Many drugs have been implicated as causes of hepatitis, although nearly all the supportive data are from case reports or case series [1]. Case reports and case series suffer from the absence of a control group, making any causal relations difficult to determine [2, 3]. In addition, it is impossible to calculate from case reports the incidence rate of liver disease [4].

We describe the results of the first case–control study to examine the relation between acute hepatitis and commonly prescribed antibiotics. Our data suggest that three of the most commonly used antibiotics are associated with acute hepatitis.

Methods

Top Methods Results Discussion Author & Article Info References

Data Sources

Data for this study were derived from the Computerized On-Line Medicaid Pharmaceutical Analysis and Surveillance System (COMPASS). The COMPASS derives its data as a by-product of the Medicaid Management Information System, which is a computed claims processing and management information system for this large health care system. The Food and Drug Administration has funded the development of COMPASS as a potential data resource for conducting postmarketing drug surveillance studies. Software has been created to store, retrieve, and abstract that fraction of each patient's record that is useful for research. At the time of this study, billing data for each patient included age, sex, state, race (except in Florida), inpatient and outpatient diagnoses recorded by International Classification of Diseases 9th Revision Clinical Modification codes (ICD-9-CM) [5], and outpatient drugs dispensed. Use of over-the-counter drugs was recorded only if the drugs had been prescribed. This system has been described more fully elsewhere [6].

Selection of Cases and Controls

A case–control study was done using COMPASS data from Michigan and Florida. Potential cases were defined as patients 20 years or older who received an inpatient diagnosis consistent with acute liver disease between 1980 and 1987. To identify the acute liver disease ICD-9-CM codes most suitable for this study, we reviewed medical records from 10 patients in a local hospital with each of the ICD-9-CM codes consistent with acute liver disease. The results of this review suggested that the inpatient codes for acute necrosis and for unspecified hepatitis (ICD-9-CM codes 570 and 573.3, respectively) were used exclusively for patients with acute hepatitis without a known cause (such as infections). Therefore, these codes were used to identify potential cases. Potential cases were excluded if a patient had an alcohol-related diagnosis any time before the study diagnosis. An "alcohol-related diagnosis" was defined as the presence of an ICD-9-CM code specific for heavy alcohol use, such as alcoholic hepatitis, Laennec cirrhosis, alcoholic pancreatitis, alcoholic cardiomyopathy, delirium tremens, and so forth. Patients with a history of chronic liver disease were also excluded.

To assure that study patients were eligible for Medicaid benefits throughout the study period, a claim for medical service was required both before and after the study period. After application of the exclusion criteria and admissibility screen, 779 potential cases of acute liver disease remained. For each case, four controls who had passed the eligibility screen and who had no recorded history of liver disease were selected from the same state. Controls were also matched to the cases for sex and age (within 10-year age ranges).

Medical Record Review

To both validate the diagnosis of acute liver disease and to exclude cases with identifiable nondrug causes, we requested the medical records of all the potential cases. To acquire the medical records, the state Medicaid agency translated the anonymized COMPASS patient and provider identification numbers into Medicaid identification numbers. After receiving permission, a nurse visited the hospital with a portable copying machine and copied the relevant parts of the medical record. All identifying information was then blocked out, and the records were mailed to the investigators for review.

We received 408 of the 779 medical records requested. The medical record review process and results have been previously described in detail [7]. Briefly, these records were abstracted into a computed database by two physician reviewers. We excluded any case without medical records and any case meeting the following criteria: 1) no diagnosis of liver disease [n = 26]; 2) an admission date different from that requested [n = 1]; 3) a diagnosis of or evidence for chronic liver disease [n = 13]; 4) a diagnosis consistent with a nondrug origin (for example, acute hepatitis A or B or choledocholithiasis) [n = 112]; or 5) a previous exposure consistent with a nondrug cause, including a history of alcohol abuse (n = 39), intravenous drug use (n = 33), or a transfusion within 180 days of the diagnosis (n = 10). A total of 239 (59%) patients met these exclusion criteria.

To avoid ascertainment bias, we also excluded patients whose liver disease was mild (n = 25) or whose liver disease was diagnosed incidentally during the hospitalization (n = 44). Mild liver disease was defined as the absence of liver enzyme test values (aspartate aminotransferase [AST], alanine aminotransferase [ALT], bilirubin, or alkaline phosphatase) exceeding two times the control value. Liver disease diagnosed incidentally during hospitalization included patients hospitalized for reasons other than symptoms from liver disease (for example, surgery).

After completion of the medical record review, 107 cases of acute idiopathic symptomatic liver disease and their 428 matched controls remained for analysis.

Study Variables

Exposure was defined as the presence in the computed data of a pharmacy bill for dispensing an antibiotic within 30 days before the index date. The index date for cases was defined as the hospitalization date for acute liver disease. The index date for controls was identical to that of their respective matched cases. The probable indication for the antibiotic was determined by review of the computer profile of exposed cases and controls. The computer profile included all billing diagnoses near the time that the antibiotic was dispensed. The diagnoses considered to be the probable indication for the antibiotic included infections or other diseases (for example, acne) for which these drugs are commonly used.

We evaluated the common antibiotics that have been implicated by case reports in the literature as causes of hepatitis. These included the erythromycins, nitrofurantoin, isoniazid, rifampicin, the sulfonamides, and the tetracyclines. For purposes of comparison, we also evaluated other antibiotics that might be used for similar indications including ampicillin, amoxicillin, and cephalosporins.

In addition to the antibiotics of interest, information on other potential confounding variables was also obtained. Specifically, we collected information on 46 drugs and 36 conditions thought to be associated with liver disease based on a review of the literature. These included use of drugs such as -methyldopa, estrogens, nonsteroidal anti-inflammatory drugs, phenytoin, phenothiazines, and quinidine, and diseases such as Epstein-Barr infection, inflammatory arthritis, biliary tract disease, and congestive heart failure. Most of these conditions were exceedingly rare in our study samples, in part because the primary record review excluded patients with an identifiable nondrug cause of their liver disease, and, in part, because many of these conditions are very rare in the general population (for example, Q Fever and Wilson disease). The drug codes and ICD-9-CM codes used for each study variable and potential confounder are available from the authors.

Data Analysis

We first described the demographics of the patient population and then the case–control distribution of antibiotic exposure and potential confounding variables. Odds ratios with 95% confidence intervals (CIs) were calculated using the method of Woolf [8] for calculating the CIs. If any cell in the table had an expected value of less than five, then exact CIs were calculated. An unmatched analysis was done because of the broad nature of the matching variables (age, sex, and state).

We then used stratified analysis to evaluate the association between liver disease and antibiotic exposure, controlling for each potential confounding variable, and calculated summary statistics using the Mantel-Haenszel procedure [9]. Evidence for interaction between antibiotic exposure and each potential confounding variable was examined.

Multiple logistic regression was then used to evaluate the relation between each antibiotic and acute liver disease, controlling for the simultaneous effects of all confounding variables. The independent variables in the model included state, age, and sex, in addition to other interactions or potential confounders identified in the univariate analysis or thought to be clinically plausible.

Because the database did not contain information on drug use during hospitalization, we conducted a subanalysis that excluded all patients hospitalized in the 30 days before the index date. We also examined the frequency of hepatitis after re-exposure to the antibiotics among the cases.

The excess risk for liver disease from each of the drugs was calculated by 1) multiplying the adjusted odds ratio for each drug times the baseline incidence rate per 10 days of acute symptomatic liver disease resulting in hospitalization [assuming a 10-day course of antibiotics]; and 2) subtracting the baseline incidence rate of acute symptomatic liver disease resulting in hospitalization. The incidence rate of acute symptomatic liver disease resulting in hospitalization was 2.2 (95% CI, 2.0 to 2.4) per 100 000 per year [10].

Results

Top Methods Results Discussion Author & Article Info References

Description of Cases and Controls

The median age of the 535 cases and controls studied was 36 years (range, 20 to 101 years); 79.4% were women. These data are similar to those from the corresponding Michigan and Florida COMPASS population, in which 74.1% were women and the median age was 36 years. Forty-nine percent of cases were white, and 35% were nonwhite; race was not documented in the medical records of the remaining cases. The indications for hospitalization for the cases are shown in Table 1. The most frequent indications included vomiting (35.5%) and abdominal pain (32.7%). A liver biopsy was done in nine patients and showed acute hepatitis or nonspecific hepatitis in five patients (two with an additional finding of cholestatic hepatitis), fatty changes in two patients, and subacute inflammation and acute hepatic necrosis in one patient each.

The results of the liver function tests for the cases are shown in Table 2. The median AST was 5.43 µkat/L (325 mg/dL), and the total bilirubin level was 2.6 µmol/L. Of the cases with a prothrombin time and a control time recorded, 14 (27.5%) had a prothrombin time that was prolonged by more than 1 second.

Risk Associated with Erythromycins

Five cases (4.7%) and 4 controls (0.9%) were exposed to erythromycin, yielding an odds ratio of 5.2 (95% CI, 1.1 to 26.6) (Table 3). Of the five cases exposed to erythromycin, their attending physician had made a diagnosis of erythromycin-related hepatitis in two patients and a diagnosis of possible drug-induced hepatitis in one. The remaining two patients were diagnosed as having "hepatitis of unknown etiology." The probable indications for erythromycin were upper respiratory infection or bronchitis (three cases and three controls), pneumonia (one control), prophylaxis for surgery (one case), and unknown (one case). When we examined the risk associated with the different types of erythromycins prescribed, the risk was elevated with erythromycin ethylsuccinate (P = 0.008). Notably, no case or control was exposed to erythromycin estolate, and exposure to the other erythromycin preparations was infrequent.

The relation between acute liver disease and erythromycin exposure was stratified by age, sex, state, and other drug exposures and is shown in Table 4. The Mantel-Haenszel adjusted odds ratios for these characteristics all closely resembled the unadjusted odds ratio. After we simultaneously controlled for the effects of age, state, sex, and other drug exposure using multiple logistic regression, the adjusted odds ratio for erythromycin exposure remained essentially unchanged: 4.8 (CI, 1.2 to 19.0). The increased risk per million patients exposed to a 10-day course of erythromycin was 2.28 cases; that is, for each million patients treated with erythromycin for 10 days, an additional 2.5 cases developed acute symptomatic liver disease resulting in hospitalization.

We repeated the above analysis after excluding the 11 cases (10.3%) and 10 controls (2.3%) with a hospitalization in the 30 days before the index date. The results were similar to those of the overall analysis: The logistic regression analysis yielded an adjusted odds ratio of 4.1 (CI, 0.96 to 17.7).

Risk Associated with Sulfonamides

Eight cases (7.5%) and three controls (0.7%) were exposed to oral sulfonamides, yielding an odds ratio of 11.4 (CI, 2.7 to 67.8). All except one control had been exposed to trimethoprim-sulfamethoxazole. Of the eight cases exposed to oral sulfonamides, two were diagnosed by the attending physician as having sulfonamide-related hepatitis; the remaining six patients were diagnosed as having "hepatitis of unknown etiology." The probable indications for sulfonamides were urinary tract infection (six cases, two controls), bronchitis (one case), and otitis media (one control).

The relation between acute liver disease and sulfonamide exposure was stratified by age, sex, state, and other drug exposures and is shown in Table 5. The Mantel-Haenszel adjusted ratios for these characteristics were similar to the unadjusted odds ratio; however, the risk differed between women and men (inter-action, P = 0.01). A significantly elevated risk was found in women, but only sparse data were available for men.

Because of the paucity of men exposed to sulfonamides, we were unable to do a multiple logistic regression analysis that included the sex-sulfonamides interaction term. We therefore created two alternative models. In the first model, we ignored the suggestion of the interaction and controlled for age, sex, and other drugs as independent variables. We thought this approach to be valid because the paucity of men exposed to sulfonamides made interpretation of the interaction P value difficult. The adjusted odds ratio from this model was 9.0 (CI, 1.9 to 35.6). In the second model, we did a multiple logistic regression analysis of women only, controlling for age, state, and other drug exposure. The adjusted odds ratio for sulfonamide exposure in women was 29.3 (CI, 3.5 to 246.3), a value that is similar to the unadjusted odds ratio in women of 35.2 (CI, 4.6 to 1567.2).

One control was exposed to triple sulfa cream, which was not considered to be an exposure in the preceding analyses. Because triple sulfa cream has been reported to cause hepatitis in one case report [11], we repeated the analysis after redefining that one control as having been exposed to sulfonamides. The odds ratio for the association between sulfonamides and acute hepatitis decreased to 8.6 (CI, 2.2 to 39.4). The results did not change appreciably after controlling for potential confounding variables.

Finally, we repeated the above analyses after excluding all cases and controls with a hospitalization in the 30 days before the index date; the results of these analyses were essentially the same as those for the entire data set.

The excess risk per million for the development of acute symptomatic liver disease resulting in hospitalization after exposure to a 10-day course of sulfonamides was 4.80 cases (assuming an adjusted odds ratio of 9.0).

Risk Associated with Tetracyclines

Five cases (4.7%) and four controls (0.9%) were exposed to tetracyclines, yielding an odds ratio of 5.2 (CI, 1.4 to 19.7). Of the five cases exposed to tetracycline, three were suspected by the attending physician of having a drug-related hepatitis; the remaining two patients were diagnosed as having "hepatitis of unknown etiology." The probable indications for tetracycline were bronchitis (two cases, one control), pelvic inflammatory disease (one case, one control), acne (one case, one control), pneumonia (one control), and unknown (one control). Of the nine cases and controls exposed to tetracycline, six were dispensed tetracycline hydrochloride, and two were dispensed doxycycline (Vibramycin, Pfizer Laboratories Division, New York, New York), and one was dispensed minocycline (Minocyn, Lederle Laboratories, Wayne, New Jersey).

The relation between acute liver disease and tetracycline exposure was stratified by age, sex, state, and other drug exposures and is shown in Table 6. The Mantel-Haenszel adjusted ratios for these characteristics were similar to the unadjusted odds ratio with the exception of other drug exposure; the odds ratio decreased to 3.8 (CI, 1.0 to 14.4). Simultaneously controlling for the effects of age, state, sex, and other drug exposure using multiple logistic regression, the adjusted odds ratio for tetracycline was 3.6 (CI, 0.9 to 14.3). The excess risk per million for the development of acute symptomatic liver disease resulting in hospitalization after exposure to a 10-day course of tetracycline was 1.56 cases.

Risk Associated with Other Antibiotics

The case–control distribution of exposure to the other antibiotics is shown in Table 7. Although associations were present for isoniazid (P < 0.008) and rifampicin (P < 0.04), the numbers of cases and controls exposed to these antibiotics were small, and adjustment for potential confounding variables was not possible. No association was seen between liver disease and either ampicillin or the cephalosporins.

Discussion

Top Methods Results Discussion Author & Article Info References

This study also provides clinically important new data regarding excess risk and the number of cases of liver disease resulting from these drugs. This quantification of the risk related to these drugs is critical to placing previous case reports into proper clinical perspective. Although we cannot directly measure the incidence rate of liver disease from these drugs using a case–control study, the increased risk can be estimated. The increased risk per million patients exposed to a 10-day course of erythromycin was 2.28 cases; the figure for sulfonamides was 4.80 cases; and the figure for tetracycline was 1.56 cases. These antibiotics are three of the most commonly prescribed drugs in the United States. In 1991, the number of erythromycin prescriptions was 28.9 million; the number of sulfonamide prescriptions was 23.7 million; and the number of tetracycline prescriptions was 29.8 million (National Prescription Audit. Plymouth Meeting, Pennsylvania: IMS America, 1991 [Used with permission]). Thus, we estimate that annually in the United States 66 cases of acute symptomatic liver disease resulting in hospitalization are due to erythromycin; 114 cases are due to sulfonamides; and 31 cases are due to tetracycline.

In the 1970s and 1980s, regulatory agencies worldwide were concerned about the liver toxicity of erythromycins. Spontaneous case reports (reports by physicians or pharmacists in the medical literature or to regulatory agencies such as the Food and Drug Administration) suggested that the risk might be greatest with erythromycin estolate. In a review of the United Kingdom experience to 1973, the Committee on Safety of Medicines found that of 48 reports of jaundice in patients taking erythromycins, 47 were in patients taking erythromycin estolate [12]. Yet, this preparation accounted for less than 50% of the sales of erythromycin. Spontaneous case reports from the United States between 1969 and 1978 found that 93% of the 418 cases of liver disease reported were in patients taking erythromycin estolate [13]. In response to these and other data [14-22], Australia, Sweden, Denmark, and The Netherlands removed the estolate form of erythromycin from the market [23]. Subsequent data from the Prescription-Event Monitoring system showed that the rate of hepatitis from erythromycin estolate was less than 1 per thousand and was similar to those of erythromycin stearate and tetracycline [24].

Our data cannot address whether the estolate preparation is more hepatotoxic than the other erythromycin preparations because none of our cases or controls was exposed to erythromycin estolate. However, this study does document that the risk associated with erythromycin is not limited to the estolate salt. We found that the risk was elevated for erythromycin ethylsuccinate and suggestive for erythromycin stearate. This finding is consistent with more recent case reports of liver disease in patients taking the ethylsuccinate preparation [25, 26]. Interestingly, two cases have been reported of patients who developed hepatitis from the estolate preparation who were later exposed to the ethylsuccinate preparation and who developed hepatitis again [27].

The erythromycin-exposed cases in our study did not exhibit a consistent clinical or liver enzyme pattern; however, other reports suggest a hypersensitivity mechanism, with patients having abdominal pain, eosinophilia, fever, and rash [1]. Liver enzymes often show a cholestatic picture with more prominent elevations of alkaline phosphatase than transaminases. Jaundice usually appears within 3 weeks and resolves with discontinuation of the drug. Chronic hepatitis has not been reported with erythromycin.

Our findings are also consistent with the large number of reports describing sulfonamide-induced hepatitis, including the recurrence of hepatitis after rechallenge [28-32]. These reports, however, were primarily in patients exposed to sulfonamides alone, whereas in our study, patients were almost exclusively exposed to trimethoprim-sulfamethoxazole. More recent reports also describe hepatitis in patients exposed to trimethoprim-sulfamethoxazole [33-38]. Although it is assumed that sulfamethoxazole is responsible for the hepatitis in these patients, one case report has described a patient who developed hepatitis from trimethoprim-sulfamethoxazole and who later developed recurrent hepatitis when rechallenged with trimethoprim alone [39].

Hepatitis from sulfonamides is often associated with fever, rash, and eosinophilia, suggesting hypersensitivity [40-43]. Cross-sensitivity among preparations has been documented [44]. In most patients, the illness resolves after withdrawal of the drug, although death [34, 36] and chronic liver disease have been reported [45, 46].

Early reports of hepatitis from tetracycline primarily included pregnant women treated with high intravenous doses [47, 48]. These patients had symptoms consistent with acute hepatitis, which was often fatal and complicated by azotemia and pancreatitis. Hepatitis most commonly occurred in patients receiving high intravenous doses but also occurred with usual doses in patients with preexisting renal dysfunction. Men and nonpregnant women were also described as having this disease [49]. The pathologic examination of the liver revealed microvesicular fat accumulation in the hepatocytes. The pathogenesis of the disease is indirect hepatoxicity resulting from inhibition of mitochondrial oxidation of fatty acids in the liver [50, 51].

More recent reports have described liver disease resulting from minocycline [52-54] taken orally as well as intravenously [55]. In the three patients taking the drug orally, hepatitis developed approximately 1 month after initiation of the drug and was fatal in one patient. Of the patients exposed to tetracycline in our study, all were taking the drug orally, one was using minocycline, and none died during that hospitalization.

Our ability to examine other antibiotics was limited by our sample size and the prevalence of use of these drugs. Although our study showed significant associations with isoniazid and rifampicin, the data were sparse and did not permit control for other potential confounding variables. These results, however, are consistent with the existing literature [1]. In contrast, we were unable to document an association between acute hepatitis and nitrofurantoin; however, this drug has been more widely reported to cause chronic hepatitis [56].

We examined several other antibiotics that might be used instead of erythromycin, sulfonamides, or tetracycline. No association was seen between hepatitis and ampicillin or the cephalosporins, although the numbers of cases and controls exposed to these drugs were small. None of the patients in our study had exposure to amoxicillin, and the study was done using data collected before the marketing of quinolones in the United States.

Several issues threaten the validity of pharmacoepidemiologic studies using secondary data sources, such as Medicaid billing data [57]. First is the validity of the case definition. Diagnoses are coded using the ICD-9-CM system; however, the provider has no incentive to use the most specific diagnosis code, and the codes themselves may be nonspecific. The coding may also be inaccurate. All studies must confirm the diagnosis by reviewing the primary medical records. Second, even when the medical records are reviewed, they may not contain all the information relevant to an accurate definition of the cases. For example, the definitive diagnostic test may not have been done for each patient. Third, review of the medical records is also necessary to obtain information on potential confounders, such as alcohol use, that are not available in the computer data. It is often not possible, however, to obtain the medical records of controls because most are outpatients. Thus, an assessment of the importance of these missing data must be made. Fourth, although drug exposure information has been documented to be accurate [58], it is impossible to document compliance. An assessment of the likelihood of compliance must be made.

Several of these limitations must be considered in our study. First, we were unable to obtain medical records for the controls and therefore could not exclude controls with alcoholism or illicit drug use. It is likely, however, that only a small proportion of controls had these problems. Second, we were unable to obtain data directly from any patients on alcohol use or over-the-counter drugs such as acetaminophen or aspirin, which are not reliably included in the database. Third, our study might have included cases with undiagnosed viral hepatitis, given that serologic tests for hepatitis A and B were not done in 16.8% of cases, and a serologic test for hepatitis C was unavailable at the time of the study. However, this would have the effect of reducing the strength of any real association. Thus, we may have underestimated the risk for liver disease associated with these antibiotics. Fourth, we were able to obtain only 59% of the medical records requested. The primary reasons for nonretrieval were administrative [7], however, and unlikely to be related to drug exposure. Fifth, the patients in this study were indigent, and the results may not be generalizable to the overall population, although we have no reason to expect that antibiotics would cause acute symptomatic liver disease only in indigent patients. Finally, it is possible that cases of idiopathic liver disease were assigned a different ICD code than the codes included in our case definition. Thus, we cannot exclude the possibility of some selection bias that resulted from not studying all the cases in our sample.

Our study took advantage of several unique features of Medicaid billing data. We did a population-based, case–control study with a randomly chosen control group drawn from approximately 1.4 million patients from two large states. Our drug exposure data were based on pharmacy billing records, which have been shown to be highly accurate [58] and which are not subject to recall or interview bias. We were able to obtain the medical records for our cases to validate the diagnosis of acute hepatitis and to exclude other common causes of liver disease such as alcohol abuse and viral infections [7].

We restricted our case definition to acute liver disease resulting in hospitalization because such disease represents clinically more important illness and to avoid ascertainment bias. In general, the patients in this study had moderately severe hepatitis, and 14 of the 107 cases had prolonged prothrombin times. Many cases of drug-induced liver disease may be asymptomatic and only be associated with minor liver enzyme abnormalities. This study does not address the association between antibiotics and asymptomatic liver enzyme elevations or mild liver disease treated on an outpatient basis.

The recognition of erythromycin-, sulfonamide-, and tetracycline-induced acute hepatitis is important because the discontinuation of the drug usually results in resolution of the illness. Furthermore, without recognition of the offending agent, repeated exposure is likely, given that these drugs have broad indications for use. Careful review of the drug history for use of these medications is mandatory in patients who have acute hepatitis.

The views expressed in this report are those of the authors and do not necessarily represent the official position of the Food and Drug Administration.