Severe Cutaneous Reactions Requiring Hospitalization in Allopurinol Initiators: A Population-Based Cohort Study

Authors

  • Seoyoung C. Kim,

    Corresponding author
    1. Brigham and Women's Hospital, Boston, Massachusetts
    • 1620 Tremont Street, Suite 3030, Boston, MA 02120
    Search for more papers by this author
    • Dr. Kim has received research support from Takeda and Pfizer and tuition support for the Pharmacoepidemiology Program at the Harvard School of Public Health funded by Pfizer and Asisa.

  • Craig Newcomb,

    1. Perelman School of Medicine, University of Pennsylvania, Philadelphia
    Search for more papers by this author
  • David Margolis,

    1. Perelman School of Medicine, University of Pennsylvania, Philadelphia
    Search for more papers by this author
  • Jason Roy,

    1. Perelman School of Medicine, University of Pennsylvania, Philadelphia
    Search for more papers by this author
    • Dr. Roy has received research support from Roche and Amgen.

  • Sean Hennessy

    1. Perelman School of Medicine, University of Pennsylvania, Philadelphia
    Search for more papers by this author
    • Dr. Hennessy has received consultant fees (less than $10,000) from Millennium, a subsidiary of Takeda.


Abstract

Objective

Rare but potentially life-threatening cutaneous adverse reactions have been associated with allopurinol, but population-based data on the incidence and mortality of such reactions are scarce.

Methods

We conducted a propensity score–matched cohort study to evaluate the incidence rate (IR) and in-hospital mortality of hospitalization for severe cutaneous adverse reactions (SCARs) in allopurinol initiators compared to non–allopurinol users, using data from 5 large Medicaid programs. The primary outcome was identified by the principal discharge diagnosis code 695.1. A Cox proportional hazards model evaluated the relative risk of SCARs associated with the use of allopurinol and determined the relative risk of SCARs associated with allopurinol dose.

Results

During a followup period of 65,625 person-years for allopurinol initiators, 45 were hospitalized with SCARs. The crude IR was 0.69 (95% confidence interval [95% CI] 0.50–0.92) per 1,000 person-years. All 45 cases occurred within 365 days and 41 (91.1%) occurred within 180 days after initiating treatment with allopurinol. Twelve patients (26.7%) died during the hospitalization. The crude IR in non–allopurinol users was 0.04 (95% CI 0.02–0.08) per 1,000 person-years. The risk of SCARs was increased in allopurinol initiators versus nonusers (hazard ratio [HR] 9.68, 95% CI 4.55–20.57). Among allopurinol initiators, the HR for high-dosage (>300 mg/day) versus low-dosage allopurinol was 1.30 (95% CI 0.31–5.36) after adjusting for age, comorbidities, and recent diuretic use.

Conclusion

Among allopurinol initiators, SCARs were found to be rare but often fatal, and occurred mostly in the first 180 days of treatment. The risk of SCARs was 10 times as high in allopurinol initiators as compared to allopurinol nonusers.

INTRODUCTION

Allopurinol is a xanthine oxidase inhibitor that reduces the production of uric acid. For the past several decades, allopurinol has been commonly used to treat patients with gout or nephrolithiasis. Allopurinol is generally well tolerated, but 2–5% of patients may develop side effects such as mild skin rash or gastrointestinal distress (1, 2). It can also cause severe hypersensitivity reactions characterized as a spectrum of clinical conditions ranging from a mild skin rash to life-threatening toxicity presenting as fever, hepatitis, vasculitis, eosinophilia, worsening renal function, and severe cutaneous adverse reactions (SCARs), including toxic epidermal necrolysis (TEN) and Stevens-Johnson syndrome (SJS) (3). Furthermore, allopurinol has been reported as one of the most common causes of SCARs (4, 5).

The precise mechanisms for the development of SCARs are still unknown, but several different factors have been postulated in its pathogenesis, mainly cell-mediated immunity directed toward allopurinol and its active metabolite, oxypurinol, genetic factors, and metabolic factors (6). Recent studies reported a strong genetic association between HLA–B*5801 and SJS and TEN induced by allopurinol (7–9). According to previous descriptive studies mainly based on small hospital case series, fewer than 1% of the patients treated with allopurinol developed hypersensitivity reactions, but the mortality was as high as 27% (3, 6). No population-based data, however, exist on the incidence or mortality from allopurinol hypersensitivity reactions, including SCARs. We examined the incidence and mortality of SCARs requiring hospitalization in patients starting allopurinol in a population-based propensity score–matched cohort to provide more accurate safety data necessary to inform clinical decision making for patients with hyperuricemia and gout.

Significance & Innovations

  • In a large population-based cohort, severe cutaneous adverse reactions requiring hospitalization in patients starting allopurinol were rare, with an incidence rate of > 1 in 1,000 person-years, but fatal.

  • To date, this is the first and largest study to examine the safety of allopurinol using population-based data and rigorous methodologies, such as the new user design and propensity score–matched analysis.

MATERIALS AND METHODS

Study design.

We conducted a retrospective cohort study of allopurinol initiators using the US Medicaid claims data from California, Florida, New York, Ohio, and Pennsylvania (1999–2005). In total, these 5 states include ∼13 million Medicaid enrollees, which account for ∼35% of the Medicaid population. The database contains clinical, demographic, and death status information for their beneficiaries as well as the Medicaid claims for covered health care services, including pharmacy benefits and hospitalizations from the time of a person's Medicaid eligibility until death. Since ∼15–17% of Medicaid beneficiaries are also enrolled in Medicare (10), Medicare data were obtained to assure complete data capture in those dually-eligible beneficiaries. The quality of the data source was assured in previous research (11).

Data use agreements were in place with the Centers for Medicare and Medicaid Services that supplied information for the study database. This study was approved by the Institutional Review Boards of both the University of Pennsylvania and Brigham and Women's Hospital, which granted waivers of informed consent and Health Insurance Portability and Accountability Act authorization.

Study patients.

We identified adult subjects who had ≥180 days of Medicaid plan eligibility and ≥1 outpatient or inpatient claim present before the first prescription of allopurinol. These criteria ensured their continuous eligibility for ≥180 days prior to the study entry, to permit us to identify new users of allopurinol and to assess their comorbidities and other medications at baseline. For the allopurinol nonuser group, Medicaid enrollees who had never received a prescription for allopurinol during the entire study period were identified. Subjects with any claims for a diagnosis of solid tumors, hematologic malignancies, and myelodysplastic syndrome or chemotherapy administration during the 180 days prior to cohort entry were excluded, since these patients have very different morbidity and mortality from patients who were receiving allopurinol for gout, renal stones, or asymptomatic hyperuricemia. Propensity score–matching methods were then used to create a group of allopurinol initiators and non–allopurinol users who were compatible with regard to the potential confounders described below (11).

Outcomes.

The primary outcome of interest was hospitalization for SCARs, defined with a principal International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) code of 695.1X after filling a prescription for allopurinol. The ICD-9-CM code 695.1X is not specific to SJS or TEN and includes several conditions with a different dermatopathology, such as erythema multiforme (EM) major and minor, SJS, and TEN (Table 1). The primary outcome in this study was intended to capture a wide range of severe cutaneous eruptions requiring hospitalization, and was not limited to a specific diagnosis of SJS or TEN. Previous validation studies reported that >90% of the cases identified with the discharge code 695.1 had a skin diagnosis including, but not limited to, EM, SJS, and TEN (12–15). Outcomes that occurred ≥100 days after filling the last prescription for allopurinol were excluded. If patients in the exposed group had multiple hospitalizations with the same diagnosis code, only the first hospitalization was counted. Our secondary outcome was death during the hospitalization for a principal diagnosis of SCARs.

Table 1. Diagnosis codes for erythematous conditions*
ICD-9-CM codeClinical condition
  • *

    ICD-9-CM = International Classification of Diseases, Ninth Revision, Clinical Modification.

695.1Erythema multiforme
 695.10Erythema multiforme, unspecified
 695.11Erythema multiforme minor
 695.12Erythema multiforme major
 695.13Stevens-Johnson syndrome
 695.14Stevens-Johnson syndrome–toxic epidermal necrolysis overlap syndrome
 695.15Toxic epidermal necrolysis
 695.19Other erythema multiforme

Patients were censored at the earliest time of the following events during the followup period: 1) occurrence of the outcome, 2) no filling for allopurinol for 180 days, 3) end of the Medicaid eligibility, 4) end of the study period, or 5) death.

Covariates.

Patient characteristics potentially related to SCARs were assessed using the data from the 180 days prior to the first prescription fill date, defined as the index date. These characteristics included demographic factors (age, sex, race, and state), health care utilization factors (acute care hospitalizations, emergency room visits, and number of physician visits and different medications), and other recorded comorbidities (hypertension, diabetes mellitus, chronic kidney disease [CKD], and liver, pulmonary, and cardiovascular disease). To quantify patients' comorbidities, we additionally calculated the Deyo-adapted Charlson Comorbidity Index based on the ICD-9-CM code from the 180 days prior to the first prescription fill date (16, 17). The comorbidity index is a summary score based on 19 major medical conditions, including myocardial infarction; pulmonary, renal, and hepatic disease; diabetes mellitus; cancer; human immunodeficiency virus infection; etc. A score of 0 represents absence of comorbidity and a higher score indicates a greater number of comorbid conditions. Furthermore, data on the use of diuretics, the daily dose of allopurinol prescribed, and the duration of allopurinol treatment were obtained.

Statistical analysis.

To control confounding by indication to a large extent, we used the propensity score–matching method embedded in a new user cohort design. A propensity score is the estimated probability of starting treatment A versus treatment B, based on preexisting patient characteristics (18, 19). Propensity score matching has been increasingly used as an effective way to adjust a large number of confounders simultaneously, even if the outcome is rare (18, 20). Logistic regression models were developed to calculate the propensity score of all of the eligible subjects in the study cohort. The propensity score is the probability of initiating allopurinol versus no treatment as a function of a number of important potential confounders such as age, sex, race, use of a lipid-lowering drug, diabetic medications, beta-blockers, angiotensin I–converting enzyme inhibitors, angiotensin II receptor blockers, digoxin, antihypertensives, diuretics, antiarrhythmics, oral steroids, Charlson Comorbidity Index, and number of prescriptions, emergency room visits, hospitalizations, and outpatient visits, which were included. Patients in each group (allopurinol initiators versus nonusers) were then matched in a 1:1 ratio using greedy matching techniques.

The number of primary outcomes was assessed in the first 30 days, 31–180 days, and 181–365 days after initiating allopurinol therapy. The incidence rate (IR) and 95% confidence interval (95% CI) of hospitalization for SCARs were calculated in both allopurinol initiators and nonusers. A Cox proportional hazards model (21) was used to examine the relative risk of hospitalization with SCARs in allopurinol initiators compared with nonusers. Among allopurinol initiators, separate unadjusted and multivariable Cox proportional hazards models (21) were used to determine the relative risk of SCARs associated with a high dosage (>300 mg/day) of allopurinol, after adjusting for age, comorbidities, and recent diuretic use. The allopurinol dose was analyzed as a time-dependent covariate. In order to avoid overfitting of the regression model, the presence of CKD was not included in the multivariable analysis. The reporting of this study follows the Strengthening the Reporting of Observational Studies in Epidemiology guidelines (22).

RESULTS

We identified 166,826 patients with ≥1 prescription filled for allopurinol and 56,482,824 Medicaid enrollees who never had a prescription for allopurinol during the entire study period. After applying the exclusion criteria, 90,358 allopurinol initiators and 8,706,492 allopurinol nonusers were identified. After 1:1 propensity score matching, our final cohort consisted of 90,358 allopurinol initiators and 90,358 nonusers. Figure 1 shows our cohort selection process.

Figure 1.

Study cohort selection. PS = propensity score.

The baseline characteristics of our propensity score–matched study population are listed in Table 2. Overall, the 2 groups were similar. The mean ± SD age was 65.5 ± 15.6 years in the allopurinol initiators and 65.6 ± 15.5 years in the nonallopurinol group. Fifty-two percent of the allopurinol initiators and nonusers were men. The mean ± SD number of prescription drugs was 6.9 ± 4.9 among the allopurinol initiators and 5.7 ± 4.8 in the allopurinol nonuser group. Approximately 12% of allopurinol initiators had ≥1 diagnosis of gout or renal stones before starting treatment with allopurinol, and ∼25% of allopurinol initiators received ≥1 diagnosis of gout or renal stones during the followup period.

Table 2. Baseline characteristics of the propensity score–matched cohort 180 days before the index date*
 Allopurinol initiators (n = 90,358)Allopurinol nonusers (n = 90,358)
  • *

    Values are the number (percentage) unless otherwise indicated. ER = emergency room; COPD = chronic obstructive pulmonary disease.

Demographic characteristics  
 Age, mean ± SD years65.5 ± 15.665.6 ± 15.5
 Men47,278 (52.3)47,278 (52.3)
 White37,403 (41.4)37,108 (41.1)
Health care utilization  
 No. of outpatient visits, mean ± SD27.0 ± 32.422.9 ± 30.6
 ≥1 ER visit32,827 (36.3)31,000 (34.3)
 No. of all prescription drugs, mean ± SD6.9 ± 4.95.7 ± 4.8
 ≥1 hospitalization22,919 (25.4)20,607 (22.8)
Comorbidities  
 Comorbidity index score, mean ± SD1.9 ± 2.11.5 ± 1.9
 Diabetes mellitus16,942 (18.8)14,727 (16.3)
 Gout9,095 (10.1)355 (0.4)
 Liver disease1,581 (1.8)1,760 (2.0)
 Kidney disease5,356 (5.9)3,117 (3.5)
 Nephrolithiasis2,024 (2.2)4,309 (4.8)
 Hypertension29,419 (32.6)23,970 (26.5)
 Heart failure12,209 (13.5)8,017 (8.9)
 Coronary artery disease10,445 (11.6)9,447 (10.5)
 COPD9,843 (10.9)9,449 (10.5)

Among the allopurinol initiators, 45 were hospitalized with SCARs over a followup period of 65,625 person-years. The crude IR of hospitalization with SCARs was 0.69 (95% CI 0.50–0.92) per 1,000 person-years. All 45 cases occurred within the first 365 days after initiation of allopurinol therapy and 41 (91.1%) occurred within 180 days (Table 3). Of those, 12 patients (26.7%) died during the hospitalization. Among allopurinol nonusers, 10 hospitalizations with SCARs occurred over a period of 237,025 person-years and only 1 patient (10%) died during the hospitalization. The crude IR of hospitalization with SCARs in non–allopurinol users was 0.04 (95% CI 0.02–0.08) per 1,000 person-years. The hazard ratio (HR) of hospitalization with SCARs compared to allopurinol nonusers was 9.68 (95% CI 4.55–20.57).

Table 3. IRs of hospitalization for severe cutaneous adverse reactions in allopurinol initiators (n = 90,358)*
 No. of eventsFollowup period, per 1,000 person-yearsIR (95% CI) per 1,000 person-years
  • *

    IR = incidence rate; 95% CI = 95% confidence interval.

Total4565.60.69 (0.50–0.92)
First 30 days207.22.77 (1.69–4.28)
31–180 days2119.91.05 (0.65–1.61)
181–365 days412.80.31 (0.85–0.80)

The results from our crude and multivariable Cox regression analyses among allopurinol initiators are shown in Table 4. Age and the comorbidity index score, but neither high dose nor the presence of CKD, were significantly associated with an increased risk of hospitalization with SCARs. In our multivariable analysis adjusting for age, comorbidities, and recent use of any diuretic, the HR of hospitalization with SCARs was 1.30 (95% CI 0.31–5.36) for the high-dose allopurinol initiators compared to the low-dose allopurinol initiators. Table 5 shows baseline demographic characteristics, the health care utilization pattern, and comorbidities of allopurinol initiators who developed SCARs and those who did not.

Table 4. Unadjusted and multivariable HRs of hospitalization for severe cutaneous adverse reactions in allopurinol initiators (n = 90,358)*
 Unadjusted HR (95% CI)Multivariable HR (95% CI)
  • *

    HR = hazard ratio; 95% CI = 95% confidence interval.

  • Modeled as a continuous variable.

Age1.04 (1.02–1.06)1.03 (1.01–1.06)
Comorbidity index score1.22 (1.09–1.35)1.18 (1.05–1.33)
High-dosage allopurinol (>300 mg/day)1.13 (0.27–4.68)1.30 (0.31–5.36)
Chronic kidney disease0.72 (0.18–2.97)
Recent diuretic use2.10 (1.10–4.01)1.49 (0.77–2.90)
Table 5. Baseline characteristics of allopurinol initiators who developed severe cutaneous adverse reactions (SCARs) versus those who did not*
 SCARs (n = 45)No SCARs (n = 90,313)
  • *

    Values are the number (percentage) unless otherwise indicated. ER = emergency room.

Demographic characteristics  
 Age, mean ± SD years73.7 ± 11.865.5 ± 15.6
 Men11 (24.4)47,267 (52.3)
 White11 (24.4)37,392 (41.4)
Health care utilization  
 No. of outpatient visits, mean ± SD43.1 ± 45.827.0 ± 32.4
 ≥1 ER visit22 (48.9)32,805 (36.3)
 No. of all prescription drugs, mean ± SD8.5 ± 4.16.9 ± 4.9
 ≥1 hospitalization22 (48.9)22,897 (25.4)
Comorbidities  
 Comorbidity index score, mean ± SD3.0 ± 1.81.9 ± 2.1
 Diabetes mellitus11 (24.4)16,931 (18.8)
 Hypertension23 (51.1)29,396 (32.6)
 Heart failure11 (24.4)12,198 (13.5)

DISCUSSION

The spectrum of cutaneous and systemic manifestations of drug reactions is diverse, ranging from self-limited exanthematous drug eruptions to severe drug hypersensitivity syndromes, EM major, SJS, or TEN. The pathophysiologic mechanisms of these SCARs are not well established. More than 100 drugs, including allopurinol, trimethoprim/sulfamethoxazole, carbamazepine, phenobarbital, cephalosporins, antifungal agents, and nonsteroidal antiinflammatory drugs, have been implicated as causes of these conditions (13). The incidence of TEN and SJS in the general population is reported to be generally low and is estimated at 0.4–1.2 and 1–6 cases per million person-years, respectively (13). Allopurinol has been repeatedly reported to have one of the highest relative risks for SCARs in the literature (4, 23), but few studies have examined its incidence and mortality in a population-based cohort. To date, little data are available to examine the risk of SCARs associated with the use of newer xanthine oxidase inhibitors.

Our study has important clinical implications for the management of hyperuricemia. We evaluated the incidence and mortality of SCARs in allopurinol initiators using the health care utilization data from 5 large Medicaid programs. Our study showed that the incidence of hospitalization with SCARs among allopurinol initiators was less than 1 per 1,000 person-years and was highest in the first 30 days of use. The risk of SCARs associated with the initiation of allopurinol treatment was 10 times greater compared to nonusers of allopurinol in the propensity score–matched cohort. The number needed to harm would be approximately 1,540, i.e., there would be 1 additional hospitalization for SCARs for each 1,540 patients newly treated with allopurinol. Although the risk seems small, the in-hospital mortality was high at approximately 27%. While prior research suggests that high-dose allopurinol, diuretic use, and CKD increase the risk of SCARs, our study, even with this large size, did not show a statistically significant association with these variables, although the point estimate for high versus low dose was elevated.

It is known that hyperuricemia is closely associated with cardiovascular disease, metabolic syndrome, and CKD (24–31). As a result, allopurinol is frequently prescribed to patients with other comorbidities and used with other drugs such as antihypertensive drugs, diuretics, antineoplastic agents, and antibiotics. It is important to examine whether the concomitant use of certain drugs increases the risk of serious adverse events in patients receiving allopurinol. Although we had extensive data on comorbidities and use of other drugs, we could not conduct multivariable analyses examining the effects of many comorbid conditions and concomitant use of individual drugs or classes of drug because of the small number of outcomes.

Other potential limitations include misclassification of exposures and outcomes. While we used pharmacy claims data, which are considered as one of the best data sources for the drug exposure to identify the exposure in this study (32), the actual patient adherence to the medication is unknown. The main outcomes in the present study were identified by a principal ICD-9-CM code of 695.1, which includes a wide range of severe cutaneous eruptions requiring hospitalization, not limited to a specific diagnosis of SJS or TEN (12–14). Therefore, the incidence of true SJS or TEN associated with the use of allopurinol is probably even lower. Our study cohort was identified as including new users of allopurinol, and was not necessarily limited to patients with gout. Although there is currently no guideline that supports the use of allopurinol in asymptomatic hyperuricemia, some of the patients in our study cohort might have had asymptomatic hyperuricemia (33, 34). Whether the risk of SCARs would differ between patients with and without gout is unknown. Lastly, our study did not find an association between CKD and SCARs. Although we used a comprehensive list of diagnosis codes to define CKD and the validity of such codes in claims data was acceptable in a prior study (35), some subjects with mild CKD might have not been captured in our study. As is the case in most epidemiologic studies, the patients were not randomly exposed to the drug in our study. Therefore, it is possible that patients with moderate to severe CKD did not receive allopurinol or were monitored more carefully before they could develop an SCAR.

Confounding bias is another important issue in observational studies for pharmacoepidemiologic research. We conducted a large population-based cohort study and attempted to minimize this bias by selecting new users of allopurinol and matching them to non–allopurinol users based on a propensity score that included many potentially important confounders, resulting in the well-balanced cohorts with respect to measured variables in the database. However, it is possible that differences still exist between the groups, resulting in residual confounding due to unmeasured confounders (e.g., history of drug reaction, mildly elevated serum creatinine, and genetic susceptibility) not included in the propensity score calculation. With regard to genetic susceptibility, a number of recent studies examined the relationship between several specific HLA–B alleles and severe allopurinol hypersensitivity reactions, and suggested that HLA–B*5801 can be a useful tool for assessing individual susceptibility to SCARs, particularly in the Asian populations (7–9, 36, 37). In a recent cohort study of Korean patients with CKD, allopurinol-induced SCARs occurred in 18% of subjects with HLA–B*5801 (37). However, because of the different frequency of the HLA–B*5801 allele in different racial and ethnic backgrounds, the role of HLA–B*5801 as a screening tool has been limited (38, 39).

In summary, the incidence of hospitalization for SCARs was 0.69 per 1,000 person-years in the first year of use among new users of allopurinol, SCARs were often fatal, and SCARs occurred mostly in the first 180 days of treatment. Our study suggests that the risk of SCARs was 10 times as high in allopurinol initiators as compared to non–allopurinol users. Future studies using a large detailed clinical data set from a large prospective inception cohort of allopurinol users is needed for a better understanding of other risk factors of SCARs such as impaired renal function and concomitant medication use.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Kim had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design.Kim, Margolis, Roy, Hennessy.

Acquisition of data.Kim, Hennessy.

Analysis and interpretation of data. Kim, Newcomb, Hennessy.

ROLE OF THE STUDY SPONSOR

This study was conducted by the authors independent of Takeda Pharmaceuticals. The sponsor was given the opportunity to make nonbinding comments on a draft of the manuscript, but the authors retained right of publication and to determine the final wording. Takeda Pharmaceuticals had no role in the study design, data collection, data analysis, or writing of the manuscript, as well as approval of the content of the submitted manuscript. Publication of this article was not contingent on the approval of Takeda Pharmaceuticals.

Acknowledgements

The authors acknowledge Qufei Wu for his data programming and statistical analysis and thank Cristin Freeman for her help in managing this research project.

Ancillary