Building evidence for diagnosis of lipedema: using a classification and regression tree (CART) algorithm to differentiate lipedema from lymphedema patients

Autores de IIS La Fe

• Isabel Forner Cordero

  Autor

Participantes ajenos a IIS La Fe

• Muñoz-Langa J.
• Morilla-Bellido L.

Grupos

• Medicina Física y Rehabilitación

Abstract

Background: Discriminating lipedema from lymphedema becomes challenging in the absence of a pathognomonic test. The objective was to find the best manifestations that discriminate between lipedema and lymphedema and to build a diagnosis algorithm. Methods: Prospective cohort study of two cohorts of patients, one with lipedema and another with lymphedema. Inclusion criteria for lipedema cohort involved bilateral lower limbs (LL) enlargement and at least three symptoms: pain/tenderness, bruising, familial history, no Stemmer’s sign, symmetrical involvement, and non-swollen feet. The lymphedema cohort included female patients with LL lymphedema. A univariate analysis was performed to determine which clinical features were different between both samples. We used a TREE procedure to create a tree-based classification model using the CART (Classification And Regression Tree) algorithm, in order to discriminate lipedema from lymphedema patients. Results: Currently, 138 lipedema and 111 lymphedema patients were included. After univariate analysis, symmetrical involvement, disproportion between upper and lower parts of the body, spare feet, bruising, spider veins, family history, and pain were significantly more present in lipedema than in lymphedema (P<0.0001). Stemmer’s sign, lymphangitis bouts, pitting and fibrosis were more representative of lymphedema (P<0.0001). The most important variables for discrimination were: disproportion (100%), spare feet (92.6%), bruising (92.3%) and symmetrical involvement (90.3%). After CART analysis, only three variables were retained in the final model: bruising, disproportion and spare feet. The model’s accuracy was 100% with a probability error of 0.0% (SE: 0.00). Conclusions: A simple clinical tree can be used to classify patients between lymphedema and lipedema. © 2025 EDIZIONI MINERVA MEDICA.

Keywords

• Adult; Aged; Algorithms; Decision Support Techniques; Diagnosis, Differential; Female; Humans; Lipedema; Lymphedema; Male; Middle Aged; Predictive Value of Tests; Prospective Studies; adult; algorithm; Article; body mass; classification; clinical feature; contusion; edema; error; family history; female; follow up; histology; human; lipedema; lower limb; lymphangitis; lymphedema; major clinical study; male; obesity; patient; prediction; predictive value; probability; prospective study; quality of life; questionnaire; receiver operating characteristic; regression analysis; risk factor; sensitivity and specificity; soreness; spider; spider vein; statistical analysis; swelling; training; univariate analysis; aged; algorithm; classification; decision support system; diagnosis; differential diagnosis; lipedema; middle aged