ISEL - Engenharia Biomédica - Dissertações de Mestrado
Permanent URI for this collection
Browse
Browsing ISEL - Engenharia Biomédica - Dissertações de Mestrado by advisor "Bento, Luís"
Now showing 1 - 2 of 2
Results Per Page
Sort Options
- Development of Predictive Models for COVID-19 Prognosis based on Patients’ Demographic and Clinical DataPublication . Von Rekowski, Cristiana; Calado, Cecília Ribeiro da Cruz; Pinto, Iola; Bento, LuísBackground – Cases of infection by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) were first reported in late December 2019. Due to the large spectrum of clinical presentations and outcomes, the disease was named Coronavirus Disease 2019 (COVID-19) and characterized as a pandemic due to the elevated number of cases worldwide, the high transmission rate and the lack of action measures. Since then, a lot of progress has been made, but the study of demographic and clinical information and the determination of possible laboratory biomarkers for COVID-19 prognosis is crucial. Purpose – Determine predictive biomarkers for COVID-19’s outcome (death or survival), in critically ill patients, using clinical, demographic and laboratory data from the intensive care unit (ICU). Methods – Demographic, clinical and laboratory data from 337 COVID-19 patients admitted to the ICU of Centro Hospitalar Universitário Lisboa Central, Portugal, between March 2020 and March 2021, was extracted from the hospital’s electronic medical record system, pre-processed, and analyzed. Comparisons were made regarding death, the need of invasive mechanic ventilation (IMV), the first three COVID-19 waves and age groups. Longitudinal data was gathered over the course of the patients stay in the ICU. To infer about the evolution of the patients' condition in the first week of ICU admission, a comparative analysis was carried out between the data from the 2nd (335 patients) and 7th days (216 patients). Comparisons of laboratory parameters between discharged and deceased patients, at these time points were performed. The associations between the several biomarkers and death were tested by means of Univariate Generalized Estimating Equations (GEEs) models. Additionally, to analyze the impact of some biomarkers in mortality, crude odds ratios were estimated and interpreted, with the corresponding 95% confidence intervals (CIs). Death event-free survival rates were obtained by the Kaplan-Meier estimator. All P values were considered statistically significant at P<0.05. Results – Deceased patients were considerably older, had more comorbidities, required more IMV, and spent less time in the hospital than discharged patients. Death rates did not differ significantly between COVID-19 waves. Patients from the 1st wave were significantly older and relied more on IMV and extracorporeal membrane oxygenation (ECMO). Most of the detected differences regarding laboratory biomarkers were found between discharged and deceased patients from the 2nd and 3rd waves, being that the deceased ones had almost always worse results. In general, worse results were obtained in the 1st wave and in the 7th day of ICU admission. In 2nd day of ICU admission, 2nd wave, higher mortality rates were observed for patients with lymphocyte (LYM) levels under normality ranges. In the 3rd wave, mortality rates were higher for patients with high sensitivity troponin I (hs-cTn I) levels above normality ranges in the 2nd day of ICU admission, with LYM levels under normality ranges in the 7th day of ICU admission, and with platelet (PLT) levels below normality ranges, either in the 2nd or 7th days of ICU admission. Through the univariate logistic regression’s results in 2nd day of ICU admission, 2nd wave, hs-cTn I, red blood cell (RBC) counts, platelet-lymphocyte ratio (PLR) and neutrophil-lymphocyte ratio (NLR) showed significant association with the risk of death. In 7th day of ICU admission, C-reactive protein (CRP), RBC counts, hematocrit (HCT), hemoglobin (HGB), white blood cell (WBC) and neutrophil (NEU) counts, eosinophil (EO) counts and NLR, revealed significant association with the risk of death. In the 2nd day of ICU admission, 3rd wave, hs-cTn I, PLT counts, lactate dehydrogenase (LDH) and CRP showed significant association with the risk of death. For the 7th day, PCT, CRP, WBC and NEU counts, LYM counts, NLR and PLT counts results were also associated with higher risks of death. Univariate GEEs models results demonstrated that, in the 1st wave, hs-cTn I, myoglobin, EO counts, results were associated with higher risks of death. In the 2nd wave, the risk of death was significantly associated with hs-cTn I, myoglobin levels, EO counts, WBC and NEU counts, LYM counts, and INR. Finally, in the 3rd wave, hs-cTn I, CK, EO counts, WBC and NEU counts, LYM counts, NLR and PLT counts, were also associated with the risk of death. Conclusion - This study provides useful information for prognostic evaluation that can be used to guide treatment and monitoring. Most importantly, it consists of valuable data that can be employed as the foundation of a variety of future research. Aside from the positive results, more research is needed to develop reliable and robust biomarkers for COVID-19’s outcomes.
- Previsão automática da mortalidade em UCI de doentes com síndrome da dificuldade respiratória aguda associada à COVID-19 utilizando radiografias de tórax e dados clínicosPublication . Galvão, Tiago Alexandre dos Santos; Domingues, Nuno Alexandre Soares; Jorge, Pedro Miguel Torres Mendes; Bento, LuísA síndrome da dificuldade respiratória aguda associada à COVID-19 (ARDS-COV19), é uma síndrome pulmonar grave que resulta em insuficiência respiratória aguda. A ARDS é complexa e heterogénea, exigindo frequentemente ventilação mecânica invasiva (VMI) em unidades de cuidados intensivos (UCI). A identificação de grupos de risco é crucial para a medicina de precisão, embora a falta de métodos de diagnóstico seja limitativo. A radiografia torácica é um exame imagiológico, qualitativo e acessível, utilizado na rotina das UCIs. É essencial o desenvolvimento de um classificador multivariado e quantitativo, baseado em radiomics, para a previsão da mortalidade destes doentes sob VMI. Para este efeito foram incluídos 110 doentes ARDS-COV19 de uma UCI, com uma idade média de 63,2 ± 11,92 anos, sendo 61,2% do sexo masculino. A mortalidade foi de 47,3%. Radiografias do 1º e 3º dia de VMI foram recolhidas, pré-processadas e concatenadas. Características de deep learning foram então extraídas, utilizando uma rede neuronal convolucional pré-treinada (CheXnet). Estas características foram acopladas a variáveis clínicas (VC), para a construção de dois modelos de aprendizagem automática, um de regressão logística (LogReg) e um perceptrão multicamada (MLP). A idade, a razão PaO2/FiO2 do 3º dia de VMI e uma característica de imagem (DLF_258) foram utilizadas nos modelos finais. Os modelos que incluíram a DLF_258, apresentaram 89% (LogReg) e 82% (MLP) de probabilidade de terem melhor exatidão, do que os modelos de VC. No grupo de teste interno (23 doentes), o modelo de LogReg obteve os melhores resultados e menor overfitting, com uma área under the ROC curve (AUC) de 0,862 95%CI [0.654, 0.969], uma exatidão de 0,783 95%CI [0.563, 0.926] e um score de F1 de 0,783 95%CI [0.563, 0.926]. Apesar dos resultados promissores, o número de amostras foi reduzido, não existindo um teste externo. A recolha de dados e posterior validação são assim essenciais.