Outcome and prognostic markers in severe drug-induced liver disease


  • Einar Björnsson,

    Corresponding author
    1. Section of Gastroenterology and Hepatology, Department of Internal Medicine, Sahlgrenska University Hospital, Gothenburg, Sweden
    • Department of Internal Medicine, Sahlgrenska University Hospital, SE-413 45 Gothenburg, Sweden
    Search for more papers by this author
    • fax: (46) 31822152

  • Rolf Olsson

    1. Section of Gastroenterology and Hepatology, Department of Internal Medicine, Sahlgrenska University Hospital, Gothenburg, Sweden
    Search for more papers by this author

  • Potential conflict of interest: Nothing to report.


The combination of high aminotransferases (hepatocellular injury) and jaundice has been reported to lead to a mortality rate of 10% to 50% for different drugs, a phenomenon known as “Hy's rule.” However, Hy's rule has never been validated, and limited data exist on predictors for outcome in hepatocellular and other forms of drug-induced liver disease. All reports of suspected hepatic adverse drug reactions received by the Swedish Adverse Drug Reactions Advisory Committee (1970-2004) were reviewed. Cases with bilirubin levels 2 or more times the upper limit of normal (ULN) were analyzed. A total of 784 cases were retrieved—409 with hepatocellular injury, 206 with cholestatic injury, and 169 with mixed liver injury. The mortality/transplantation rate was 9.2%, and bilirubin (median 18.7 × ULN [IQR 12.6-25]; range 4.5-42) was higher (P < .0001) in the deceased/transplant recipients compared with the surviving patients (median 5.5 × ULN [IQR 3.3-9.5]; range 2.0-38). A total of 7.8% with cholestatic and 2.4% with a mixed pattern died. The mortality rate in hepatocellular injury for different drugs varied from 40% (6 of 15) for halothane to 0% (0 of 32) for erythromycin, in total 12.7%. Using logistic regression analysis, age, aspartate aminotransferase (AST) and bilirubin were found to independently predict death or liver transplantation in the hepatocellular group, whereas among patients with cholestatic/mixed liver injury, bilirubin was the only independent predictor. In conclusion, hepatocellular jaundice has a high but variable mortality rate, depending on the drug involved. The AST and bilirubin levels are the most important predictors of death or liver transplantation. (HEPATOLOGY 2005;42:481–489.)

Drug-induced liver disease (DILD) is a potential complication with many drugs. Acute hepatic injury due to drugs has been reported to occur in 5% to 10% of patients hospitalized for jaundice.1–4 Furthermore, drugs are the most common cause of fulminant hepatic failure, both in the United States and Europe.5–7 In reports from the United States and Sweden, idiosyncratic drug reactions were the presumptive causes in 13% to 17% of cases of acute liver failure.5, 6 The prognosis for patients with acute liver failure due to idiosyncratic drug reactions is usually poor, with 60% to 80% mortality without liver transplantation.8, 9 During the last decade, drug-induced liver injury has led to the withdrawal of a number of drugs from the market.10–14

The combination of high serum aminotransferases (hepatocellular injury) and jaundice has in earlier studies been reported to result in a mortality of 10% to 50% for different drugs.11–13 These observations have been named “Hy's rule” after Hyman Zimmerman, who first described them. The rule states that if both drug-induced hepatocellular (HC) injury and jaundice occur at the same time without biliary obstruction, mortality of at least 10% can be expected.11, 13, 14 Hy's rule defined as DILD with serum alanine aminotransferase (ALT) levels 3 or more times the upper limit of normal (ULN) + serum bilirubin levels 2 or more times the ULN has been advocated by the U.S. Food and Drug Administration for use in the assessment of the hepatotoxicity of newly developed drugs.13, 15 However, this rule has never been scientifically validated. The sensitivity and specificity of clinical jaundice for the outcome in patients with drug-induced HC injury is thus unknown. The most important predictors of outcome in drug-induced liver disease with HC injury have not been analyzed in a large number of patients; furthermore, information about the prognosis in other forms of DILD (e.g., DILD with cholestatic or mixed patterns) is limited.

In Sweden, a systematic monitoring system of DILD has been in use since 1966, with regular causality assessment offering the opportunity to evaluate a large number of patients with DILD. The aim of this study was to analyze the outcome of patients with severe DILD associated with jaundice with HC, cholestatic (CS), or mixed liver injury to evaluate the validity of Hy's rule and to analyze the most important predictors for outcome.


ULN, upper limit of normal; AST, aspartate aminotransferase; DILD, drug-induced liver disease; HC, hepatocellular; ALT, alanine aminotransferase; CS, cholestatic; SADRAC, Swedish Adverse Drug Reaction Advisory Committee; RUCAM, Roussel Uclaf causality assessment method; ALP, alkaline phosphatase.

Materials and Methods

All reports of suspected drug-induced liver injury received by the Swedish Adverse Drug Reactions Advisory Committee (SADRAC) between 1970 and 2004 have been computerized and are available for legally acceptable users with a password online. Since 1975, the reporting of fatal, otherwise serious, and new reactions is compulsory. The quality of the reports may vary, because there is no standardized report form. Full medical records, including results of laboratory tests, imaging studies, biopsies, and autopsies, are requested for all fatal cases and for the majority of serious cases. The opinion of an expert hepatologist with extensive experience in DILD is requested in difficult assessments. This assessment is based on a clinical judgement, not on any published method for assessment of causality.

We retrieved those reports in which a possible or probable relationship had been assessed by SADRAC to exist. Our analysis was restricted to patients with serum bilirubin levels more than twice the ULN. Furthermore, in patients with a HC type of injury, our analysis was restricted to those patients with ALT levels 3 or more times the ULN as well as serum bilirubin levels 2 or more times the ULN.

The altogether 838 reports fulfilling these criteria were evaluated using international consensus criteria (Roussel Uclaf causality assessment method [RUCAM])16, 17 to assess the probability of a causal relationship between drug exposure and liver disease (Fig. 1). The causality assessment is based on information about the time of onset of the reaction from the beginning of the drug, the development of liver tests after cessation of the drug, the presence of risk factors, and known hepatotoxicity of the suspected drug and concomitant drug or drugs.16 Furthermore, the investigations performed to exclude nondrug causes for the reaction are looked for. Thus, abnormal liver tests developing shortly after beginning of a new drug, rapid decline of these after stopping the drug, and exclusion of other causes give high scores compatible with the drug as a possible, probable, or highly probable cause of the reaction.16 If the report does not receive a high enough score to consider a causal relationship with the suspected drug the reaction is according to the criteria unlikely or the relationship was excluded.16 Each author scored approximately half of the cases. We performed assessment of 50 cases independently and found very low intraobserver variability with no disagreement in the assessment of cases into unlikely or not.

Figure 1.

Diagram of the patients included in the study. SADRAC, Swedish Adverse Drug Reaction Advisory Committee; ULN, upper limit of normal.

In case more than one drug could possibly have been responsible for the liver disease, we gave no minus scores for “concomitant drug(s),” because the main purpose of the present investigation was to study prognostic predictors in DILD in general, not outcomes with specific drugs. On the other hand, the reports only rarely stated whether or not the patient was an ethanol user (giving plus scores), and if so, it was only to report that the patient abused alcohol. Data on “search for nondrug causes” (e.g., an anti–hepatitis C virus test) were often missing, particularly in the early cases, causing a relatively low likelihood of a relationship in many cases.

Because many patients had been exposed to several drugs at the time of the appearance of the liver injury, it is not always possible to deduce which drug is most likely responsible. In such cases, we judged the reaction to have been potentially caused by more than one drug. On the other hand, if there was a close temporal relationship between the liver injury and treatment with only one of many drugs with which the patient was treated, this drug has been considered to be the only suspected one.

Intoxications with acetaminophen were excluded from this analysis, because in contrast to other cases of DILD, they represent a direct, dose-dependent, and predictable type of liver injury for which predictors of outcome have previously been studied extensively. Such cases are usually reported to the Swedish Poisons Information Centre and only exceptionally to SADRAC.

The computerized reports include all relevant facts from medical records and the results of laboratory investigations. The following information was retrieved from the reports: year of exposure; drug(s) suspected to be responsible; age and sex of the patient; duration of treatment; type of liver injury; results of AST, ALT, alkaline phosphatase (ALP), and bilirubin tests; search for nondrug causes; and outcome of the patient (recovery, death or liver transplantation). The outcome of the SADRAC database was also compared with a central registry of patients undergoing liver transplantation in Sweden. Eleven of 13 patients who underwent transplantation for idiosyncratic drug reactions had been reported to SADRAC, and the rest of the patients were also included in our analysis.

Classification of the liver injury was also based on International Consensus Criteria.16–17 In the current study, the maximum (peak) values for serum bilirubin, AST, ALT, and ALP were used for analysis.


For descriptive purposes, sensitivity, specificity, positive predictive values, and negative predictive values were calculated. The Fisher exact test was used to test differences between groups regarding dichotomous variables; the Mann-Whitney test was used for continuous variables. Stepwise logistic regression was performed for multivariate purposes to predict death. All tests were two-tailed and were conducted at a 5% significance level.


During the period 1970-2004, SADRAC received a total of 4,396 reports of suspected drug-induced liver injury. A total of 3,841 (87.4%) were considered to have a possible or probable causal relationship with drug exposure. Of these 3,841 cases, a total of 873 cases (22.9%) with bilirubin levels 2 or more times the ULN and ALT levels 3 or more times the ULN were retrieved. Among these 873 cases, 35 cases had to be excluded because of a lack of important clinical or laboratory data, leaving 838 cases for analysis (Fig. 1).

As a result of the RUCAM, another 42 cases had to be excluded because they did not fulfill the RUCAM criteria for at least a possible relationship, and 12 reports on acetaminophen-associated liver injury were also excluded (see Materials and Methods). Thus, a total of 784 cases were included in the final analysis. According to the RUCAM, 360 cases (46%) had a possible relationship, 331 (42%) had a probable relationship, and 93 (12%) had a highly probable relationship. The incidence of DILD and the different types of drugs reported during the different time periods is demonstrated in Table 1. The low numbers during the early time period may be explained by a lower incidence of reports in general during the earlier years as well as by a poorer quality of the reports, precluding a possible or probable assessment. The 784 cases with a possible/probable/highly probable causal relationship to drug(s) include a total of 409 cases with HC injury, 206 with CS injury, and 169 with mixed liver injury (Table 2). In a total of 633 cases, one drug was suspected of being responsible for the liver injury, whereas in 151 cases more than one drug could potentially have been responsible. Table 2 shows age and sex, duration of treatment, and peak liver test values in patients with different types of DILD.

Table 1. Incidence of Drug-Induced Liver Disease in Reports With Bilirubin ≥2 × ULN With a Possible/Probable/Highly Probable Relationship in Time Periods (1970–2004)
  1. NOTE. HC, CS, and mixed injury were considered. The major types of drugs associated with DILD and concomitant jaundice during the different time periods are listed. The number of cases and the percentage (in parentheses) are given for each type of drug.

  2. Abbreviation: NSAIDs, nonsteroidal anti-inflammatory drugs.

Deaths/TX996 (4)21 (9.5)17 (8.1)10 (5.5)
Antibiotics1 (10)2 (18)47 (31)77 (35)59 (28)57 (31)
Anesthetics6 (60)1 (9)7 (4.7)4 (1.8)00
NSAIDs1 (10)09 (6)15 (6.8)11 (5.3)16 (8.8)
Analgetics02 (18)7 (4.7)5 (2.3)4 (1.9)2 (1.1)
Others02 (18)51 (34)79 (35.6)77 (36.8)64 (35.2)
Multiple03 (27)20 (13.3)33 (14.9)55 (26.3)40 (22)
Antiepileptic1 (10)1 (9)4 (2.7)7 (3.2)3 (1.4)1 (0.5)
Tuberculostatics1 (10)05 (3.3)2 (0.9)02 (1.1)
Table 2. Clinical and Laboratory Data in the Total Study Population as Well as in the Subgroups According to Type of Liver Injury
  1. NOTE. The liver laboratory values are expressed as multiples of the ULN (median, with IQR in parentheses). The laboratory parameters are all peak values. The number of patients with available data of bilirubin, AST, ALT, and ALP are provided in percentages.

  2. Abbreviations: F, female; M, male.

Age58 (42–74)55 (39–69)69 (49–80)59 (43–72)
Sex (F/M)452/332237/172121/8796/73
Duration of treatment (d)21 (10–55)28 (10–66)21 (10–35)18 (10–35)
Bilirubin (100%)6.2 (3.5–10.7)6.0 (3.3–10.6)6.6 (4.0–12.4)5.7 (3.3–8.8)
AST (94%)7.2 (3.5–19.5)17.1 (7.2–34.3)3.4 (2.0–5.7)4.8 (2.9–8.0)
ALT (99%)12.5 (5.9–28.1)25.6 (14.3–44.3)3.9 (2.6–6.6)8.7 (6.5–13.4)
ALP (99%)2.1 (1.4–3.4)1.4 (0.9–2.1)3.8 (2.6–6.0)2.6 (1.9–3.7)

The patients with CS injury were significantly older than those with a mixed or HC pattern (P = .003 and P < .0001, respectively) (Table 2). A higher proportion of females was observed in all the different subgroups, with approximately 58% females in all groups (Table 2). Patients with CS injury had higher ULN bilirubin levels compared with those with a mixed type (P = .01) and HC injury (P value not significant) (Table 2). For obvious reasons, the AST, ALT, and ALP values differed between those with HC injury compared with those with CS or a mixed pattern, because these differences constituted the basis for the classification into the different types of injury.

A total of 72 (9.2%) of the 784 patients died from liver failure or underwent liver transplantation (Table 3). The following drugs were associated with death/liver transplantation: flucloxacillin (n = 7), halothane (n = 6), diclofenac (n = 4), naproxen (n = 4), isoniazide (n = 3), disulfiram (n = 3), chlorpromazine (n = 2), ciprofloxacin (n = 2), enalapril (n = 2), trimethoprim/sulfametoxazol (n = 2), or other drugs (n = 1 each) (ranitidin, rofecoxib, sulfasalazine, carbamazepine, ticlopidine, simvastatin, acetylsalicylic acid, atorvastatin, bleomycin, ibuprufen, danazol, dicloxacillin, dikumarol, dextropropoxiphen, donepezil, estradiol, pyrimethamin/sulfadoxine, phenytoin, fluorouracil, interferon, chlormezanon, nefazodon, omeprazol). In 14 cases multiple drugs were involved, any of which could therefore have caused the reaction.

Table 3. Outcome of All Patients as Well as Those With Hepatocellular, Cholestatic, or Mixed Liver Injury
  1. NOTE. The number of transplant recipients (13) is included within the total of 72 patients from the total study group and of those 50 within the hepatocellular group.

Died from liver failure/liver transplantation72 (9.2%)52 (12.7%)16 (7.8%)4 (2.4%)
Liver transplantation131300

Another 5 patients hospitalized for drug-induced jaundice died of other causes, namely cardiac death (2 cases) or multiorgan failure (3 cases). Patients with HC injury had the highest mortality (12.7%) compared with 7.8% of patients with CS drug-induced jaundice (P value not significant) and 2.4% of those with mixed liver injury pattern (P < .0001 compared with HC injury). Patients with HC injury had a significantly higher mortality than those with CS and mixed injury combined (12.7% vs. 5.3%; P = .0005). A total of 13 patients, all with HC injury, underwent transplantation. The reason no transplantations were performed in the CS/mixed injury group was either that most of the patients were more than 70 years of age and considered too old or had contraindication for transplantation or that these cases occurred before liver transplantation was available in Sweden (1986).

Comparison Between Nonsurvivors and Survivors

The patients who died (or underwent transplantation) were older than those who recovered (Table 4). The proportion of females and males was similar in those who recovered and in those who did not and no difference was observed in the duration of treatment in the two groups (Table 4). Deceased patients or patients who underwent transplantation had higher serum bilirubin, AST, and ALT values and AST/ALT ratio than surviving patients, whereas ALP levels were similar in the two groups (Table 4). A comparison between the survivors and nonsurvivors within the HC group revealed no differences regarding sex, duration of treatment, or ALP (Table 5). Patients with HC injury who died or underwent transplantation were older than those who recovered and had higher AST and ALT levels, AST/ALT ratio, and bilirubin levels (Table 5). In the CS/mixed group, bilirubin levels and AST/ALT ratio were significantly higher in deceased patients and transplant recipients (Table 5).

Table 4. Comparison Between Patients Who Died or Underwent Transplantation and Patients Who Survived Drug-Induced Liver Injury
 Died/TransplantationRecoveredP Value
  1. NOTE. The laboratory parameters are all peak values. Results are expressed as medians, with IQRs in parentheses.

  2. Abbreviations: F, female; M, male; NS, not significant.

Age65 (47–77)58 (41–74).04
Sex (F/M)43/29409/303NS
Duration of treatment25 (10–94)21 (10–49)NS
Bilirubin18.7 (12.6–25.3)5.5 (3.3–9.5)<.0001
AST34 (14.0–59)6.7 (3.4–17.1)<.0001
ALT31 (15.7–56)11.4 (5.7–24)<.0001
ALP1.9 (1.1–3.3)2.1 (1.4–3.4)NS
AST/ALT ratio1.1 (0.8–1.4)0.6 (0.4–0.9)<.0001
Table 5. Comparison Between Patients With Hepatocellular Injury and Patients With Cholestatic/Mixed Liver Injury Among Those Who Died or Underwent Transplantation and Those Who Survived Drug-Induced Liver Injury
 Hepatocellular InjuryCholestatic/Mixed Injury
  • NOTE. The results are expressed as medians, IQRs in parentheses. The laboratory parameters are all peak values.

  • *

    P < .001.

  • P < .0001.

  • P < .05.

Age64 (48–73)53 (38–68)*75 (43–83)63 (47–77)
Sex, F/M (%)34/18 (65/35)204/153 (57/43)10/10 (50/50)205/150 (57/43)
Duration23 (6–90)28 (10–60)28 (15–112)21 (10–35)
Bilirubin17.3 (12.6–24.6)5.2 (3.1–9.2)24.3 (12.7–31.8)6.0 (3.6–9.8)
AST49.3 (30–69)13.9 (6–27)3.3 (3.0–6.9)4.0 (2.3–6.6)
ALT44.3 (30–61)22.8 (13–39.7)4.6 (2.6–11.3)6.3 (3.7–10)
ALP1.5 (0.9–2.4)1.4 (0.9–2.0)4.3 (2.1–9.0)3.1 (2.2–4.8)
AST/ALT ratio1.1 (0.9–1.4)0.6 (0.4–0.9)0.8 (0.6–1.4)0.7 (0.5–1.0)

Using a forward stepwise analysis, AST (P < .0001, odds ratio [OR] 1.013 for each increase × ULN) and bilirubin (P < .0001, OR 1.092) were found to independently predict death or transplantation in the total study population. Within the HC group, age (P = .0063, OR 1.032 for each year increase), AST (P < .0001, OR 1.012), and bilirubin (P < .0001, OR 1.202) were independent predictors of death or transplantation. Among patients with CS or mixed injury, bilirubin (P < .0001, OR 1.194) was an independent predictor of death.

Because in the univariate analysis ALT and AST/ALT ratio were also shown to be significantly related to death or transplantation, and because there was a strong correlation between AST and ALT, we calculated the OR for combinations of not only AST and bilirubin, but also for ALT and bilirubin (i.e., what is included in Hy's rule) and AST/ALT and bilirubin in HC injuries, as well as the OR for bilirubin in CS injuries. Based on these logistic regression models, we found ALT levels 3 times the ULN plus bilirubin levels 2 times the ULN (which is in agreement with Hy's rule) in HC injury to have a positive predictive value of 0.13, a negative predictive value of 1.00, a sensitivity of 1.00, and a specificity of 0.04 for prediction of death or transplantation. Table 6 describes the logistic regression models considered above, as well as predictive values, sensitivities, and specificities for other combinations of risk factors.

Table 6. Results of Logistic Regression
ModelOR95% CIP ValueLimits*PPVNPVSensitivitySpecificity
  • NOTE. ORs are given for each increase of ULN with one unit (AST, ALT, and bilirubin) or for each increase of the AST/ALT ratio with >1, respectively, for positive and negative predictive values, sensitivity, and specificity, for each model when predicting DILD.

  • Abbreviations: PPV, positive predictive value; NPV, negative predictive value.

  • *

    Cutoff limits when calculating PPV, NPV, sensitivity, and specificity.

HC Injury        
  ALT1.0141.005–1.023.00223× ULN    
  Bilirubin1.2001.146–1.256<.00012× ULN0.
  AST1.0111.005–1.017.00013× ULN    
  Bilirubin1.1981.143–1.256<.00012× ULN0.
  AST/ALT >11.7561.216–2.537.0027>1    
  Bilirubin1.1981.144–1.255<.00012× ULN0.330.950.690.81
CS injury or mixed type        
  Bilirubin1.1931.132–1.258<.00012× ULN0.250.940.040.99
  Bilirubin1.1931.132–1.258<.00015× ULN0.500.940.041.00

The Role of Specific Drugs

Hepatocellular Injury.

The highest number of reports of drug-induced fatal jaundice and HC injury with a probable or possible causal relationship was related to halothane (Table 7). Unfortunately, a total of 9 out of the 25 reported cases, 7 of whom died, had to be excluded because of missing laboratory data (mostly cases during the earliest part of the time period), and one did not fulfill the RUCAM criteria for a possible relationship. Consequently, only 15 cases with complete laboratory and clinical data remained for full analysis of data with a least possible relationship, and 6 of these died (40%). The second most commonly reported drug associated with mortality was disulfiram (31 reports total). Four reports of disulfiram-associated hepatoxicity had to be excluded (1 of whom died) because of a lack of laboratory data. Three (11%) of the remaining 27 cases died (Table 7). Mortality ranged from 40% of the halothane and naproxen cases to zero in reports related to many other drugs (Table 7). Notably, all 32 patients with probable or possible erythromycin-associated HC injury survived this adverse drug reaction (Table 7). Because of the absence of fatal cases among erythromycin-treated patients, we compared the laboratory data in the erythromycin and nonerythromycin cases. As seen in Table 8, the test results in the identified markers of poor outcome were significantly less abnormal in the erythromycin cases than in the nonerythromycin cases.

Table 7. Number of Cases with Hepatocellular, Cholestatic, or Mixed-Type Liver Injury and Proportion of Cases Leading to Death or Transplantation for Drugs With at Least Three Reports of Drug-Induced Liver Disease, With a Possible or Probable Causality Assessment
HepatocellularDeath/LTCholestatic or MixedDeath/LTTotal Study GroupDeath/LT
  1. Abbreviation: NSAIDs, nonsteroidal anti-inflammatory drugs.

 Erythromycin320 Flucloxacillin1046 Flucloxacillin1297
 Flucloxacillin251 Trimethoprim/sulfametoxazol141 Erythromycin420
 Isoniazid73 Erythromycin100 Trimethoprim/sulfametoxazol212
 Trimethoprim/sulfametoxazol62 Ciprofloxacin41 Isoniazide73
 Ciprofloxacin31 Amoxicillin/Clavulonic acid40 Ciprofloxacin72
 Nitrofurantoin30 Trimethoprim40 Dicloxacillin31
Anesthetics156 Dicloxacillin31 Pivmecillinam31
 Halothane156 Cloxacillin30Anesthetics156
NSAIDs298 Rifampicin30 Halothane156
 Naproxen104 Diclofenac72 Diclofenac204
 Ibuprufen41 Sulindac50 Naproxen114
 Rofecoxib21 Naproxen30 Ibuprufen41
Other drugs778Other drugs616 Rofecoxib31
 Disulfiram243 Carbamazepine92Other drugs10619
 Carbamazepine81 Chlorpromazine42 Disulfiram273
 Ranitidin71 Terbenafin50 Carbamazepine173
 Chlorpromazine50 Clomipramine50 Ranitidin101
 Azathioprine40 Natriumaurothiomalate50 Enalapril82
 Flutamide40 Ticlopidine41 Chlorpromazine82
 Omeprazol41 Dextropropoxiphen40 Sulfasalazine71
 Nefazodon31 Sulfasalazine40 Omeprazol61
 Sulfasalazine31 Disulfiram30 Cyclophosphamid52
 Thiamazol30 Atorvastatin31 Ticlopidine51
 Mianserin30 Thiamazol30 Atorvastatin41
 Clomipramine30 Ranitidin30 Simvastatin41
 Cimetidine30 Ethinylestradiol30 Nefazodon41
 Ticlopidine30 Desogestrel30≥1 drug suspected15111
≥1 drug suspected828 Glibenclamide30   
   ≥1 drug suspected693   
Table 8. Comparison of Liver Laboratory Tests and Age in Patients With Erythromycin (n = 32) Versus Nonerythromycin (n = 377) Hepatocellular Injury
 Erythromycin CasesNonerythromycin CasesP Value
  1. NOTE. Values are expressed as medians, with IQRs in parentheses.

Bilirubin × ULN2.93 (2.4–4.0)6.38 (3.6–11.1)<.0001
AST × ULN11.0 (5.4–17.1)18.0 (7.3–35.7).0048
ALT × ULN11.6 (8.9–25.6)27.1 (15.7–45.7).0001
Age, yrs40 (32–48)56 (40–69)<.0001

In 82 cases of HC jaundice, the patients were being treated with more than one drug with temporal relationship to the diagnosis of liver injury and positive dechallenge, any of which could therefore have caused the reaction. Eight (9.8%) of these patients died, which was not significantly different from the 12.9% mortality among patients with only one suspected drug.

CS and Mixed Liver Injury.

The highest number of reports of CS or mixed type of jaundice with a probable or possible causal relationship to a drug was related to flucloxacillin (Table 7). Mortality ranged from 0% with most of the drugs to 50% (2 out of 4 chlorpromazine-associated cases). Both diclofenac and carbamazepine were associated with CS/mixed injury in addition to a similar number of reactions with a HC pattern (Table 7). Similar to the patients with HC injury considered to be related to erythromycin, all 10 patients with probable or possible erythromycin-associated CS/mixed injury recovered. The drug suspected of the reaction, its indication, duration of treatment, and liver tests for the 20 patients with CS/mixed injury associated with fatal outcome are demonstrated in Table 9.

Table 9. Individual Subjects With Cholestatic/Mixed-Type Liver Injury Associated With Fatal Outcome
  1. NOTE. Cholestatic reaction was defined as R ≤ 2 (R [ratio] which is ×N for serum activity of ALT/×N for serum activity of ALP). The pattern was regarded as mixed when both ALT (above 2 N) and ALP are increased, and 2 < R < 5.17 The four subjects at the bottom with an ALT/ALP ratio ≥2 had a mixed pattern.

  2. Abbreviations: M, male; F, female.

M 64EnalaprilHeart failure60d3.115.71.725.21.8
M 16DiclofenacBack pain360d17.110.38.612.41.2
M 83ChlorpromazinePsychosis28d3.25.313.64.60.38
F 85FlucloxacillinCellulitis26d1.
F 45MultipleMultiple14d4.03.07.727.40.39
F 78FlucloxacillinCellulitis18d13.212.316.115.40.76
F 85FlucloxacillinCellulitis21d19.424.419.221.21.3
M 77DicloxacillinOsteomyelitis87d3.13.13.1321.0
F 83FlucloxacillinCellulitis10d 5.99.436.40.63
F 41Trimethoprim/sulfametoxazolHIV180d7.02.45.813.80.41
M 78FlucloxacillinOral abscess21d14.317.123.8181.32
M 39MultipleProstate cancer32d1.
F 58EstradiolOsteoporosis365d4.13.01.7435.21.72
M 73ChlorpromazinePsychosis13d1.42.24.412.90.5
F 39AtorvastatinHyperlipemia120d2.
M 40DiclofenacArthralgia6d3.
M 69CiprofloxacinCellulitis23d3.
M 90FlucloxacillinCellulitis11d6.
F 77TiclopidineUremia35d6.115.74.839.13.3
M 83DanazolThrombocytopenia270d3.

In 69 cases of CS/mixed jaundice, the patients were being treated with more than one drug, with temporal relationship to the diagnosis of liver injury and positive dechallenge, any of which could therefore have caused the reaction. Only 3 of these patients (4.3%) died, which was not significantly different from the 6.5% mortality when only one drug was suspected.


Because most of the knowledge of the severity of DILD comes from case reports and case series,12, 14 the assessment of the prognosis of patients with DILD suffers from lack of prognostic markers.13, 18–20 Hyman Zimmerman, the legendary pioneer researcher in the field of DILD, observed that the combination of HC injury (high aminotransferases) and jaundice induced by a drug was associated with a poor prognosis, with a fatality rate of 10% to 50% for the different drugs involved (Hy's rule).11–14 In a previous National Institutes of Health conference, a consensus was reached that a new drug should be stopped in a person with previously normal liver tests if AST and ALT levels were more than 3 times the ULN and bilirubin levels were more than twice the ULN.15 Concomitant jaundice and hepatocellular injury observed in clinical trials of a new drug have thus been considered to cause serious trouble concerning safety in the postmarketing phase, when a much larger number of patients are exposed to the drug.13 Although still used by the U.S. Food and Drug Administration in the assessment of the hepatotoxicity of different drugs, a validation of Hy's rule has been requested, because the information about its sensitivity and specificity are largely lacking.13 Although the findings of Zimmerman and coworkers are important contributions to our knowledge of DILD, the mortality rate according to Hy's rule has been reported for only a few compounds, such as isoniazid, iproniazid, cinchophen, dantrolene, halothane, methyldopa, and tienillic acid.21–25 Isoniazid is probably the only compound out of those analyzed and reported previously by Zimmerman et al. to cause DILD, which still is in widespread use nowadays.

Our analysis is unique because it is performed in a large cohort of patients with severe DILD, giving the opportunity to elucidate the most important predictors for outcome. It is well recognized that adverse drug reactions are significantly underreported. The true incidence of hepatic adverse drug reactions has until recently been largely obscure. However, a recent, careful prospective survey of drug-induced liver injury in the general population in France suggests that at the most, only 1 out of 16 cases of DILD in France is actually reported.26 Spontaneous reporting was found to be at least twice as high in Sweden compared with figures from France.27 It is of course possible that severe adverse drug reactions are reported more frequently, although this has not been assessed in a well-designed study. Further limitations of the present study are its retrospective nature and the lack of international normalized ratio data.

In the current study, approximately 20% of patients with probable or possible DILD had serum bilirubin levels 2 or more times the ULN, the limit suggested by the U.S. Food and Drug Administration to correspond to the presence of jaundice. Most of the patients in this study with DILD and concomitant jaundice had a good prognosis with complete recovery, but approximately 10% of the patients either died or underwent liver transplantation. However, our figures on the “drug-specific incidence” of icteric DILD cannot be used as an estimate of the “drug-specific risk of icteric DILD,” because the figures are not sales-related. The small number of patients precludes an assessment of the inherent risk of developing serious liver reaction once a DILD has occurred, at least for most of the drugs.

Our results are in accordance with Hy's rule insofar as HC jaundice was found to have a mortality of approximately 10%. A total of 72 (9.2%) of 784 patients in the total study group either died or underwent transplantation. However, because a significant number of halothane reports (7 suspected lethal reactions) and 1 suspected disulfiram case associated with mortality had to be excluded because of a lack of clinical and laboratory data, the true proportion of deaths and transplantations could be underreported. If these cases had been included, the overall mortality would not have been more than 9.9% in the total study population.

However, the mortality rate of our patients fulfilling Hy's rule seemed to be highly variable when related to which drug was associated with the DILD. Thus, a notable finding in the present study is that all of the 32 patients with erythromycin-induced HC jaundice (fulfilling Hy's rule) recovered. Although erythromycin is among the top five reported hepatic adverse drug reactions in Sweden, with approximately 200 reports between 1966 and 2004, not a single case has been associated with a fatal outcome.28 All patients with erythromycin-induced liver injury recovered in the current study. This was probably due to relatively mild erythromycin-associated DILD. The prognosis among patients fulfilling RUCAM criteria for DILD but in whom the specific drug could not be identified because of multiple drug exposure was not different from those patients with one specified drug suspected to have caused the reaction.

In the current study, the International Consensus Criteria were applied to the classification of hepatic reaction into HC, CS, and mixed adverse drug reactions.16–17 These criteria were first published in 1990, and because most of the work by Zimmerman et al. was published before that time, these publications did not apply the modern strict criteria for the classification of hepatic reactions.11, 14, 21–25 As a consequence, many of the cases with DILD-associated HC jaundice reported by Zimmerman et al. would possibly be classified as mixed type of DILD using the International Consensus Criteria.16–17

CS injury due to drugs has been reported to have a good prognosis, and fatalities were considered to be the results of underlying disease rather than hepatic injury.11 However, in our cohort, patients with CS injury also had a high mortality rate of 7.8%, whereas those with mixed injury had a mortality rate of 2.4%. The mortality rate of 5.4% among patients suffering from flucloxacillin-induced liver injury is in line with previous reports from Sweden and Australia.29–32 In a recent study from Japan, age was not found to differ between patients with DILD who developed fulminant hepatic failure versus those who did not.33 However, in the current study, nonsurvivors with the HC type of DILD were significantly older than those who recovered. Age was found to independently predict death or transplantation in the HC group. Furthermore, in HC injury as well as in CS/mixed injury, the peak level of bilirubin was found to be an independent risk factor for death or liver transplantation. Interestingly, forward stepwise analysis identified AST but not ALT to be an independent predictor of bad outcome in the total study. This is in line with results reported by Ohmori et al.33 in patients with DILD showing higher aminotransferase levels, especially AST in patients who progressed to fulminant hepatic failure compared with those who did not. Furthermore, our data are consistent with those of Gitlin,34 who observed a higher AST/ALT ratio in fatal cases than in survivors with severe acute viral hepatitis. A hypothetical explanation put forward by the author was that when the serum ALT level is exceeded by the serum AST level, this reflects additional AST release from the hepatocyte mitochondrial AST component as a consequence of more severe hepatocyte damage. In contrast to the results obtained by Ohmori et al., we did not find longer duration of drug therapy to be a risk factor for bad outcome.

In summary, in accordance with Hy's rule, HC jaundice has a high but variable mortality rate, depending on the drug involved. AST and bilirubin levels are the most important predictors of death or liver transplantation.