Full text of DOI:10.1053/gast.2002.33650

GASTROENTEROLOGY 2002;122:1822–1828
Diabetes Increases the Risk of Acute Hepatic Failure
HASHEM B. EL–SERAG* and JAMES E. EVERHART‡
*Sections of Gastroenterology and Health Services Research, Houston Department of Veterans Affairs Medical Center; Baylor College of
Medicine, Houston, Texas; and ‡National Institute of Diabetes and Digestive and Kidney Diseases, Bethesda, Maryland
Background & Aims: It is unclear whether patients with
diabetes are at an increased risk of developing acute
liver failure (ALF). We performed a large cohort study to
examine the occurrence of ALF by using the databases
of the Department of Veterans Affairs. Methods: We
identified all patients with a hospital discharge diagno-
sis of diabetes (ICD-9 codes: 250 [1–9][0–4]) from
1985 to 1990 and randomly assigned patients without
diabetes for comparison (3:1 ratio). We excluded pa-
tients with concomitant liver disease as far back as
1980. After excluding the first year of follow-up, the
remaining patients were observed through 2000 for the
occurrence of ALF (ICD-9 570). The cumulative risk and
the relative risk of ALF were determined by Kaplan–
Meier and Cox Proportional Hazard survival analysis,
respectively. Results: We included 173,643 patients
with diabetes and 650,620 patients without diabetes.
Patients with diabetes were significantly older (62 vs. 54
years) and were less likely to be white (28% vs. 24%).
The cumulative risk of ALF was significantly higher
among patients with diabetes (incidence rate, 2.31 per
10,000 vs. 1.44 per 10,000 person-years; P < 0.0001).
In the Cox proportional hazard model, diabetes was
associated with a relative risk of 1.44 (95% CI, 1.26–
1.63; P < 0.0001) for ALF while controlling for comor-
bidity index, age, sex, ethnicity, and period of service.
This risk remained significantly increased after exclud-
ing patients with liver disease or viral hepatitis recorded
during follow-up or those with ALF recorded after the
introduction of troglitazone (relative risk ؍ 1.40; P <
0.0001). Conclusions: Diabetes increases the risk of
ALF. The increase in ALF is independent of recognized
underlying chronic liver disease or viral hepatitis.
among patients with other causes of chronic liver disease,
such as hepatitis C virus or alcoholic cirrhosis.8 Recent
articles have shown fulminant hepatic failure among
diabetics that is related to the use of the hypoglycemic
agent troglitazone.9–12 However, apart from these case
reports, it remains unknown whether diabetes increases
the risk for fulminant hepatic failure among troglitazone
nonusers.
The Department of Veterans Affairs (VA) is one of the
largest health care providers in the United States. The
computerized database of the VA, the Patient Treatment
File (PTF), registers patients’ information from 172 VA
hospitals throughout the United States. By using infor-
mation contained in these databases, we have examined
the incidence of acute liver failure in a large cohort of
hospitalized veterans with diabetes.
Methods
Databases
The study population was assembled from hospitalized
veterans registered within the nationwide PTF. The PTF com-
prises a multitude of annual data files in which discharge
diagnoses are recorded for each inpatient hospital visit since
1970. Individual patients can be traced through the annual
files of the PTF through their unique social security numbers.
Since 1981, discharge diagnoses have been coded according to
the 9th revision of the Clinical Modification of International
Classification of Diseases (ICD-9-CM).13 The PTF does not
contain information about pharmacy or laboratory test results.
The Beneficiary Identification and Records Locator Sub-
system (BIRLS) file contains claims and benefits paid to vet-
erans and their beneficiaries. The Mini-BIRLS Death File of
the PTF contains 1 record for each veteran in the BIRLS file
that has a date of death recorded. Because of various incentives,
up to 90% of deaths among veterans are captured by the
BIRLS file.14–16
ersons with diabetes mellitus are at an increased risk
P
of developing a variety of liver diseases. Diabetes
predisposes to nonalcoholic fatty liver disease, including
nonalcoholic steatohepatitis. Nonalcoholic steatohepati-
tis may progress in 5%–20% of cases to cirrhosis.1
Diabetes and insulin resistance have been reported to be
independent risk factors for advanced fibrosis and cirrho-
sis among patients with nonalcoholic steatohepatitis.2–4
The risk of developing hepatocellular carcinoma may also
be increased in persons with diabetes,5–7 particularly
Abbreviations used in this paper: BIRLS, Beneficiary Identification
and Records Locator Subsystem; ICD, International Classification of
Diseases; PTF, Patient Treatment File; VA, Department of Veterans
Affairs.
© 2002 by the American Gastroenterological Association
0016-5085/02/$35.00
doi:10.1053/gast.2002.33650

June 2002
DIABETES AND ACUTE HEPATIC FAILURE 1823
496, 500–505, and 506.4); congestive heart failure (428.0–
428.9); dementia (290.0–290.9); diabetes with complica-
tions (250–250.3); hemiplegia or paraplegia (344.1 and 342–
342.9); myocardial infarction (410.0–410.9 and 412); peptic
ulcer disease (531–534.9); peripheral vascular disease (443.9,
441.0–441.9, 785.4, and V43.4); renal disease (582–582.9,
583–583.7, 585, 586, and 588–588.9); metastatic solid tu-
mors (196–199.1); any malignancy, including lymphoma and
leukemia (140–172.9, 174–195.8, and 200–208.9); and
rheumatologic disease (710.0, 710.1, 710.4, 714.0–714.2,
714.81, and 725). In constructing the index, we did not
include diabetes only (absent by definition from the control
group) or liver disease (absent in both groups by definition).
The composite score for the Deyo comorbidity index at the
initial hospitalization was compared between patients with
diabetes and those without. The comorbidity index was also
included as a covariate in the multiple regression analyses.
Study Population
The study population comprised all hospitalized pa-
tients with diabetes, as defined by discharge diagnosis codes
(ICD-9 250 [1–9][0–4]),13 during the 5-year period from
October 1985 to October 1990. For comparison, we randomly
assigned 3 patients without diabetes for every patient with
diabetes matched on the year of hospitalization. We included
only adults (older than 20 years). A method of random selec-
tion without replacement was used to ensure that no individ-
ual control subject was selected more than once. The sampling
frame for groups of patients with and without diabetes was
individual patients, rather than hospitalizations. Patients with
multiple hospitalizations were counted only once, and their
earliest hospitalization date was chosen to be the index hospi-
talization.
By using the social security number as a unique identifier,
the remaining patients were prospectively observed in the
annual PTF and BIRLS death files until the end of fiscal year
2000. These files were searched for the occurrence of diagnoses
of acute liver failure (ICD-9 570), as well as death (in BIRLS),
in both groups of patients. We also searched the PTF for other
causes of acute liver failure and conditions that could be
confused with acute liver failure. These included hepatitis C
(ICD-9 codes 070.41, 070.44, 070.51, 070.54, and V02.62);
hepatitis B (070.20–070.23, 070.30–070.33, and V02.61);
heart failure (428.0, 428.1, and 428.9); hepatitis A (070.0 and
070.1); liver abscess (572.0 and 572.1); fatty liver (571.0 and
571.8); chronic liver disease, including alcoholic liver disease
and cirrhosis (571.1–571.3, 571.40, 571.41, 571.49, 571.5,
571.6, 571.9, 572.2, 572.3, and 572.8); alcoholism (all codes
starting with 303, and 790.3); abnormal liver enzymes
(790.4); hemochromatosis (275.0); and hepatocellular carci-
noma (155.0).
To increase the probability of capturing new (incident) cases
of acute liver failure, we excluded from both groups patients
with acute or chronic liver disease recorded during the index
hospitalization or during any previous hospitalization dating
back to 1980. We also excluded all patients who had any
diagnosis of acute or chronic liver disorder during the first year
that followed their index hospitalization. During subsequent
follow-up (after the first year), the presence of liver disease
(chronic liver disease, including alcohol-related liver disease,
cirrhosis, viral hepatitis [B or C], fatty liver, liver abscess, and
hepatocellular carcinoma) was identified and recorded. The
time at risk of acute liver failure was measured from 1 year
after the index hospitalization until October 2000 or the
development of acute liver failure or death.
Statistical Analyses
Baseline demographic features were compared between
2
patients with and those without diabetes; ␹ tests were used
for univariate comparisons between dichotomous variables,
whereas unpaired t tests were used to compare continuous
variables. The cumulative risk of acute liver failure was esti-
mated in a Kaplan–Meier survival analysis. The log-rank test
was used to test the statistical significance for differences in the
rates of acute liver failure between patients with and those
without diabetes. The occurrence of acute liver failure was
modeled as the outcome variable in a Cox proportional hazard
survival analysis that examined the risk associated with diabe-
tes while controlling for age, sex, ethnicity, period of military
service, hepatitis C virus, hepatitis B virus, chronic liver
disease, and the Deyo comorbidity index. Separate models were
used to examine the same associations after excluding patients
with any liver disease, alcoholism, abnormal liver tests, or
those with risk factors for acute liver failure (viral hepatitis and
heart failure) that were recorded during the follow-up period,
including those recorded in the index hospitalization. The
models were tested for the presence of statistical interactions
among the potential risk factors. Wald’s ␹2 tests were used to
test for the significance of the influence for each risk factor.
Hazard rate ratios and their 95% confidence intervals were
presented for each estimate to represent the relative risk. The
log-log survival plots were used to examine the proportional
hazards assumption, which was met in all models. All analyses
were performed with SAS (SAS Institute, Cary, NC).18
Results
The inpatient files were searched for a number of conditions
during the initial hospitalization to calculate the Deyo comor-
bidity index, which is a modification of the Charlson disease
severity index.17 This index is commonly used to adjust for the
effect of comorbidities when using administrative data in
comparative studies. These conditions included the following:
acquired immunodeficiency syndrome (042–044); cerebrovas-
cular disease (430–438); chronic pulmonary disease (490–
We identified 257,649 patients with diabetes and
772,947 patients without diabetes who were hospitalized
in VA facilities between October 1985 and October
1990. Of these, 216,831 patients with diabetes and
765,853 patients without diabetes did not have liver
disease in their hospitalization records as far back as
1980. After excluding all patients in whom liver disease

1824 EL–SERAG AND EVERHART
GASTROENTEROLOGY Vol. 122, No. 7
was recorded during the first year of follow-up, 173,643
patients with diabetes and 650,620 patients without
diabetes remained and were included in the principal
analysis of acute liver failure. The demographic features
for these patients are shown in Table 1. Patients with
diabetes were an average of 8 years older and were less
likely to be white or belong to a Vietnam period of
service (not necessarily serving in Vietnam) than patients
without diabetes. The majority (99.5%) of diabetic pa-
tients had type II diabetes mellitus. The translation of
the Charlson comorbidity index into ICD-9-CM codes
includes 14 diagnostic categories (see Methods). With a
2
Mantel-Haenszel ␹ test, there was no significant differ-
Figure 1. The cumulative risk of acute liver failure among veteran
patients hospitalized during 1985–1990. No patient had acute or
chronic liver disease recorded before, during, or 1 year after their
index hospitalization. The follow-up period ended in October 2001.
The upper line represents acute liver failure recorded after the first
year of index hospitalization in patients with diabetes, whereas the
lower line represents the same outcome in patients without diabetes.
Acute liver failure was significantly higher in patients with diabetes
(P Ͻ 0.0001).
ence between cases and controls in these comorbid con-
ditions at the initial hospitalization (2.96; P ϭ 0.09). For
example, myocardial infarction was diagnosed in 8.7% of
cases with diabetes and 8.6% of controls without diabe-
tes; congestive heart failure in 4.9% of cases and 4.9% of
controls; peripheral vascular disease in 1.7% of cases and
1.6% of controls; and cerebrovascular disease in 0.6% of
cases and controls. An average comorbidity score that
reflects the presence or absence of all of the 14 diagnostic
categories was calculated for each person in the study;
this variable was included in the regression analyses
described below.
During follow-up of 1,494,995 patient-years with
diabetes, 346 patients developed acute liver failure (in-
cidence rate, 2.31 per 10,000 person-years); during fol-
low-up of 6,556,350 patient-years without diabetes, 942
patients developed acute liver failure (incidence rate,
1.44 per 10,000 person-years). During follow-up that
ended in October 2000, the Kaplan–Meier survival anal-
ysis showed a significantly higher cumulative incidence
of acute liver failure among patients with diabetes com-
pared with those without diabetes (0.30% vs. 0.19%;
P Ͻ 0.0001; Figure 1). The 6-week mortality after
hospitalization with acute liver failure was 60% for
patients with diabetes and 63% for those without dia-
betes (P Ͼ 0.05).
In the Cox proportional hazard analysis, diabetes was
associated with a relative risk of acute liver failure of 1.44
(95% confidence interval, 1.26% to 1.63%; P Ͻ
0.0001). This adjusted relative risk was obtained while
controlling for differences in age, sex, ethnicity, Deyo
comorbidity index, and period of service (Table 2). In a
separate model, we also controlled for the presence of
chronic liver disease, including alcoholic liver disease and
cirrhosis, that was recorded during follow-up. In that
model, diabetes remained an independent risk for acute
liver failure, with an adjusted relative risk of 1.44. In the
Table 2. Risk Factors for Acute Liver Failure: Results of
Multivariate Cox Proportional Hazard Analysis in a
Cohort of 824,263 Hospitalized Veterans During
1985–1990 With No Prior Liver Disease, While
Adjusting for the Comorbidity Index
Table 1. A Comparison of Demographic Characteristics and
Major Risk Factors Between Cases With Diabetes
and Controls Without Diabetes
Adjusted
hazard
ratio
95%
Confidence
interval
P
Variable
value
Cases
Control
P
Variable
(n ϭ 173,643)
(n ϭ 650,620)
value
Diabetes
Older age (per 10 yr)
Women
1.44
1.88
1.12
1.26–1.63 Ͻ0.0001
1.40–2.51 Ͻ0.0001
Age, yr, mean
(ϮSD)
0.79–1.61
0.5261
61.7 (Ϯ10)
54.5 (Ϯ12)
Ͻ0.0001
Hispanic (vs. non-Hispanic,
white)
Black (vs. non-Hispanic, white)
Vietnam era (vs. non-Vietnam)
Chronic liver disease
(including cirrhosis)a
Men
Ethnicity:
Non-Hispanic,
White
Period of military
service
170,944 (98.4%) 635,005 (97.6%) Ͻ0.0001
124,361 (71.6%) 496,290 (76.3%) Ͻ0.0001
23,308 (13.4%) 175,949 (27.0%) Ͻ0.0001
0.83
0.96
1.06
0.56–1.22
0.83–1.11
0.90–1.23
0.3395
0.5801
0.5045
14.31
12.55–16.3 Ͻ0.0001
(Vietnam era)
aRecorded during follow-up.

June 2002
DIABETES AND ACUTE HEPATIC FAILURE 1825
Table 3. Risk Factors for Acute Liver Failure: Results of
Multivariate Cox Proportional Hazard Analysis in a
Cohort of 797,551 Hospitalized Veterans During
1985–1990 With No Prior Chronic Liver Disease
(Including Alcoholic), Liver Abscess, Hepatocellular
Carcinoma, Hemochromatosis, Viral Hepatitis, or
Concomitant Congestive Heart Failure
P Ͻ 0.0001). All models were also adjusted for the Deyo
comorbidity index.
Troglitazone, a hypoglycemic drug linked to fulmi-
nant hepatic failure, was introduced to the VA pharmacy
in 1998. When the follow-up of the study cohort was
restricted to the end of 1997, the rate of acute liver
failure remained higher among patients with diabetes in
Kaplan–Meier analysis (0.28% vs. 0.17%; P Ͻ 0.0001)
and in the Cox proportional hazard model (1.40, 1.22–
1.61; P Ͻ 0.0001) (Figure 2).
Adjusted
hazard
ratio
95%
Confidence
interval
P
Variable
value
Diabetes
Older age (per 10 yr)
Women
1.43
1.92
1.30
1.25–1.67 Ͻ0.0001
1.27–2.88 Ͻ0.0001
0.89–1.92
0.1898
Discussion
Hispanic (vs. non-Hispanic,
white)
Black (vs. non-Hispanic, white)
Vietnam-era veteran
(vs. non-Vietnam)
Comorbidity index
0.77
1.03
0.49–1.22
0.87–1.22
0.2674
0.7270
We are unaware of other studies that have specif-
ically examined the incidence of acute liver failure in
diabetic patients. In this large cohort study, we found
that diabetes mellitus was associated with an approxi-
mately 1.5-fold increase in the risk of acute hepatic
failure. This increased risk was not fully explained by the
presence of underlying liver disease or concomitant
known risk factors, such as viral hepatitis, chronic liver
disease, alcoholism, congestive heart failure, or other
major nonhepatic comorbid illness. The risk of acute
liver failure associated with diabetes also predated the
introduction of troglitazone.
1.00
1.03
0.79–1.19
0.97–1.09
0.7791
0.3203
same analysis, a diagnosis of chronic liver disease proved
to be a strong predictor of acute liver failure, with a
relative risk of 14.3 (Table 3). Chronic liver disease was
also a strong risk factor in a model that did not include
diabetes as a covariate (relative risk ϭ 13.5). Increasing
age was also associated with acute liver failure. After
controlling for the older age of the diabetic group, the
effect of Vietnam period of service was no longer signif-
icant.
Diabetes can be a result of severe liver disease. For
example, it is reported that 10% to 20% of patients with
cirrhosis may have overt diabetes mellitus, and a larger
fraction have impaired glucose tolerance.19,20 Moreover,
Of patients who developed acute liver failure, 20%
had chronic liver disease (including cirrhosis and alco-
holic liver disease; ICD-9 571.1–571.3, 571.40, 571.41,
571.49, 571.5, 571.6, 571.9, 572.2, 572.3, and 572.8)
recorded after the first year of follow-up, 2% had fatty
liver, 1% had hepatocellular carcinoma, 12% had alco-
holism, 1% had liver abscess, and 0.2% had abnormal
liver enzymes with no documented specific liver disease.
However, most patients who developed acute liver failure
(70%) had no documented past or concomitant liver
disease or alcoholism. Patients with diabetes and acute
liver failure had a concurrent diagnosis of congestive
heart failure (31%) more often than patients without
diabetes who had acute liver failure (22%). When the
Cox proportional hazard analysis was restricted to pa-
tients with no liver disease (chronic liver disease that
included cirrhosis and alcoholic liver disease, liver ab-
scess, hepatocellular carcinoma, hemochromatosis, and
abnormal liver enzymes), viral hepatitis, or alcoholism
before the hospitalization with acute hepatic failure and
no concurrent diagnosis of congestive heart failure, the
adjusted relative risk for diabetes remained significantly
increased (1.43, 1.25–1.67; P Ͻ 0.0001). Further exclu-
sion of patients with fatty liver did not change the results
appreciably (1.41; 95% confidence interval, 1.22–1.63;
Figure 2. The cumulative risk of acute liver failure among veteran
patients hospitalized during 1985–1990. No patient had acute or
chronic liver disease recorded before, during, or 1 year after their
index hospitalization. The follow-up period was restricted up to Octo-
ber 1997 (before the introduction of troglitazone to the VA pharmacy
in 1998). The upper line represents acute liver failure recorded after
the first year of index hospitalization in patients with diabetes,
whereas the lower line represents the same outcome in patients
without diabetes. Acute liver failure was significantly higher in patients
with diabetes (P Ͻ 0.0001).

1826 EL–SERAG AND EVERHART
GASTROENTEROLOGY Vol. 122, No. 7
uncontrolled diabetes was also reported to accompany
fulminant liver failure.21 Therefore, to avoid inclusion of
patients with diabetes secondary to liver disease, we
excluded all cases of liver disease before the identification
of all patients, as well as those recorded during the first
year of follow-up. The use of the VA PTF facilitated the
identification of a large cohort of patients with diabetes
and the random identification of controls without diabe-
tes from the same hospitalization files. By including
many thousands of patients, the study design allowed the
detection of a modest increased risk of acute liver failure
that might be missed in a smaller study.
chronic liver disease, liver abscess, fatty liver, hepatocel-
lular carcinoma, and infectious hepatitis.
This study shows that people with diabetes are at
increased risk of acute hepatic failure, but the reasons for
the increased risk are uncertain. From other reports of
acute hepatic failure, we speculate that most of the
patients would have had drug-related liver injury. The
largest series of acute liver failure in the United States
showed that drugs were the most common cause of liver
failure.22 Although viral hepatitis is also a common cause
of acute liver failure, our results changed little with
exclusion of these conditions with acute liver failure
diagnosis. Furthermore, some causes of liver failure, such
as Wilson disease and autoimmune hepatitis, would not
be expected in this generally older population of men. If
drug-related hepatotoxicity is the main reason for the
increased risk among persons with diabetes, then diabet-
ics are either more susceptible to liver injury from drugs
or they have greater exposure to hepatotoxic medications.
It is quite possible that both increased susceptibility and
exposure contributed to the increased risk. The fatty liver
and steatohepatitis associated with diabetes are known to
cause mitochondrial injury25 and other hepatocellular
dysfunction,1 which would increase susceptibility to in-
jury. Although excluding persons with recorded fatty
liver did not change our results, this diagnosis may be
underreported. In the setting of insulin deficiency or
insulin resistance, there is increased lipolysis with release
of free fatty acids, which are taken up by the liver. Free
fatty acids are potentially cytotoxic and cause mitochon-
drial swelling, increased lysosomal fragility, and impair-
ment of cellular membranes.26 Patients with diabetes
have higher levels of plasma free fatty acids as compared
with nondiabetics.27 Diabetics may also have a greater
exposure to dose-dependent or idiosyncratic hepatotox-
ins, because they typically take several medications, both
for diabetes and for associated conditions such as hyper-
tension, heart disease, and hyperlipidemia. In the general
population, more than half of people with diabetes took
3 or more prescription medications.28 Because the VA
database did not contain pharmacy information, we
could not establish which, if any, medications were caus-
ing the increased risk. We do know that one hypogly-
cemic agent, troglitazone, was implicated in several cases
of acute liver failure in the late 1990s.9–12 However, in
this study, limiting our follow-up to the time before the
availability of troglitazone had no effect on the relative
risk of diabetes for acute liver failure.
Misdiagnosis of both the risk factor of interest (dia-
betes) and outcome (acute liver failure) are potential
concerns in large administrative data sets, in which di-
agnoses cannot be verified. Although the vast majority of
patients with a diagnosis of diabetes actually have dia-
betes, the disease is frequently not recognized and is
underreported on medical records. Thus, there were cer-
tainly patients in the nondiabetic group who actually had
diabetes. This misclassification would tend to diminish
the true effect of diabetes. Although the validity of the
ICD-9 code as an indicator of acute liver failure is not
known, the high mortality after hospitalization with
acute liver failure (more than 60% during the first 6
weeks) is consistent with carefully documented results22
and argues against an erroneous inclusion of a significant
number of cases with mild or chronic liver disease. In
addition, the same code was used to define acute liver
failure in patients with and without diabetes, and the
mortality rates for these groups were not different. Al-
though it is unlikely that acute liver failure is overlooked
as a diagnosis, it is possible that patients receive this
diagnosis in whom the liver injury is secondary to an-
other disorder. Such misclassification with acute liver
failure would most likely occur with congestive heart
failure and “shock liver,” which is more common among
diabetics. Therefore, we also performed an analysis that
excluded patients with concurrent diagnoses of acute
liver failure and congestive heart failure. The relative risk
was slightly reduced but remained highly statistically
significant. Other potential confounders include chronic
liver disease, liver abscess, fatty liver, and hepatocellular
carcinoma, all of which are known to be more common
among diabetics and also increase the risk of acute he-
patic failure. Patients with diabetes have also been re-
ported to have higher rates of infectious hepatitis, espe-
cially hepatitis C virus and hepatitis B virus, possibly
related to needle use.23,24 In this study, however, diabetes
remained an independent risk factor of acute hepatic
failure even after we excluded patients with concurrent
Differences in comorbid illness between diabetics and
nondiabetics could affect the occurrence of acute liver
failure, either directly or indirectly, through the use of

June 2002
DIABETES AND ACUTE HEPATIC FAILURE 1827
ease: predictors of nonalcoholic steatohepatitis and liver fibrosis
in the severely obese. Gastroenterology 2001;121:91–100.
3. Matteoni CA, Younossi ZM, Gramlich T, Boparai N, Liu YC, McCul-
lough AJ. Nonalcoholic fatty liver disease: a spectrum of clinical and
pathological severity. Gastroenterology 1999;116:1413–1419.
4. Marchesini G, Brizi M, Morselli-Labate AM, Bianchi G, Bugianesi
E, McCullough AJ, Forlani G, Melchionda N. Association of non-
alcoholic fatty liver disease with insulin resistance. Am J Med
1999;107:450–455.
5. La Vecchia C, Negri E, Decarli A, Franceschi S. Diabetes mellitus
and the risk of primary liver cancer. Int J Cancer 1997;73:204–
207.
6. Adami HO, Chow WH, Nyren O, Berne C, Linet MS, Ekbom A, Wolk
A, McLaughlin JK, Fraumeni JF Jr. Excess risk of primary liver
cancer in patients with diabetes mellitus. J Natl Cancer Inst
1996;88:1472–1477.
7. Lagiou P, Kuper H, Stuver SO, Tzonou A, Trichopoulos D, Adami
HO. Role of diabetes mellitus in the etiology of hepatocellular
carcinoma. J Natl Cancer Inst 2000;92:1096–1099.
8. El-Serag HB, Richardson P, Everhart JE. The role of diabetes in
hepatocellular carcinoma among veterans: a case-control study.
Am J Gastroenterol 2001;96:2462–2467.
9. Murphy EJ, Davern TJ, Shakil AO, Shick L, Masharani U, Chow H,
Freise C, Lee WM, Bass NM. Troglitazone-induced fulminant he-
patic failure. Acute Liver Failure Study Group. Dig Dis Sci 2000;
45:549–553.
10. Booth AM, Caldwell SH, Iezzoni JC. Troglitazone-associated he-
patic failure. Am J Gastroenterol 2000;95:557–558.
11. Menon KVN, Angulo P, Lindor KD. Severe cholestatic hepatitis
from troglitazone in a patient with nonalcoholic steatohepatitis
and diabetes mellitus. Am J Gastroenterol 2001;96:1631–1634.
12. Li H, Heller DS, Leevy CB, Zierer KG, Klein KM. Troglitazone-
induced fulminant hepatitis: a report of a case with autopsy
findings. J Diabetes Complications 2000;14:175–177.
13. Department of Health and Human Services. The international
classification of diseases. 9th rev., clinical modification, 3rd ed.
Vol. 1: Diseases: tabular list. DHHS publication no. PHS 89-
1260. Government Printing Office, Washington, DC, 1989.
14. Boyko EJ, Koepsell TD, Gaziano JM, Horner RD, Feussner JR. US
Department of Veterans Affairs medical care system as a re-
source to epidemiologists. Am J Epidemiol 2000;151:307–314.
15. Page WF. VA mortality reporting for World War II army veterans.
Am J Epidemiol 1992;82:124–125.
multiple potentially hepatotoxic drugs. For every patient
in the study, we calculated the Deyo comorbidity index,
which is a modification of Charlson comorbidity index.17
The presence of major nonhepatic diseases was assessed in
this study in both diabetics and nondiabetics. There were
no statistically significant differences between the 2
groups in the composite score of the comorbidity index.
Moreover, including the comorbidity index as a covariate
in the regression model had little effect on the results,
with diabetes remaining as an independent risk factor for
acute liver failure.
To our knowledge, this study provides the first esti-
mate of the incidence of acute liver failure in a defined
population in the United States. Although hospitalized,
predominantly male veterans are not representative of the
United States population, the incidence rate of 1.44 per
10,000 person-years for nondiabetic veterans provides a
benchmark for other studies of acute liver failure. Al-
though the absolute rate of acute liver failure was low,
over the course of the study, 130 excess cases of acute
liver failure occurred among diabetics. Considering a
60% death rate with acute liver failure, there was an
excess of 78 deaths because of this condition among
diabetics during the period of the study.
In addition to diabetes, increasing age and the pres-
ence of chronic liver disease increased the rate of acute
liver failure. As with diabetes, older persons may have
both greater susceptibility to acute hepatic failure and
more exposures. The increased risk associated with a
history of chronic liver disease shows the potential con-
cern with making an accurate diagnosis of acute hepatic
failure. Whereas acute hepatic failure may be more likely
in patients with chronic liver disease, it is also possible
that the final stages of chronic end-stage liver disease can
be confused with acute hepatic failure when there is not
sufficient information about the preexisting chronic dis-
ease. This is a diagnostic challenge that should be ad-
dressed in future large-scale studies of acute hepatic
failure.
In summary, we found diabetes mellitus to be associ-
ated with an increase in the risk of acute liver failure.
Older age and the presence of chronic liver disease fur-
ther increase the risk of this highly fatal condition.
Periodic monitoring of liver enzymes and caution in the
use of potentially hepatotoxic drugs may be warranted in
patients with diabetes. Studies are needed to examine the
possible mechanisms of acute liver failure in diabetes.
16. Page WF, Mahan CM, Kang HK. Vital status ascertainment
through the files of the department of Veterans Affairs and the
social security administration. Ann Epidemiol 1996;6:102–109.
17. Deyo RA, Cherkin DC, Ciol MA. Adapting a clinical comorbidity
index for use of the ICD-9-CM in administrative databases. J Clin
Epidemiol 1992;45:613–619.
18. SAS Institute Inc. SAS/STAT user’s guide: basics. Version 6. SAS
Institute Inc., Cary, NC, 1990.
19. Kingston ME, Ali MA, Atiyeh M, Donnelly RJ. Diabetes mellitus in
chronic active hepatitis. Gastroenterology 1984;87:688–694.
20. Gentile S, Loguercio C, Marmo R, Carbone L, Del Vecchio Blanco
C. Incidence of altered glucose tolerance in cirrhosis. Diabetes
Res Clin Pract 1993;22:37–44.
21. Matz R. Diabetes with liver failure. Lancet 1972;1:851–852.
22. Schiodt FV, Atillasoy E, Shakil AO, Schiff ER, Caldwell C, Kowdley
KV, Stribling R, Crippin JS, Flamm S, Somberg KA, Rosen H,
McCashland TM, Hay JE, Lee WM. Etiology and outcome for 295
patients with acute liver failure in the United States. Liver Transpl
Surg 1999;5:29–34.
References
1. Reid AE. Nonalcoholic steatohepatitis. Gastroenterology 2001;
121:710–723.
23. Silverman JF, Pories WJ, Caro JF. Liver pathology in diabetes
mellitus and morbid obesity. Pathol Annu 1989;24:275–302.
24. Khuri KJ, Shammaa MH, Abourizk N. Hepatitis B virus markers in
diabetes mellitus. Diabetes Care 1985;8:250–253.
2. Dixon JB, Bhathal PS, O’Brien PE. Nonalcoholic fatty liver dis-

1828 EL–SERAG AND EVERHART
GASTROENTEROLOGY Vol. 122, No. 7
25. Pessayre D, Berson A, Fromenty B, Mansouri A. Mitochondria in
steatohepatitis. Semin Liver Dis 2001;21:57–69.
patients with type 2 diabetes. Diabetes Care 2000;23:754–
758.
26. Acosta D, Wenzel DG. Injury produced by free fatty acids to
lysosomes and mitochondria in cultured heart muscles and en-
dothelial cells. Atherosclerosis 1974;20:417–426.
27. Bogardus C, Lillioja S, Howard BV, Reaven G, Mott D. Rela-
tionships between insulin secretion, insulin action, and fast-
ing plasma glucose concentration in nondiabetic and noninsu-
lin-dependent diabetic subjects. J Clin Invest 1984;74:1238–
1246.
Received November 20, 2001. Accepted February 28, 2002.
Address requests for reprints to: HashemB. El-Serag, M.D., M.P.H.,
The Houston Veterans Affairs Medical Center (152), 2002 Holcombe
Blvd., Houston, Texas 77030. e-mail: hasheme@bcm.tmc.edu; fax:
(713) 748-7359.
Dr. El-Serag is a Veterans Affairs Health Services Research and
Development Awardee (RCD00-013-2).
28. Harris MI. Health care and health status and outcomes for