An ensemble prediction model for COVID-19 mortality risk

Background: It's critical to identify COVID-19 patients with a higher death risk at early stage to give them better hospitalization or intensive care. However, thus far, none of the machine learning models has been shown to be successful in an independent cohort. We aim to develop a machine learning model which could accurately predict death risk of COVID-19 patients at an early stage in other independent cohorts. Methods: We used a cohort containing 4711 patients whose clinical features associated with patient physiological conditions or lab test data associated with inflammation, hepatorenal function, cardiovascular function and so on to identify key features. To do so, we first developed a novel data preprocessing approach to clean up clinical features and then developed an ensemble machine learning method to identify key features. Results: Finally, we identified 14 key clinical features whose combination reached a good predictive performance of AUC 0.907. Most importantly, we successfully validated these key features in a large independent cohort containing 15,790 patients. Conclusions: Our study shows that 14 key features are robust and useful in predicting the risk of death in patients confirmed SARS-CoV-2 infection at an early stage, and potentially useful in clinical settings to help in making clinical decisions.

Efficacy and Safety of Anticoagulation Treatment in COVID-19 Patient Subgroups Identified by Clinical-Based Stratification and Unsupervised Machine Learning: A Matched Cohort Study

Objective: To explore the efficacy of anticoagulation in improving outcomes and safety of Coronavirus disease 2019 (COVID-19) patients in subgroups identified by clinical-based stratification and unsupervised machine learning.Methods: This single-center retrospective cohort study unselectively reviewed 2,272 patients with COVID-19 admitted to the Tongji Hospital between Jan 25 and Mar 23, 2020. The association between AC treatment and outcomes was investigated in the propensity score (PS) matched cohort and the full cohort by inverse probability of treatment weighting (IPTW) analysis. Subgroup analysis, identified by clinical-based stratification or unsupervised machine learning, was used to identify sub-phenotypes with meaningful clinical features and the target patients benefiting most from AC.Results: AC treatment was associated with lower in-hospital death risk either in the PS matched cohort or by IPTW analysis in the full cohort. A higher incidence of clinically relevant non-major bleeding (CRNMB) was observed in the AC group, but not major bleeding. Clinical subgroup analysis showed that, at admission, severe cases of COVID-19 clinical classification, mild acute respiratory distress syndrome (ARDS) cases, and patients with a D-dimer level ≥0.5 μg/mL, may benefit from AC. During the hospital stay, critical cases and severe ARDS cases may benefit from AC. Unsupervised machine learning analysis established a four-class clustering model. Clusters 1 and 2 were non-critical cases and might not benefit from AC, while clusters 3 and 4 were critical patients. Patients in cluster 3 might benefit from AC with no increase in bleeding events. While patients in cluster 4, who were characterized by multiple organ dysfunction (neurologic, circulation, coagulation, kidney and liver dysfunction) and elevated inflammation biomarkers, did not benefit from AC.Conclusions: AC treatment was associated with lower in-hospital death risk, especially in critically ill COVID-19 patients. Unsupervised learning analysis revealed that the most critically ill patients with multiple organ dysfunction and excessive inflammation might not benefit from AC. More attention should be paid to bleeding events (especially CRNMB) when using AC.

Evaluation and Comparison of the STIMUL Extended and Simplified Risk Scores for Predicting Two-Year Death in Patients Following ST-Segment Elevation Myocardial Infarction

Background and Objectives: The management of ST-segment elevation myocardial infarction (STEMI) requires a patient’s long-term risk to be estimated. The objective of this study was to develop extended and simplified models of two-year death risk estimation following STEMI that include and exclude cardiac troponins as prognostic factors and to compare their performance with each other. Materials and Methods: Extended and simplified multivariable logistic regression models were elaborated using 1103 patients with STEMI enrolled and followed up in the STIMUL (ST-segment elevation Myocardial Infarctions in Ukraine and their Lethality) registry. Results: The extended STIMUL risk score includes seven independent risk factors: age; Killip class ≥ II at admission; resuscitated cardiac arrest; non-reperfused infarct-related artery; troponin I ≥ 150.0 ng/L; diabetes mellitus; and history of congestive heart failure. The exclusion of cardiac troponin in the simplified model did not influence the predictive value of each factor. Both models divide patients into low, moderate, and high risk groups with a C-statistic of 0.89 (95% CI 0.84–0.93; p < 0.001) for the extended STIMUL model and a C-statistic of 0.86 (95% CI 0.83–0.99; p < 0.001) for the simplified model. However, the addition of the level of troponin I to the model increased its prognostic value by 10.7%. Conclusions: The STIMUL extended and simplified risk estimation models perform well in the prediction of two-year death risk following STEMI. The simplified version may be useful when clinicians do not know the value of cardiac troponins among the population of STEMI patients.

Blood Hemoglobin Substantially Modulates the Impact of Gender, Morbid Obesity, and Hyperglycemia on COVID-19 Death Risk: A Multicenter Study in Italy and Spain

BackgroundHyperglycemia and obesity are associated with a worse prognosis in subjects with COVID-19 independently. Their interaction as well as the potential modulating effects of additional confounding factors is poorly known. Therefore, we aimed to identify and evaluate confounding factors affecting the prognostic value of obesity and hyperglycemia in relation to mortality and admission to the intensive care unit (ICU) due to COVID-19.MethodsConsecutive patients admitted in two Hospitals from Italy (Bologna and Rome) and three from Spain (Barcelona and Girona) as well as subjects from Primary Health Care centers. Mortality from COVID-19 and risk for ICU admission were evaluated using logistic regression analyses and machine learning (ML) algorithms.ResultsAs expected, among 3,065 consecutive patients, both obesity and hyperglycemia were independent predictors of ICU admission. A ML variable selection strategy confirmed these results and identified hyperglycemia, blood hemoglobin and serum bilirubin associated with increased mortality risk. In subjects with blood hemoglobin levels above the median, hyperglycemic and morbidly obese subjects had increased mortality risk than normoglycemic individuals or non-obese subjects. However, no differences were observed among individuals with hemoglobin levels below the median. This was particularly evident in men: those with severe hyperglycemia and hemoglobin concentrations above the median had 30 times increased mortality risk compared with men without hyperglycemia. Importantly, the protective effect of female sex was lost in subjects with increased hemoglobin levels.ConclusionsBlood hemoglobin substantially modulates the influence of hyperglycemia on increased mortality risk in patients with COVID-19. Monitoring hemoglobin concentrations seem of utmost importance in the clinical settings to help clinicians in the identification of patients at increased death risk.

197. A Predictive Model for the Prompt Identification of Suspected Sepsis Patients with High Death Risk

Background How to improve the high mortality rate of sepsis? The prompt identification of at-risk patients, and the interdisciplinary sepsis treatment protocol implementation are interventions that can reverse such unacceptable outcome. The objective of our study is to summarize main results of the protocol for the management of sepsis and septic shock, implemented at Biocor Instituto, a general hospital in Belo Horizonte, a 3,000,000 inhabitants city from Brazil. Methods Prospective cohort study of patients with sepsis, evaluated between May/2018-Apr/2020. Univariate and multivariate analysis by logistic regression to identify risk factors for hospital death. Results Over 28 months, 220 patients were included in sepsis protocol: 121 hospital deaths, a crude mortality = 121/220 = 55% (95%C.I. = [48%;62%]). 136 patients (62%) came from the emergency room. In 97 cases (44%) it was possible to isolate 111 microorganisms, with a predominance of Klebsiella, E.coli, and S.aureus. 75% of the cases (165) had definition of APACHE, with the absolute majority of these (88%) having APACHE between 25 and 40. Most patients (52%) received antibiotic (ATB) in 15 minutes and only 4% received ATB after 60 minutes of waiting time. In 198 patients (90%) it was possible to identify the focus of sepsis, with a predominance of pulmonary (47%), urinary (21%) and abdominal (15%). Hospital mortality varied from 30 to 62%, when the focus was pulmonary (p-value = 0.045). In univariate analysis (Figure 1), pulmonary sepsis, creatinine, lactate, and APACHE were significantly associated with hospital death. The time for ATB administration was typically close to 20 minutes, and time to receive the therapeutic antibiotic were not associated with the patient’s death. By using the logistic model (Figure 2) to assign cases of predicted hospital death for probabilities &gt;= 0.5 and controls for probabilities &lt; 0.5, the prediction model had a sensitivity of 0.68 (0.59–0.76), a specificity of 0.58 (0.48–0.67), an area under the curve of the receiver operating characteristic curve of 0.75 (0.68–0.82). There was no significant difference between observed versus expected mortality by APACHE (Figure 3). Figure 1. Univariate analysis to identify risk factors for hospital death. Figure 2. Logistic model for predicting hospital death. Figure 3. Observed X Expected/severity-adjusted mortality (APACHE). Conclusion The logistic model developed uses only creatinine and lactate data to predict suspected sepsis patients with high death risk. Disclosures All Authors: No reported disclosures

Social Participation and Survival in Widowed Persons: Results of the Taiwan Longitudinal Study on Aging

It has been considered that widowed persons have a higher risk of death. This study intended to explore whether social participation could improve this trend. A longitudinal study database was constructed to explore the trend of survival and its change with social participation in widowed persons. The Taiwan Longitudinal Study on Aging (TLSA), based on four consecutive waves of longitudinal follow-up data in 1999, 2003, 2007, and 2011 was linked with the National Death Registry from 1999 through 2012. In total, there were 1417 widowed persons and 4500 nonwidowed persons included in this study, excluding divorced and never-married people. The survival trend analysis was carried out with social participation as the main predictive factor stratified for comparative analysis. Our results showed that the widowed were older than the nonwidowed, were female-dominant, had a lower education level, were more economically stressed, and were less likely to engage in regular exercise, and thus showed generally poorer health; for example, being more vulnerable to having chronic diseases, disability with the Activities of Daily Living (ADL), cognitive impairment with the Short Portable Mental State Questionnaire (SPMSQ), and depression with The Center for Epidemiological Studies-Depression (CES-D). The death risk of the widowed was significantly higher than that of the nonwidowed, but the death trend for those with social participation was significantly lower than that of their counterparts in both the widowed and nonwidowed. After matching with gender and age for widowed persons, the widowed with social participation had a significantly lower risk of death (adjusted hazard ratio (HR), 0.83; 95% confidence interval (CI), 0.71–0.98) compared to the widowed without social participation. It was concluded that social participation can improve the death risk for the widowed, and it is worthily included in health promotion plans and social welfare services for widowed persons.

Differences in Cancer Death Risk Long After ACS Among Selected Urban and Rural Areas in North Italy: The ABC-7a† Study on Heart Disease

BackgroundAn increased risk of cancer death has been demonstrated for patients diagnosed with acute coronary syndrome (ACS). We are investigating possible geographic risk disparities.MethodsThis prospective study included 541 ACS patients who were admitted to hospitals and discharged alive in three provinces of Italy’s Veneto region. The patients were classified as residing in urban or rural areas in each province.ResultsWith 3 exceptions, all patients completed the 22-year follow-up or were followed until death. Urban (46%) and rural (54%) residents shared most of their baseline demographic and clinical characteristics. Pre-existing malignancy was noted in 15 patients, whereas 106 patients developed cancer during the follow-up period, which represented 6232 person-years. No difference in the cancer death risk was found between the urban and rural areas or between southern and northern provinces (hazard ratio [HR] 1.1; 95% confidence interval [CI] 0.7–1.7; p = 0.59 and HR 1.1; 95% CI 0.9–1.4; p = 0.29, respectively) according to the unadjusted Cox regression analysis. Geographic areas, however, showed a strong positive interaction, with risk increasing from the urban to rural areas from southern to northern provinces (HR 1.9; 95% CI 1.1–3.0; p = 0.01). The fully adjusted Cox regression and Fine-Gray competing risk regression models provided similar results. Interestingly, these results persisted, and even strengthened, after exclusion of the 22 patients who developed malignancy and survived to the end of follow-up. We did not observe an urban/rural difference in non-neoplastic death risk or a significant interaction between the geographic areas.ConclusionOur analysis reveals that the cancer death risk among unselected ACS patients in Italy’s Veneto region significantly differs by geography. The northern rural area has the highest risk. These results highlight the importance of implementing a preventive policy based on area-specific knowledge.