Bibliographic citations
Nieto, W., (2023). Modelos de predicción para la letalidad por COVID-19: análisis secundario del Sistema Nacional de Vigilancia Epidemiológica del Ministerio de Salud de Perú [Universidad Peruana Cayetano Heredia]. https://hdl.handle.net/20.500.12866/14920
Nieto, W., Modelos de predicción para la letalidad por COVID-19: análisis secundario del Sistema Nacional de Vigilancia Epidemiológica del Ministerio de Salud de Perú []. PE: Universidad Peruana Cayetano Heredia; 2023. https://hdl.handle.net/20.500.12866/14920
@mastersthesis{renati/911038,
title = "Modelos de predicción para la letalidad por COVID-19: análisis secundario del Sistema Nacional de Vigilancia Epidemiológica del Ministerio de Salud de Perú",
author = "Nieto Gutierrez, Wendy Carolina",
publisher = "Universidad Peruana Cayetano Heredia",
year = "2023"
}
Aim: To develop and evaluate models for predicting COVID-19 fatality, we considered population data from the disease surveillance system in Peru. Methods: Retrospective cohort study. The study population consisted of registered and confirmed COVID-19 cases in the surveillance system of nine provinces in Lima. The total population was divided into an 80% random sample, where models were constructed using different strategies for variable selection (1: previous studies; 2: Lasso method; 3: statistical significance; 4: post-hoc). The remaining 20% underwent internal validation. Selection of final models was based on comparing the achieved performance and biological coherence of associations. Results: Information from 17,678 cases was used for model creation, and 4,420 for validation. Constructed models had comparable performances; however, model 1 (13 items) was selected due to its lower number of variables and slightly higher performance, along with the summary model obtained from strategy 4 (3 items). The performances of both models remained consistent during validation, even across different scenarios. Conclusion: The constructed models exhibited comparable performances; nevertheless, we recommend two prediction models with optimal area under the curve values that demonstrated superiority due to their biological plausibility and fewer included variables. Future studies should corroborate the performance and validate utility in current contexts.
This item is licensed under a Creative Commons License