Gezondheid en Tijd: Innovaties in Immuunmonitoring en Survivalclustering

This thesis explores the application and development of machine learning and statistical methodologies for the analysis of medical longitudinal data, which comprises data accumulated over time. Longitudinal data analysis is a crucial tool in the medical domain. By conducting longitudinal follow-ups, healthcare practitioners and researchers can assess the effects of medical treatments or forecast prospective health outcomes. Therefore, robust methods to effectively analyze such data are critically important. The primary objective of this thesis is to enhance the utility of longitudinal medical data analysis by developing and applying advanced machine learning and statistical methodologies. The presented research is divided into two main parts. The first part focuses on the longitudinal monitoring of humoral immunity using novel multiplex immunoassays and developing statistical approaches to optimize information extraction. This part examines three conditions: infections with human T-lymphotropic virus (HTLV), COVID-19, and Chagas disease. For HTLV, the objective was to develop a high-performance multiplex immunoassay interpretation algorithm to confirm and discriminate between HTLV types 1 and 2, as these types can have different clinical outcomes and implications for treatment. In the COVID-19 study, a statistical methodology, called the DF50 method, was developed to transform dilution sequences from a SARS-CoV-2 multiplex immunoassay into quantitative antibody profiles. This method allows for a comparison of immune responses between vaccinated and naturally infected individuals. For Chagas disease, the same DF50 quantification method was applied to monitor humoral immunity as a surrogate for treatment success. This approach is particularly valuable in Chagas disease due to the chronic nature of the infection and the absence of a well-defined clinical endpoint, which hinders clinical trials for new medications. The second part of the thesis introduces a novel machine learning method for clustering in survival analysis, called SurvivalLVQ. Survival analysis, commonly used in medical research, models the time until an event of interest occurs. Traditional methods typically focus on estimating risk scores or modeling survival distributions individually. In contrast, our approach clusters patients based on both their characteristics and survival outcomes. This method can uncover patterns and subgroups within the data, providing valuable insights for developing more personalized and effective treatment strategies. The findings from the research presented in this thesis demonstrate that combining multiplex immunoassays with advanced statistical methods can effectively track humoral immunity over time. Additionally, the newly proposed SurvivalLVQ method can uncover hidden patient subgroups. These methodologies have implications for both clinical practice and medical research, offering innovative ways to monitor infectious diseases and analyze survival data.