Assessment of Need for Improved Identification of a Culprit Drug in Stevens-Johnson Syndrome/Toxic Epidermal Necrolysis

Key Points

Question

Is there a need to improve culprit drug identification in Stevens-Johnson syndrome/toxic epidermal necrolysis (SJS/TEN)?

Findings

In this retrospective cohort study of 48 patients with SJS/TEN overlap and toxic epidermal necrolysis, a need was found for improved culprit drug identification, as a clear culprit drug was identified in a minority of cases, drug allergies were overlabeled in most cases, and the culprit drug was potentially missed in a subset of cases. Theoretical incorporation of specific parameters improved identification.

Meaning

The results of this cohort study suggest that a diagnostic test is needed for SJS/TEN culprit drug identification, but in its absence, physicians should incorporate a systematized unbiased approach to identify a culprit drug.

Abstract

Importance

Stevens-Johnson syndrome/toxic epidermal necrolysis (SJS/TEN) is a severe hypersensitivity reaction. Identifying a culprit drug is critical for patient care, yet identification is based on clinical judgment. Data are limited on the accuracy in or approach to identifying a culprit drug.

Objective

To evaluate patient allergy list outcomes, current approaches in identifying culprit drugs, and potential methods of improving culprit drug identification.

Design, Setting, and Participants

This retrospective cohort study spanned 18 years (January 2000 to July 2018), was conducted at Brigham and Women’s Hospital and Massachusetts General Hospital (Boston), and included patients with clinically and histologically confirmed cases of SJS/TEN overlap and TEN.

Main Outcomes and Measures

This study descriptively analyzed potential culprits to SJS/TEN, patients’ allergy lists, and currently used approaches that led to those lists. It then tested the theoretical contribution of incorporating various parameters to allergy list outcomes.

Results

Of 48 patients (29 women [60.4%]; 4 Asian [8.3%], 6 Black [12.5%], 5 Hispanic [10.4%], and 25 White [52.1%] individuals; median age, 40 years [range, 1-82 years]), the mean (SD) number of drugs taken per patient at disease onset was 6.5 (4.7). Physicians labeled 17 patients as allergic to a single culprit drug. Comparatively, 104 drugs were added to allergy lists across all patients. Physicians’ approaches relied largely on heuristic identification of high-notoriety drugs and the timing of drug exposure. Use of a vetted database for drug risk improved sensitivity. Algorithm for Drug Causality for Epidermal Necrolysis scoring was discordant in 28 cases, labeling an additional 9 drugs missed by physicians and clearing 43 drugs labeled as allergens by physicians. Human leukocyte antigen testing could have potentially affected 20 cases. Consideration of infection as a culprit was limited.

Conclusions and Relevance

The results of this cohort study suggest that currently used approaches to identify culprit drugs in SJS/TEN are associated with overlabeling patients allergic to likely nonculprit drugs and less commonly missed possible culprit drugs. Incorporation of a systematized unbiased approach could potentially improve culprit drug identification, although ultimately a diagnostic test is necessary.

This cohort study examines patient allergy list outcomes, current approaches in identifying culprit drugs, and potential methods of improving culprit drug identification for patients with Stevens-Johnson syndrome/toxic epidermal necrolysis.

Introduction

Stevens-Johnson syndrome (SJS)/toxic epidermal necrolysis (TEN) is a severe mucocutaneous desquamative disease with high morbidity and mortality, no clear treatment, and potential for long-term sequelae.^1,2,3 It is categorized by extent of body surface area (BSA)–detached epidermis, with SJS less than 10%, SJS/TEN overlap of 10% to 30%, and TEN greater than 30%.⁴ Stevens-Johnson syndrome/TEN is a delayed-type hypersensitivity reaction that is most commonly drug-induced, with the remainder infection-induced or idiopathic.^5,6,7,8,9

Prompt identification and discontinuation of a culprit drug is critical to improving patient outcomes and preventing recurrence.¹⁰ Identification is difficult because there is no laboratory test or other criterion standard (in the absence of rechallenge) to identify a culprit drug, and patients take on average 6 medications at the time of their reaction.⁷ Consequently, many patients may be labeled as allergic to multiple agents.^11,12,13 While failure to identify a culprit drug could have severe consequences, overlabeling (labeling a patient as allergic to drug(s) they can safely tolerate) is not insignificant. The patient may receive a less efficacious, more toxic, and/or more expensive agent than necessary,^14,15,16 and in some cases may be left without treatment for their underlying disease.

Algorithm for Drug Causality for Epidermal Necrolysis (ALDEN) was published in 2010 to improve culprit drug identification.⁷ ALDEN has shown high sensitivity, strong correlation with case-control analysis,⁷ and the best interrater reliability compared with alternative algorithms.¹⁷ However, it has limitations^17,18 and anecdotally is not widely used. Research during the last decade has generated insights that may aid identification. Human leukocyte antigen (HLA) alleles predisposing to SJS/TEN induced by specific drugs have been increasingly identified since 2004,^19,20,21 although they can vary in prevalence across racial and ethnic populations.²² Data also suggest that reduced drug metabolism or excretion could be associated with an increased risk of SJS/TEN.^9,23 It is unclear to what extent physicians use this new knowledge and how it would contribute to culprit drug identification. This study aimed to investigate whether a need exists to improve culprit drug identification, physicians’ approach(es) to identify culprit drug, and the potential benefit of incorporating additional parameters into their approach.

Methods

This retrospective cohort study of SJS/TEN overlap and TEN at Brigham and Women’s Hospital and Massachusetts General Hospital (Boston) spanned from 2000 to 2018. This study was approved by the institutional review board of Mass General Brigham with exemption from consent and followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.²⁴ A flow diagram for case selection is shown in Figure 1. Potential cases were identified by pathology database searches using the terms Stevens-Johnson Syndrome, Toxic Epidermal Necrolysis, Stevens-Johnson Syndrome/Toxic epidermal necrolysis, SJS, TEN, or SJS/TEN. Each case was diagnosed clinically by a board-certified dermatologist, confirmed pathologically at the time of disease, and then was retrospectively reviewed by a dermatologist with expertise in SJS/TEN (S.J.D.) to exclude any questionable cases. Only cases with a 10% or greater total BSA blistered/detached were included to ensure exclusion of erythema multiforme and Mycoplasma pneumoniae mucositis. Patients with history of hematopoietic cell transplant were excluded to avoid confounding by graft-vs-host disease. Medical record data included patient sex, age at diagnosis, patient-reported race and ethnicity, comorbidities, concurrent infection, prodromal symptoms, heart rate, BSA blistered/sloughed, mucosal involvement (ocular, oropharyngeal, or genital), intubation, reduced liver and kidney function before SJS/TEN onset, systemic treatment, treatment in a burn unit, all medications taken within 3 months of onset (including timing of and prior exposure history), and drugs listed as an allergy. Dermatology clinical assessments and recommendations were recorded that paid particular attention to inclusion and completeness of a drug chart, use of any scoring algorithm and mention of drug half-life, drug-drug interactions, altered drug metabolism, drug notoriety, HLA, and race and ethnicity.

Figure 1. Flow Diagram Detailing Case Identification.

EM indicates erythema multiforme; SJS, Stevens-Johnson syndrome; TBSA, total body surface area.

ALDEN was applied to every drug in every case (the scoring algorithm is described in eTable 1 in Supplement 1).⁷ Drug half-life was obtained from DrugBank,²⁵ drug-drug interactions through Lexicomp,²⁶ drug notorieties through the RegiSCAR 2017 revisions,²⁷ and drug class and similarities through the World Health Organization’s anatomical therapeutic chemical code.²⁸ Index day (day of SJS/TEN onset) was defined as either (1) date of rash onset or (2) date of prodromal symptom onset that within 3 days evolved to skin/mucous membrane erosion or blister formation⁷ and is specified in the Results section when used.

Results

Forty-eight patients with SJS/TEN overlap (n = 26) and TEN (n = 22) were identified, with 29 female individuals (60.4%) and 19 male individuals (39.6%) and a median age of 40 years (range, 1-82 years) (Table 1). Mean (SD) BSA blistered/sloughed was 40% (26.5%). The mean (SD) Severity-of-Illness Score for Toxic Epidermal Necrolysis score was 2.5 (1.2). Forty-one patients (85.4%) had involvement of 2 or more mucosal surfaces. Twenty-seven patients (56.3%) underwent intubation. Forty-one patients (85.4%) were treated in a burn unit, and 40 (83.3%) received 1 or more systemic therapies. Mortality was 27%.

Table 1. Patient Demographic and Clinical Data.

Characteristic	Patient, No. (%) (n = 48)
Median age at diagnosis (range), y	40 (1-82)
Sex
Female	29 (60.4)
Male	19 (39.6)
Patient-reported race and ethnicity
Asian	4 (8.3)
Black	6 (12.5)
Hispanic	5 (10.4)
White	25 (52.1)
NA or Other	8 (16.7)
Clinical characteristics (n = 48)
Mean TBSAa	40% (26.5%)
No. with ≥2 mucosae involvedb	41 (85.4)
No. who underwent intubationc	27 (56.3)
Mean SCORTEN scored	2.5 (1.2)
No. treated in burn unit	41 (85.4)
Systemic treatment (n = 48)e
IVIG only	13 (27.1)
Steroids only	14 (29.2)
IVIG + steroids	10 (20.8)
Cyclosporine	1 (2.1)
Cyclosporine + steroids	1 (2.1)
Cyclosprine + steroids + IVIG	1 (2.1)
Outcome (n = 45)f
Died	12 (26.7)

Abbreviations: IVIG, intravenous immunoglobulin; NA, not available; SCORTEN, Severity-of-Illness Score for Toxic Epidermal Necrolysis; TBSA, total body surface area.

^a Lowest percentage TBSA confirmed.

^b Only oral, ocular, and genital involvement confirmed in notes was counted; if not available in notes, it was counted as 0.

^c Underwent intubation for any reason.

^d SCORTEN data were available for 38 patients.

^e Steroid treatment included cases in which steroids were administered for a reason other than Stevens-Johnson syndrome/toxic epidermal necrolysis (eg, autoimmune disease, chronic obstructive pulmonary disorder).

^f Three patients were lost to follow-up.

Potential Culprit Exposure

We investigated the number of potential triggers of each patient’s reaction. The mean (SD) number of systemic drugs taken during the 3 months before SJS/TEN onset (using rash as the index day) was 6.5 (1.2). Only 5 patients (10.4%) were taking 1 drug, and 14 patients (29.2%) were taking more than 10 drugs (Table 2). This number excluded inhaled medications, topical medications, eye drops, 1 pneumovax injection, 1 intravenous contrast exposure, and 1 longstanding intrauterine device. Although SJS/TEN is classically attributed to drugs, approximately 2% to 40% are reportedly infection-induced or idiopathic.^5,6,7,8,9 Sixteen patients (33.3%) had diagnosed concurrent infection, and 2 (4.2%) had suspected non–M pneumonia infection immediately preceding or concurrent with SJS/TEN onset (Table 2). These patients all received 1 or more medications prior to rash onset. Additionally, 8 patients (16.7%) had features of upper respiratory infection before SJS/TEN skin involvement (Table 2), which could have been initial features of SJS/TEN or symptoms of infection. Upper respiratory infection symptoms in 5 of these patients (10.4%) were presumed infectious by a physician and treated with systemic medication before SJS/TEN rash onset.

Table 2. Systemic Medication Exposure and Infection History Preceding SJS/TEN Rash Onset.

Characteristic	No. of patients (%)
No. of drugs
1	5 (10.4)
2	4 (8.3)
3	7 (14.6)
4-9	18 (37.5)
≥10	14 (29.2)
Diagnosed infection	16 (33.3)
Suspected infection	2 (4.2)
URTI symptoms of unclear etiology	8 (16.7)

Abbreviations: SJS/TEN, Stevens-Johnson syndrome/toxic epidermal necrolysis; URTI, upper respiratory tract infection.

Allergy List Outcomes

All 48 patients had at least 1 drug labeled as an allergy (eTable 2 in Supplement 1). A single culprit drug was labeled in 17 cases, but physicians communicated certainty in only 7. Overall, 104 drugs were labeled as allergies across all 48 cases (eTable 2 in Supplement 1). In 18 cases, multiple antibiotics were labeled as allergies, and in 4 cases, multiple antiepileptics were labeled as allergies. In some cases, entire drug classes were labeled as allergies in addition to the specific drug taken by the patient (eg, β lactams for amoxicillin + clavulanic acid). The dermatology service attributed 1 case to Epstein-Barr virus (EBV) but marked that the patient was allergic to 2 medications regardless.

Physician’s Approach to Identifying a Culprit Drug

Given that each case is triggered by only 1 drug (possibly 0 if infection induced), 48 cases should yield 48 culprit drugs. More than double were labeled as allergies. Therefore, we investigated the approach physicians take to identify culprit drugs. Physicians appeared to use primarily 2 factors: drug notoriety and timing of exposure compared with SJS/TEN onset. Identifying high-risk medications seemed heuristic, with 1 or more drugs in question noted in the record as a common culprit without reference to published or vetted data regarding risk. Regarding timing, drug charts when present in medical records were incomplete, as they focused predominantly on high-notoriety drugs. The index day used was typically the day of rash onset, although this was not always explicitly stated and it was not clear if/how possible prodromal symptoms were considered. Drug half-life was mentioned in 1 case in which the dermatology service specifically noted an antibiotic was unlikely causal given its clearance before disease onset, yet still recommended adding it to the allergy list. There was no other mention of drug half-life, drug metabolism or clearance, or drug-drug interactions in any other case.

Of the 19 cases occurring after 2010, none documented consideration of ALDEN or other scoring systems. Testing with HLA was discussed and tested in 1 case, but testing was only available for an allele that was unlikely to be positive based on the patient’s reported race and ethnicity. The negative result was not discussed in the medical record. A potential role for M pneumoniae and herpes simplex virus (HSV) was commonly acknowledged by the dermatology service. Patients with potential signs/symptoms underwent testing for M pneumoniae (n = 4), HSV (n = 2), or both (n = 3). All patients had negative test results. Of the 18 patients with known preceding/concurrent infection, only 1 who was EBV positive was discussed as potentially relevant in the record, although the patient was still marked as allergic to 2 drugs.

Potential Improvement of Allergy List With Incorporation of Specific Parameters

According to the RegiSCAR database,²⁷ dermatologists identified all high-risk and definite-risk medications, but did not include 8 medications previously reported as culprit drugs in the literature (Table 3). Using rash as the index day, ALDEN-labeled allergy lists were discordant with physician-labeled lists in 28 cases (58.3%; Figure 2). Of the 17 cases (35.4%) that physicians labeled a single culprit drug, ALDEN agreed in 13. ALDEN labeled a single culprit drug in an additional 10 cases, 1 of which was not considered a potential culprit by physicians. ALDEN labeled an additional 9 possible culprit drugs that were not labeled by physicians. Seven of these had notoriety ratings of 0 according to the RegiSCAR database²⁷ but scored highly on ALDEN due to pharmacokinetics and the timing of exposure.

Table 3. Summary of Potential Association With Allergy Lists in Incorporating Specific Parameters.

Factor	No. of drugs potentially removed	No. of drugs potentially addeda
RegiSCAR drug notoriety database	0	8
ALDEN (composite score)b	43	9
Drug pharmacokinetics	17	0
Timing: drug treatment initiated after rash onset	3	0
HLA testing	0	20

Abbreviations: ALDEN, Assessment by Algorithm for Drug Causality for Epidermal Necrolysis; HLA, human leukocyte antigen.

^a In the case of HLA testing, it was the number of drugs with potentially confirmed continued avoidance.

^b Using rash as the index day.

Figure 2. Results of Theoretical Assessment by Algorithm for Drug Causality for Epidermal Necrolysis (ALDEN) Compared With Physician Assessment Using Rash Day as the Index Day.

ALDEN-labeled allergy lists were discordant on 28 cases compared with physician-labeled allergy lists. ALDEN narrowed allergy lists in 20 cases; labeling an additional 10 cases as single-culprit drugs. It labeled drugs as allergies that were not labeled as allergies by physicians in 9 cases. The number of cases is shown in each box.

Forty-three drugs labeled as allergies by physicians scored “unlikely” by ALDEN (Table 3). Seventeen of these were secondary to drug half-life. Drug-drug interactions and altered metabolism were identified, although they did not change outcomes for any drug in any case. Time since initial drug exposure to index day exonerated 3 drugs when rash onset was used as the index day. Compared with using rash onset as the index day, a repeated ALDEN analysis using the date of prodrome onset that within 3 days evolved to skin/mucous membrane erosion or blister formation⁷ altered ALDEN results in 4 cases (8.3%; eTable 2 in Supplement 2). Twenty-three drugs were eliminated based on the composite ALDEN score (rather than individual ALDEN parameters). ALDEN excluded all drugs in 5 cases.

Twenty-eight cases (58.3%) had 1 or more drugs added to the allergy list that had a published HLA allele or other gene variation associated with SJS/TEN (eTable 3 in Supplement 1). Given the severity of SJS/TEN, a positive test result supports continued drug avoidance and can thus be helpful in cases for which the culprit drug is unclear.²⁹ There were 20 cases for which 1 of multiple drugs added to the allergy list had an associated HLA allele and could have potentially benefitted from testing (Table 3).

Discussion

It is important to identify a culprit drug in SJS/TEN as accurately as possible. In the absence of a criterion standard laboratory test, the onus is on clinicians to make that determination. This study’s data suggest that culprit drug identification is commonly very challenging. Patients were taking several drugs during the period preceding SJS/TEN, often more than 1 of high notoriety and fitting the timeline typically considered applicable. Moreover, many patients had confirmed concurrent or possible infection, which potentially could cause SJS/TEN.^5,6,7,8,9

Identifying high-risk drugs appeared predominantly heuristic, although dermatologists identified all high-risk (score 3) and definite-risk (score 2) drugs according to the RegiSCAR database. Scores of −1, 2, and 3 are based on case-controlled studies so potentially provide higher-quality evidence for or against a drug’s potential culpability.²⁷ Eight drugs previously reported in the literature as causal in SJS/TEN as score 1 were not included in physician-labeled allergy lists. Thus, incorporating a vetted database may improve sensitivity, although prior reporting in the literature does not necessarily equal risk, as it displays reporting bias. Indeed, ALDEN identified several drugs as potential culprits despite being previously unpublished.

Physicians used the timing of drug exposure to exculpate medications administered after disease onset. Controversy regarding the definition of index day potentially complicates this approach, although analysis using 2 different definitions of index day affected only 4 cases in this study. Although changing the index day used did not change the outcome dramatically in this study, it is likely best to consider the index day definition for each patient. Timing is similarly important in identifying whether recently discontinued medications could be culprit. Retrospective incorporation of drug half-life theoretically exonerated 17 drugs. Although drug-drug interactions and reduced metabolism/clearance are additional factors that affect drug levels, they did not alter the allergy lists in the present cohort. This analysis did not consider metabolite-induced disease, which is not included in ALDEN, but could affect pharmacokinetics/clearance of antigen. Increased drug levels can potentially trigger SJS/TEN.^9,23 To our knowledge, there are limited data in this regard, and ALDEN does not account for it, but it may be prudent to reconsider the culpability of a long-standing drug if there was potential recent perturbation of its level.

Physicians largely recognized a potential role for HSV and M pneumoniae in disease. Testing was only performed for a few patients, presumably because the patients had signs and symptoms of those infections. Consideration of infection was limited to only 1 other case in which EBV was positive before the onset of SJS/TEN (yet the patient was marked as allergic to 2 drugs). Notably, ALDEN does not account for infectious etiologies. It may be that more patients have an infectious etiology that goes undetected because most patients are taking multiple medications at the time of their reaction, and medications are preferentially deemed as culprit. Complicating matters further, it is possible that concurrent infection triggers drug-induced SJS/TEN.³⁰ Further research is necessary in this domain.

Physicians erred on the side of caution as to not risk reexposure to a potential culprit drug, which can have severe consequences for the patient. Despite this, potential culprit drugs may have been missed when compared with ALDEN, although in the absence of a criterion standard test, we cannot know for sure. Comparatively, most patients were overlabeled on their allergy lists. The consequences of overlabeling can be significant for patients and the medical system.^31,32 Overlabeling likely occurs in other forms of drug hypersensitivity reactions as well, further magnifying the problem.

Limitations

This study had limitations. First, the retrospective approach limits available data to those explicitly recorded in the medical record. It is possible that the dermatology team did create more complete drug charts, search a vetted list of high-risk drugs, performed ALDEN, or discussed HLA testing but did not document it in the record. The allergy lists overall argue against this, although this is certainly not definitive. Second, the study spanned from 2000 to 2018, with many cases predating the research advancements tested in this study. Although we analyzed a theoretical benefit of incorporating these advancements, a large prospective study is necessary to best test this. Regarding generalizability, the demographic and clinical data parallel prior studies, including female predominance,³³ mortality rate,³⁴ and average number of drugs taken at time of reaction.⁷ Additionally, the number of patients with SJS/TEN overlap and TEN mirrors that of a prior unrelated study at our institution.³⁵ Whether physician approach at identifying culprit drugs is generalizable across institutions within the US and across countries is unknown. The 2 study hospitals are tertiary care centers with long-standing inpatient consultative dermatology services, with several consult attending physicians having trained at other inpatient/consultative–heavy dermatology departments. Therefore, we suspect that the findings of this study would be reproducible across many other institutions within the US, although to what extent is unclear. In other countries, the approaches used by physicians and/or the effect of the tested parameters could be markedly affected not only by training and standard of care, but also by population characteristics (common drug culprits could vary with underlying disease burden and HLA prevalence) and health care resources (which could affect access to infection and genetic testing).

There are also practical limitations to incorporating some of the tested components into clinical practice. Gene and HLA variations and infection testing may have limited availability and can be cost-prohibitive. Clear data on the use of HLA testing regarding causality in SJS/TEN are limited, although if a patient has a positive test result for a high-risk allele of a potential culprit drug, that drug should be avoided going forward. Moreover, the prevalence of particular HLA alleles can vary by race and ethnicity, affecting the cost-effectiveness of testing. It may also be impractical to incorporate the previously described parameters fully at initial presentation. There does appear to be a need for closer follow-up postdischarge with a physician experienced in SJS/TEN.³⁶ Review of potential culprit drugs and allergy list revision could be incorporated into that follow-up and would benefit from detailed medical record documentation.

Conclusions

The results of this cohort study underscore the need for a laboratory test to identify culprit drugs. In its absence, a systematic unbiased approach, such as ALDEN or the RegiSCAR database, with possibly HLA testing, should be considered to ensure the true culprit drug is not missed and exonerate as many nonculprit drugs as possible.

Supplement 1.

eTable 1. ALDEN scoring system modified from the original

eTable 2. Physician-labeled vs ALDEN-labeled allergy lists

eTable 3. Drug:HLA associations in SJS/TEN by ethnicity

eReferences

Supplement 2.

Data sharing statement