Clinical features and therapeutic outcomes in pediatric patients with Castleman's disease: a retrospective cohort...

Castleman disease (CD) is a rare lymphoproliferative disorder classified by anatomic site and clinical presentation. Pediatric CD is exceedingly uncommon, with only limited cases reported in the literature. This cohort study encompassed 30 pediatric patients (aged ≤ 18 years) diagnosed with CD. Statistical analyses and advanced modeling were conducted using SPSS version 25.0 and R version 4.3.2. Data collected included demographic information, histopathological findings, laboratory results, and imaging characteristics. Among 30 pediatric CD patients, 19 had unicentric CD (UCD) and 11 had multicentric CD (MCD). UCD patients (12 males, 7 females) had a median diagnosis age of 11 years (IQR, 7.25-13.75; range, 3-18), median diagnostic delay of 2.4 months (IQR, 1.1-8.5), and median follow-up of 16.5 months (IQR, 6.75-27); MCD patients (5 males, 6 females) had a median diagnosis age of 15 years (IQR, 12-16; range, 11-16); median diagnostic delay of 1.77 months (IQR, 0.63-4.17); and median follow-up of 9.5 months (IQR, 5.75-26.5). Compared with MCD, UCD patients had significantly higher MCV (85.04 ± 6.60 fL), ALC (2.67 ± 0.93 × 10⁹/L), Hb (130.03 ± 13.22 g/L), and serum albumin (45.12 ± 4.01 g/L), while MCD patients had higher diagnostic age, PLT (281.37 ± 67.06 × 10⁹/L), globulin (24.35 ± 3.54 g/L), and CRP. UCD had significantly lower incidences of multicentric involvement, hepatosplenomegaly, and systemic symptoms (p < 0.05). Histopathologically, hyaline-vascular subtype predominated in UCD, and mixed hyaline-vascular-plasma-cell subtype was more common in MCD (p = 0.004). Surgery was the primary treatment for UCD, while chemotherapy predominated for MCD (p < 0.001). Night sweats, dizziness, and headache did not affect diagnostic delay or maximum lesion cross-sectional area (p > 0.05). Firth's penalized univariate exact logistic regression showed MCV, Hb, PLT, and somatic symptoms approached statistical significance for outcomes; LASSO regression identified fatigue, MCV, and medical history as core prognostic predictors, while Ridge regression indicated positive signs, fatigue, and MCV as high-risk variables, with globulin, albumin, and complications exerting protective effects, and both models had good predictive efficacy.

Significant differences (p < 0.05) existed between single-center and multicenter CD cohorts in clinical and laboratory parameters. The single-center group had higher MCV, ALC, Hb, and albumin, and lower age, PLT, and globulin. It also had significantly lower rates of fever, hepatosplenomegaly, surgical biopsy, and systemic symptoms at initial diagnosis. Pathologically, hyaline vascular type predominated in the single-center group, while the multicenter group had mixed subtypes. The single-center group primarily underwent surgery, and the multicenter group favored chemotherapy. Subgroup analysis of symptoms showed no significant differences in delayed diagnosis time or maximum lesion cross-sectional area (p > 0.05). Firth's exact logistic regression and LASSO/Ridge penalized regression identified MCV, fatigue, and medical history as potential CD prognostic factors, with penalized models showing good predictive performance in this small-sample study.

• Castleman's disease (CD) is a rare lymphoproliferative disease, which can be classified according to anatomical location and clinical manifestations. Children's CD is extremely rare and the literature reports are limited. It is mainly divided into unicentric (UCD) and multicentric (MCD). • The common pathological subtypes of CD are hyaline vascular type, plasma cell type and mixed type. UCD is mainly treated by surgery, and MCD is mainly treated by chemotherapy. Night sweats, dizziness, and headache do not affect the diagnostic delay and maximum cross-sectional area of children with CD.

• Thirty children with CD were included in this study, and it was clear that there were significant differences between UCD and MCD in demography, laboratory indicators, clinical phenotype and pathological subtype, and UCD was more benign. • Through a variety of regression models, core prognostic predictors such as MCV, fatigue, and past medical history are screened out, and high-risk and protective factors are identified, and the model has good predictive efficacy in small samples.