| 摘要: | Title of Thesis: Predicting Severe COVID-19 and Assessing Long-Term Cardiovascular Complications Compared to Community-Acquired Pneumonia
Author: Nguyen Thi Kim Hien
Thesis Advised by: Professor. Feng-Jen Tsai; Professor. Jason C. Hsu
Background: Since an outbreak was announced in December 2019, the Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2) has affected over 774 million people worldwide and caused more than 7 million deaths as of March 2024. The rapid increase in the number of patients has significantly strained the health system. Furthermore, there are apprehensions regarding the long-term well-being of a substantial proportion of individuals who have recuperated from COVID-19.
Purpose: To develop novel predictive models to assess the probability of severe outcomes or mortality in COVID-19 patients upon hospital admission; To investigate the long-term effects of COVID-19 on the cardiovascular system in recovered patients.
Methodology: Retrospective cohort studies were used in study 1 to approach the real-world data in Taiwan. Seven machine learning models were performed on two settings to predict the short-term composite severe outcome in hospitalized COVID-19 patients, including ventilator use, intubation, intensive care unit admission, and mortality. Study 2 employed the Cox proportional hazard model to estimate the occurrence and risk of major adverse cardiovascular events, such as myocardial infarction, stroke, heart failure, and cardiovascular death after being infected with COVID-19 for a duration of 12 weeks. The second study also compared these outcomes with patients diagnosed with community-acquired pneumonia (CAP).?
Results: Machine learning models were trained and tested on data from 22,192 in-patients diagnosed with COVID-19 between January 1, 2021, and May 31, 2022. Using 90 features, the model with the light gradient boosting machine algorithm achieved the highest AUROC value (0.939), with an accuracy of 85.5%, sensitivity of 0.897, and specificity of 0.853. The top 30 features, including age, vaccination status, neutrophil count, sodium levels, and platelet count, were selected for a simplified model. Using the extreme gradient boosting algorithm, this model achieved an AUROC of 0.935, accuracy of 89.9%, sensitivity of 0.843, and specificity of 0.902. Additonally, Between January 1, 2022, and December 31, 2022, 4,095 participants were included in the second study (1,822 COVID-19 patients and 1,727 CAP patients). After matching, each group had 1,353 patients. The data showed no significant differences in the incidence of MACE, myocardial infarction, and heart failure between the two groups. COVID-19 diagnosis was not statistically associated with an increased risk of MACE (HR, 0.903), heart failure (HR, 0.3744), or myocardial infarction (HR, 1). However, the incidence of stroke and the risk for cardiovascular death (HR, 0.313) were significantly lower in COVID-19 patients.
Conclusion: The initial study demonstrated that machine learning models can accurately predict severe outcomes or death in COVID-19 patients, aiding treatment and resource allocation. The second study found no increased long-term cardiovascular risk in COVID-19 patients compared to those with community-acquired pneumonia (CAP). These insights are valuable for healthcare planning, but further research is needed to validate the findings and explore risk factor differences between the groups.
Keywords: COVID-19, TMUCRD, MACE, CAP, machine learning |
