int Med),
Singaporea; Department of Dermatology, Singapore General FAMS (Dermatology), Department of Dermatology, Singapore
Hospitalb; and Health Services Research Unit, Division of General Hospital, Outram Road, Singapore 169608. E-mail: lee.
Medicine, Singapore General Hospital.c [email protected].
Dr Koh and Dr Chai contributed equally to this study. Published online July 16, 2019.
Funding sources: None. 0190-9622/$36.00
Conflicts of interest: None disclosed. Ó 2019 by the American Academy of Dermatology, Inc.
Accepted for publication May 29, 2019. https://doi.org/10.1016/j.jaad.2019.05.096

1
2 Koh et al J AM ACAD DERMATOL
n 2019

Stevens-Johnson syndrome (SJS) and toxic METHODS

epidermal necrolysis (TEN) are severe cutaneous Patients and data collection
adverse reactions characterized by epidermal This cohort study was conducted over a 14-year
detachment and mucositis.1 SJS-TEN is a disease period (2003-2016) at Singapore General Hospital,
continuum differentiated on the basis of the percent- the national referral center for SJS-TEN. We adopted
age of body surface area (BSA) detached: [30% in a hybrid model of care for SJS-TEN in which patients
TEN, 10%-30% in SJS-TEN overlap, and\10% in SJS.2 are nursed in the burns unit (if BSA involved is
Drug exposures are the lead- [10%) or in the general ward
ing triggers of SJS-TEN. The (if BSA involved is \10%)
incidence of TEN has been CAPSULE SUMMARY with dermatologists being
estimated to be 1-2 cases/1 the primary provider.
million person-years.3-5 d
Sepsis is associated with high mortality Intensive care facilities are
Mortality ranges from 10% in Stevens-Johnson syndrome and toxic available in burn units.
for SJS and up to 50% for epidermal necrolysis. Diagnosis and classification
TEN.6 Supportive care in a d
The Bacteremia Risk Score, which of epidermal necrolysis was
reference center and the assesses the presence of hemoglobin based on established criteria2
early withdrawal of the #10 g/dL, cardiovascular disease, and with supportive histologic
culprit drug remains the $10% body surface area involvement on evidence. On admission,
gold standard of care.7 admission, is useful in predicting the supportive care is instituted,
The most common cause development of bacteremia. consisting of specialized
of death in SJS-TEN is sepsis, Hypothermia and procalcitonin $1 g/L nursing, fluid and electrolyte
which accounts for 50% of help in the timely detection of resuscitation, nutritional sup-
deaths.8,9 Bacteremia in- bacteremia during hospitalization. plementation, and medical
creases the risk of dying by involvement of other spe-
3-fold, and the risk factors cialties as needed. In terms
for bacteremia include age [40 years, white blood of skin care, we prescribed irrigation with normal
cell count [10,000 cells/mm3, and BSA $30% on saline and utilized tulle gras as primary dressing and
admission.10 The routine use of prophylactic antibi- Gamgee as secondary dressing to absorb exudates.
otics is not recommended.7,11 However, empiric These were held in place with a vest, with alternate
antibiotic use should be started early if sepsis is day wound reviews. No prophylactic antibiotics,
suspected.7,12 Despite such recommendations, operative debridement, skin grafting, or use of skin
various challenges exist. Standard markers of sepsis, substitutes were done. Immunomodulatory treat-
such as fever, leukocytosis, and elevated C-reactive ment prescribed during the study period included
protein (CRP), are not specific for infection in SJS- cyclosporine13 and intravenous immunoglobulin.14
TEN. Likewise, a time-lag exists between the sam- To determine risk factors associated with bacter-
pling of blood cultures and confirmatory results. emia, data on patient demographics, clinical fea-
Two studies from the reference center in France tures, laboratory investigations, and SCORTEN
showed that Staphylococcus aureus and (SCORe for TEN) collected within the first 24 hours
Pseudomonas aeruginosa were the 2 most common of hospitalization were recorded and analyzed.15
bacteria species, but whether similar species would To determine the clinical and biochemical
be identified in other centers or settings, particularly markers associated with positive blood cultures,
in burn units or intensive care units (ICUs),9,10 and if temperature and laboratory results taken within
the microbial species changes across the period of 24 hours of blood culture collection were recorded
hospitalization remains unclear. Such challenges and analyzed.
result in delayed diagnosis of sepsis and initiation
of antibiotics or the inappropriate overuse of anti-
biotics with the accompanying problems of Microbiologic samples and definitions
multidrug-resistant organisms. Blood cultures consisted of standard aerobic and
Therefore, in our study, we aimed to validate anaerobic paired samples (two 10-mL blood vials)
potential risk factors for bacteremia that were pre- taken at admission, when there was clinical suspi-
sent in patients upon admission, identify clinical and cion of bacteremia or every 48 hours until re-
laboratory findings that might enable rapid predic- epithelialization.
tion of positive blood cultures, and highlight the A patient was defined as having bacteremia when
trends in bacteremia in SJS-TEN patients managed in a recognized pathogen was identified in at least 1
a reference center. blood culture or when the same normal-appearing
J AM ACAD DERMATOL Koh et al 3
VOLUME jj, NUMBER j

Table I. Baseline characteristics of SJS-TEN patients

Abbreviations used:
Parameters Value, N = 176
BRS: Bacteremia Risk Score
BSA: body surface area Age, y, median (IQR) 57 (44-71)
CI: confidence interval Sex
CRP: C-reactive protein Female 104 (59.1)
ICU: intensive care unit
OR: odds ratio Male 72 (40.9)
SCORTEN: severity-of-illness score for TEN Ethnicity
SJS: Stevens-Johnson syndrome Chinese 121 (68.8)
TEN: toxic epidermal necrolysis Malay 41 (23.3)
Others 14 (8.0)
Disease classification
SJS 59 (33.5)
skin commensal (mainly coagulase-negative staphy-
SJS-TEN overlap 51 (29.0)
lococci) was isolated from [1 blood culture within a
TEN 66 (37.5)
48-hour period. Polymicrobial bacteremia was SCORTEN, median (IQR) 2 (1-3)
defined as the recovery of $2 pathogens from the Medical history
same blood culture. Isolation of different species Hypertension 84 (47.7)
from different samples or isolates of the same species Hyperlipidemia 43 (24.4)
from samples with an intervening negative sample Diabetes 43 (24.4)
(indicating resolution of the initial bacteremia, fol- Cardiovascular disease 39 (22.2)
lowed by a recurrence) were considered separate Liver disease 8 (4.5)
bacteremia episodes. Renal disease 27 (15.3)
Malignancy 34 (19.3)
Autoimmune disease 19 (10.8)
Data presentation and statistical analysis AIDS 6 (3.4)
Data were presented as medians for continuous Values are n (%), unless stated otherwise.
data and percentages for categorical data. For IQR, Interquartile range; SCORTEN, SCORe for TEN; SJS, Stevens-
univariate analyses, categorical data were analyzed Johnson syndrome; TEN, toxic epidermal necrolysis.
by using x2 analysis or Fisher’s exact test and
continuous data was analyzed by using the Mann-
Whitney U test. Cutoff values were defined accord-
Subsequently, scores were constructed by con-
ing to either a clinically significant threshold (eg,
verting the regression coefficients of independently
hemoglobin #10 g/dL) or an established threshold
significant variables in the multivariate model into
used in the literature (eg, BSA $10%15). Variables
integers. A total score, named Bacteremia Risk Score
with P \ .10 in univariate analyses, which were
(BRS), was then calculated by summing up the
clinically relevant, were selected for entry into a
assigned scores for each predictive factor.
multivariate logistic regression model. A penalized
We considered P \ .05 statistically significant.
regression method with stepdown variable selec-
Statistical analysis was performed with IBM SPSS
tion at P \ .01 was used to alleviate model
version 25 (Armonk, NY) and R version 3.4 (R
overfitting, which enabled the development of a
Foundation for Statistical Computing, Vienna,
more accurate risk prediction model.16 This method
Austria). This study was approved by the Singhealth
is also recommended in the TRIPOD checklist for
Institutional Review Board (CIRB Ref 2014/2011).
developing and validating prediction models.17
SCORTEN15 was not entered into the multivariate
model because it is a composite score encompass- RESULTS
ing other variables. Patient characteristics
Internal validation was performed by using the The study included 176 patients, comprising 59
bootstrap technique with 200 resamples. The cases of SJS (33.5%), 51 cases of SJS-TEN overlap
bootstrap-corrected C-statistic and bias-corrected (29.0%), and 66 cases of TEN (37.5%). The median
calibration curve were generated to assess discrim- age was 57 (interquartile range 44-71) years.
ination and calibration, respectively. The mean ab- Baseline characteristics of patients are summarized
solute error was also reported to measure the in Table I.
predictive accuracy of the model. A C-statistic 0.7-
0.9 indicates fair-to-good discrimination, and a Epidemiology of bacteremia
calibration curve close to the ideal y = x line indicates Bacteremia developed in 52 (29.5%) patients
good calibration. during their hospitalization; 86 cases of bacteremia
4 Koh et al J AM ACAD DERMATOL
n 2019

Fig 1. Median time of onset of bacteremia caused by each microorganism. Size of circles
represents the number of organisms. Enclosed within brackets are the 25th and 75th percentiles
of time since date of blistering (in days) for each organism. CoNS, Coagulase-negative
staphylococci; MRSA, methicillin-resistant Staphylococcus aureus; MSSA, methicillin-sensitive S.
aureus.

were recorded in these 52 patients over 3181 patient- (n = 34) (P = .797) and intravenous immunoglobulin
days (27.0/1000 patient-days). Twenty of 176 (n = 78) (P = .064) were not found to be significantly
(11.4%) patients had polymicrobial infections, with associated with bacteremia.
14 patients (8.0%) experiencing 1 polymicrobial
episode each and 3 patients (1.7%) experiencing 2 Admission risk factors associated with
polymicrobial episodes each. In total, 112 microor- bacteremia
ganisms were identified, mainly A. baumannii Univariate analysis of the clinical factors recorded
(27.7%, n = 31), S. aureus (21.4%, n = 24), other within the first 24 hours of admission indicated that
gram-positive pathogens (18.8%, n = 21), and BSA $10%, hemoglobin #10 g/dL, and existing
Enterobacteriaceae (15.2%, n = 17) (Fig 1). cardiovascular disease and hypertension were signif-
The median time that elapsed between detach- icantly associated with bacteremia (P \.1) (Table II).
ment to bacteremia was 9 (interquartile range 6-14) After multivariate analysis, the final model obtained
days. contained 3 significant risk factors: hemoglobin
#10 g/dL (OR 2.4, 95% CI 2.2-2.6), existing cardio-
Patient outcomes vascular disease (OR 2.1, 95% CI 2.0-2.3), and BSA
The overall in-hospital mortality rate was 23.9%. $10% (OR 14.3, 95% CI 13.4-15.2) (Table III). The
Patients with bacteremia were more likely to require model developed with these 3 covariates has good
ICU admission (odds ratio [OR] 6.8, 95% confidence discrimination (bootstrap-corrected C-statistic 0.76)
interval [CI] 3.1-15.0), invasive mechanical ventila- and a good performance on the basis of the
tion (OR 6.6, 95% CI 2.3-18.4), and dialysis (OR 18.3, calibration plot. The mean absolute error was low
95% CI 3.9-85.3) and had longer hospital lengths of (0.029), indicating the model was a good fit for the
stay (P \.0005) and higher mortality (OR 4.4, 95% CI data.
2.1-9.1). The BRS was calculated on the basis of the score
After adjusting for SCORTEN, which has been assigned to each factor (Table III). A plot of patients’
shown to be a significant risk factor for bacteremia predicted probability versus their total risk score was
(Table II), patients who received cyclosporine examined (not shown herein), and the following 4
J AM ACAD DERMATOL Koh et al 5
VOLUME jj, NUMBER j

Table II. Patient and clinical characteristics on admission by bacteremia group

Bacteremia
Factor No, n = 124 Yes, n = 52 P value
Demographics
Age $40 y 98 (79.0) 43 (82.7) .579
Sex
Female 75 (60.5) 29 (55.8) .562
Male 49 (39.5) 23 (44.2)
Ethnicity
Chinese 82 (66.1) 39 (75.0) .458
Malay 32 (25.8) 9 (17.3)
Other 10 (8.1) 4 (7.7)
Parameters
BSA $10% 67 (54.0) 50 (96.2) .001*
SCORTEN, median (IQR) 2 (1-3) 3 (2-4) \.0005*
Laboratory findingsy
Urea $10 mmol/L 28 (23.1) 18 (34.6) .117
HCO3 #20 mmol/L 28 (23.5) 18 (34.6) .133
Glucose $14 mmol/L 8 (7.5) 4 (8.7) .755
Sodium \125 or [145 mmol/L 9 (7.4) 5 (9.6) .761
Potassium \3 or [5 mmol/L 13 (10.7) 9 (17.3) .227
Creatinine $75 mmol/L 67 (54.5) 30 (57.7) .695
Bilirubin $68.4 mol/L 3 (2.6) 0 (0) .557
Hemoglobin #10 g/dL 23 (18.9) 21 (41.2) .002*
Total WBC $11 3 109/L 31 (25.4) 12 (24.0) .846
Platelets #150 3 109/L 25 (20.5) 14 (27.5) .318
Absolute eosinophil count, $0.8 3 109/L 3 (4.8) 2 (5.7) 1.000
Atypical lymphocytes present 13 (10.5) 5 (9.6) .862
Medical history
Chronic steroid therapy 22 (17.7) 6 (11.5) .305
Malignancy 24 (19.4) 10 (19.2) .985
AIDS 5 (4.0) 1 (1.9) .672
Liver disease 6 (4.8) 2 (3.8) 1.000
Renal disease 16 (12.9) 11 (21.2) .166
Cardiovascular disease 21 (16.9) 18 (34.6) .010*
Hypertension 52 (41.9) 32 (61.5) .018*
Hyperlipidemia 27 (21.8) 16 (30.8) .205
Diabetes 27 (21.8) 16 (30.8) .205
Autoimmune disease 12 (9.7) 7 (13.5) .460
Drug allergy 27 (21.8) 11 (21.2) .927
SJS-TEN 3 (2.4) 0 (0) .556

Values are n (%), unless stated otherwise.

BSA, Body surface area; CRP, C-reactive protein; HCO3, bicarbonate; IQR, interquartile range; SCORTEN, SCORe for TEN; SJS, Stevens-Johnson
syndrome; TEN, toxic epidermal necrolysis; WBC, white blood cell.
*P value \ .1; variable is significant for predicting development of bacteremia during the hospital stay.
y
Laboratory findings for each patient were recorded within first 24 hours of each hospital admission.

risk categories for in-hospital bacteremia occurrence Predictive markers for positive blood cultures
were created: low (BRS 0-1), moderate (BRS 2, 4), Univariate analysis of temperature and laboratory
high (BRS 5), and very high (BRS 6). The mean variables obtained within 24 hours of blood culture
predicted probability of bacteremia was 3.1% for the collection indicated that hypothermia (temperature
low-risk category, 31.4% for the moderate-risk cate- #36.08C), CRP, and procalcitonin were significantly
gory, 52.4% for the high-risk category, and 71.6% for associated with bacteremia (P \ .1) (Table IV).
the very higherisk category. A cutoff score of BRS Hypothermia, CRP $100 mg/L, and procalcitonin
$2 is strongly predictive (negative predictive value $1 g/L were selected for entry into multivariate
98.2% [95% CI 92.1%-99.8%] and positive predictive analysis, which showed that concurrent hypother-
value 43.1% [95% CI 34.4%-52.2%]). mia (OR 2.4, 95% CI 1.1-5.3; P = .03) and
6 Koh et al J AM ACAD DERMATOL
n 2019

Table III. Multivariate analysis of hospital the BRS. SJS-TEN patients can be risk stratified into 1
admission factors for predictors of bacteremia of 4 groups, with corresponding risk of bacteremia
during hospital stay and corresponding score ranging 3%-72%. The utility of risk stratification is
assigned for each significant factor in the BRS manifold; risk stratification aids in the determination
Beta
of appropriate monitoring and blood culture sam-
Factor OR (95% CI) P value coefficient Score pling, identification of patients who might need early
Hemoglobin 2.4 (2.2-2.6) \.0001 0.88 1 treatment, and prognostication.
#10 g/dL We found higher SCORTEN to be significantly
Cardiovascular 2.1 (2.0-2.3) \.0001 0.74 1 associated with increased risk of bacteremia
disease (P \ .0005). SCORTEN, a severity-of-illness score,
BSA $10% 14.3 (13.4-15.2) \.0001 2.66 4 has been widely used to predict mortality in SJS-TEN.
Sepsis is the main cause of mortality in SJS-TEN,
Low risk (BRS 0-1): 3.1% predicted probability of bacteremia.
accounting for ;50% of deaths. The close-knit
Moderate risk (BRS 2, 4): 31.4% predicted probability of
bacteremia. High risk (BRS 5): 52.4% predicted probability of relationship between sepsis and mortality in SJS-
bacteremia. Very high risk (BRS 6): 71.6% predicted probability of TEN explains our findings. Our results also affirm the
bacteremia. value of SCORTEN in offering important prognostic
BRS, Bacteremia Risk Score; BSA, body surface area; CI, confidence information. However, BRS remains superior to
interval; OR, odds ratio.
SCORTEN in predicting bacteremia (nonbootstrap-
corrected C-statistic BRS 0.781 vs SCORTEN 0.698;
procalcitonin $1 g/L (OR 2.4, 95% CI 1.1-4.8; HosmereLemeshow test BRS P = .990 vs SCORTEN
P = .02) to be predictive of blood culture positivity. P = .816). These findings highlight the strength and
importance of BRS in identifying patients at risk of
bacteremia.
DISCUSSION
Our study showed that bacteremia affects up to
Timely detection of bacteremia
30% of SJS-TEN patients and is associated with
The routine use of prophylactic antibiotics is not
poorer outcomes of mortality, ICU admission, the
recommended because this practice has not been
need for dialysis, and longer lengths of stay.
shown to be beneficial and is also associated with
Recommendations for sepsis management include
longer lengths of hospital stay.20,21 In many cen-
early transfer to a reference center, reverse barrier
ters,12 including ours, frequent blood culture sam-
nursing, avoidance of prophylactic antibiotics, regu-
pling is performed to pick up bacteremia early,
lar sampling for culturing, and early institution of
notwithstanding a lag-time of at least 48 hours before
empiric antibiotics. Nonetheless, several challenges
confirmatory results are available. Predictive clinical
remain, including identification of high-risk patients,
and laboratory markers for sepsis are needed to
timely detection of bacteremia, and appropriate
balance early empiric antibiotic treatment versus
choice of empiric antibiotics on the basis of local
prophylactic or inappropriate antibiotic use and the
trends in microbial agents. Our study has tried to
risk of emergence of multidrug-resistant organisms.
answer some of these existing gaps.
The utility of standard indicators, such as leukocy-
tosis and elevated CRP and procalcitonin, remains
Identification of high-risk patients on unclear in this group of patients. In a small case series
admission of 8 patients with severe adverse reactions without
Among various clinical and laboratory markers infections, procalcitonin was found to be mildly
evaluated, BSA $10%, hemoglobin #10 g/dL, and elevated in an SJS patient (0.53 g/L) and a DRESS
existing cardiovascular disease were found to be (drug reaction with eosinophilia and systemic symp-
significant predictors of bacteremia. This finding toms) patient (3.96 g/L).22 In a recent series of 42
complements prior results from the French reference SJS-TEN patients, those with systemic infections had
center showing BSA to be a possible risk factor.10 higher levels of procalcitonin than those with cuta-
Cardiovascular disease and anemia were unique risk neous infections and no infections.23 Utilizing a
factors identified in our cohort. Wang et al previously procalcitonin cutoff of 0.65 ng/mL, estimated sensi-
showed that coronary artery disease increases the tivity was 85% and specificity 90%. In this study, CRP
risk of sepsis in a longitudinal cohort of community was not useful.
adults.18 Likewise, the odds of hospital-acquired In our current study, we evaluated several
bacteremia is increased by the presence of anemia.19 clinical and laboratory markers at the point of
Utilizing these 3 risk factors, we were able to blood culture sampling and correlated them to
construct a validated and discriminatory risk scored eventual blood culture positivity. On multivariate
J AM ACAD DERMATOL Koh et al 7
VOLUME jj, NUMBER j

Table IV. Univariate analysis of temperature and laboratory variables obtained within 24 hours of blood culture
collection
Bacteremia
Factor No Yes P value
Fever $38.38C 87 (29.6) 15 (21.4) .172
Hypothermia \368C 83 (28.5) 31 (44.3) .011
WBC count, 103 L, median (IQR) 7.9 (5.4-12.3) 9.3 (5.6-13.5) .218
CRP, mg/L, median (IQR) 89.9 (35.1-165.0) 122.5 (66.2-202.0) .006
Procalcitonin, g/L, median (IQR) 0.7 (0.2-2.6) 1.3 (0.5-3.6) .052

Values are n (%), unless stated otherwise.

CRP, C-reactive protein; IQR, interquartile range; WBC, white blood cell.

analysis, only 2 factors (hypothermia and procalci- The median time to bacteremia was 9 days from the
tonin $1 g/L) were significantly associated with onset of disease, supporting the fact that the risk of
blood culture positivity. Although fever is generally bacteremia was highest during the active phase of the
used to support diagnosis of bacteremia and disease. We have also demonstrated a temporal
sepsis,24,25 this factor is not predictive in SJS-TEN pattern to the emergence of pathogens. In general,
patients. Anecdotal evidence has suggested that the earliest pathogens were more likely gram-negative
hypothermia might be useful as a marker of bacteria and Enterobacteriaceae, followed by gram-
infection,20 and our study validated this observa- positive bacteria and nosocomial infections, such as
tion. The presence of both hypothermia and Acinetobacter, Candida, and Stenotrophomonas. The
procalcitonin $1 g/L has a specificity of 86.2% exception was Pseudomonas, which tended to appear
and sensitivity of 25.0% in predicting bacteremia. later than the other species of gram-negative organ-
The procalcitonin threshold that we utilized in this isms. Such information might help practitioners pre-
study is similar to that used in another study on the dict possible pathogens and tailor empiric
prediction of bacteremia.26 antimicrobial treatments accordingly.

Epidemiology of bacteremia Limitations

Prior studies have identified S. aureus, P. aerugi- Our study had some limitations. This study was
nosa, and Enterobacteriaceae as the most common retrospective and included the flaws inherent in that
causes of bacteremia in SJS-TEN patients.9,10,21,27 design choice. As the national reference center for
Although S. aureus, Enterobacteriaceae, and SJS-TEN, our data are prone to referral bias. In the
Pseudomonas sp. were common in our cohort, setting of a burns unit and burns ICU, the microbial
accounting for 21%, 15%, and 6% of cases, respec- data might be skewed toward nosocomial organ-
tively, the most common isolate was A. baumannii, isms. Our bacteremia risk scoring system, though
which accounted for 28% of cases. Such variation internally validated, would need to be prospectively
might result from the patients being managed in a evaluated. These limitations might have affected the
burns unit with ICU facilities or from the local generalizability of results to other centers and
tropical environment. A 5-year review among 352 populations. Nonetheless, there were certain
burns patients showed the most common isolated strengths. Our cohort is one of the largest series of
organisms were A. baumannii, P. aeruginosa, and SJS-TEN patients used for evaluating the risk of
methicillin-resistant S. aureus, which is similar to our bacteremia. Cases were managed in the same center,
findings.28 A. baumannii is an emerging nosocomial under the same protocol, thereby obviating the bias
pathogen, particularly in burns units and ICUs.28-30 attributed to center effect. This study yielded data
In countries with temperate climates, A. baumannii obtained from patients nursed predominantly in a
also shows seasonal variability, being more burns unit, thus, this study might have practical
commonly isolated during the warmer, summer utility for other patients managed in a similar care
months, potentially explaining in part the higher setting.
prevalence in warmer climates. Risk factors for
developing multidrug-resistant A. baumannii Conclusions
include invasive procedures (such as central venous Although there has been considerable interest and
catherization), ICU admission, mechanical ventila- emphasis on the use of immunomodulatory agents in
tion, and use of broad-spectrum antibiotics.29,30 the treatment of SJS-TEN, a gap remains regarding
8 Koh et al J AM ACAD DERMATOL
n 2019

the optimization of supportive care, wound care, and in a specialized referral center. J Am Acad Dermatol. 2017;76:
sepsis management for this disease.31 Our study has 106-113.
14. Lee HY, Lim YL, Thirumoorthy T, Pang SM. The role of
yielded additional perspectives, including a strategy intravenous immunoglobulin in toxic epidermal necrolysis: a
for identifying high-risk patients, the timely predic- retrospective analysis of 64 patients managed in a specialized
tion of bacteremia via markers (such as hypothermia center. Br J Dermatol. 2013;169:1304-1309.
and elevated procalcitonin), as well as highlighting 15. Bastuji-Garin S, Fouchard N, Bertocchi M, Roujeau JC,
the temporal trends and epidemiology of bacteremia Revuz J, Wolkenstein P. SCORTEN: a severity-of-illness score
for toxic epidermal necrolysis. J Invest Dermatol. 2000;115:
in SJS-TEN patients. Taken together, such informa- 149-153.
tion would enable the early detection and manage- 16. Pavlou M, Ambler G, Seaman SR, et al. How to develop a more
ment of sepsis, which remains the major cause of accurate risk prediction model when there are few events. BMJ
mortality. (Clinical Research Ed). 2015;351:h3868.
17. Collins GS, Reitsma JB, Altman DG, Moons KG. Transparent
REFERENCES reporting of a multivariable prediction model for individual
1. Miliszewski MA, Kirchhof MG, Sikora S, Papp A, Dutz JP. prognosis or diagnosis (TRIPOD): the TRIPOD statement. BMJ
Stevens-Johnson syndrome and toxic epidermal necrolysis: an (Clinical Research Ed). 2015;350:g7594.
analysis of triggers and implications for improving prevention. 18. Wang HE, Shapiro NI, Griffin R, Safford MM, Judd S, Howard G.
Am J Med. 2016;129:1221-1225. Chronic medical conditions and risk of sepsis. PLoS One. 2012;
2. Bastuji-Garin S, Rzany B, Stern RS, Shear NH, Naldi L, 7:e48307.
Roujeau JC. Clinical classification of cases of toxic epidermal 19. Jensen AG, Wachmann CH, Poulsen KB, et al. Risk factors for
necrolysis, Stevens-Johnson syndrome, and erythema multi- hospital-acquired Staphylococcus aureus bacteremia. Arch
forme. Arch Dermatol. 1993;129:92-96. Intern Med. 1999;159:1437-1444.
3. Roujeau JC, Guillaume JC, Fabre JP, Penso D, Flechet ML, 20. Roujeau JC, Chosidow O, Saiag P, Guillaume JC. Toxic
Girre JP. Toxic epidermal necrolysis (Lyell syndrome). Inci- epidermal necrolysis (Lyell syndrome). J Am Acad Dermatol.
dence and drug etiology in France, 1981-1985. Arch Dermatol. 1990;23:1039-1058.
1990;126:37-42. 21. Diao M, Thapa C, Ran X, Ran Y, Lv X. A retrospective analysis of
4. Rzany B, Mockenhaupt M, Baur S, et al. Epidemiology of infections and antibiotic treatment in patients with Stevens-
erythema exsudativum multiforme majus, Stevens-Johnson Johnson syndrome and toxic epidermal necrolysis. J Derma-
syndrome, and toxic epidermal necrolysis in Germany (1990- tolog Treat. 2018:1-19.
1992): structure and results of a population-based registry. J 22. Sfia M, Boeckler P, Lipsker D. High procalcitonin levels in
Clin Epidemiol. 1996;49:769-773. patients with severe drug reactions. Arch Dermatol. 2007;143:
5. Chan HL, Stern RS, Arndt KA, et al. The incidence of erythema 1591.
multiforme, Stevens-Johnson syndrome, and toxic epidermal 23. Wang Q, Tian XB, Liu W, Zhang LX. Procalcitonin as a
necrolysis. A population-based study with particular reference diagnostic indicator for systemic bacterial infections in pa-
to reactions caused by drugs among outpatients. Arch tients with Stevens-Johnson syndrome/toxic epidermal nec-
Dermatol. 1990;126:43-47. rolysis. J Dermatol. 2018;45:989-993.
6. Sekula P, Dunant A, Mockenhaupt M, et al. Comprehensive 24. Gaeta GB, Fusco FM, Nardiello S. Fever of unknown origin: a
survival analysis of a cohort of patients with Stevens-Johnson systematic review of the literature for 1995-2004. Nucl Med
syndrome and toxic epidermal necrolysis. J Invest Dermatol. Commun. 2006;27:205-211.
2013;133:1197-1204. 25. Bleeker-Rovers CP, Vos FJ, de Kleijn EM, et al. A prospective
7. Creamer D, Walsh SA, Dziewulski P, et al. UK guidelines for the multicenter study on fever of unknown origin: the yield of a
management of Stevens-Johnson syndrome/toxic epidermal structured diagnostic protocol. Medicine. 2007;86:26-38.
necrolysis in adults 2016. Br J Dermatol. 2016;174:1194-1227. 26. Riedel S, Melendez JH, An AT, Rosenbaum JE, Zenilman JM.
8. Yamane Y, Matsukura S, Watanabe Y, et al. Retrospective Procalcitonin as a marker for the detection of bacteremia and
analysis of Stevens-Johnson syndrome and toxic epidermal sepsis in the emergency department. Am J Clin Pathol. 2011;
necrolysis in 87 Japanese patientsetreatment and outcome. 135:182-189.
Allergol Int. 2016;65:74-81. 27. McGee T, Munster A. Toxic epidermal necrolysis syndrome:
9. Revuz J, Penso D, Roujeau JC, et al. Toxic epidermal necrolysis. mortality rate reduced with early referral to regional burn
Clinical findings and prognosis factors in 87 patients. Arch center. Plast Reconstr Surg. 1998;102:1018-1022.
Dermatol. 1987;123:1160-1165. 28. Li L, Dai JX, Xu L, et al. Antimicrobial resistance and pathogen
10. de Prost N, Ingen-Housz-Oro S, Duong T, et al. Bacteremia in distribution in hospitalized burn patients: a multicenter study
Stevens-Johnson syndrome and toxic epidermal necrolysis: in Southeast China. Medicine. 2018;97:e11977.
epidemiology, risk factors, and predictive value of skin 29. Garcia-Garmendia JL, Ortiz-Leyba C, Garnacho-Montero J, et al.
cultures. Medicine. 2010;89:28-36. Risk factors for Acinetobacter baumannii nosocomial bacter-
11. Duong TA, Valeyrie-Allanore L, Wolkenstein P, Chosidow O. emia in critically ill patients: a cohort study. Clin Infect Dis.
Severe cutaneous adverse reactions to drugs. Lancet. 2017; 2001;33:939-946.
390:1996-2011. 30. Jung JY, Park MS, Kim SE, et al. Risk factors for multi-drug
12. Fromowitz JS, Ramos-Caro FA, Flowers FP. Practical guidelines for resistant Acinetobacter baumannii bacteremia in patients with
the management of toxic epidermal necrolysis and Stevens- colonization in the intensive care unit. BMC Infect Dis. 2010;10:
Johnson syndrome. Int J Dermatol. 2007;46:1092-1094. 228.
13. Lee HY, Fook-Chong S, Koh HY, Thirumoorthy T, Pang SM. 31. Lee HY. Wound management strategies in Stevens-Johnson
Cyclosporine treatment for Stevens-Johnson syndrome/toxic syndrome/toxic epidermal necrolysis: an unmet need. J Am
epidermal necrolysis: retrospective analysis of a cohort treated Acad Dermatol. 2018;79:e87-e88.