This section collects peer-reviewed journal articles and conference papers related to the UIOMOLS project and its methodological developments in clinical Natural Language Processing, causal machine learning, and statistical analysis of Electronic Health Records.
Alessandro Albano, Chiara Di Maria, Mariangela Sciandra, Antonella Plaia
First published: 25 August 2025
🔗 https://doi.org/10.1002/asmb.70038
This paper introduces a causal machine learning framework based on Causal Forests to identify diagnostic language patterns within Electronic Health Records (EHRs), providing interpretable insights into clinically relevant textual features.
Marco Speciale, Alessandro Albano, Mariangela Sciandra, Antonella Plaia
Scientific Meeting of the Italian Statistical Society, 16 June 2025
Springer Nature Switzerland, pp. 351–356
Abstract
The effective integration of real and synthetic clinical data in multiple languages is essential to advance healthcare research. This study proposes a statistical framework leveraging cross-lingual embeddings to validate semantic alignment between authentic Italian EHRs and synthetic English clinical notes. Using state-of-the-art models (E5 and BGE), texts are encoded and analysed via Fuzzy C-Means clustering and multidimensional scaling. The results highlight language-specific structures alongside robust cross-lingual alignment, demonstrating the potential of synthetic data augmentation in addressing resource scarcity.
Alessandro Albano, Chiara Di Maria, Mariangela Sciandra, Antonella Plaia
Scientific Meeting of the Italian Statistical Society, 16 June 2025
Springer Nature Switzerland, pp. 227–232
Abstract
This work proposes a methodology to detect comorbidities from clinical discharge notes using propensity scores derived from textual features. Textual information is summarised and incorporated into logistic regression models to assess disease associations. The study compares TF-IDF representations and text embeddings using LASSO, XGBoost, and multilayer perceptrons (MLP). Applied to diabetes and Chronic Kidney Disease, the results demonstrate the value of NLP-based representations in epidemiological analysis.
Alessandro Albano, Chiara Di Maria, Mariangela Sciandra, Antonella Plaia
Scientific Meeting of the Italian Statistical Society, 17 June 2024
Springer Nature Switzerland, pp. 34–39
Abstract
This paper proposes a causal framework for analysing medical texts using the MIMIC-III dataset. By applying Causal Forests, the study identifies linguistic factors that causally influence the probability of specific diagnoses. The results reveal significant causal relationships between clinical terminology and hypothyroidism diagnosis, highlighting the importance of causal inference in medical NLP.
Alessandro Albano, Mariangela Sciandra, Antonella Plaia
Book of Short Papers – CLADAG 2023, p. 17
Abstract
This paper presents an ensemble method for text classification in the presence of multiple rare classes within medical record data. The study focuses on classifying clinical notes into multiple disease categories, including rare diseases.
The proposed ensemble approach combines the predictions of three machine learning models — Support Vector Machine (SVM), Random Forest, and Naive Bayes — to improve diagnostic prediction accuracy.
The results show that the ensemble strategy outperforms individual models in terms of classification performance. The methodology was evaluated on a dataset of 50,000 clinical notes containing multiple rare classes, demonstrating the effectiveness of ensemble learning in complex medical text classification tasks.