Prognostic Prediction Machine Learning Models for Elderly Pneumonia Patients - A Data-Driven Approach

Aug 3, 2026

Integrating diverse datasets with different predictors while keeping the same target variable strengthens both the performance and reliability of machine learning prognostic models. This thesis applies the idea to healthcare data, focusing on pneumonia in elderly patients in the Veneto region, by combining invasive bacterial diseases (IBD) notiﬁcation forms with hospital discharge records (HDR) under deterministic feature-matching enabled by pseudonymization. Models based on SVMs, random forests, and artiﬁcial neural networks are trained separately and on the merged dataset, with the combined approach achieving the highest test ROC-AUC (0.861).

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages99

Reading Time4 h 9 min

AccessFree to download

Abstract

Introduction

Related works

AI models in healthcare

Machine learning techniques used

Datasets exploration and creation

Invasive bacterial diseases (IBD) dataset

Hospital discharge records (HDR) dataset

Combined dataset

Models

Baseline models

Artiﬁcial neural networks

Support vector machines

Random Forest

Experiments and results

IBD dataset

HDR dataset

Combined dataset

Best performing models

Conclusions

References

Acknowledgments

University of Padova Department of Mathematics ”Tullio Levi-Civita” Master Thesis in Data Science Prognostic Prediction Machine Learning Models for Elderly Pneumonia Patients Supervisor Master Candidate Nicolò Navarin Claudio Palmeri University of Padova Co-supervisor Vincenzo Baldo University of Padova Academic Year 2024-2025 ii This thesis is dedicated to all the people who helped me and supported me during my academic career. Without you, Elisabetta, Corrado, Silvia, Luca, and many more, this would not have been possible. iv Abstract Integrating di昀昀erent datasets with di昀昀erent predictors and the same target variable can enhance the performances and the reliability of machine learning models trained on them. This work applied this concept to healthcare data, speci昀椀cally pneumonia in elderly patients in the Veneto region, by combining the noti昀椀cation forms from the regional surveillance system for invasive bacterial diseases (IBD) dataset with the hospital discharge records dataset (HDR) . Due to the datasets’ pseudonymization, deterministic techniques were used to exploit the fact that some features are shared to identify the matches between them. Predictive prognostic models were independently developed and evaluated for IBD, HDR, and the combined dataset. The models were implemented using machine learning techniques such as support vector machines, random decision forests, and arti昀椀cial neural networks. The best model trained on the combined dataset yielded higher performances than the models trained only on one of the two datasets. More precisely, the best model trained only on the IBD dataset, had an area under the ROC curve equal to 0.685 on the test set; the best model trained only on the HDR dataset had an area under the ROC curve equal to 0.793 on the test set; while the best model trained on the combined dataset had an area under the ROC curve equal to 0.861 on the test set. All of these models, especially the one trained on the combined dataset, could be used to identify higher-risk patients so that the hospital’s personnel could increase their likelihood of survival by surveilling them more. vi Contents Abstract v List of figures ix Listing of acronyms xiii 1 Introduction 1 2 Related works 5 2.1 AI models in healthcare ............................ 5 2.1.1 Early applications of AI in healthcare ................. 5 2.1.2 Recent developments ......................... 7 2.2 Machine learning techniques used ....................... 9 2.2.1 Support vector machines ....................... 9 2.2.2 Random decision forests . . . . . . . . . . . . . . . . . . . . . . . 11 2.2.3 Arti昀椀cial Neural Networks ...................... 11 3 Datasets exploration and creation 13 3.1 Invasive bacterial diseases (IBD) dataset .................... 13 3.1.1 Year of diagnosis ........................... 14 3.1.2 Sex .................................. 15 3.1.3 Age of the patient ........................... 16 3.1.4 Associated invasive diseases ...................... 18 3.1.5 Sample material ............................ 19 3.1.6 Time of noti昀椀cation ......................... 21 3.1.7 Notifying hospital .......................... 22 3.1.8 Bacteria serotype ........................... 22 3.1.9 Vaccination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 3.1.10 Antibiograms ............................. 27 3.2 Hospital discharge records (HDR) dataset .................. 27 3.2.1 Year of diagnosis ........................... 28 3.2.2 Hospital’s data ............................ 28 3.2.3 Age of the patient and date ofbirth .................. 30 3.2.4 Patient’s anagraphic data ....................... 31 3.2.5 Patient hospitalization ........................ 34 3.2.6 Patient diagnosis and procedures ................... 38 3.3 Combined dataset ............................... 42 4 Models 45 4.1 Baseline models ................

FAQs

What datasets are combined for the prognostic prediction task?

The work combines the invasive bacterial diseases (IBD) notiﬁcation forms from the regional surveillance system with the hospital discharge records (HDR) dataset for elderly pneumonia patients in Veneto.

How are the datasets matched despite pseudonymization?

Because of dataset pseudonymization, deterministic techniques are used to exploit shared features and identify matches between records across datasets.

Which machine learning methods are used and what is the best performance?

Models are built using support vector machines, random decision forests, and artiﬁcial neural networks. The best model trained on the combined dataset achieves a test ROC-AUC of 0.861, outperforming models trained on only one dataset.

Aug 3, 2026

Upload by River Wang