Seasonal Variations in Drug-Related Erythema Multiforme: A Time Series Analysis Using the JADER Database

Abstract

Erythema multiforme (EM) is a rare, immune-mediated skin condition triggered primarily by infections, drugs, and autoimmune diseases. Although its seasonal variations have been reported, with peaks in spring and summer, comprehensive analyses remain limited. In this study, we aimed to investigate the seasonal patterns and drug associations of EM using the Japanese Adverse Drug Event Report database. A time series analysis was performed using EM-related adverse event onset dates reported between January 2005 and December 2019. The periodic patterns and residual autocorrelation were evaluated by applying seasonal and trend decomposition using loess (STL) and autocorrelation function (ACF) analyses to the time series data of monthly EM reports. A total of 3843 cases of EM were analyzed, with 43.1% being male and 56.3% female. STL decomposition identified June as the peak month for EM cases, with a higher incidence observed from spring to summer than in winter. Male patients exhibited greater seasonal variations, with a higher incidence in summer. Reports of anti-infective and antineoplastic drugs increased from spring to summer and declined in winter in male patients with EM, suggesting a potential seasonal trend. Female patients with EM also showed seasonal variations, albeit less pronounced than in male patients. The current study identified seasonal variations in drug-related EM, with reports peaking in June, which is consistent with previous observations of higher EM incidence in spring and summer. The findings indicated potential sex-related disparities in drug-related EM, underscoring the necessity for additional research to elucidate its mechanisms and patterns.

INTRODUCTION

Erythema multiforme (EM) results from an immune-mediated reaction that primarily affects the skin and sometimes the mucosa.^1–5) Lesions usually erupt over a 3- to 5-d period and, in some cases, produce a mild pruritus or a burning sensation.^6,7) EM lesions present symmetrically on the extremities (especially on the extensor surfaces) and spread centripetally. The prognosis of mild cases of EM is typically favorable, with spontaneous recovery observed in many instances. EM, Stevens–Johnson syndrome (SJS), and toxic epidermal necrolysis (TEN) often have overlapping clinical presentations. Owing to this, EM was previously considered to be on the same pathological spectrum as SJS and TEN; however, it is now recognized as a distinct disease.⁷⁾

EM is primarily caused by cell-mediated immune responses, with the main causes being infections, drugs, vaccines, and autoimmune diseases.^1–7) In some instances, the cause cannot be identified; however, most cases of EM are a consequence of infectious diseases, predominantly herpes simplex virus type 1 (HSV-1), followed by Mycoplasma pneumoniae, the latter being particularly prevalent in children.^8–10) Many drugs have been associated with EM, the most common ones being nonsteroidal anti-inflammatory drugs, antiepileptics, and antibiotics.^1–5) Antibiotics associated with EM include sulfonamides, penicillins, erythromycin, nitrofurantoin, and tetracyclines. Other medications include barbiturates, phenothiazines, statins, and tumor necrosis factor-α inhibitors.^1–5) EM has been associated with autoimmune diseases, such as inflammatory bowel disease¹¹⁾ and systemic lupus erythematosus,¹²⁾ as well as malignancies, specifically leukemia and lymphoma, though less commonly.¹⁰⁾ Additionally, certain HLA alleles could have a genetic predisposition to EM.¹³⁾

EM affects less than 1% of the population^4–6) and is most common in young adults (younger than 40 years).^3–6) To date, no association with race^4,5) and no consistent evidence of sex-related differences have been reported.^4,6,14,15) There have been several reports on the seasonality of EM. Sakurai reported that, in Japan, EM with distinct facial and extremity skin lesions mainly occurs in winter, whereas EM with skin lesions confined to the extremities most frequently occurs in summer.¹⁶⁾ Brito et al. reported that hospitalizations due to EM in Brazil are more frequent during summer.¹⁷⁾ In a retrospective review of electronic medical records from the mid-Atlantic region of the United States, Harvell and Selig found that EM occurred more frequently in spring and summer.¹⁸⁾ Currently, evidence on sex-related differences and the seasonality of EM is limited, emphasizing the need for comprehensive investigations to better understand its temporal patterns.

Japan experiences significant temperature variations and seasonal changes throughout the year, which can lead to associated changes in certain medical conditions. Previous studies have utilized spontaneous reporting databases to explore seasonal variations in adverse events (AEs). The Japanese Adverse Drug Event Report (JADER) database made public by the Pharmaceuticals and Medical Devices Agency (PMDA),¹⁹⁾ includes detailed information on the dates of AEs and offers an opportunity for time series analysis. For example, Matsumoto et al. conducted a retrospective analysis of seasonal variation in AEs associated with SGLT2 inhibitors in the JADER database.²⁰⁾ Nakao et al. identified seasonal variations in drug-related photosensitivity using the JADER database.²¹⁾ Similarly, Marrero et al. analyzed the FDA Adverse Event Reporting System to uncover seasonal and regional variations in photosensitivity.²²⁾

The current study aimed to clarify the seasonal patterns in EM reports, providing new insights into the triggers, particularly sex-related differences, and temporal trends associated with this rare condition.

MATERIALS AND METHODS

Data Source

The JADER database, from April 2004 to September 2024, was downloaded from the PMDA and the Japanese Regulatory Authority website.²⁰⁾ All JADER data were cleaned and anonymized by the PMDA. The database consists of 4 data tables: 1) patient demographic information (DEMO), 2) drug information (DRUG), 3) AEs (REAC), and 4) primary illness (HIST). The JADER database does not contain the codes for identifying case reports (A1.11); therefore, we could not exclude duplicate patient cases (https://www.pmda.go.jp/files/000145474.pdf). In the DRUG table, the causality of each drug is assigned a code according to its association with the AEs, such as a “suspected drug,” “concomitant drug,” or “interacting drug.” Our analysis was limited to cases that had been reported as suspected drug cases. In the time series analysis, reports of AE onsets from January 2005 to December 2019 were used. Data from 2020 onwards were excluded to minimize potential bias due to the impact of coronavirus disease 2019 and the overreporting of AEs related to the severe acute respiratory syndrome coronavirus 2 RNA vaccine (Supplementary Fig. S1). Patients with AEs but with unknown onset dates were excluded.

Monthly temperature and relative humidity data were obtained from the Japan Meteorological Agency website, which provides detailed statistics on climate and weather patterns in Japan.²³⁾ The Anatomical Therapeutic Chemical (ATC) classification system was used to categorize drugs based on their anatomical and therapeutic properties.²⁴⁾

Approach to Duplicate Case Reports in the JADER Database

As previously alluded, the JADER database does not contain codes for identifying case reports (A1.11); therefore, we could not exclude duplicate patient cases (https://www.pmda.go.jp/files/000145474.pdf). We evaluated potential duplicate reports in the JADER database by examining the consistency of patient demographics and report characteristics. Key variables used for comparison included age, sex, and date of AE onset. An exact match was defined as a case in which age, sex, body weight, suspected drug, and reason for use were all identical. A partial match was defined as a case in which age, sex, and weight were completely matched, and the suspected drug and reason for use were not identical but showed similarity. For both exact and partial matches, we further analyzed the data by dividing cases into those with identical AE onset dates and those with differing onset dates. The similarity between suspected drugs and reasons for use was evaluated using the Jaro–Winkler similarity metric, a well-established method for approximate string matching.²⁵⁾ A similarity score of 1.0 was interpreted as an exact match, while scores between 0.7 and less than 1.0 were interpreted as partial matches.

Definition of AEs

AEs in the JADER database were defined based on the Medical Dictionary for Regulatory Activities (MedDRA) version 27.0.²⁶⁾ For case extraction, we focused on the preferred terms (PTs) of EM (PT code: 10015218).

Seasonal Decomposition and Trend Analysis Using Locally Estimated Scatterplot Smoothing (STL)

Monthly time series data on AEs from January 2005 to December 2019 were analyzed using STL to examine long-term trends, seasonal patterns, and residual variations.²⁷⁾ The decomposition was conducted using Python’s seasonal decomposition function from the statsmodels library,^28,29) employing an additive decomposition model with a 12-month cycle. In this method, each observed value (Y_t) is expressed as the sum of 3 components:

  

$$Y_t=T_t+S_t+R_t$$

where T_t is the trend component representing the long-term progression of the data; S_t is the seasonal component capturing periodic fluctuations within a 12-month cycle; and R_t the residual component reflecting random noise and unexplained variations. The trend component (T_t) obtained from the decomposition was analyzed to identify long-term patterns in the data. To evaluate the statistical significance of the observed trends, the Mann–Kendall test, a nonparametric method that is robust against missing values and non-normality, was applied. The seasonal component (S_t) was analyzed to investigate the periodic patterns, with the peak month defined as the month with the highest seasonal value and identified to determine the month with the highest frequency of AEs. The residual component (R_t) was analyzed to assess whether it exhibited characteristics of white noise. The Ljung–Box test was applied to the residuals at multiple lags, including 12 and 24 months, to evaluate autocorrelation.

Analysis of Periodicity Using an Autocorrelation Function (ACF)

Monthly time series data on AEs from January 2005 to December 2019 were analyzed to assess seasonality using an ACF.³⁰⁾ The analysis was conducted using Python’s statsmodels library.^28,29) The ACF was computed to examine the correlation of AE frequencies at different time lags, with a particular focus on periodic patterns and seasonality.³⁰⁾ ACF plots were generated to visually inspect the patterns, allowing the identification of any recurring cycle in the data. To statistically evaluate autocorrelation, the Ljung–Box test was applied to the ACF values at lags of 12 and 24 months to assess the significance of the observed correlations and determine the presence of seasonal periodicity.

Assessment of Actual Prescription Data and JADER Reports

To compare actual prescription volumes with the number of reports in the JADER database, we used the Japan Medical Data Center (JMDC) claims database from April 2014 to March 2022.³¹⁾ The analysis was conducted on patients with neoplasms, including those who were also treated with antibiotics, defined by International Classification of Diseases, Tenth Revision (ICD-10) major category codes C00–D48.³²⁾ We investigated the use of sorafenib, amoxicillin, and amoxicillin/clavulanic acid. It is important to note that the number of prescriptions for amoxicillin and amoxicillin/clavulanic acid only includes cases with ICD-10 major category codes C00–D48, and not the total number of prescriptions for all cases.

From the JADER database, we extracted reports in which anticancer agents were listed as suspected drugs. We defined cases corresponding to ICD-10 major category codes C00–D48 as cases in which anticancer drugs were reported as the suspected drug. The study using the JMDC claims database included only reports on patients with cancer, and the observation period was from April 2014 to March 2022, consistent with insurance claims data. Pearson’s correlation coefficient was calculated to assess the linear association between monthly prescription counts and monthly spontaneous reports for each drug.³³⁾

RESULTS

The JADER database began to collect data in 2004. The initial period of the database, when the number of reports was limited, was excluded from the analysis. Further, patients lacking an identified onset date of AEs were excluded. This resulted in a time series analysis of a 15-year period from January 2005 to December 2019. A total of 3843 reports of EM were reported, of which 3821 (99.4%) had information on sex, with 1656 (43.1%) being male and 2165 (56.3%) female. Moreover, 3783 reports (98.5%) had information on age; 2130 reports (56.3%) were individuals 60 years or older and 1653 reports (43.7%) were those aged <60 years.

Potential duplicate entries within the JADER database were evaluated. Of 3843 EM reports, 327 had identical onset dates and were classified as exact matches, 205 reports (5.3%), or partial matches, 122 reports (3.2%). Four hundred and sixty-six reports had differing onset dates and were classified as exact matches, 178 reports (4.6%), or partial matches, 288 reports (7.5%), based on Jaro–Winkler similarity scores (Supplementary Table S1).

The data presented in Fig. 1 indicate monthly patterns in EM reports and the corresponding temperatures in Tokyo. Figure 1a shows the monthly number of EM reports reported from 2005 to 2019, which varied with the average temperatures recorded in Tokyo during the same period. Figure 1b summarizes the trends by aggregating the total number of EM reports per month over 15 years. The highest number of EM reports, including both males and females, was reported in June (435 reports); the numbers were lowest in January (233 reports). August recorded the highest average temperature (27.9°C), while January recorded the lowest (5.9°C).

Fig. 1. (a) Monthly Trends in EM Reports and the Proportion of EM Relative to Total Reports from 2005 to 2019; (b) Monthly Distribution of EM Reports, Average Temperature, and Relative Humidity; (c) Monthly Reporting Ratio (EM Reports/Total Reports), Average Temperature, and Relative Humidity

EM: erythema multiforme.

The data were subjected to STL decomposition to separate the components into trends, seasonal variations, and residuals (Table 1 and Fig. 2). The Mann–Kendall test applied to the trend component of all cases yielded a Z-value of 3.73 and a p-value of 0.00019, indicating a statistically significant upward trend in EM cases. Stratified analyses for male and female cases demonstrated statistically significant upward trends, with Z-values of 3.06 (p = 0.00224) and 6.05 (p = 1.44 × 10⁻⁹), respectively.

Table 1. Statistical Parameters from Seasonal-Trend Decomposition Using Locally Estimated Scatterplot Smoothing (STL) of Monthly Erythema Multiforme (EM) Reports, Including Ljung–Box Test Results for Residuals and Overall Autocorrelation

Category	Counts (EM/total, %)	Seasonal component from STL			Trend component from STL		Residual component from STL		Autocorrelation function
		Peak montha)	Trough montha)	Amplitudeb)	Z-valuec)	p-Valuec)	p-Value (lag 12)d)	p-Value (lag 24)d)	p-Value (lag 12)e)	p-Value (lag 24)e)
All	3843/655521 (0.6)	6	1	13.521	3.794	>0.001	>0.001	0.002	>0.001	>0.001
Male	1656/334175 (0.5)	6	1	7.158	3.074	0.002	>0.001	0.003	>0.001	>0.001
Female	2165/315622 (0.7)	6	2	6.973	6.067	>0.001	>0.001	>0.001	>0.001	>0.001

a) Peak month and trough month refer to the months with the highest and lowest values, respectively, in the seasonal component derived by STL analysis. b) Amplitude represents the difference, in terms of the monthly report count, between the maximum and minimum values of the seasonal component derived by STL analysis. c) The Z-value and p-value were obtained from the Mann–Kendall test applied to the trend component derived by STL analysis. d) The p-values were obtained from the Ljung–Box test (lag 12 and lag 24), which was applied to the residual component derived by STL analysis. e) The p-values were obtained from the Ljung–Box test (lag 12 and lag 24), based on the autocorrelation function for the entire time series of counts of monthly EM reports.

Fig. 2. Seasonal and Trend Decomposition Using Locally Estimated Scatterplot Smoothing (STL)

Panel (a) shows the seasonal decomposition of aggregated data (all cases), while panels (b) and (c) show the seasonal decomposition for male and female cases, respectively. From top to bottom, the panels display the original data, the long-term trend, the seasonal component, and the residuals. The Mann–Kendall test Z-values were 3.73 (p = 0.00019) for all cases, 3.06 (p = 0.00224) for males, and 6.05 (p = 1.44 × 10⁻⁹) for females; Ljung–Box p-values for the residuals were 1.00 × 10⁻⁴ (lag 12) and 0.00195 (lag 24) for all cases, 2.14 × 10⁻⁴ (lag 12) and 0.00304 (lag 24) for males, and 7.10 × 10⁻⁶ (lag 12) and 9.06 × 10⁻⁶ (lag 24) for females. The decomposition analysis results are summarized in Table 1.

Seasonal decomposition further confirmed June as the peak month for EM cases, as identified by the seasonal component capturing periodic fluctuations in the 15-year dataset. The residual component of STL decomposition, which represents the noise remaining after removing the trend and seasonal components, was analyzed using the Ljung–Box test to evaluate the white noise characteristics. For all cases, the Ljung–Box p-values at lags of 12 and 24 were 1.00 × 10⁻⁴ and 0.00195, respectively, indicating some degree of autocorrelation in the residual component. For male cases, the Ljung–Box p-values were 2.14 × 10⁻⁴ (lag 12) and 0.00304 (lag 24), while for female cases, they were 7.10 × 10⁻⁶ (lag 12) and 9.06 × 10⁻⁶ (lag 24). These results indicate a stronger autocorrelation in the residual component of female patients than that of male patients.

ACF analysis was conducted to investigate the periodic patterns in the residual series (Table 1 and Fig. 3). For all cases, the ACF values were 0.333 at the 12-month lag and 0.158 at the 24-month lag, indicating the peaks at these intervals, consistent with a 12-month cycle. The Ljung–Box p-values at lags of 12 and 24 were 7.73 × 10⁻²⁸ and 1.49 × 10⁻³³, respectively, indicating statistically significant autocorrelation at these intervals. In male patients, the ACF values were 0.168 at the 12-month lag and 0.022 at the 24-month lag. The Ljung–Box p-values at lags of 12 and 24 were 7.48 × 10⁻¹³ and 2.04 × 10⁻¹¹, respectively, also demonstrating significant autocorrelation. For female cases, the ACF values were 0.291 at the 12-month lag and 0.159 at the 24-month lag, with Ljung–Box p-values of 2.99 × 10⁻²⁰ and 6.14 × 10⁻³¹, respectively. While the Ljung–Box test suggested a stronger statistical autocorrelation in female patients, the overall ACF plot indicated male patients as possibly exhibiting more consistent periodic patterns.

Fig. 3. ACF Plot for Monthly Erythema Multiforme Reports

Panel (a) shows the ACF plot for the aggregated data (all cases), while panels (b) and (c) show the ACF plots for male and female cases, respectively. The plots display autocorrelation coefficients across different time lags, with shaded areas indicating the 95% confidence intervals. For all cases, the Ljung–Box p-values at lags of 12 and 24 were 7.73 × 10⁻²⁸ and 1.49 × 10⁻³³, respectively; for male cases, the Ljung–Box p-values were 7.48 × 10⁻¹³ and 2.04 × 10⁻¹¹, respectively; for female cases, the Ljung–Box p-values were 2.99 × 10⁻²⁰ and 6.14 × 10⁻³¹, respectively. The decomposition analysis results are summarized in Table 1. ACF: autocorrelation function.

A total of 7195 compounds were reported as suspected drugs for EM, with 3130 (43.5%) reported in males and 4035 (56.1%) reported in females (Table 2). Among the ATC classification categories, category J (anti-infectives for systemic use) was the most frequently reported, with 1886 drugs, including 857 (45.4%) in males and 1023 (54.2%) in females. The second most frequently reported category was category L (antineoplastic and immunomodulatory agents), with 1524 drugs reported, including 728 (47.8%) in males and 780 (51.2%) in females.

Table 2. The Suspected Drugs Reported in Relation to Erythema Multiforme Are Categorized According to the ATC Classification and Presented Separately for Males and Females

ATC classification	Category details	All counts	Male counts	Male ratio (%)	Female counts	Female ratio (%)
	Total	7195	3130	43.5	4035	56.1
A	Alimentary tract and metabolism	703	285	40.5	418	59.5
B	Blood and blood-forming organs	186	83	44.6	103	55.4
C	Cardiovascular system	339	142	41.9	197	58.1
D	Dermatologicals	78	35	44.9	41	52.6
G	Genitourinary system and sex hormones	53	24	45.3	29	54.7
H	Systemic hormonal preparations, excluding sex hormones and insulins	74	39	52.7	35	47.3
J	Anti-infectives for systemic use	1886	857	45.4	1023	54.2
L	Antineoplastic and immunomodulating agents	1524	728	47.8	780	51.2
M	Musculoskeletal system	626	240	38.3	386	61.7
N	Nervous system	824	285	34.6	538	65.3
P	Antiparasitic products, insecticides, and repellents	25	6	24.0	19	76.0
R	Respiratory system	262	129	49.2	130	49.6
S	Sensory organs	3	1	33.3	2	66.7
V	Various	35	17	48.6	18	51.4
	Others (non-ATC-classified drugs)	577	259	44.9	316	54.8

Within these categories, some specific drugs were identified as major contributors. Sorafenib was the most frequently reported drug in ATC category L (392 reports) (Table 3); 213 (54.3%) and 176 (44.9%) reports were reported in males and females, respectively. The second most frequently reported drug was celecoxib in category M (musculoskeletal system), with 232 reports, including 65 men (28.0%) and 167 women (72.0%). The third most frequently reported drug was amoxicillin in category J, with 211 reports, including 76 men (36.0%) and 135 women (64.0%).

Table 3. The Top 20 Drugs Reported to Be Associated with Erythema Multiforme, with Counts and Percentages Provided Separately for Males and Females

Drug	ATC code	All counts	Male counts	Male ratio (%)	Female counts	Female ratio (%)
Total		7195	3130	43.5	4035	56.1
Sorafenib	L01EX02	392	213	54.3	176	44.9
Celecoxib	M01AH01	232	65	28.0	167	72.0
Amoxicillin	J01CA04	211	76	36.0	135	64.0
Ribavirin	J05AP01	172	98	57.0	74	43.0
Lamotrigine	N03AX09	165	46	27.9	119	72.1
Clarithromycin	J01FA09	155	52	33.5	103	66.5
Peginterferon alfa-2b	L03AB04	151	90	59.6	61	40.4
Telaprevir	J05AP02	144	89	61.8	55	38.2
Loxoprofen	M02AA31	113	49	43.4	64	56.6
Acetaminophen	N02BE01	111	50	45.0	61	55.0
Carbamazepine	N03AF01	107	49	45.8	57	53.3
Nivolumab	L01FF01	107	64	59.8	42	39.3
Regorafenib	L01EX05	95	12	12.6	82	86.3
Lansoprazole	A02BC03	90	27	30.0	63	70.0
Allopurinol	M04AA01	74	48	64.9	26	35.1
Rebamipide	A02BX14	72	26	36.1	46	63.9
Carbocisteine	R05CB03	72	30	41.7	41	56.9
Sulfamethoxazole/trimethoprim	J01EE01	64	29	45.3	35	54.7
Itraconazole	J02AC02	62	28	45.2	34	54.8
Amoxicillin/clavulanic acid	J01CR02	60	30	50.0	29	48.3
Others		4546	1959	43.1	2565	56.4

The correlation between the actual prescription composition of the JMDC claims database and the number of JADER reports was calculated. The monthly number of spontaneous reports of sorafenib, amoxicillin, and amoxicillin/clavulanic acid in patients with cancer from the JADER database showed no significant correlation with monthly prescription counts obtained from the insurance claims database (Pearson’s r: −0.0415, p = 0.898) (Supplementary Table S2).

DISCUSSION

This study was the first attempt to analyze seasonal variations in drug-related EM using the JADER database. Prior literature indicates an elevated prevalence of EM during the spring and summer months.^17,18) EM remains a rare AE, and robust evidence regarding its seasonal patterns is limited. Through a 15-year chronological analysis of monthly EM reports (January 2005–December 2019), we applied STL decomposition and ACF analysis to thoroughly investigate the potential seasonal trends, leveraging the time series nature of the data.

The results of monthly aggregates indicated June as having the highest number of EM reports. This aligned with the seasonal component of STL decomposition, which also identified June as the peak. In contrast, the lowest number of reports was recorded in January, which corresponds to the winter season with the lowest annual temperatures. These findings highlighted the potential impact of seasonal variables, such as temperature, on the observed outcomes.

Sunlight exposure is known to influence the development and/or progression of EM. Photodistributed erythema multiforme is a subtype of photoallergy caused by drugs such as ketoprofen and itraconazole, characterized by lesions confined to sun-exposed areas, with a clear detachment from unexposed areas.³⁴⁾ Recurrent herpes-associated EM can be triggered by sun exposure, similar to recurrent HSV.⁸⁾ Pathogenetically, the etiologies of herpes-associated EM and drug-related EM are different; however, they are indistinguishable clinically.³⁵⁾

Seasonal patterns differed between sexes. Male patients exhibited a distinct seasonal pattern with a 12-month periodicity, showing an elevated incidence during the summer months and a nadir during winter. Female patients showed less seasonal variation compared to male patients. This discrepancy could be attributable to the characteristics of the reported drugs. In ATC category J, sex-related differences were observed in the seasonality of EM reports (Supplementary Table S3 and Supplementary Figs. S2 and S3). Although seasonality was not detected through seasonal decomposition or ACF evaluation in ATC category L, trends were observed in the shapes of the ACF plots (Supplementary Table S3 and Supplementary Figs. S4 and S5). The drugs included in these categories, such as antibiotics, antiviral agents, antineoplastic drugs, immunostimulants, and immunosuppressants, possibly contribute to the seasonality of drug-related EM. Moreover, discrepancies in drug usage patterns between male and female patients could account for the observed differences in seasonal trends.

Trends were observed in the aggregated data for specific drugs (Supplementary Table S4); however, the number of reports for individual drugs was insufficient for time series analyses, such as STL decomposition or ACF analysis, limiting the evaluation of drug-specific seasonal patterns.

Although identifying the underlying diseases is difficult, the JADER database includes fields for indications of suspected drugs. The frequently reported indications for suspected drugs included chronic hepatitis C, nasopharyngitis, liver cancer, and renal cell carcinoma in male patients and Helicobacter infection, chronic hepatitis C, nasopharyngitis, and epilepsy in female patients. These findings suggested that sex-specific differences in the underlying conditions might contribute to the observed variation in drug usage and, consequently, to seasonal trends in EM reports.

Importantly, the majority of EM reports are attributed to infectious diseases. Infections, including HSV-1, are major triggers of EM.^1–5) The differences in HSV-1 infection pathology between male and female patients remain unclear, since the prevalence of HSV-1 in the trigeminal ganglia does not appear to differ significantly by sex, with 88% of females and 90% of males testing positive for HSV-1.³⁶⁾ As reported by the WHO, approximately 3.8 billion people under the age of 50, or 64.2% of the global population, were infected with HSV-1 as of 2020, with the majority of these infections occurring during childhood.³⁷⁾ In contrast, HSV-2, another reported trigger of EM, is transmitted more efficiently from male to female individuals, resulting in a higher infection rate in females.³⁷⁾ Sex differences in susceptibility and severity of viral infections, including those caused by influenza A, Epstein–Barr virus, hepatitis B and C, and HSV, have been attributed to the influence of sex hormones on immune cell function and cytokine synthesis.^38–40) The mechanisms underlying HSV-associated EM are gradually being elucidated. However, in the context of EM following HSV infection, HSV detection is not always successful,³⁾ thereby complicating the clinical assessment of EM etiology. It is thought that herpes simplex encephalitis is not seasonal,^41,42) and there is no evidence to suggest that there is a seasonal pattern in HSV infection or reactivation.

Despite our comprehensive analysis, the precise reasons for the observed male-specific seasonality of EM remain unclear. Although the potential influence of sex differences on the underlying conditions and drug use patterns was considered, these factors alone could not fully explain the observed variation. To understand the causal relationships better, future research involving electronic medical records and claims databases is essential.

Examining the interplay between drug-related EM and seasonal triggers, such as infections or environmental factors, could yield additional insights. Understanding these factors would be critical to improve our knowledge of drug-related EM and formulate targeted preventive strategies.

The current study findings offer valuable insights into seasonal variations of drug-related EM and demonstrate the importance of integrating diverse analytical methods to explore the complex patterns in AE data. The findings provide a foundation for future research and contribute to our understanding of EM triggers.

LIMITATIONS

This study has inherent limitations related to spontaneous reporting systems, such as overreporting, underreporting, missing data, exclusion of healthy subjects (lack of a control population or reference group), and the presence of confounding factors. External factors, such as safety communications and market dynamics, may also influence reporting patterns. Future research using more granular data, including detailed patient demographics, clinical conditions, and environmental factors, is required to uncover the mechanisms underlying these findings. EM is more common in young people,^3–6) and there is a concern that this study might have underreported this population group. This study may reflect the incidence of infectious diseases; although reported as drug-related, potential infectious diseases may be included as a cause of EM. However, our research method did not allow us to clarify whether the patient had a history of an infectious disease.

The number of JADER cases analyzed in this study was larger in the elderly (59–79 years) than in the young (10–49 years) (Supplementary Table S5). Seasonal patterns of EM were observed for both younger and older age groups (Supplementary Table S6 and Supplementary Figs. S6 and S7). June did not uniformly account for the highest number or proportion of reports across the fiscal years spanning 2005 to 2019 (Supplementary Table S7); therefore, we performed a time series analysis to identify cyclical trends in the number and ratio of monthly reports. It is an important and interesting question whether each drug is used under similar conditions in Japan in a fiscal year; nonetheless, we did not examine this aspect in this study. A possible method to determine and discuss trends in drug prescribing using past JMDC claims data should thus be considered in the future.

Although the number of EM reports submitted each month may be influenced by fluctuations in the total number of reports in the JADER database, we consider such seasonal variation in overall reporting volume to be minimal (Supplementary Table S7). Therefore, the seasonal trend observed in this study—characterized by a peak in June—is considered to be meaningful and not merely an artifact of overall reporting patterns. Furthermore, as a result of performing STL and ACF analysis on the monthly reporting ratio data, the same trend was obtained as with the simple EM report count data (Supplementary Table S6 and Supplementary Figs. S8 and S9). The volume of drug prescriptions, including agents such as antibiotics, varies seasonally, which may influence the number of EM reports; although we could not entirely eliminate this concern, we confirmed a low correlation for some drugs—specifically, there was little correlation between actual prescriptions and JADER reports for sorafenib, the most frequently reported EM suspected drug (Supplementary Table S2)—and, regarding antibiotics, the seasonal fluctuations in prescriptions in Japan differed from the seasonal pattern of EM reports in JADER.⁴³⁾ However, a more detailed analysis is required in the future.

An inherent limitation in working with JADER data is the potential for duplicate reporting, which cannot be entirely avoided. Although the JADER database is anonymized and therefore not accurate, we estimated the degree of agreement based on the background of patients with EM, the drugs used, and the reasons for use. Among the EM reports, 18.1% showed a perfect match for sex, age, weight, suspected drug, and reason for use, while 10.7% showed a perfect match for sex, age, and weight, but only a partial match for suspected drug and reason for use (Supplementary Table S1). Given the inherent challenges in accurately distinguishing duplicate reports, the analysis was conducted using unadjusted report frequencies.

We present the monthly average temperature for Tokyo in Fig. 1a. However, in Japan, temperatures vary considerably by region, even within the same month. For example, the temperature in Tokyo in May can be similar to that in Okinawa (located in the south of Japan) in January or February. The geographic locations of the JADER cases are unknown; therefore, this study does not directly demonstrate an association with temperature.

The contributors only report AEs according to the International Council for Harmonisation E2B, the international safety reporting guidelines, and rely on the definitions provided by MedDRA. It was difficult to confirm the criteria used for EM events by volunteers at the time of reporting. However, the reports in the JADER database have been included by healthcare “professionals.” Additionally, our analysis was limited to cases in which the drug was designated as a “suspected drug,” which may have excluded relevant cases where causality was not clearly established. While our findings suggest potential associations between certain drugs and EM, further validation through clinical or pharmacoepidemiological studies is warranted.

CONCLUSION

Drug-related EM exhibits seasonal variations, particularly in male patients. EM reports in males demonstrated a seasonal pattern with a higher incidence during the summer months; conversely, those in females showed minimal seasonal variation, indicating a year-round occurrence. The observed differences between male and female patients may have been influenced by the reporting of specific drug categories, including antibiotics, antiviral agents, antineoplastic drugs, immunostimulants, and immunosuppressants. However, the JADER database has inherent limitations, such as incomplete reporting of underlying medical conditions and concomitant medications. Further studies using more detailed datasets are necessary to validate these findings and explore the mechanisms driving the observed sex-specific and seasonal variations in EM.

Acknowledgments

This research was partially supported by JSPS KAKENHI (Grant Numbers: 21K06646 and 21K11100).

Conflict of Interest

The authors declare no conflict of interest.

Supplementary Materials

This article contains supplementary materials.

REFERENCES

• 1)  Soares A, Sokumbi O. Recent updates in the treatment of erythema multiforme. Medicina (Kaunas), 57, 921 (2021).
• 2)  MSD manual professional version. “Erythema Multiforme.”: <https://www.msdmanuals.com/professional/dermatologic-disorders/hypersensitivity-and-reactive-skin-disorders/erythema-multiforme>, accessed 16 April, 2025.
• 3)  Lamoreux MR, Sternbach MR, Hsu WT. Erythema multiforme. Am. Fam. Physician, 74, 1883–1888 (2006).
• 4)  DermNet. “Erythema Multiforme.”: <https://dermnetnz.org/topics/erythema-multiforme>, accessed 16 April, 2025.
• 5)  Trayes KP, Love G, Studdiford JS. Erythema multiforme: recognition and management. Am. Fam. Physician, 100, 82–88 (2019).
• 6)  Sokumbi O, Wetter DA. Clinical features, diagnosis, and treatment of erythema multiforme: a review for the practicing dermatologist. Int. J. Dermatol., 51, 889–902 (2012).
• 7)  Paulino L, Hamblin DJ, Osondu N, Amini R. Variants of erythema multiforme: a case report and literature review. Cureus, 10, e3459 (2018).
• 8)  Aurelian L, Ono F, Burnett J. Herpes simplex virus (HSV)-associated erythema multiforme (HAEM): a viral disease with an autoimmune component. Dermatol. Online J., 9, 1 (2003).
• 9)  Staikūnienė J, Stanevičiūtė J. Long-term valacyclovir treatment and immune modulation for herpes-associated erythema multiforme. Cent. Eur. J. Immunol., 40, 387–390 (2015).
• 10)  Siedner-Weintraub Y, Gross I, David A, Reif S, Molho-Pessach V. Paediatric erythema multiforme: epidemiological, clinical and laboratory characteristics. Acta Derm. Venereol., 97, 489–492 (2017).
• 11)  Chapman RS, Forsyth A, MacQueen A. Erythema multiforme in association with active ulcerative colitis and Crohn’s disease. Dermatology, 154, 32–38 (1977).
• 12)  Yachoui R, Cronin PM. Systemic lupus erythematosus associated with erythema multiforme-like lesions. Case Rep. Rheumatol., 2013, 212145 (2013).
• 13)  Ohtani T, Deguchi M, Aiba S. Erythema multiforme-like lesions associated with lesional infiltration of tumor cells occurring with adult T-cell lymphoma/leukemia. Int. J. Dermatol., 47, 390–392 (2008).
• 14)  Heinze A, Tollefson M, Holland KE, Chiu YE. Characteristics of pediatric recurrent erythema multiforme. Pediatr. Dermatol., 35, 97–103 (2018).
• 15)  Shabahang L. Characteristics of adult outpatients with erythema multiforme. Pak. J. Biol. Sci., 13, 1106–1109 (2010).
• 16)  Sakurai M. Erythema multiforme in children: unusual clinical features with seasonal occurrence. J. Dermatol., 16, 361–368 (1989).
• 17)  Brito LAR, Nascimento ACM, de Marque C, Miot HA. Seasonality of the hospitalizations at a dermatologic ward (2007–2017). An. Bras. Dermatol., 93, 755–758 (2018).
• 18)  Harvell JD, Selig DJ. Seasonal variations in dermatologic and dermatopathologic diagnoses: a retrospective 15-year analysis of dermatopathologic data. Int. J. Dermatol., 55, 1115–1118 (2016).
• 19)  Pharmaceuticals and Medical Devices Agency (PMDA). “Suspected adverse drug reaction information.”: <https://www.pmda.go.jp/safety/info-services/drugs/adr-info/suspected-adr/0003.html>, accessed 16 April, 2025.
• 20)  Matsumoto K, Goto F, Maezawa M, Nakao S, Miyasaka K, Hirofuji S, Shiota K, Ichihara N, Yamashita M, Nokura Y, Yamazaki T, Sugishita K, Tanaka H, Tamaki H, Ishiguro M, Iguchi K, Nakamura M. A retrospective study of seasonal variation in sodium-glucose co-transporter 2 inhibitor-related adverse events using the Japanese adverse drug event report database. Sci. Rep., 14, 30072 (2024).
• 21)  Nakao S, Hatahira H, Sasaoka S, Hasegawa S, Motooka Y, Ueda N, Abe J, Fukuda A, Naganuma M, Kanoh H, Seishima M, Ishiguro M, Kinosada Y, Nakamura M. Evaluation of drug-induced photosensitivity using the Japanese Adverse Drug Event Report (JADER) database. Biol. Pharm. Bull., 40, 2158–2165 (2017).
• 22)  Marrero O, Hung EY, Hauben M. Seasonal and geographic variation in adverse event reporting. Drugs Real World Outcomes, 3, 297–306 (2016).
• 23)  Japan Meteorological Agency. “Monthly statistical data.”: <https://www.data.jma.go.jp/obd/stats/etrn/view/monthly_s3.php?prec_no=44&block_no=47662>, accessed 16 April, 2025.
• 24)  World Health Organization. “ATC/DDD toolkit: ATC classification.”: <https://www.who.int/tools/atc-ddd-toolkit/atc-classification>, accessed 16 April, 2025.
• 25)  Winkler WE. String comparator metrics and enhanced decision rules in the Fellegi-Sunter model of record linkage. Proceedings of the Section on Survey Research Methods, American Statistical Association, Alexandria, Virginia, U.S.A., pp.354–359 (1990).
• 26)  International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use (ICH). “Medical Dictionary for Regulatory Activities (MedDRA).”: <https://www.ich.org/page/meddra>, accessed 16 April, 2025.
• 27)  Cleveland RB, Cleveland WS, McRae JE, Terpenning I. STL: a seasonal-trend decomposition procedure based on LOESS. J. Off. Stat., 6, 3–73 (1990).
• 28)  Seabold S, Perktold J. Statsmodels: econometric and statistical modeling with python. “Proceedings of the 9th Python in Science Conference.”: <https://www.statsmodels.org/>, accessed 16 April, 2025.
• 29)  Pal A, Prakash PKS. Practical time-series analysis: master time series data processing, visualization, and modeling using python. 1st ed., Packt, Birmingham, UK, pp. 7–45 (2017).
• 30)  Hyndman RJ, Athanasopoulos G. Forecasting: principles and practice. 3rd ed., OTexts, Melbourne, Australia (2021). <https://otexts.com/fpp3>, accessed 16 April, 2025.
• 31)  JMDC. “JMDC Claims Database.”: <https://www.jmdc.co.jp/en/jmdc-claims-database/>, accessed 16 April, 2025.
• 32)  World Health Organization (WHO). “International Statistical Classification of Diseases and Related Health Problems (10th Revision).”: <https://www.who.int/classifications/icd/en/>, accessed 16 April, 2025.
• 33)  Pearson K. Notes on regression and inheritance in the case of two parents. Proc. R. Soc. Lond., 58, 240–242 (1895).
• 34)  Di Bartolomeo L, Irrera N, Campo GM, Borgia F, Motolese A, Vaccaro F, Squadrito F, Altavilla D, Condorelli AG, Motolese A, Vaccaro M. Drug-induced photosensitivity: clinical types of phototoxicity and photoallergy and pathogenetic mechanisms. Front. Allergy, 3, 876695 (2022).
• 35)  Shivhare P, Haidry N, Sharma A, Agrawal D, Gupta A, Subramanyam S. Drug-induced erythema multiforme. Case Rep. Dermatol. Med., 2023, 8706006 (2023).
• 36)  Hill JM, Ball MJ, Neumann DM, Azcuy AM, Bhattacharjee PS, Bouhanik S, Clement C, Lukiw WJ, Foster TP, Kumar M, Kaufman HE, Thompson HW. The high prevalence of herpes simplex virus type 1 DNA in human trigeminal ganglia is not a function of age or gender. J. Virol., 82, 8230–8234 (2008).
• 37)  World Health Organization. “Herpes simplex virus.”: <https://www.who.int/news-room/fact-sheets/detail/herpes-simplex-virus>, accessed 16 April, 2025.
• 38)  Klein SL. Sex influences immune responses to viruses, and efficacy of prophylaxis and treatments for viral diseases. BioEssays, 34, 1050–1059 (2012).
• 39)  Geurs TL, Hill EB, Lippold DM, French AR. Sex differences in murine susceptibility to systemic viral infections. J. Autoimmun., 38, J245–J253 (2012).
• 40)  Lundberg P, Welander P, Openshaw H, Nalbandian C, Edwards C, Moldawer L, Cantin E. A locus on mouse chromosome 6 that determines resistance to herpes simplex virus also influences reactivation, while an unlinked locus augments resistance of female mice. J. Virol., 77, 11661–11673 (2003).
• 41)  Whitley RJ, Soong SJ, Linneman C Jr, Liu C, Pazin G, Alford CA. Herpes simplex encephalitis. Clinical Assessment. JAMA, 247, 317–320 (1982).
• 42)  Sköldenberg B, Alestig K, Burman L, Forkman A, Lövgren K, Norrby R, Stiernstedt G, Forsgren M, Bergström T, Dahlqvist E, Frydén A, Norlin K, Olding-Stenkvist I, Uhnoo I, De Vahl K. Acyclovir versus vidarabine in herpes simplex encephalitis. Randomised multicentre study in consecutive Swedish patients. Lancet, 324, 707–711 (1984).
• 43)  Sugawara T, Ohnichi Y, Gu Y, Kawanohara H, Taniguchi K, Okabe N. Study on antibiotic consumption based on pharmacy surveillance. Japanese Journal of Environmental Infections, 27, 195–198 (2012).