COVID-19 infection wave mortality from surveillance data in the Philippines using machine learning

Abstract

ABSTRACTObjectiveThe Philippines has had several COVID-19 infection waves brought about by different strains and variants of SARS-CoV-2. This study aimed to describe COVID-19 outcomes by infection waves using machine learning.MethodsWe used a cross-sectional surveillance data review design using the DOH COVID DataDrop data set as of September 24, 2022. We divided the data set into infection wave data sets based on the predominant COVID-19 variant(s) of concern during the identified time intervals: ancestral strain (A0), Alpha/Beta variant (AB), Delta variant (D), and Omicron variant (O). Descriptive statistics and machine learning models were generated from each infection wave data set.ResultsOur final data set consisted of 3 896 206 cases and ten attributes including one label attribute. Overall, 98.39% of cases recovered while 1.61% died. The Delta wave reported the most deaths (43.52%), while the Omicron wave reported the least (10.36%). The highest CFR was observed during the ancestral wave (2.49%), while the lowest was seen during the Omicron wave (0.61%). Higher age groups generally had higher CFRs across all infection waves. The A0, AB and D models had up to four levels with two or three splits for each node. The O model had eight levels, with up to 16 splits in some nodes. Of the ten attributes, only age was included in all the decision tree models, while region of residence was included in the O model. F-score and specificity were highest using naïve Bayes in all four data sets. Area under the curve (AUC) was highest in the naïve Bayes models for the A0, AB and D models, while sensitivity was highest in the decision tree models for the A0, AB and O models.DiscussionThe ancestral, Alpha/Beta and Delta variants seem to have similar transmission and mortality profiles. The Omicron variant caused lesser deaths despite being more transmissible. Age remained a significant predictor of death regardless of infection wave. We recommend constant timely analysis of available data especially during public health events and emergencies.