To determine the validity, magnitude, precision, and applicability of data on the rates at which drugs cause adverse cutaneous reactions.
Systematic review of the medical literature.
The MEDLINE database was searched (1966-August 2000) for studies that contain information on the rates of cutaneous reactions to drugs. The bibliographies of retrieved articles and review articles were also examined to find relevant studies.
Main Outcome Measures
Studies that included primary data on cutaneous reaction rates to drugs were evaluated for their validity, magnitude, precision, and applicability, using guidelines derived from existing guidelines for the evaluation of articles about harm and prognosis.
Nine studies met the study criteria. Five of the studies were based on prospectively collected data on medical inpatients, 2 were retrospective studies based on chart or computerized medical record review, and 2 were based on spontaneous reports and consumption data. The morbilliform drug exanthem and urticaria were the most common cutaneous reactions to drugs. Reaction rates varied from 0% to 8% and were highest for antibiotics (in the range of 1% to 8% for several classes of antibiotics).
Despite differences in the methods of the studies reviewed and their time of execution, there is remarkable agreement in the results. Reaction rates (and 95% confidence intervals) are available for many commonly used drugs.
RASHES ARE among the most common adverse reactions to drugs. Since most drug-induced eruptions appear within the first week after the drug therapy is started, attributing an eruption to a specific drug is often straightforward. Antibiotics and allopurinol are well-known exceptions to this temporal rule and may induce rashes up to 2 weeks after therapy is started. Determining that a particular drug causes an eruption is often difficult when the patient is taking multiple drugs. Data on the rates at which drugs induce rashes are useful in this situation. But what are the data, how valid are they, what are their magnitude and precision, and how are they applied to the care of individual patients?
To answer questions regarding data on the rates at which drugs induce rashes, I performed a PubMed search of MEDLINE (1966-August 2000) using the search string "((((("pharmaceutical preparations"[MeSH Terms] OR drug[Text Word]) AND ("exanthema"[MeSH Terms] OR rash[Text Word])) OR (cutaneous[All Fields] AND reaction[All Fields])) AND (rate[All Fields] OR (((("epidemiology"[Subheading] OR "incidence"[MeSH Terms]) OR "epidemiology"[MeSH Terms]) OR "Incidence"[MeSH Terms]) OR incidence[Text Word]))) AND notpubref[sb]). I screened the titles and abstracts for studies that might contain primary data on the rates of cutaneous reactions to drugs (ie, contained information on rates or incidence of rashes or drug eruptions, or were population-based studies of rashes or drug eruptions). Sensitive PubMed Clinical Queries searches for articles about etiology and prognosis were also conducted using the same search string. The bibliographies of review articles that were included in the results of the searches were also screened for relevant articles. The titles or abstracts of 9 of 1544 articles seemed to contain information about rates of cutaneous reactions to drugs.1-9 These 9 articles and an article recently published in the Archives of Dermatology10 were examined in detail. The references of these articles and of review articles yielded 6 additional articles that were reviewed in detail.11-17
Studies that included primary data on cutaneous reaction rates to drugs were evaluated for their validity, magnitude, precision, and applicability. Guidelines for these evaluations of studies on the rates at which drugs cause rashes were derived from existing guidelines for the evaluation of articles about harm and prognosis.18,19
The validity of a study is determined by the quality of its design and execution. To determine the rates of cutaneous reactions to a drug, a study should have the following features: (1) a representative and well-defined sample of patients; (2) sufficiently long and complete follow-up; (3) a description of methods used to associate drug exposure and the development of rash; and (4) a correct temporal relationship between exposure and the development of rash.
The calculated reaction rates and the 95% confidence intervals (CIs) determined the magnitude and precision of the studies. The rates of cutaneous reactions to drugs were abstracted from the studies or were calculated if sufficient primary data were provided to allow for calculation. Exact 95% CIs were calculated based on binomial distribution.20 The data presented were limited to reaction rates based on at least 1000 patients exposed to a drug because the 95% CI is profoundly influenced by the number of patients exposed.1 For the rates that were derived from data based on at least 1000 exposed patients, 95% CIs were sufficiently narrow, but they become wider when derived from data of smaller numbers of exposed patients. Criteria for applicability included a well-described study population so that individual patients could be compared with the study populations, an estimation of the rate of cutaneous reactions to a drug (with a narrow 95% CI), and a finding that the data could help determine whether to stop a drug therapy or reassure a patient.18,19
Nine studies contained primary data on the rates of cutaneous reactions to drugs that met the study criteria.1-3,5,6,10,11,16,17 Five of the studies were based on prospectively collected data on medical inpatients,1,5,6,11,16 2 were retrospective studies based on chart or computerized medical record review,2,10 and 2 were based on patients seen or spontaneous reports and consumption data.3,17 The quality of the studies reviewed is given in Table 1.
The Boston Collaborative Drug Surveillance Program (BCDSP) collected data on adverse events (including rashes) on 37 665 consecutive medical inpatients monitored between 1966 and 1982.1,11 Nurses monitored consecutive patients admitted to the medical services of participating hospitals. Reported rashes were limited to those thought to be allergic in nature (morbilliform drug exanthem, urticaria, and generalized pruritus). Cutaneous reaction rates were determined in a sequential process that relied on identifying rash-producing drugs based on (1) a rate of association of a drug with the development of a rash more than twice the mean rate for all drugs and (2) the clustering of rashes in the first 7 days after a drug therapy was started. Preliminary reaction rates for rash-producing drugs were determined based on rates of reactions observed in patients receiving only 1 rash-producing drug. Weighted attribution of rashes to specific drugs was used for patients who developed rashes while taking more than 1 rash-producing drug. After all rashes were assigned to specific drugs, reaction rates were determined for each drug by dividing the number of attributed rashes by the total number of recipients of each drug.
The morbilliform drug exanthem and urticaria accounted for 95% and 5% of skin reactions, respectively. Rashes occurred in 2% to 3% of patients who were taking an average of 8 or 9 different drugs. Cutaneous reaction rates for drugs that were received by more than 1000 patients and had reaction rates greater than 1% are given in Table 2. Data from the BCDSP also identified drugs that were least likely to cause rashes. Drugs received by more than 1000 patients with no reported reactions are given in Table 3. For these drugs the 1-sided upper 95% CI is no more than 0.36% for any of the drugs listed. Reaction rates were determined for more than 60 drugs using data collected by the BCDSP. The reaction rates for drugs with rates less than 1% and for drugs received by fewer than 1000 patients were published.1,11
The Comprehensive Hospital Drug Monitoring (CHDM) program collected data on adverse events (including rashes) on consecutive medical inpatients monitored in teaching hospitals in Switzerland.5,6,16 Cutaneous reaction rates were determined by dividing the number of patients receiving a particular drug who developed a rash by the total number of patients who received that drug. Attribution of the type of rash and its attribution to a particular drug were determined by the drug monitoring team of investigators. Only reactions judged to be definite (proven by rechallenge) or probable (more likely to be caused by a drug than any other cause) were included. Reactions attributable to more than 1 drug were excluded.
Based on CHDM data on 19 653 consecutive medical inpatients monitored between 1974 and 1982 in 2 teaching hospitals in Bern, Switzerland, Sonntag et al16 reported the incidence of exanthemas definitely or probably due to antibiotics and allopurinol. Exanthemas occurred in 318 (8.0%; 95% CI, 7.2%-8.9%) of 3980 patients and 74 (2.8%; 95% CI, 2.2%-3.5%) of 2619 patients exposed to aminopenicillins (amoxicillin, epicillin, or ampicillin) and trimethoprim-sulfamethoxale (co-trimoxazole), respectively (Table 4). For other penicillins and for cephalosporins, reactions occurred in 38 (4.7%; 95% CI, 3.3%-6.4%) of 808 patients and 8 (1.9%; 95% CI, 0.8-3.7) of 427 patients, respectively.16
Also based on CHDM data on 37 392 monitored patients admitted between 1974 and 1989, Oberholzer et al6 reported the incidence of allergic or pseudoallergic reactions to nonsteroidal anti-inflammatory drugs (NSAIDs) and analgesics (eg, aspirin, propoxyphene, or pyrazolones).6 The drug exanthem (73%, 69 patients) and urticaria (14%, 13 patients) accounted for most reactions. Reactions occurred in 95 (0.5%) of 19 082 patients treated with NSAIDs or analgesics. Excluding patients with multiple reactions, the reaction rates were 35 (0.69%: 95% CI, 0.48%-0.96%) of 5076 patients and 50 (0.21%; 95% CI, 0.15%-0.27%) of 24 390 patients for NSAIDs and analgesics, respectively (Table 4).
Also based on CHDM data on 48 005 monitored patients admitted to divisions of general internal medicine of 3 Swiss hospitals between 1974 and 1993, Hunziker et al5 conducted a study of adverse skin reactions over a 20-year period. Skin reactions were classified as "maculopapular" exanthem, urticaria, vasculitis, and other. The maculopapular exanthem and urticaria accounted for 91% and 6% of the skin reactions, respectively. Cutaneous reaction rates were highest for penicillins, sulfonamides (especially co-trimoxazole), and NSAIDs (Table 4). The reaction rates for antibiotics received by fewer than 1000 patients were published.5 Reaction rates for drugs other than antibiotics and NSAIDs were not reported, although they may be available in the author's database.
van der Linden et al2 conducted a retrospective cohort study in a dynamic population using data from computer-based patient records of Dutch general practitioners. Their objective was to determine skin reaction rates to antibacterial agents in general practice. The study population consisted of 13 679 antibacterial drug users. Diagnosis and exposure were based on coded diagnoses and chart review. Cutaneous reaction rates were determined by dividing the number of patients receiving a particular drug who developed a rash by the total number of patients who received that drug.
Rash, generalized pruritus, urticaria, and other reactions occurred in 56%, 13%, 14%, and 16% of patients, respectively. One percent of antibacterial drug users developed a rash. Cutaneous reaction rates for drugs that were received by more than 1000 patients are given in Table 5. The reaction rates for drugs received by fewer than 1000 patients were published.2
Ibia et al10 report on the rates of rashes induced by antibiotics based on a retrospective review of 5923 charts from a private pediatric group practice in northern Virginia. Patients identified as having an adverse drug reaction to an antibiotic were contacted via a survey questionnaire. Cutaneous reaction rates were determined by dividing the number of patients receiving a particular drug who developed a rash by the total number of patients who received a prescription for that drug. Urticaria and the "macular and/or papular" drug exanthem accounted for 47% of rashes. The data indicate that on a per prescription basis, significantly more rashes were documented for cefaclor (4.8%) than for penicillins (2.7%), sulfonamides (3.5%), and other cephalosporins (1%) (Table 6). The reaction rates for drugs received by fewer than 1000 patients were published.10
Swanbeck and Dahlberg17 used data on all patients with suspected cutaneous drug reactions presenting at the Department of Dermatology at Sahlgren Hospital in Gothenburg, Sweden, from June 1984 to May 1988. Reaction rates were estimated by dividing the number of patients who had a cutaneous drug reaction while taking a particular drug or class of drugs by the drug consumption (sold-defined daily doses). Of 440 suspected cutaneous drug reactions, 210 (48%) were macular and "maculopapular" eruptions and 98 (22%) were urticaria. Results were reported as figures corrected for frequency of use (ie, number of reactions divided by the sold-defined daily doses) and were 50, 28, 23, 15, and 11 for gold compounds, co-trimoxazole, trimethoprim, cephalosporins, and penicillins, respectively.
Naldi et al3 used data from a database that holds all of the spontaneous reports of adverse drug reactions from 4 Italian regions.3 Data collected between January 1996 and December 1997 were analyzed. Reaction rates were estimated by dividing the number of spontaneous reports by drug consumption data expressed as daily defined doses per 1000 inhabitants per day. Of the 2595 reported adverse skin reactions, 28%, 30%, 25%, and 17% were exanthems, urticaria, other nonserious reactions, and other serious reactions, respectively. Antimicrobial agents and NSAIDs were the most commonly reported agents and accounted for more than 70% of reports. Cotrimoxazole was associated with the highest rate (number of reports per consumption) (80), followed by cephalosporins (69), fluoroquinolones (65), penicillins (45), and macrolides (35). There were no striking differences in the reporting rate and reaction profile of individual drugs in the various categories, with the possible exception of the fluoroquinolones. The reported rates of reactions were 176 for pefloxacin, 92 for ciprofloxacin, 85 for lomefloxacin, and less than 60 for ofloxacin, rufloxacin, and norfloxacin. Aspirin and dipyrone were associated with more than twice the number of reactions of other NSAIDs. The rates were 46 and 38 for aspirin and dipyrone, respectively.
Several conclusions can be reached from the information collected and presented. Despite differences in the methods of the studies and their times of execution, there is remarkable agreement in results. Data from the BCDSP,1,11 the CHDM,5,6,16 the private pediatric practice in the United States,10 the Dutch study based on an electronic medical record system,2 the Swedish study based on cases and consumption data,17 and the Italian spontaneous reporting system3 all point to similar conclusions. The morbilliform drug exanthem and urticaria are the most common cutaneous reactions to drugs. The reaction rates to antibiotics are high (in the range of 1% to 8% for several classes of antibiotics). The NSAIDs are a common cause of rashes, especially urticaria. All of the studies provide reasonable estimates of cutaneous reaction rates that can be of use to practicing dermatologists.
Several of the studies reported data regarding the effects of age and gender on the development of drug rashes. As with reaction rates, there was generally good agreement among studies. Cutaneous reaction rates were 35% higher in females than in males in the BCDSP studies;1,11 there was no correlation between age and reaction rates. In the CHDM data reported by Sonntag et al,16 the female-male ratios of subjects with rashes were 0.9, 1.05, and 1.63 for those younger than 10 years, those younger than 14 years, and adults (18 years or older), respectively. The 14- to 17-year-old age group was included in the study but was not separately reported because there was no significant age-related observation in this group. The authors noted that the female-male ratios may have been confounded by greater drug consumption by women and the greater number of women among the elderly. Rates of reactions per prescription were higher in boys than girls for children younger than 3 years (female-male ratio, 0.79) but higher in girls than boys for children older than 9 years (female-male ratio, 4.72) in the data from the private pediatric practice in the United States.10 Most drug rashes occurred in children 6 years or younger in that study (92% and 82% of the drug rashes occurred in boys and girls 6 years or younger, respectively).10 Reaction rates increased with age and were higher in females (female-male ratio, 1.58) in the data from the Italian spontaneous reporting system.3
Few of the studies reported data on the relationship between underlying medical conditions and the development of rashes or on the influence of a multiple course of drug treatment on the development of rashes. There was no correlation between diagnoses or survival and the development of rashes in the data from the BCDSP.1,11 The incidence of skin reactions to the first treatment was no different than that during the second or subsequent treatments for most antibacterial drugs with the exception of amoxicillin-clavulanic acid in the data from the Dutch study based on an electronic medical record system.2 Reactions to amoxicillin-clavulanic acid were higher on first exposure than on second or subsequent exposures (2.2 per 1000 patient-days vs 0.3 per 1000 patient-days, respectively).
The present study has several limitations. As a single-authored study, it may contain the biases of the author. This reservation may have its greatest impact on the assessment of the quality of studies (Table 1). The method used to assess the quality of studies has not been previously used and has not been validated, although it is based on work done on studies of harm and prognosis.18,19 Many of the validity and magnitude and precision criteria have been recognized and used by others to assess studies of adverse reactions to drugs.21 The studies of Sonntag et al,16 Oberholzer et al,6 and Hunziker et al5 are likely to contain duplicate data. The criteria used to classify clinical patterns of drug eruptions were not precisely defined in any of the studies. Pattern definitions were based on the clinical judgment of monitors, medical care providers, or investigators. Finally, important studies may have been missed. The foibles of electronic searching were certainly evident as one BCDSP study1 was detected by the MEDLINE strategy used but another BCDSP study was not.11 Several references that contain relevant data were not detected by electronic searches but were identified by reviewing references11-17 or were already known by the author.10
Reporting rates derived from cases seen or spontaneous reports and consumption data may be useful for comparisons but they do not provide reaction rates.3,17 The relationship between the number of spontaneous reports and the actual occurrence of adverse reactions is unknown and subject to many biases. Consumption data do not take into account long-term, continuous prescriptions (where adverse cutaneous reactions are not expected) and new or intermittent prescriptions (where cutaneous adverse reactions are most likely to occur).3,21
Data were provided in several of these studies for drugs that were not in widespread use during the collection of data by the prospective drug monitoring studies (BCDSP and CHDM). The estimated reaction rate for cefaclor was 4.8% in children.10 The estimated reaction rate for fluoroquinolones was 1.6%.2
The strongest data to establish accurate estimates of rates of cutaneous reactions to drugs comes from prospectively collected data in population-based studies that monitor large numbers of patients and capture all adverse events that occur.21 Such studies are difficult to perform, time consuming, and very expensive. Few such studies have been conducted.1,5,6,11,16 Therefore, there is a data gap, especially for drugs in limited use up to or introduced after 1993. Several studies, including the ones reviewed herein, are helping us to mind the gap.
How are data on the rates at which drugs cause rashes applied when the patient is taking multiple drugs? To demonstrate how the data may be useful in the care of individual patients with drug eruptions, I simulated a patient taking 3 drugs (A, B, and C) with reactions rates of 3%, 0.5%, and 0.1%, respectively, using resampling (Figure 1).22 The simulation is performed by constructing an urn containing 1000 balls (964, 30, 5, and 1 of which are labeled with 0, 1, 2, and 3, respectively). Balls labeled 0 represent patients who will not develop a rash to A, B or C, whereas balls labeled 1, 2, and 3 represent patients who will develop rashes due to the administration of drugs A, B, and C, respectively.
A sample of 1000 balls is then randomly selected from the urn with replacement. "With replacement" means that a selected ball is taken, identified, counted, and then returned to the urn before the next ball is selected. After 1000 balls are selected, the numbers of balls labeled with 1, 2, or 3 are counted. They represent the numbers of patients who developed rashes due to the administration of drugs A, B, and C, respectively. The proportion of rashes due to A, B, and C are then determined. The sampling procedure is repeated 1000 times (ie, 1000 samples of 1000 balls) and the median proportions of rashes caused by A, B, and C and their respective 95% CIs are determined. The proportion of rashes produced by A, B, or C in this simulation is equivalent to the probabilities that A, B, or C will produce rashes. This simulation was performed using Resampling Stats (Resampling Stats Inc, Arlington, Va), a computer program that performs resampling using a simple programming language.22
As indicated (Figure 1), the probabilities (95% CIs) of the rashes being caused by A, B, and C were 0.84 (0.71-0.95), 0.13 (0.03-0.25), and 0.03 (0-0.09), respectively. The probability that drug A caused the eruption is high enough that its administration should be stopped unless it is essential for the patient's health. Whether therapy with drug B should be stopped would require an assessment of its importance, the anticipated duration of its use, and the seriousness of the drug eruption. Therapy with drug C could be continued because the likelihood of it being the cause of the eruption is sufficiently small. This approach can be used to convert reaction rates to probabilities for any drug combination with known or estimated reaction rates.
A cooperative effort of the Clinical Epidemiology Unit of the Istituto Dermopatico dell' Immacolata–Istituto di Ricovero e Cura a Carattere Scientifico (IDI-IRCCS) and the Archives of Dermatology.
Accepted for publication February 27, 2001.
Michael Bigby, MD, Beth Israel Deaconess Medical Center, 330 Brookline Ave, Boston, MA 02215 (e-mail: email@example.com).
K Drug-induced cutaneous reactions: a report from the Boston Collaborative Drug Surveillance Program on 15 438 consecutive inpatients, 1975 to 1982. JAMA.
1986;2563358- 3363Google ScholarCrossref
van der Linden
PDvan der Lei
BH Skin reactions to antibacterial agents in general practice. J Clin Epidemiol.
1998;51703- 708Google ScholarCrossref
et al. Cutaneous reactions to drugs: an analysis of spontaneous reports in four Italian regions. Br J Clin Pharmacol.
1999;48839- 846Google ScholarCrossref
MB Cutaneous adverse drug reactions in a hospital setting. N Z Med J.
1995;108165- 166Google Scholar
R Comprehensive hospital drug monitoring (CHDM): adverse skin reactions, a 20-year survey. Allergy.
1997;52388- 393Google ScholarCrossref
et al. Incidence of drug side effects by symptoms and syndromes: from the experiences of the Comprehensive Hospital Drug Monitoring and the Swiss Drug Side Effect Center—as an example, allergic and pseudo-allergic reactions with mild analgesics and NSAID. Ther Umsch.
1993;5013- 19Google Scholar
AK Surveillance of drug induced diseases in children. Indian J Pediatr.
1994;61357- 365Google ScholarCrossref
S Adverse drug reactions among inpatients in a north Indian referral hospital. Natl Med J India.
2000;1316- 18Google Scholar
Gruppo Italiano Studi Epidemiologici in Dermatologia (GISED), Cutaneous reactions to alimentary tract medications: results of a seven-year surveillance program and review of the literature. Dermatology.
1996;19311- 16Google ScholarCrossref
BL Antibiotic rashes in children: a survey in a private practice setting. Arch Dermatol.
2000;136849- 854Google ScholarCrossref
H Rates of cutaneous reactions to drugs: a report from the Boston Collaborative Drug Surveillance Program. JAMA.
1976;235918- 922Google ScholarCrossref
C Drug eruptions in Bangkok: a 1-year study at Ramathibod. Int J Dermatol.
1998;37747- 751Google ScholarCrossref
PN Incidence of adverse drug reactions in hospitalized patients: a meta-analysis of prospective studies. JAMA.
1998;2791200- 1205Google ScholarCrossref
J Adverse reactions in children during long-term antimicrobial therapy. Pediatr Infect Dis J.
1996;15404- 408Google ScholarCrossref
DY Hypersensitivity reactions to antibiotics commonly used in children. Pediatr Infect Dis J.
1995;14221- 231Google ScholarCrossref
et al. Exanthema during frequent use of antibiotics and antibacterial drugs (penicillin, especially aminopenicillin, cephalosporin and cotrimoxazole) as well as allopurinol: results of the Berne Comprehensive Hospital Drug Monitoring program. Schweiz Med Wochenshr.
1986;116142- 145Google Scholar
E Cutaneous drug reactions: an attempt to quantitative estimation. Arch Dermatol Res.
1992;284215- 218Google ScholarCrossref
Vfor the Evidence-Based Medicine Working Group, Users' guides to the medical literature, Pt IV: how to use an article about harm. JAMA.
1994;2711615- 1619Google ScholarCrossref
Pfor the Evidence-Based Medicine Working Group, Users' guides to the medical literature, Pt V: how to use an article about prognosis. JAMA.
1994;272234- 237Google ScholarCrossref
H Value of epidemiologic studies in determining the true incidence of adverse events: the nonsteroidal anti-inflammatory drug story. Arch Intern Med.
1997;1572129- 2136Google ScholarCrossref
P The new biostatistics of resampling. MD Computing.
1995;12115- 121Google Scholar