Abstract Number: PB1366
Meeting: ISTH 2020 Congress
Background: Predict the severity of new bleeding in patients affected by Immune Thrombocytopenia (ITP) is a challenge. There is no consensus on severity of new bleeding in these patients. Therefore, artificial neural networks (ANN) – relating clinical factors with consequences of ITP – may have great potential to assist in treatment.
Aims: To obtain an ANN model capable of predicting the severity of new bleeds based on blood count and patient history.
Methods: Data from 631 C consultations and 631 C+1 consultations – of 118 patients with ITP were collected between 2007 and 2018. Twenty six clinical and laboratorial parameters were considered: sex, age, comorbidities (hypertension, diabetes, pregnancy, neoplasia, chemotherapy or radiotherapy, infection and/or fever, lupus, dyslipidemia, stroke and/or acute myocardial infarction, venous thromboembolism, others), hepatitis B, hepatitis C, HIV, ANF, anticardiolipin, lupus anticoagulant, hemoglobin, erythrocytes, leukocytes, platelets, mean platelet volume, bleeding presence and bleeding severity (between 1 and 5). All these variables considered in C consultation were used as ANN inputs. The bleeding severity in C+1 consultation were used as ANN output. The data were randomly divided into train (70%), test (15%), and validation (15%) data. ANNs with the number of neurons from 5 to 30 were tested, trained with the Levenberg-Marquardt method, and minimizing the mean square error (MSE) between the clinical parameter and the calculated.
Results: The best result was obtained with a two hidden layers ANN, with 25 neurons in both layers. The correlation was 0.8854 (train), 1.0000 (test) and 0.9998 (validation). This model was able to predict severity of new bleeding in 99.21% of cases with a relative error less than 1%.
Conclusions: The results obtained show that ANN have great potential to assist in the treatment of ITP, and must be refined and validate in other group of patients to be used in clinical practice.
To cite this abstract in AMA style:de Carvalho LP, Colella MP, Annichino-Bizzacchi JM, Martins TD. Artificial Neural Network for Prediction of Hemorrhagic Severity in Patients with Immune Thrombocytopenia Purpura [abstract]. Res Pract Thromb Haemost. 2020; 4 (Suppl 1). https://abstracts.isth.org/abstract/artificial-neural-network-for-prediction-of-hemorrhagic-severity-in-patients-with-immune-thrombocytopenia-purpura/. Accessed October 1, 2023.
« Back to ISTH 2020 Congress
ISTH Congress Abstracts - https://abstracts.isth.org/abstract/artificial-neural-network-for-prediction-of-hemorrhagic-severity-in-patients-with-immune-thrombocytopenia-purpura/