RT info:eu-repo/semantics/doctoralThesis
T1 Aplicación de Técnicas de Machine Learning en la Predicción de Hospitalizaciones y Reingresos de pacientes con Esquizofrenia en Castilla y León
A1 Góngora Alonso, Susel
A2 Universidad de Valladolid. Escuela de Doctorado
K1 Artificial intelligence
K1 Schizophrenia
K1 Esquizofrenia
K1 Machine Learning
K1 Hospitalizaciones
K1 Risk factors
K1 Reingreso
K1 Random Forest
K1 Castilla y León
K1 1203.04 Inteligencia Artificial
AB Schizophrenia is a severe mental disorder characterized by symptoms such as hallucinations, delusions, thought and behavior disorders. People with schizophrenia are associated with an increased risk of substance abuse, suicide, and mortality compared to the general population. They present hospitalization rates of 20-40% in a year, which results in high costs in the health system and affects the life quality of patients and family members. In Spain, hospital stay accounts for 37.6% of total healthcare costs. The use of Machine Learning (ML) techniques makes it possible to analyze data patterns using statistical methods and to create models that learn and generalize the behavior of the data. In Castilla y León (CyL), reducing the number of hospitalizations and readmissions is of great importance for psychiatric services. Therefore, in this Doctoral Thesis it is hypothesized that the application of ML algorithms helps to identify risk factors for hospitalization and predict readmission of patients with schizophrenia. Consequently, the main objective of this research is to develop and evaluate new predictive models using ML algorithms, in order to help in the prediction of hospitalizations and readmissions of patients with schizophrenia in CyL. To achieve this objective, 11,126 administrative records were used, corresponding to 5,412 hospitalized patients with schizophrenia from 11 public hospitals in CyL, in two different time periods. The records are global data, not based on the clinical psychopathology of the patient; they include demographic information, characteristics of hospitalization episodes, diagnoses and procedures concerning the hospitalized patient. These records were automatically analyzed using ML classification techniques, and predictive models were created to predict the readmission risk of these patients. In this sense, a methodological approach was proposed where a preprocessing and feature selection phase is applied where the predictive variables of the research were determined. The cross-validation method was used in the validation of the models and the ROC curves for their interpretation. Finally, a web application was developed to transfer the main contribution of this Doctoral Thesis to clinical practice.The different models created based on their performance metrics were compared, and the Random Forest (RF) algorithm was found to be the best predictor of the readmission risk of patients with schizophrenia in CyL. This RF model achieved an accuracy of 0.817 and an area under the ROC curve (AUC) of 0.879. These values suggest that the model has a reasonable discrimination capacity to predict the readmission of these patients. Variables such as age, length of stay, V-code diagnoses, substance abuse, and mental disorders were identified as the most predictive variables of the model. These variables indicate possible risk factors associated with the readmission of patients with schizophrenia. Therefore, the results obtained in this Doctoral Thesis suggest that ML algorithms such as RF have the ability to learn complex features from the data and predict the risk of readmission of hospitalized patients with schizophrenia in CyL. It is considered that the developed models can help decision-making, improving the quality of patient care and developing preventive treatments in function of reducing the number of hospitalizations. In addition, the implementation of the web application developed in this research, in public hospitals in CyL, can be very useful to health personnel in terms of reducing the costs associated with these hospitalizations.
YR 2023
FD 2023
LK https://uvadoc.uva.es/handle/10324/62641
UL https://uvadoc.uva.es/handle/10324/62641
LA spa
NO Escuela de Doctorado
DS UVaDOC
RD 18-dic-2024