MAJOR ARTICLE

Diabetes Mellitus and Pyogenic Liver Abscess:

Risk and Prognosis
Reimar W. Thomsen,1 Peter Jepsen,1,2 and Henrik T. Srensen1,3
1
Department of Clinical Epidemiology, Aarhus University Hospital, Aalborg, and 2Department of Medicine V, Aarhus University Hospital, Aarhus,
Denmark; and 3Department of Epidemiology, School of Public Health, Boston University, Boston, Massachusetts

Background. Pyogenic liver abscess (PLA) is a rare, life-threatening disease with an increasing rate of incidence.
Case reports from East Asia suggest that diabetes mellitus is an important risk factor, but formal evidence is
limited.
Methods. We performed a case-control study with participants drawn from the entire population of Denmark.
Cases of PLA were defined as occurring in all patients who received a first-time diagnosis of PLA on hospital
discharge between 1977 and 2002, as identified in the nationwide Danish National Patient Registry. Fifty sex- and
age-matched population control subjects were selected for each patient with PLA. We computed the relative risk
of PLA associated with diabetes using conditional logistic regression and controlling for major potential con-
founders. We further examined whether diabetes increased the relative risk of death until 30 days after hospital
discharge among patients with PLA.
Results. We identified 1448 patients who experienced a first hospitalization for PLA during the study period
(median age, 64 years; male sex, 54.2%). Persons with diabetes had a 3.6-fold increased risk of experiencing PLA,
compared with population control subjects (adjusted relative risk, 3.6; 95% confidence interval, 2.94.5]. In
addition, patients with PLA who had diabetes had a higher 30-day postdischarge mortality rate, compared with
patients with PLA who did not have diabetes (24.8% vs. 18.0%). After controlling for other prognostic factors,
the relative risk of death for patients with PLA and diabetes was 1.3 (95% confidence interval, 0.92.1).
Conclusions. Diabetes is a strong, potentially modifiable risk factor for PLA. PLA is associated with a similarly
poor prognosis for patients with diabetes and for other patients.

Pyogenic liver abscess (PLA) is a rare, life-threatening
disease that has an increasing incidence rate in the
United States and Europe [13]. In Denmark, from
1977 to 2002, the incidence rate of PLA increased from
6 cases per 1 million person-years to 18 cases per 1
million person-years for men and from 8 cases per 1
million person-years to 12 cases per 1 million person-
years for women [2]. During the same period, mortality
rates decreased from 40%50% to 10%; however, this
change is perhaps largely explained by the use of more-
sensitive diagnostic tools [2, 3]. Obstructive biliary dis-
ease, abdominal infections, previous surgical proce-
Received 17 October 2006; accepted 10 January 2007; electronically published
28 March 2007.
Reprints or correspondence: Dr. Reimar Wernich Thomsen, Dept. of Clinical
Epidemiology, Aarhus University Hospital, Forskningens Hus, Sdr. Skovvej 15,
Postbox 365, DK-9100 Aalborg, Denmark (r.thomsen@rn.dk).
Clinical Infectious Diseases 2007; 44:11941201
 2007 by the Infectious Diseases Society of America. All rights reserved.
1058-4838/2007/4409-0010$15.00
DOI: 10.1086/513201
dures, and immunocompromising medical conditions
(including cancer and liver cirrhosis) have been re-
ported as being risk factors for PLA [4, 5]. However,
the evidence is based mainly on case reports, and up
to 50% of patients with PLA have none of these risk
factors (cryptogenic PLA) [6, 7]. Knowledge about
risk and prognostic factors is needed to develop strat-
egies to prevent PLA and to improve the outcome of
this severe infection.
Large population-based epidemiological studies in-
creasingly suggest that diabetes mellitus is an important
risk and prognostic factor for severe gram-negative in-
fections, including bacteremia [8]. However, the asso-
ciation between diabetes and PLA remains controver-
sial. The few available studies have been hampered by
the absence of control groups, small sample size, lack
of covariate information, and incomplete follow-up [5,
911]. The evidence for an association between diabetes
and PLA comes primarily from case series of Klebsiella
species PLA in East Asia [12, 13]. No studies in pop-
ulations in Europe and North America have focused

1194 CID 2007:44 (1 May) Thomsen et al.

specifically on the association between diabetes and PLA risk.
Because PLA is rare, large datasets are needed to address the
issue [5]. Using a nationwide database that included all hos-
pitalizations, we examined diabetes as a risk factor and a prog-
nostic factor for PLA in Denmark over a 26-year period.
METHODS
Setting
The study drew on the entire population of Denmark (5.4
million inhabitants), whose tax-supported national health ser-
vice provides all residents with free access to hospitals and
primary medical care [14]. The unique civil registry number
assigned to every Danish resident since 1968 permitted data
linkage across nationwide administrative and health registries
and allowed us to establish a complete hospitalization history
for each individual under study.
The Danish National Patient Registry contains computerized
records of all patient discharges from Danish hospitals since
1977 [15]. Files for each hospitalization include information
on the patients civil registry number, dates of hospital admis-
sion and discharge, surgical procedures performed, and up to
20 diagnoses coded by physicians on the date of discharge
according to the Danish version of the International Classifi-
cation of Diseases (ICD; 8th revision used until the end of
1993, 10th revision used thereafter).
Patients with PLA and Population Control Subjects
The study included all patients who were discharged from a
hospital (living or deceased) with a diagnosis of PLA between
1 January 1977 and 31 December 2002. Only the date of each
patients first hospital discharge diagnosis of PLA (the index
date) was considered. On the index date for each PLA patient,
we randomly selected 50 control subjects, matched by sex and
age (same year and month of birth), from the Danish popu-
lation as a whole by means of electronic linkage with the Danish
Civil Registration System. Selection of control subjects was done
on the basis of risk set sampling, ensuring that the estimated
exposure OR in our case-control study was an unbiased esti-
mate of the relative risk [16].
Diabetes
Diabetes was considered to be present in case patients with PLA
and control subjects if a hospital discharge diagnosis of insulin-
dependent diabetes, noninsulin-dependent diabetes, unspe-
cified diabetes, or diabetic retinopathy was recorded on or at
any time before the index date of the first discharge diagnosis
of PLA.
Confounders
Diabetes and risk of PLA. We compiled a complete hospi-
talization history for each patient with PLA and each popu-
lation control subject to assess medical or surgical risk factors
for PLA that also may be associated with diabetes [5, 7, 9, 10,
17, 18]. The following medical conditions were considered to
be risk factors for PLA if they had been recorded as discharge
diagnoses before or on the index date: benign biliary obstruc-
tion, liver cancer (including metastases to the liver), biliary tract
cancer, pancreatic cancer, other gastrointestinal cancers, all
other solid-organ cancers, hematological cancers, liver cirrhosis,
inflammatory bowel disease, alcoholism, ongoing abdominal
infection (acute cholecystitis or cholangitis without cholelithi-
asis, diverticulitis, appendicitis, or peritonitis), and ongoing
systemic infection (sepsis or endocarditis). For cases of infec-
tion, only discharge diagnoses from hospitalizations ending !30
days before the index date were considered.
The following surgical procedures were considered to be risk
factors for PLA, provided that they had been performed !6
months before the index date: surgical procedures involving
the liver, biliary tract (including gall bladder), pancreas, or
spleen; upper endoscopy; endoscopy of the biliary tract (in-
cluding endoscopic retrograde cholangiopancreatography);
other gastrointestinal surgery; and any other surgical proce-
dures. Because procedures performed during a hospitalization
for PLA may have been related to PLA diagnosis or treatment,
procedures performed before and after the date of admission
were considered separately. Among the latter, we defined a cat-
egory of probably PLA-related procedures (a list of all ICD
codes used is provided in the Appendix).
Diabetes and 30-day postdischarge mortality following
PLA. To adjust for the burden of comorbidity in mortality
analyses, we computed Charlson index scores for patients with
diabetes and PLA and other patients with PLA using records
of all hospital discharge diagnoses made up to and including
the index date [19]. Diabetes was removed from the Charlson
index, because it defined the exposure in this study. We defined
3 comorbidity index levels: low (score, 0), medium (score, 1
2), and high (score, 12). Alcoholism-related diagnoses were
handled as a separate variable, because most are not included
in the Charlson index.
Statistical Analyses
Diabetes and risk of PLA. For the case-control analysis of
PLA risk, we used conditional logistic regression to compute
ORs with 95% CIs as a measure of the relative risk of PLA
according to diabetes status, adjusted for medical and surgical
risk factors. Analyses were conducted both with and without
adjusting for surgical procedures performed during the PLA-
related hospitalization, excluding probable PLA-related
procedures.
We further computed ORs stratified by sex and age group
(039 years, 4064 years, 6579 years, and 80 years) and ORs
restricted to patients with PLA and control subjects for whom
Diabetes and Pyogenic Liver Abscess CID 2007:44 (1 May) 1195
 Table 1. Characteristics of case patients with pyogenic liver abscess (PLA) and of
sex- and age-matched population control subjects, Denmark, 19772002.

No. (%) of patients

Case patients Control subjects
Characteristic (n p 1448) (n p 72,332)
Diabetes mellitusa
Yes 162 (11.2) 1855 (2.6)
Type 1 (% of those with diabetes) 11 (6.8) 114 (6.1)
Type 2 (% of those with diabetes) 151 (93.2) 1741 (93.9)
No 1286 (88.8) 70,477 (97.4)
Male sex 785 (54.2) 39,182 (54.2)
Age, years
039 156 (10.8) 7800 (10.8)
4064 569 (39.3) 28,250 (39.1)
6579 538 (37.2) 27,050 (37.4)
80 185 (12.8) 9232 (12.8)
Medical risk factorsa
Benign biliary obstruction 285 (19.7) 1643 (2.3)
Liver cancer 56 (3.9) 30 (0.0)
Biliary tract cancer 39 (2.7) 7 (0.0)
Pancreatic cancer 45 (3.1) 29 (0.0)
Other gastrointestinal cancer 73 (5.0) 759 (1.0)
Any other solid-organ cancer 90 (6.2) 2838 (3.9)
Hematological cancer 17 (1.2) 270 (0.4)
Liver cirrhosis 43 (3.0) 196 (0.3)
Inflammatory bowel disease 32 (2.2) 329 (0.5)
Alcoholism-related disorders 106 (7.3) 1059 (1.5)
Abdominal infectionb 210 (14.5) 11 (0.0)
b,c
Systemic infection 123 (8.5) 9 (0.0)
Surgical risk factorsd
Procedures involving the liver 44 (3.0) 14 (0.0)
Procedures involving the biliary tract or gall bladder 113 (7.8) 60 (0.1)
Procedures involving the pancreas 15 (1.0) 3 (0.0)
Procedures involving the spleen 3 (0.2) 2 (0.0)
Upper endoscopy 133 (9.2) 375 (0.5)
Biliary tract endoscopy (including ERCP) 95 (6.6) 25 (0.0)
Any other gastrointestinal surgery 152 (10.5) 527 (0.7)
Any other surgery 281 (19.4) 3353 (4.6)

NOTE. ERCP, endoscopic retrograde cholangiopancreatography.

a
On the index date of the discharge diagnosis of PLA, except for infections.
b
For infections, only diagnoses from hospitalizations within 30 days before the index date were
considered.
c
Sepsis and endocarditis.
d
Excluding procedures performed during hospitalization for PLA.

no surgical procedures were performed at any time or who had
no recorded PLA risk factors. Because the 1-to-50 technique
of matching case patients with control subjects could not be
retained for these analyses, we used ordinary logistic regression
adjusted for age, sex, calendar year, and the potential con-
founders listed previously. To examine the public health impact
of diabetes on the overall risk of PLA, we calculated the pop-
ulation-attributable risk for a diagnosis of diabetesthat is, the
proportion of all cases of PLA that may be attributable to
diabetes [16].
The likelihood of receiving a diagnosis of diabetes might have
been higher among case patients than control subjects because
of more hospitalizations (including the hospitalization for
PLA). To address this issue, we performed a sensitivity analysis.
A previous report showed that only 63% of individuals with
known diabetes in Denmark could be identified through pre-
vious discharge diagnoses [20]. In the sensitivity analysis, we
assumed that our findings were strongly biased and that the
diabetes detection rate was 100% among the patients with PLA
but only 63% among the control subjects. Using these rates,

1196 CID 2007:44 (1 May) Thomsen et al.

we recalculated the diabetes-PLA risk estimate as it would have
been if the diabetes detection rate was 100% among both case
patients and control subjects.
Diabetes and 30-day postdischarge mortality following
PLA. For the analysis of mortality, we obtained data from the
Danish Civil Registration System, which records all changes in
vital status and migration, including date of death [21]. Six
cases of PLA from the case-control study were excluded because
of an invalid date of death recorded in the registry. Our out-
come measure was the cumulative mortality rate from admis-
sion to 30 days after discharge. Logistic regression analyses were
used to compute ORs with 95% CIs as a measure of the relative
risk of death at day 30 after hospital discharge among patients
with diabetes and PLA, compared with other patients with PLA,
adjusting for sex, age group, comorbidity index level, alcohol-
ism-related disorders, and time period of hospitalization (1977
1988, 19891996, or 19972002). Analyses were stratified by
sex, age, comorbidity, and time period. All statistical analyses
were performed using Stata software, version 9.2 (StataCorp).
RESULTS
Diabetes and Risk of PLA
Descriptive data. The study involved 1448 patients with PLA.
Their median age (interquartile range) was 64 years (5075
years), and 54.2% were men. Table 1 shows the characteristics
of the patients with PLA and the 72,332 population control
subjects. A total of 162 patients with PLA (11.2%) had diabetes,
compared with 1855 control subjects (2.6%). Among the di-
abetic PLA patients, 22 (13.6%) received a diagnosis of diabetes
for the first time during their hospitalization for PLA; none
were recorded as having ketoacidosis. The most common med-
ical risk factors for cases of PLA were benign biliary obstruction
(285 patients [19.7%]), any type of cancer (257 patients
[17.8%]), and ongoing abdominal infection (210 patients
[14.5%]). A total of 464 patients with PLA (32.0%) had un-
dergone a surgical procedure within the prior 6 months; if
procedures performed during hospitalizations for PLA were
also counted, 811 (56.0%) of patients with PLA had undergone
a surgical procedure, compared with only 3867 (5.3%) of con-
trol subjects. Commonly encountered surgical risk factors were
upper endoscopy and biliary tract procedures (table 1).
Risk factor analysis. The crude OR for PLA in persons
with diabetes was 5.0 (95% CI, 4.25.9). After controlling for
confounding factors, the OR decreased to 3.6 (95% CI, 2.9
4.5). Inclusion of surgical procedures performed during PLA
hospitalization in the model left the OR unchanged (table 2).
Benign biliary obstruction, alcoholism, and liver cirrhosis were
the strongest confounders of the association between diabetes
and PLA risk, reducing the crude OR from 5.0 to 4.2, 4.4, and
4.6, respectively, when presence of these conditions was in-
cluded one by one in the analysis. Thus, 15% of the apparent
effect of diabetes was caused by a higher prevalence of benign
biliary obstruction, and 12% was caused by a higher prevalence
of alcoholism among subjects with diabetes.
In contrast, the risk increase conferred by diabetes apparently
was not mediated by other ongoing abdominal or systemic
infections, because adjustment for them left the crude OR vir-
tually unchanged at 4.9 and 4.8, respectively (data not shown).
Adjusted ORs for patients with PLA and diabetes were similarly
high, even among patients who had not undergone surgical
procedures and patients who were not identified as having any
PLA risk factors (table 2). The relative risk increase conferred
by diabetes tended to be highest in the youngest age groups:
ORs for PLA were 56 times higher among subjects with di-
abetes who were aged !65 years. Assuming that the association
between diabetes and PLA is causal, and given a prevalence of
diabetes of 2.6% among control subjects, the population-at-
tributable risk of PLA from diabetes was 6.3%.
Sensitivity analysis. Under the very conservative assump-
tion that we were able to identify only 63% of control subjects
with diabetes but 100% of PLA case patients with diabetes, the
prevalence of diabetes among control subjects would have been
1855/0.63 p 2944 (4.1%) of 72,332 persons. This prevalence
adjustment decreases the unadjusted OR for PLA in persons
with diabetes from 5.0 to 3.0. Similarly, when we restricted our
exposure measurement to a diagnosis of diabetes received only
before the date of hospital admission for PLA, the crude OR
became 4.2 (95% CI, 3.55.0) and was 2.9 (95% CI, 2.23.7)
in the adjusted analysis.
Diabetes and 30-Day Postdischarge Mortality Attributable
to PLA
Descriptive data. A total of 16.8% of patients with diabetes
and PLA died during hospitalization, compared with 13.2% of
other patients with PLA. At 30 days after hospital discharge,
the cumulative mortality for those 2 groups was 24.2% and
17.8%, respectively. Overall, the mortality rate was lower among
men than among women, increased sharply with age and co-
morbidity level, and was substantially higher in the early years
of the 26-year study period (table 3).
Prognostic factor analysis. The crude OR for death on
postdischarge day 30 for patients with diabetes and PLA was
1.5 (95% CI, 1.02.2). After controlling for other prognostic
factors, the OR for death decreased to 1.3 (95% CI, 0.92.1)
(table 3). The effect of diabetes on PLA prognosis in early study
years, when PLA outcome was poor, was similar to that ob-
served in more recent time periods. In relative terms, diabetes
had a particularly strong impact on mortality among patients
with PLA who were !40 years of age (adjusted OR, 13.7; 95%
CI, 0.8238.8). Although the estimates showed considerable
statistical imprecision, ORs for death due to diabetes were con-
sistently elevated across all strata examined (table 3).
Diabetes and Pyogenic Liver Abscess CID 2007:44 (1 May) 1197
 Table 2. Crude and adjusted ORs for pyogenic liver abscess (PLA) according to
the presence of diabetes mellitus.

a b
Crude OR Adjusted OR
Characteristic (95% CI) (95% CI)
No diabetes 1.0 1.0
Diabetes present, overall 5.0 (4.25.9) 3.6 (2.94.5)
Diabetes present, modified analyses
Also adjusting for surgical procedures performed
c
during PLA hospitalization 5.0 (4.25.9) 3.6 (2.84.6)
Restricted to patients who had not undergone
surgical procedures 4.6 (3.66.0) 3.5 (2.54.7)
Restricted to patients with no risk factors for PLA 4.0 (2.75.9) 4.2 (2.86.3)
Sex
Male 5.7 (4.67.1) 3.8 (2.85.2)
Female 4.1 (3.15.4) 3.3 (2.34.7)
Age, years
039 4.5 (1.612.6) 5.4 (1.618.2)
4064 8.5 (6.511.2) 5.9 (4.18.4)
6579 3.9 (3.05.1) 2.6 (1.83.8)
80 3.2 (2.05.0) 2.5 (1.44.4)
a
Crude OR for presence of diabetes in patients with PLA, compared with sex- and age-matched
control subjects.
b
OR adjusted by conditional logistic regression analysis for medical risk factors recorded before
or on the date of the PLA discharge diagnosis and for surgical risk factors recorded within 6 months
of the date of PLA discharge diagnosis, excluding procedures performed during the hospitalization
for PLA.
c
Excluding probably PLA-related procedures (see Appendix).

DISCUSSION derestimation of mortality attributable to PLA. However, the

This large nationwide study, which spans a 26-year period,
shows that diabetes is a strong risk factor for PLA. In addition,
it shows that mortality rates among individuals with diabetes
who are hospitalized for PLA are at least as high as those for
patients without diabetes who are hospitalized with PLA.
Our study has several strengths. Because nearly all inpatient
treatment in Denmark is provided by the national health ser-
vice, our design is virtually population based for the identifi-
cation of case patients and control subjects. We were able to
adjust for a wide range of potential confounders through access
to independent medical databases providing a complete medical
history and a record of all surgical procedures performed
among study participants.
Some limitations also deserve discussion. The validity of all
of our estimates depends on the accuracy of the hospital dis-
charge diagnoses used to identify cases of PLA and diabetes.
Patients with PLA usually have severe symptoms, including
abdominal pain, fever, and malaise, and are thus unlikely to
avoid hospitalization [9, 17]. Still, the onset of PLA may be
insidious, with unspecific symptoms, and patients may die be-
fore a diagnosis is made [10, 22]. Another potential weakness
stems from the clinical practice of keeping patients with dia-
betes under close surveillance for infection. This could poten-
tially lead to overestimation of the risk and, probably, to un-
high mortality observed among patients with diabetes and PLA
argues against closer surveillance. We consider it unlikely that
increased surveillance alone explains the consistently high in-
creases in risk that were observed.
Another concern is possible misclassification of PLA. How-
ever, a previous Danish series [3] showed that 51 (98%) of 52
microbiologically confirmed cases of PLA were correctly iden-
tified in the regional component of the National Patient Reg-
istry used in our study and that misclassification of PLA is
unlikely to be associated with diabetes. We also showed that,
even assuming considerably less complete ascertainment of di-
abetes among control subjects than among case patients with
PLA, adjusted risk estimates for diabetes remained 23 times
higher.
Similarly, we find it unlikely that unmeasured or unknown
confounders could explain risk estimates of the magnitude ob-
served. We were able to adjust for a wide range of important
PLA risk factors that only modestly decreased the relative risk
for diabetes. Misclassification of data on confounders might
have led to some residual confounding. However, registration
of previous diagnoses and procedures should be at least as
complete for patients with diabetes as it is for other individuals,
leading to conservative risk estimates.
Our findings corroborate those of a recent population-based

1198 CID 2007:44 (1 May) Thomsen et al.

Table 3. Association between diabetes mellitus and 30-day tors [17]. Our findingsthat gastrointestinal and other malig-
postdischarge mortality in 1442 patients with pyogenic liver ab- nancies, as well as alcoholism, were frequent underlying risk
scess (PLA). factors among cases of PLAare consistent with the findings
of the Canadian study [5]. Diabetes appears to increase PLA
30-day Adjusted OR,
No. (%) postdischarge diabetes vs. other risk substantially, regardless of the presence or absence of other
a
Characteristic of patients mortality, % (95% CI) risk factors. Major mechanisms for PLA development are he-
Overall 1442 (100) 18.5 1.3 (0.92.1) matogenous seeding of the liver, either through the portal sys-
Age, years tem or the greater circulation, or local spread from infections
039 156 (10.8) 3.8 13.7 (0.8238.8) within the peritoneal cavity [6]. Diabetes is a documented risk
4064 566 (39.3) 13.4 1.0 (0.42.2) factor for gram-negative bacteremia, including episodes derived
6579 538 (37.3) 22.9 1.3 (0.72.4) from abdominal foci of infection [8]. Underlying biological
80 182 (12.6) 34.1 2.0 (0.75.5) mechanisms may include tissue hyperglycemia and predilection
Sex for certain microorganisms, including Escherichia coli and Kleb-
Male 781 (54.2) 15.2 1.1 (0.62.1)
siella species [12, 24, 25]. Unfortunately, we lacked the system-
Female 661 (45.8) 22.4 1.5 (0.82.9)
b atic microbiological data needed to investigate this issue further.
Comorbidity index
Hansen et al. [26] studied a subgroup of our cohort and found
Low (0) 741 (51.4) 10.7 1.3 (0.63.0)
Medium (12) 478 (33.1) 22.2 1.5 (0.82.9)
that enteric gram-negative rods accounted for 45% of PLA-
High (12) 223 (15.5) 36.8 1.1 (0.52.4) related isolates, anaerobic bacteria accounted for 31%, and
Time period gram-positive cocci accounted for 19%. These findings are
19771988 472 (32.7) 33.9 1.3 (0.62.5) comparable with those of other Western studies of PLA [7, 17].
19891996 473 (32.8) 11.0 1.4 (0.63.3) Of note, the prevalence of PLA due to Klebsiella species is
19972002 497 (34.5) 11.1 1.3 (0.62.9) reported to be currently increasing in the United States [11,
NOTE. Six cases of PLA from the case-control study were excluded be- 27]. The role of diabetes for this observed shift in PLA epi-
cause of an invalid date of death in the registry. demiology remains to be elucidated.
a
Relative mortality adjusted by logistic regression analysis for age, sex, Our in-hospital mortality figures were high, but they were
level of comorbidity, alcoholism-related disorders, and time period of diagnosis
(except when stratified by variable). comparable with findings from previous, predominantly uni-
b
Level of Charlson index score; see Methods. versity hospitalbased studies [5, 9, 13, 17]. The decrease in
mortality attributable to PLA between the early 1980s and re-
cent years that was documented in our study is consistent with
epidemiologic study conducted in the Calgary Health Region previous observations [3, 18].
of Canada. In that region, the relative risk for development of We found that diabetes was associated with statistically non-
PLA among individuals with diabetes was estimated at 11.1 significantly increased mortality. Pathophysiological mecha-
(95% CI, 6.319.0) [5]. However, the study relied on estimated nisms might include harmful effects of hyperglycemia [28],
diabetes prevalence in the background population and did not general diabetic angiopathy, and decreased immunity. Recently,
adjust for confounding risk factors that were found to play an Chen et al. [29] found that diabetes was associated with a 7.7-
important role in our study. fold (95% CI, 2.129-fold) increased risk for developing met-
The average diabetes prevalence of 11% in our PLA cohort astatic infections from PLA. There is increasing evidence that
that spanned the years 19772002 was comparable to the 13% diabetes is a factor associated with a poor prognosis among
17% prevalence of diabetes reported for cohorts of individuals patients with gram-negative bacteremia [8] or conditions as-
with PLA in the United States and Europe during the 1990s sociated with abdominal sepsis, such as gastrointestinal per-
[911]. The earlier studies included !100 patients with PLA in foration [19]. We speculate that latent diabetic organ disease
most cases, however, and had no control groups for compar- may render patients with diabetes particularly vulnerable to
ison. These studies estimates are markedly lower than the di- multiorgan failure associated with gram-negative sepsis.
abetes prevalence found in recent Taiwanese case series [12, 13, We conclude that diabetes is a strong, potentially modifiable
23]. These series, which encompass several hundred patients risk factor for PLA. The prognosis for patients with PLA who
with PLA, have reported a prevalence of diabetes of 75% have diabetes is as poor as the prognosis is for other patients
among patients with PLA who are infected with Klebsiella pneu- who have PLA.
moniae, the dominant microbial agent causing PLA in this re-
gion [12, 13, 23]. Acknowledgments
Our study confirmed others findings that truly crypto-
Financial support. The Western Danish Research Forum for Health
genic PLA is rare: only one-quarter (24.5%) of patients with Sciences and Klinisk Epidemiologisk Forskningsfond.
PLA in our study lacked recorded medical or surgical risk fac- Conflicts of interest. All authors: no conflicts.

Diabetes and Pyogenic Liver Abscess CID 2007:44 (1 May) 1199

APPENDIX
DIAGNOSIS CODES USED IN THE STUDY, IN
ACCORDANCE WITH THE DANISH VERSION OF
THE INTERNATIONAL CLASSIFICATION OF
DISEASES (ICD; 8TH REVISION [ICD-8] USED
UNTIL THE END OF 1993, 10TH [ICD-10]
REVISION USED THEREAFTER).
Diabetes mellitus (ICD-8: 249 and 250; ICD-10: E10, E11, E14,
and H36.0) and pyogenic liver abscess (ICD-8: 572.01 and
572.09; ICD-10: K75.0).
Medical risk factors. Benign biliary obstruction (ICD-8:
574; ICD-10: K80), liver cancer (including liver metastases;
ICD-8: 155, 197.79, and 197.89; ICD-10: C22 and C78.7), bil-
iary tract cancer (ICD-8: 156; ICD-10: C23 and C24), pancre-
atic cancer (ICD-8: 157; ICD-10: C25), other gastrointestinal
cancers (ICD-8: 150154, 158, 159, 195.09, 196.29, 197.49,
197.59, 197.69, and 197.99; ICD-10: C15C21, C26, C76.2,
C77.2, C78.4, C78.5, C78.6, and C78.8), other solid-organ can-
cers (ICD-8: 140149, 160194, 195.19195.99, 196.09, 196.19,
196.39196.99, 197.09197.39, 198, and 199; ICD-10: C00
C14, C30C75, C76.0, C76.1, C76.3C76.8, C77.0, C77.1,
C77.3C77.9, C78.0C78.3, C79, and C80), hematological can-
cers (ICD-8: 200203, 204207, and 275.59; ICD-10: C81C85,
C88, C90, C91C95, and C96), liver cirrhosis (ICD-8: 571.09,
571.90571.92, and 571.99; ICD-10: K70.3, K71.7, and K74.3
K74.6), inflammatory bowel disease (ICD-8: 563; ICD-10: K50
and K51), alcoholism-related disorders (ICD-8: 291, 303, 979,
980, 577.10, 571.09, and 571.10; ICD-10: F10, G31.2, G62.1,
G72.1, I 42.6, K29.2, K86.0, K70, R78.0, T51, and Z72.1 [for
adjustment in the prognosis analysis, the codes for alcoholic
liver disease {K70, 571.09, and 571.10} were excluded, because
they were already included in the Charlson index]), ongoing
abdominal infection (acute cholecystitis or cholangitis without
cholelithiasis, diverticulitis, appendicitis, or peritonitis; ICD-8:
575, 562.1119, 540, and 567; ICD-10: K81.0, K83.0, K57.0,
K57.2, K57.4, K57.8, K35, and K65), and ongoing systemic
infection (sepsis or endocarditis; ICD-8: 038 and 421; ICD-10:
A40, A41, and I33).
Surgical risk factors. Probably PLA-related procedures (in-
cluding liver punctures, biopsies, and drainage; KJJA10A,
KTJJ00, KJWC00, 91560, 91900, 47280, 47120, and 49000),
other surgical procedures performed on the liver (KJJ [exclud-
ing KJJA10A], KTJJ [excluding KTJJ00], and 4700047291 [ex-
cluding 47280 and 47120]), surgical procedures performed on
the biliary tract (including the gall bladder; KJK, KTJK, 47320
48299, and 92301), surgical procedures performed on the pan-
creas (KJL, KJTL, 4832048895, 91920, and 92305), surgical
procedures performed on the spleen (KJL, KJTL, 4832048895,
1200 CID 2007:44 (1 May) Thomsen et al.
91920, and 92305), upper endoscopy (KUJD and 91010
91040), endoscopy of the biliary tract (including endoscopic
retrograde cholangiopancreatography; KUJK and 91050
91059), any other gastrointestinal surgery (KJA-KJH, KTJA-
KTJG, KTJW, 4002046990, 92260, 92280, 92300, and 92320
92370), and any other surgical procedure (all other procedure
codes).
References
1. Branum GD, Tyson GS, Branum MA, Meyers WC. Hepatic abscess:
changes in etiology, diagnosis, and management. Ann Surg 1990; 212:
65562.
2. Huang CJ, Pitt HA, Lipsett PA, et al. Pyogenic hepatic abscess: changing
trends over 42 years. Ann Surg 1996; 223:6007.
3. Jepsen P, Vilstrup H, Schnheyder HC, Srensen HT. A nationwide
study of the incidence and 30-day mortality rate of pyogenic liver
abscess in Denmark, 19772002. Aliment Pharmacol Ther 2005; 21:
11858.
4. Mlle I, Thulstrup AM, Vilstrup H, Srensen HT. Increased risk and
case fatality rate of pyogenic liver abscess in patients with liver cirrhosis:
a nationwide study in Denmark. Gut 2001; 48:2603.
5. Kaplan GG, Gregson DB, Laupland KB. Population-based study of the
epidemiology of and the risk factors for pyogenic liver abscess. Clin
Gastroenterol Hepatol 2004; 2:10328.
6. The liver in infections. In: Sherlock S, Doley J, eds. Diseases of the
liver and biliary system. Oxford: Blackwell Science, 1997:497500.
7. Krige JEJ, Beckingham IJ. ABC of diseases of liver, pancreas, and biliary
system: liver abscesses and hydatid disease. BMJ 2001; 322:53740.
8. Thomsen RW, Hundborg HH, Lervang HH, Johnsen SP, Schnheyder
HC, Srensen HT. Diabetes mellitus as a risk and prognostic factor
for community-acquired bacteremia due to enterobacteria: a 10-year,
population-based study among adults. Clin Infect Dis 2005; 40:62831.
9. lvarez Perez JA, Gonzalez JJ, Baldonedo RF, et al. Clinical course,
treatment, and multivariate analysis of risk factors for pyogenic liver
abscess. Am J Surg 2001; 181:17786.
10. Hernandez JL, Ramos C. Pyogenic hepatic abscess: clues for diagnosis
in the emergency room. Clin Microbiol Infect 2001; 7:56770.
11. Rahimian J, Wilson T, Oram V, Holzman RS. Pyogenic liver abscess:
recent trends in etiology and mortality. Clin Infect Dis 2004; 39:16549.
12. Wang JH, Liu YC, Lee SS, et al. Primary liver abscess due to Klebsiella
pneumoniae in Taiwan. Clin Infect Dis 1998; 26:14348.
13. Yang CC, Yen CH, Ho MW, Wang JH. Comparison of pyogenic liver
abscess caused by nonKlebsiella pneumoniae and Klebsiella pneumon-
iae. J Microbiol Immunol Infect 2004; 37:17684.
14. Ministry of the Interior and Health. Health care in Denmark. Copen-
hagen: Ministry of the Interior and Health. Available at: http://
www.im.dk/publikationer/healthcare_in_dk/index.htm. Accessed 10
January 2007.
15. Nickelsen TN. Data validity and coverage in the Danish National
Health Registry: a literature review. Ugeskr Laeger 2001; 164:337.
16. Szklo M, Nieto FJ. Epidemiology: beyond the basics. 1st ed. Gaithers-
burg, MD: Aspen Publishers, 2000:345.
17. Mohsen AH, Green ST, Read RC, McKendrick MW. Liver abscess in
adults: ten years experience in a UK centre. QJM 2002; 95:797802.
18. Petri A, Hohn J, Hodi Z, Wolfard A, Balogh A. Pyogenic liver abscess
20 years experience: comparison of results of treatment in two periods.
Langenbecks Arch Surg 2002; 387:2731.
19. Thomsen RW, Riis A, Christensen S, Nrgaard M, Srensen HT. Di-
abetes and 30-day mortality from peptic ulcer bleeding and perforation:
a Danish population-based cohort study. Diabetes Care 2006; 29:
80510.
20. Drivsholm TB, Frederiksen K, de Fine Olivarius N, Odegaard B, Kris-
tensen JK. The prevalence of diabetes in Denmark: development of a
 method for a registry-based assessment. Ugeskr Laeger 2003; 165:
288791.
21. Frank L. Epidemiology: when an entire country is a cohort. Science
2000; 287:23989.
22. Smoger SH, Mitchell CK, McClave SA. Pyogenic liver abscesses: a
comparison of older and younger patients. Age Ageing 1998; 27:4438.
23. Ko WC, Paterson DL, Sagnimeni AJ, et al. Community-acquired Kleb-
siella pneumoniae bacteremia: global differences in clinical patterns.
Emerg Infect Dis 2002; 8:1606.
24. Hansen DS, Gottschau A, Kolmos HJ. Epidemiology of Klebsiella bac-
teraemia: a case control study using Escherichia coli bacteraemia as
control. J Hosp Infect 1998; 38:11932.
25. Geerlings SE, Meiland R, van Lith EC, Brouwer EC, Gaastra W, Hoe-
pelman AI. Adherence of type 1-fimbriated Escherichia coli to uroe-
pithelial cells: more in diabetic women than in control subjects. Di-
abetes Care 2002; 25:14059.
26. Hansen PS, Schnheyder HC. Pyogenic hepatic abscess: a 10-year pop-
ulation-based retrospective study. APMIS 1998; 106:396402.
27. Lederman ER, Crum NF. Pyogenic liver abscess with a focus on Kleb-
siella pneumoniae as a primary pathogen: an emerging disease with
unique clinical characteristics. Am J Gastroenterol 2005; 100:32231.
28. Van den Berghe G, Wouters P, Weekers F, et al. Intensive insulin therapy
in critically ill patients. N Engl J Med 2001; 345:135967.
29. Chen SC, Lee YT, Lai KC, et al. Risk factors for developing metastatic
infection from pyogenic liver abscesses. Swiss Med Wkly 2006; 136:
11926.
Diabetes and Pyogenic Liver Abscess CID 2007:44 (1 May) 1201