Participant characteristics were summarized descriptively. Comparisons between patients discharged home, admitted to the medical ward, or admitted directly to the ICU were made with Wilcoxson rank sum and Pearson chi-square tests for continuous and categorical variables, respectively. Impact of timing during the pandemic was assessed as days since data collection started (March 8, 2020).
All tests were 2-sided and a P value < .05 was considered statistically significant. All variables were initially assessed for significance using univariable analysis comparing: Patients discharged home versus admitted to the medical ward and; Patients admitted to the medical ward versus ICU (see Tables S1 and S2, Supplemental Digital Content,http://links.lww.com/MD/I601 , which shows the results of univariable analysis). A Multivariable logistic regression was fitted separately comparing: Patients discharged home versus admitted to the medical ward and; Patients admitted to the medical ward versus ICU. We opted for 2 logistic regression models to reflect the distinct clinical decision making processes in the ED (i.e., “discharge home” vs “admit to medical ward,” and “admit to medical ward” vs “admit to ICU”).”
Our key associations of interest were race, ethnicity, ADI, English as a primary language, homelessness, and illicit substance use (opiates, cocaine, methamphetamine); variables also included age, gender, and clinical comorbidities, including body mass index (mg/kg2) and clinical severity. We evaluated disease severity using clinical severity scores (sequential organ failure assessment, Charlson comorbidity index) and laboratory markers found in other risk severity scores,[27 (link),28 ] specifically, C-reactive protein (mg/L), ferritin (ug/L), D-dimer (ng/mL), creatine kinase (U/L), troponin (ng/L), procalcitonin (ng/mL), absolute lymphocyte count (K/mL), and blood urea nitrogen (mg/dL). Timing of admission was calculated as days after the first date of data collection (March 8, 2020). In our regression, we controlled for timing of admission and included the square of timing of admission to evaluate how the effect changed over time. To build our regression models, we first included a priori variables based on clinical understanding (i.e., age, sex, sequential organ failure assessment, C-reactive protein, ferritin, and troponin), and then added variables that were significant on univariable analysis.” Variables were excluded if they showed significant co-linearity (variance inflation factors over 10). We used stepwise, backward selection for our logistic regression model, using a P value of over 0.2 as a cutoff to remove variables. Potential interaction between significant variables was explored.
Additionally, we divided differences in number of admissions in 3 groups to visually evaluate changes in admission over time. Groups were created as general phases of the surge in SARS-CoV-2 admissions in our hospital, representing changes in comfort with diagnosis and clinical management of COVID-19. Changes in admission patterns over time were assessed using the Jonckheere–Terpstra test for trend. All data were analyzed using Stata Statistical Software (Release 16. College Station, TX: StataCorp LLC).
All tests were 2-sided and a P value < .05 was considered statistically significant. All variables were initially assessed for significance using univariable analysis comparing: Patients discharged home versus admitted to the medical ward and; Patients admitted to the medical ward versus ICU (see Tables S1 and S2, Supplemental Digital Content,
Our key associations of interest were race, ethnicity, ADI, English as a primary language, homelessness, and illicit substance use (opiates, cocaine, methamphetamine); variables also included age, gender, and clinical comorbidities, including body mass index (mg/kg2) and clinical severity. We evaluated disease severity using clinical severity scores (sequential organ failure assessment, Charlson comorbidity index) and laboratory markers found in other risk severity scores,[27 (link),28 ] specifically, C-reactive protein (mg/L), ferritin (ug/L), D-dimer (ng/mL), creatine kinase (U/L), troponin (ng/L), procalcitonin (ng/mL), absolute lymphocyte count (K/mL), and blood urea nitrogen (mg/dL). Timing of admission was calculated as days after the first date of data collection (March 8, 2020). In our regression, we controlled for timing of admission and included the square of timing of admission to evaluate how the effect changed over time. To build our regression models, we first included a priori variables based on clinical understanding (i.e., age, sex, sequential organ failure assessment, C-reactive protein, ferritin, and troponin), and then added variables that were significant on univariable analysis.” Variables were excluded if they showed significant co-linearity (variance inflation factors over 10). We used stepwise, backward selection for our logistic regression model, using a P value of over 0.2 as a cutoff to remove variables. Potential interaction between significant variables was explored.
Additionally, we divided differences in number of admissions in 3 groups to visually evaluate changes in admission over time. Groups were created as general phases of the surge in SARS-CoV-2 admissions in our hospital, representing changes in comfort with diagnosis and clinical management of COVID-19. Changes in admission patterns over time were assessed using the Jonckheere–Terpstra test for trend. All data were analyzed using Stata Statistical Software (Release 16. College Station, TX: StataCorp LLC).
Free full text: Click here
