Machine learning models for predicting cure or relapse in post kala azar dermal leishmaniasis (PKDL)

Resumen

Post-Kala-Azar Dermal Leishmaniasis (PKDL) is a neglected tropical disease that emerges in some patients after treatment for visceral leishmaniasis. Its clinical management is challenging due to complex immune responses and the potential for relapse. This Master Thesis aims to explore patterns in PKDL patient data through both unsupervised and supervised machine learning methods. Initially focused on exploratory clustering, the study evolved to include classification models after identifying patients who experienced relapse or required rescue treatment.

Acurated set of clinical, biochemical, hematological, and immunological biomarkers was analyzed, with particular emphasis on cytokines, organ function indicators, and parasite load. Two new outcome variables—Rescue and Relapse—were engineered to better capture patient trajectories. Through this data-driven approach, the project seeks to identify predictive factors of treatment failure and deepen the understanding of disease progression in PKDL. The findings are interpreted in light of recent literature and aim to support more effective treatment strategies.