Clinical differentiation between acute generalized exanthematous pustulosis and generalized pustular psoriasis: A multi-center study

To the Editor: Acute generalized exanthematous pustulosis (AGEP) and generalized pustular psoriasis (GPP) both present with widespread sterile pustules, sometimes forming large pustular lakes, and share overlapping clinical and histological features.[1,2] However, they differ significantly in cause and prevalence. AGEP is a rare, severe drug-induced reaction mediated by T cells, affecting 1–5 people per million. In contrast, GPP is a rare but potentially life-threatening form of psoriasis, observed in about 14 individuals per million. Approximately 50% of GPP patients have pre-existing psoriasis vulgaris (PsV), and many endure delayed diagnosis due to clinical similarity, leading to prolonged symptoms and impaired quality of life. Although genetic analyses reveal minimal molecular distinctions between these entities (PDK4 and ZBTB16),[3] their distinct triggers and disease courses make accurate diagnosis essential for proper treatment. The 2024 International Psoriasis Expert Consensus emphasized the importance of clinical diagnosis for GPP, removing the need for mandatory histopathology to allow for quicker identification.[4] In our study, we performed a comprehensive comparison of the clinical features, laboratory markers, and treatment responses in AGEP and GPP patient groups. Additionally, we evaluated the diagnostic accuracy of two models—(1) univariate binary logistic regression and (2) Gaussian Naive Bayes. This combined clinical and computational approach offers clinicians both practical diagnostic criteria and validated tools to improve differential diagnosis and guide treatment decisions for these clinically similar but pathogenetically distinct pustular diseases. This retrospective study included 132 hospitalized patients (93 with GPP and 39 with AGEP), treated at the Affiliated Hospital of Xuzhou Medical University and the First People’s Hospital of Lianyungang between December 2018 and October 2024. Collected data included age, gender, disease duration, history of PsV, triggering factors, combined symptoms, quality of life, pain and itch scores, medication use, laboratory results, and histopathological findings. All diagnoses were made by experienced dermatologists using characteristic clinical features, laboratory data, and pathology. The study was approved by the Medical Ethics Committee of Xuzhou Medical University (No. XYFY2024-564-01). All participants signed an informed consent form. Data were analyzed using SPSS Statistics software (version 27.0, IBM Corp, Armonk, USA). Categorical variables were presented as percentages, while continuous variables were expressed as mean with standard deviation. For continuous data, comparisons were made using the independent samples t-test, Mann–Whitney U-test, or one-way analysis of variance (ANOVA). Categorical variables were analyzed using the chi-square test or Fisher’s exact test. A P-value of <0.05 was considered statistically significant. To identify clinical characteristics with significant group differences, univariate binary logistic regression was used to calculate odds ratios (ORs) for variables that showed significance in the chi-squared or Fisher’s tests. To assess the predictive performance of the classification model, Gaussian Naive Bayes was implemented using Python (version 3.8.0). Evaluation metrics included the confusion matrix and the area under the receiver operating characteristic curve (area under curve [AUC]). To reduce the influence of prior information on the results, the prior distribution was set as P (Y = 0) = 0.3300 and P (Y = 1) = 0.6712,[5] based on literature reporting the incidence rate. Posterior probabilities were computed according to Bayes’ theorem: P(Y = k|X)∝P(Y = k)∏i=1nP(Xi|Y = k) , where Y represents the grouping variable (GPP = 0, AGEP = 1) and Xi denotes the feature selected after dimensionality reduction using mutual information and random forest feature importance. Model performance was validated using five-fold cross-validation, and metrics such as accuracy and AUC were used to evaluate generalizability. Compared to patients with GPP, those with AGEP were generally older at disease onset, had shorter disease durations, and incurred lower hospitalization costs. Statistically significant differences (P <0.05) were observed between the two groups in age, triggers, family history of psoriasis, joint and nail involvement, presence of fever, disease course, and medical expenses. Interestingly, 82.1% (32/39) of AGEP cases initially developed lesions on the trunk and skin folds—a distribution pattern that differed significantly from that seen in GPP [Supplementary Table 1, https://links.lww.com/CM9/C608]. Laboratory findings also revealed notable differences: GPP patients had significantly higher rates of elevated alkaline phosphatase (ALP), reduced calcium levels, and hypoproteinemia, along with more frequent elevations in aspartate aminotransferase (AST). Inflammatory markers such as procalcitonin (PCT) and C-reactive protein (CRP) were also significantly higher in GPP compared to AGEP (P <0.01). After treatment, AGEP patients experienced less pain, pruritus, and quality-of-life impairment, suggesting a better overall prognosis [Supplementary Table 2, https://links.lww.com/CM9/C608]. Our study identified key distinctions between GPP and AGEP, with all AGEP cases (100%) linked to drug-induced triggers. Univariate binary logistic regression revealed that a history of psoriasis (OR = 153.55, P <0.001), recurrent disease episodes (OR = 82.71, P <0.001), and hypocalcemia (OR = 11.75, P <0.001) were the strongest predictors of GPP. Elevated ALP (OR = 7.57, P <0.001) and hypoalbuminemia (OR = 2.28, P = 0.034) were also associated with GPP, though to a lesser extent. While CRP levels differed significantly between groups (P = 0.025), their predictive value was minimal, with ORs close to 1 (GPP: OR = 1.01; AGEP: OR = 0.99) [Supplementary Table 3, https://links.lww.com/CM9/C608]. Using a mutual information threshold >0.2 and feature importance rankings from a random forest model, five key predictors were selected: triggering factors, comorbid psoriasis, history of relapse, hypocalcemia, and elevated ALP. A Gaussian Naive Bayes model built on these features achieved high accuracy (0.948; 95% confidence interval [CI]: 0.899–0.967). The classification results from this model aligned closely with those from the univariate logistic regression, confirming the robustness and consistency of these predictive indicators [Supplementary Figure 1, https://links.lww.com/CM9/C608]. Among patients with GPP, 47.3% (44/93) required additional treatment with low-dose glucocorticoids (0.5–1 mg/kg per day) or immunosuppressants (cyclosporine or methotrexate) due to inadequate response to retinoids or persistent fever. In contrast, for patients with AGEP, suspected causative drugs were discontinued immediately upon admission. Systemic glucocorticoids (1–1.5 mg/kg per day) were started, and immunosuppressants were used as needed based on clinical presentation. AGEP patients showed rapid clinical improvement, with body temperature normalizing in an average of 22.3 ± 6.2 hours and pustules typically drying within two days, reflecting a significantly shorter recovery time than GPP. At late follow-up, 81.7% (76/93) of GPP patients reported recurrence after discharge [Supplementary Table 4 and Supplementary Figures 2 and 3, https://links.lww.com/CM9/C608]. Although previous studies have compared AGEP and GPP in terms of histopathology and gene expression, definitive diagnostic criteria to distinguish the two conditions are still missing. Recent research has even proposed that AGEP may represent a subtype of GPP within the spectrum of Interleukin-36 (IL-36)-mediated sterile pustulosis.[3] However, genetic and histopathological testing is often time-consuming and costly, which may delay diagnosis and treatment. Therefore, this study aimed to identify distinguishing features between AGEP and GPP using available clinical signs and laboratory data. By applying both univariate binary logistic regression and Gaussian Naive Bayes models, we validated key diagnostic indicators to improve diagnostic accuracy and support clinical decision-making. It was previously shown that AGEP typically presents with a sudden onset, short disease duration, and polymorphic skin lesions following recent drug exposure. These lesions often resolve spontaneously after the offending drug is stopped, with a low risk of recurrence and minimal association with PsV or arthritis. In our cohort, all 39 patients with AGEP developed widespread erythematous macules with pustules after drug exposure. Only 5.1% (2/39) had a prior history of PsV, and all recovered completely within two weeks after discontinuation of the triggering drug and treatment with glucocorticoids. Notably, none of the patients with AGEP showed joint or nail involvement. In contrast, GPP is a distinct subtype of psoriasis that is closely associated with PsV and driven by the IL-17/IL-23 inflammatory pathway. In our study, 62.4% (58/93) of patients with GPP exhibited joint, mucosal, or nail involvement, highlighting key clinical differences between the two conditions. Liver dysfunction is frequently observed in patients with GPP, with previous studies reporting that up to 90% show abnormalities in one or more liver enzymes, often correlating with the severity of skin involvement.[5] In our study, the incidence of elevated ALP in patients with GPP exceeded typical levels reported in the literature. In contrast, AGEP, being a drug-induced hypersensitivity reaction that rarely affects internal organs, showed significantly fewer liver enzyme abnormalities.[2] Patients with GPP also exhibited elevated levels of PCT and CRP, along with increased rates of hypocalcemia and hypoproteinemia. These abnormalities likely reflect the intense systemic inflammation characteristic of GPP, as well as its chronic nature, long-term pharmacologic treatment, or disrupted vitamin D metabolism.[6,7] Consistent with earlier reports of high relapse rates after treatment,[8] our data showed that 81.7% of patients with GPP experienced recurrence after discharge. Using both Gaussian Naive Bayes and univariate binary logistic regression models, we identified key distinguishing features between GPP and AGEP, including triggering factors, a history of psoriasis, recurrence, hypocalcemia, and elevated ALP. A total of 26 cases of drug-induced GPP were identified in this study, all of which initially presented with clinical manifestations resembling AGEP, thereby complicating differential diagnosis. However, the subsequent confirmation of underlying plaque psoriasis and favorable responses to retinoids or biologics (e.g., spesolimab) ultimately supported a diagnosis of GPP. A recent study by Benezederet al[3] reveals remarkable immunological parallels between drug-induced GPP and AGEP, showing nearly identical immune response patterns and cellular profiles through genomic analyses. These findings reinforce the emerging concept of AGEP as a drug-induced variant within the IL-36-mediated pustular psoriasis spectrum. The therapeutic success of IL-36 antagonists in severe AGEP cases with GPP-like features further supports this nosological relationship, while also highlighting their promise as targeted treatments for this condition. While histopathology retains diagnostic value, the typically self-limiting nature of AGEP and the rapid resolution of symptoms in most cases often render invasive skin biopsies unnecessary. Several limitations should be considered when interpreting our findings. First, clinical data for discharged patients were incomplete, as only admission-related indicators were recorded, restricting our ability to evaluate treatment outcomes over time. Second, diagnostic interpretation must be integrated with clinical presentation, laboratory results, and response to therapy. Lastly, the Bayesian model employed in this study assumes conditional independence among features, which may underestimate interactions between variables. In summary, this study is to apply both Bayesian modeling and logistic regression to differentiate GPP from AGEP. We identified key predictive factors—including triggering factors, a history of psoriasis, recurrence, hypocalcemia, and elevated ALP—and demonstrated that models incorporating these variables achieved robust diagnostic performance. Our findings contribute to more precise diagnostic criteria for pustular dermatoses and lay the groundwork for future large-scale studies aimed at improving diagnostic accuracy and advancing personalized treatment strategies. Funding None. Conflicts of interest None.