학술논문
Potential and limitations of machine meta-learning (ensemble) methods for predicting COVID-19 mortality in a large inhospital Brazilian dataset
Document Type
Original Paper
Author
de Paiva, Bruno Barbosa Miranda; Pereira, Polianna Delfino; de Andrade, Claudio Moisés Valiense; Gomes, Virginia Mara Reis; Souza-Silva, Maira Viana Rego; Martins, Karina Paula Medeiros Prado; Sales, Thaís Lorenna Souza; de Carvalho, Rafael Lima Rodrigues; Pires, Magda Carvalho; Ramos, Lucas Emanuel Ferreira; Silva, Rafael Tavares; de Freitas Martins Vieira, Alessandra; Nunes, Aline Gabrielle Sousa; de Oliveira Jorge, Alzira; de Oliveira Maurílio, Amanda; Scotton, Ana Luiza Bahia Alves; da Silva, Carla Thais Candida Alves; Cimini, Christiane Corrêa Rodrigues; Ponce, Daniela; Pereira, Elayne Crestani; Manenti, Euler Roberto Fernandes; Rodrigues, Fernanda d’Athayde; Anschau, Fernando; Botoni, Fernando Antônio; Bartolazzi, Frederico; Grizende, Genna Maira Santos; Noal, Helena Carolina; Duani, Helena; Gomes, Isabela Moraes; Costa, Jamille Hemétrio Salles Martins; di Sabatino Santos Guimarães, Júlia; Tupinambás, Julia Teixeira; Rugolo, Juliana Machado; Batista, Joanna d’Arc Lyra; de Alvarenga, Joice Coutinho; Chatkin, José Miguel; Ruschel, Karen Brasil; Zandoná, Liege Barella; Pinheiro, Lílian Santos; Menezes, Luanna Silva Monteiro; de Oliveira, Lucas Moyses Carvalho; Kopittke, Luciane; Assis, Luisa Argolo; Marques, Luiza Margoto; Raposo, Magda Cesar; Floriani, Maiara Anschau; Bicalho, Maria Aparecida Camargos; Nogueira, Matheus Carvalho Alves; de Oliveira, Neimy Ramos; Ziegelmann, Patricia Klarmann; Paraiso, Pedro Gibson; de Lima Martelli, Petrônio José; Senger, Roberta; Menezes, Rochele Mosmann; Francisco, Saionara Cristina; Araújo, Silvia Ferreira; Kurtz, Tatiana; Fereguetti, Tatiani Oliveira; de Oliveira, Thainara Conceição; Ribeiro, Yara Cristina Neves Marques Barbosa; Ramires, Yuri Carlotto; Lima, Maria Clara Pontello Barbosa; Carneiro, Marcelo; Bezerra, Adriana Falangola Benjamin; Schwarzbold, Alexandre Vargas; de Moura Costa, André Soares; Farace, Barbara Lopes; Silveira, Daniel Vitorio; de Almeida Cenci, Evelin Paola; Lucas, Fernanda Barbosa; Aranha, Fernando Graça; Bastos, Gisele Alsina Nader; Vietta, Giovanna Grunewald; Nascimento, Guilherme Fagundes; Vianna, Heloisa Reniers; Guimarães, Henrique Cerqueira; de Morais, Julia Drumond Parreiras; Moreira, Leila Beltrami; de Oliveira, Leonardo Seixas; de Deus Sousa, Lucas; de Souza Viana, Luciano; de Souza Cabral, Máderson Alvares; Ferreira, Maria Angélica Pires; de Godoy, Mariana Frizzo; de Figueiredo, Meire Pereira; Guimarães-Junior, Milton Henriques; de Paula de Sordi, Mônica Aparecida; da Cunha Severino Sampaio, Natália; Assaf, Pedro Ledic; Lutkmeier, Raquel; Valacio, Reginaldo Aparecido; Finger, Renan Goulart; de Freitas, Rufino; Guimarães, Silvana Mangeon Meirelles; Oliveira, Talita Fischer; Diniz, Thulio Henrique Oliveira; Gonçalves, Marcos André; Marcolino, Milena Soriano
Source
Scientific Reports. 13(1)
Subject
Language
English
ISSN
2045-2322
Abstract
The majority of early prediction scores and methods to predict COVID-19 mortality are bound by methodological flaws and technological limitations (e.g., the use of a single prediction model). Our aim is to provide a thorough comparative study that tackles those methodological issues, considering multiple techniques to build mortality prediction models, including modern machine learning (neural) algorithms and traditional statistical techniques, as well as meta-learning (ensemble) approaches. This study used a dataset from a multicenter cohort of 10,897 adult Brazilian COVID-19 patients, admitted from March/2020 to November/2021, including patients [median age 60 (interquartile range 48–71), 46% women]. We also proposed new original population-based meta-features that have not been devised in the literature. Stacking has shown to achieve the best results reported in the literature for the death prediction task, improving over previous state-of-the-art by more than 46% in Recall for predicting death, with AUROC 0.826 and MacroF1 of 65.4%. The newly proposed meta-features were highly discriminative of death, but fell short in producing large improvements in final prediction performance, demonstrating that we are possibly on the limits of the prediction capabilities that can be achieved with the current set of ML techniques and (meta-)features. Finally, we investigated how the trained models perform on different hospitals, showing that there are indeed large differences in classifier performance between different hospitals, further making the case that errors are produced by factors that cannot be modeled with the current predictors.