Association Between Drug Use and Subsequent Diagnosis of Lupus Erythematosus

Key Points

Question  What is the association between drug use and a subsequent diagnosis of cutaneous lupus erythematosus or systemic lupus erythematosus?

Findings  In this nationwide case-control study, many significant associations were found between the use of a variety of drugs and a subsequent diagnosis of cutaneous lupus erythematosus or systemic lupus erythematosus. Relatively well-established causal associations together with some new plausible causal associations were observed.

Meaning  The findings indicate that physicians must be aware that a newly diagnosed case of cutaneous lupus erythematosus or systemic lupus erythematosus could be drug induced.

Importance  It has been estimated that up to 30% of all subacute cutaneous lupus erythematosus (CLE) cases and up to 15% of systemic lupus erythematosus (SLE) cases are drug induced. Based on numerous case reports and several epidemiologic studies, more than 100 drugs from more than 10 drug classes are suspected to cause drug-induced lupus erythematosus.

Objective  To examine the association between drug use and a subsequent diagnosis of CLE or SLE based on a systematic screening process of the drugs in the Anatomical Therapeutic Chemical classification system in a nationwide setting.

Design, Setting, and Participants  A matched case-control study was conducted using all incident cases of CLE and SLE registered in the Danish National Patient Register between January 1, 2000, and December 31, 2017. Patients with CLE and patients with SLE were matched (1:10) on age and sex, with individuals from the general population serving as controls.

Exposures  To select which drugs to examine for an association with CLE or SLE, a screening process of all drugs was performed, including drugs filled at pharmacies and drugs administered in hospitals.

Main Outcomes and Measures  Odds ratios (ORs) were calculated for the association between exposures to certain drugs and the subsequent diagnosis of CLE or SLE.

Results  In all, 3148 patients with CLE (n = 1298; 1022 women [78.7%]; median age at diagnosis, 50.5 years [interquartile range, 39.4-62.2 years]) or SLE (n = 1850; 1537 women [83.1%]; median age at diagnosis, 45.0 years [interquartile range, 33.6-56.4 years]) and 31 480 controls (25 590 women [81.3%]; median age, 47.5 years [interquartile range, 35.9-59.5 years]) were found. Many significant associations between drug use and a subsequent diagnosis of CLE and SLE were observed. Many associations were likely due to protopathic bias. However, new plausible causal associations were observed between CLE or SLE and some drugs, including fexofenadine hydrochloride (SLE: OR, 2.61 [95% CI, 1.80-3.80]; CLE: OR, 5.05 [95% CI, 3.51-7.26]), levothyroxine sodium (SLE: OR, 2.46 [95% CI, 1.97-3.07]; CLE: OR, 1.30 [95% CI, 0.96-1.75]), metoclopramide hydrochloride (SLE: OR, 3.38 [95% CI, 2.47-4.64]; CLE: OR, 1.47 [95% CI, 0.85-2.54]), and metronidazole hydrochloride (SLE: OR, 1.57 [95% CI, 1.09-2.27]; CLE: OR, 1.93 [95% CI, 1.25-2.97]).

Conclusions and Relevance  The study’s findings suggest that physicians should be cognizant about whether a new case of CLE or SLE could be drug induced. Furthermore, the results highlight that the reported associations in the published literature may be due to publication or protopathic bias.

Drug-induced lupus erythematosus (DILE) is best defined as an adverse reaction of treatment with a certain drug taken continuously over time, inducing lupus-like symptoms. Symptoms must resolve when treatment is stopped to meet the definition of DILE.¹ Most authors agree on this definition of DILE, yet no explicit clinical or serologic criteria for DILE exist, to our knowledge.

Drug-induced lupus erythematosus occurs owing to a complex process of drug-induced breakdown of self-tolerance, rather than an allergic drug reaction, and prolonged exposure to the suspected drug is usually required for the condition to develop.² Currently, more than 100 drugs are suspected to induce DILE.³ The list of drugs continues to grow as new drugs arise and physicians become more aware of DILE.

To our knowledge, no study has examined the association between drug use and a subsequent diagnosis of systemic lupus erythematosus (SLE) or cutaneous lupus erythematosus (CLE) based on a systematic screening process of the drugs in a large-scale setting. We therefore performed a drug-screening study based on the Anatomical Therapeutic Chemical (ATC) classification system using nationwide administrative prescription data from Denmark.

This case-control study was conducted using Danish administrative registers, which are unique because they cover the entire country, date far back in time, and enable cross-linkage using an individual Civil Personal Register number assigned to all residents at birth or immigration.⁴ In addition, the Danish health care system provides free and unencumbered health care for all residents. An additional description of the data sources can be found in the eMethods in the Supplement. The current study was approved by the Danish Data Protection Agency. Ethical approval is not required for register studies in Denmark. No informed consent was obtained, as the civil personal register number is encrypted in the registers, making each citizen appear anonymous.

A nationwide observational case-control study was performed to examine the association between drug use and subsequent diagnosis of lupus erythematosus (LE). Patients were 18 years or older with a first-time diagnosis of CLE or SLE during the study period from January 1, 2000, until December 31, 2017 (see the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision [ICD-10] codes used in eTable 1 in the Supplement). Only incident cases were included to ensure that the specific drug preceded the LE diagnosis. The index date was the date of LE diagnosis. Cases were matched on age and sex in a 1:10 ratio with individuals from the general population in Denmark serving as controls. The index date for the controls was the same as the index date for the corresponding cases.

Exposure for cases and controls was defined as recorded use of a particular drug prior to the index date. For the purpose of selecting which drugs to use in the subsequent analyses, a screening process for drug exposure in cases was conducted. All drugs filled at all Danish pharmacies were included, together with all registered treatments from hospitals (eg, drugs dispensed for outpatient treatment, such as antineoplastic drugs). If at least 100 cases were registered or prescribed a specific drug or drug class 12 months prior to the index date, the specific drug or drug class was included in the following analyses. This was done to avoid spurious associations randomly occurring based on only a small number of cases. The screening process was performed according to the levels in the ATC classification system. The structure in the ATC classification system is built on a hierarchy of 5 different levels, which is described in detail in the eMethods in the Supplement. A detailed list of the screened drug categories is found in eTable 2 in the Supplement.

Conditional logistic regression analyses were performed calculating odds ratios (ORs) representing the relative risk of being treated with a specific drug among the cases compared with the controls. Exposure for cases and controls was ascertained if a registration of the suspected drug was performed during the period 6 months prior to index date.

Stratification according to disease (CLE and SLE) and subtypes (subacute CLE [SCLE] and discoid LE) was performed, owing to clinical relevance, as different drugs are thought to induce CLE and SLE. Data management and statistical analyses were performed using SAS, version 9.4 (SAS Institute Inc) and Stata/MP, version 15 (StataCorp LLC).

To ensure accuracy of the diagnosis of CLE or SLE, we used a stricter definition of these patients by including only patients with CLE or SLE diagnosed from the dermatology, nephrology, or rheumatology departments. In another analysis, we included only patients with an ICD-10 code for drug-induced SLE.

The duration of exposure was shortened to 3 months prior to the index date, to ensure that patients were still treated with the drug on the date of diagnosis, as most drugs are prescribed in a 3-month supply. In addition, the duration of exposure was altered to handle long-term exposure, defined in 3 different ways: (1) long-term intermittent drug exposure was defined as at least 1 registration during the period between the index date and 12 months prior to the index date, (2) long-term persistent drug use was described as a registration every third month for at least 12 months, and (3) long-term exposure was defined as a registration every sixth month for at least 2 years. In post hoc analyses, we tried to eliminate the delay between the first symptom and the final diagnosis by defining the duration of exposure as the period between 3 months prior to the index date and 12 months prior to the index date.

In post hoc analyses, we calculated the ORs of being treated with tumor necrosis factor (TNF) inhibitors and terbinafine before receiving a diagnosis of LE. These drugs initially failed to meet the study inclusion criteria; however, since it is relatively well established that these drugs may cause LE,^3,5 they were included to serve as a positive outcome and aid in interpretation of the observed effect sizes of the other study results.

The study population consisted of 3148 patients with CLE (n = 1298; 1022 women [78.7%]; median age at diagnosis, 50.5 years [interquartile range, 39.4-62.2 years]) or SLE (n = 1850; 1537 women [83.1%]; median age at diagnosis, 45.0 years [interquartile range, 33.6-56.4 years]) and 31 480 matched controls (25 590 women [81.3%]; median age, 47.5 years [interquartile range, 35.9-59.5 years]) from the general population in Denmark. The characteristics of the cases and controls are presented in Table 1.⁶

The overall association between drug exposure and subsequent diagnosis of CLE or SLE is illustrated in Figure 1. At the fourth level in the ATC classification system, 46 drug classes were analyzed for the association between drug use 6 months prior to the index date and a subsequent diagnosis of LE (Figure 2 and Figure 3).

The associations between drug use and subsequent diagnosis of CLE or SLE are further detailed in Table 2,^1,3,5,7-11 where a comment is added on whether the association is causal or more likely due to bias. In the eFigure in the Supplement, the drugs with the strongest association for patients with LE are presented, with the top 5 being hydroxychloroquine sulfate, prednisolone sodium phosphate, folic acid, methotrexate sodium, and warfarin sodium. All these drugs are associated with the treatment of LE. New plausible associations were also observed between LE and some drugs, including fexofenadine hydrochloride (SLE: OR, 2.61 [95% CI, 1.80-3.80]; CLE: OR, 5.05 [95% CI, 3.51-7.26]), levothyroxine sodium (SLE: OR, 2.46 [95% CI, 1.97-3.07]; CLE: OR, 1.30 [95% CI, 0.96-1.75]), metoclopramide hydrochloride (SLE: OR, 3.38 [95% CI, 2.47-4.64]; CLE: OR, 1.47 [95% CI, 0.85-2.54]), and metronidazole hydrochloride (SLE: OR, 1.57 [95% CI, 1.09-2.27]; CLE: OR, 1.93 [95% CI, 1.25-2.97]). When stratifying according to subtype, we found 19 significant associations between drug classes and SCLE and 31 significant associations for discoid LE (eTable 3 in the Supplement).

The potentially new, significant associations that we found were metoclopramide hydrochloride and levothyroxine sodium for SLE and metronidazole hydrochloride and fexofenadine hydrochloride for both CLE and SLE. Overall, the results of these drugs were not altered in the sensitivity analyses; the association for fexofenadine hydrochloride was even stronger when stratifying CLE in SCLE (OR, 8.88 [95% CI, 4.87-16.21]). In sensitivity analyses, the duration of exposure was altered to handle long-term exposure. Persistent long-term exposure was associated with fewer exposed cases and controls, resulting in several insignificant relative risk estimates (eTable 4 in the Supplement). When the duration of exposure was altered to long-term intermittent, no major changes were seen (eTable 4 in the Supplement). In another analysis, the exposure period was shortened; however, the overall findings were not altered (eTable 4 in the Supplement). Changing the exposure period to 3 months prior to the index date and to end 12 months prior to the index date was associated with more exposed cases and controls and a trend that the relative risk estimates were slightly lower (eTable 4 in the Supplement). Stricter definitions of CLE and SLE resulted in 2529 cases, and the relative risk estimates were found to be almost the same as in our main analysis (eTable 5 in the Supplement). Finally, the analyses were repeated for patients with a diagnosis of drug-induced SLE, which resulted in only 11 significant associations (eTable 6 in the Supplement). In the post hoc analyses for TNF inhibitors and terbinafine, increased ORs were found (terbinafine: OR, 4.11 [95% CI, 2.63-6.41] for CLE and 1.85 [95% CI, 2.63-6.41] for SLE; TNF inhibitors: OR, 4.96 [95% CI, 2.30-10.71] for CLE and 3.64 [95% CI, 1.93-6.89] for SLE).

We observed many significant associations between the use of a variety of drugs and a subsequent diagnosis of LE. Our findings raise the important question of whether the suspected drugs reported in the growing number of published case reports have an increased risk of inducing DILE or whether they are reflecting publication or protopathic bias. However, in our study, new plausible causal associations were observed between LE and some drugs, including fexofenadine hydrochloride, levothyroxine sodium, metoclopramide hydrochloride, and metronidazole hydrochloride.

In this study, there is a potentially high risk of protopathic bias (ie, the risk of being prescribed certain drugs owing to early symptoms of LE [eg, being prescribed nonsteroidal anti-inflammatory drugs for pain in the joints]). Therefore, for each drug, it must be considered whether causality for inducing DILE is plausible or the association is a sign of treating early symptoms or comorbidities.

A Swedish register–based study also investigated the association between drug exposure and subsequent diagnosis of SCLE³; however, the study included only 234 patients with SCLE, which resulted in relatively few significant associations and large 95% CIs. Another case-control study investigated the association between suspected drugs and risk of SLE or SCLE, but it also lacked power owing to few exposed cases.¹²

The potentially new significant associations that we found were metoclopramide hydrochloride and levothyroxine sodium for SLE and metronidazole hydrochloride and fexofenadine hydrochloride for both CLE and SLE. Overall, the results of these drugs were not altered in the sensitivity analyses; the association for fexofenadine hydrochloride was even stronger when stratifying CLE in SCLE (OR, 8.88 [95% CI, 4.87-16.21]). To our knowledge, these drugs have not been previously reported as inducing DILE, and protopathic bias is less likely. However, hypothyroidism is reported in 4% of patients with SLE,¹³ which could partially explain the association between levothyroxine sodium and SLE.

Several limitations need to be considered when interpreting this study. As with all observational studies, we cannot conclude on causality, even though only incident cases were included. Moreover, the risk of protopathic bias was high in our study. This kind of bias can be difficult to avoid; however, we tried to avoid protopathic bias by conducting a sensitivity analysis that excluded the first 3 months prior to the index date. However, because the onset of symptoms of SLE may occur up to several years before final diagnosis,¹⁴ the problem of protopathic bias is not necessarily eliminated, which is a concern of the study. Although drug-induced SCLE is common, only a few drugs are reported to induce chronic CLE.¹⁵ In addition, we investigated only first-time diagnoses of CLE or SLE and did not take into account whether patients with CLE progressed to SLE later.

In line with authoritative recommendations,¹⁶ no adjustment for multiple comparisons was performed because this was an explorative screening study investigating new potential exposures. The screening process was also an attempt to limit the amount of comparisons. We tried to use terbinafine and TNF inhibitors to test the robustness of our results. Increased relative risks were found for both drugs and with a stronger association for CLE, which may indicate a causal association. However, terbinafine could also be prescribed more frequently for patients with CLE because the rash could be misinterpreted as a fungal infection. Lupus erythematosus can also be exacerbated by drugs¹⁷; however, in this study, we investigated drug exposure only before the first diagnosis of LE, which is why extrapolation of our results to include drug exacerbation should be performed with caution.

This study also has some strengths. By Danish law, it is required that all prescriptions filled from Danish community pharmacies are registered in the Danish National Prescription Register; therefore, no bias regarding drug use must be present.¹⁸ All in-hospital medication use was included in the study as well. Another strength was the use of nationwide registers, resulting in many cases.

Many significant associations between drug use and a subsequent diagnosis of CLE or SLE were observed. The list of drugs reported to induce DILE is constantly changing as new drugs arise. Therefore, it is important that physicians are updated on the changing drug list and remain aware of potential new drug associations, but at the same time, they must be aware of publication and protopathic bias.

Accepted for Publication: June 5, 2020.

Corresponding Author: Jeanette Halskou Haugaard, MD, Department of Dermatology, Allergy, and Venerology, Herlev and Gentofte University Hospital, Gentofte Hospitalsvej 15, First Floor, 2900 Hellerup, Denmark (jeanette.halskou.haugaard@regionh.dk).

Published Online: September 2, 2020. doi:10.1001/jamadermatol.2020.2786

Author Contributions: Drs Haugaard and Egeberg had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Drs Dreyer and Egeberg are co–last authors.

Concept and design: Haugaard, Kofoed, Dreyer, Egeberg.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Haugaard, Kofoed.

Critical revision of the manuscript for important intellectual content: Kofoed, Gislason, Dreyer, Egeberg.

Statistical analysis: Haugaard.

Obtained funding: Haugaard, Kofoed, Dreyer, Egeberg.

Administrative, technical, or material support: Kofoed, Gislason, Egeberg.

Supervision: Kofoed, Gislason, Dreyer, Egeberg.

Conflict of Interest Disclosures: Dr Kofoed reported receiving personal fees from Leo Pharma, AbbVie, Takeda Pharma, Bristol-Myers Squibb, Galderma Nordic, and Eli Lilly outside the submitted work. Dr Dreyer reported receiving grants from Bristol-Myers Squibb outside the submitted work. Dr Egeberg reported receiving grants and personal fees from Pfizer, Eli Lilly, AbbVie, and Novartis; grants from Danish National Psoriasis Foundation and Kgl Hofbundtmager Aage Bang Foundation; and personal fees from Bristol-Myers Squibb, Leo Pharma, Samsung Bioepis Co Ltd, Galderma, and Janssen Pharmaceuticals during the conduct of the study. No other disclosures were reported.