We included 171 patients with complete data for all variables (53 non-survivors and 118 survivors) in the multivariable logistic regression model. We found that older age, higher SOFA score, and d-dimer greater than 1 μg/mL at admission were associated with increased odds of death (table 3). When adjusting for study centre, our generalised linear model showed similar results (appendix p 5).
For survivors, the median duration of viral shedding was 20·0 days (IQR 17·0–24·0) from illness onset, but the virus was continuously detectable until death in non-survivors (table 2; figure 1). The shortest observed duration of viral shedding among survivors was 8 days, whereas the longest was 37 days. Among 29 patients who received lopinavir/ritonavir and were discharged, the median time from illness onset to initiation of antiviral treatment was 14·0 days (IQR 10·0–17·0) and the median duration of viral shedding was 22·0 days (18·0–24·0). The median duration of viral shedding was 19·0 days (17·0–22·0) in patients with severe disease status and 24·0 days (22·0–30·0) in patients with critical disease status.
Figure 1. Clinical courses of major symptoms and outcomes and duration of viral shedding from illness onset in patients hospitalised with COVID-19
Figure shows median duration of symptoms and onset of complications and outcomes. ICU=intensive care unit. SARS-CoV-2=severe acute respiratory syndrome coronavirus 2. ARDS=acute respiratory distress syndrome. COVID-19=coronavirus disease 2019.
Major laboratory markers were tracked from illness onset (figure 2). Baseline lymphocyte count was significantly higher in survivors than non-survivors; in survivors, lymphocyte count was lowest on day 7 after illness onset and improved during hospitalisation, whereas severe lymphopenia was observed until death in non-survivors. Levels of d-dimer, high-sensitivity cardiac troponin I, serum ferritin, lactate dehydrogenase, and IL-6 were clearly elevated in non-survivors compared with survivors throughout the clinical course, and increased with illness deterioration (figure 2). In non-survivors, high-sensitivity cardiac troponin I increased rapidly from day 16 after disease onset, whereas lactate dehydrogenase increased for both survivors and non-survivors in the early stage of illness, but decreased from day 13 for survivors.
Figure 2. Temporal changes in laboratory markers from illness onset in patients hospitalised with COVID-19
Figure shows temporal changes in d-dimer (A), lymphocytes (B), IL-6 (C), serum ferritin (D), high-sensitivity cardiac troponin I (E), and lactate dehydrogenase (F). Differences between survivors and non-survivors were significant for all timepoints shown, except for day 4 after illness onset for d-dimer, IL-6, and high-sensitivity cardiac troponin I. For serum ferritin (D), the median values after day 16 exceeded the upper limit of detection, as indicated by the dashed line. COVID-19=coronavirus disease 2019. IL-6=interleukin-6.
Median time from illness onset to dyspnoea was similar in survivors and non-survivors, with a median duration of dyspnoea of 13·0 days (9·0–16·5) for survivors (table 2; figure 1). In survivors, the median duration of fever was 12·0 days (8·0–13·0) and cough persisted for 19·0 days (IQR 12·0–23·0; figure 1). 62 (45%) survivors still had cough on discharge and 39 (72%) non-survivors still had cough at the time of death. The dynamic profiles of fever, cough, and dyspnoea are shown in the appendix (p 6). Sepsis developed at a median of 9·0 days (7·0–13·0) after illness onset among all patients, followed by ARDS (12·0 days [8·0–15·0]), acute cardiac injury (15·0 days [10·0–17·0]), acute kidney injury (15·0 days [13·0–19·5]), and secondary infection (17·0 days [13·0–19·0]). The initiation time and duration of systematic corticosteroid use was also similar between the two groups. Among non-survivors, the median time from illness onset was 10·0 days (7·0–14·0) to sepsis, 12·0 days (8·0–15·0) to ARDS, 14·5 days (9·5–17·0) to acute cardiac injury, and 17·0 days (13·0–19·0) to secondary infection (figure 1; table 2). Among survivors, secondary infection, acute kidney injury, and acute cardiac injury were observed in one patient each, occurring 9 days (acute kidney injury), 14 days (secondary infection), and 21 days (acute cardiac injury) after illness onset.
The median time from dyspnoea to intubation was 10·0 days (IQR 5·0–12·5) for patients who received invasive mechanical ventilation and the time from invasive mechanical ventilation to occurrence of ventilator-associated pneumonia was 8·0 days (2·0–9·0; figure 1).Continuous and categorical variables were presented as median (IQR) and n (%), respectively. We used the Mann-Whitney U test, χ2 test, or Fisher's exact test to compare differences between survivors and non-survivors where appropriate. To explore the risk factors associated with in-hospital death, univariable and multivariable logistic regression models were used. Considering the total number of deaths (n=54) in our study and to avoid overfitting in the model, five variables were chosen for multivariable analysis on the basis of previous findings and clinical constraints. Previous studies have shown blood levels of d-dimer and Sequential Organ Failure Assessment (SOFA) scores to be higher in critically ill or fatal cases, whereas lymphopenia and cardiovascular disease have been less commonly observed in non-critical or surviving patients with SARS-COV-2 infection.5, 6, 12Similar risk factors, including older age, have been reported associated with adverse clinical outcomes in adults with SARS and Middle East respiratory syndrome (MERS).3, 13 Some laboratory findings, including alanine aminotransferase (ALT), lactate dehydrogenase, high-sensitivity cardiac troponin I, creatine kinase, d-dimer, serum ferritin, and IL-6, might be unavailable in emergency circumstances. Therefore, we chose lymphocyte count, d-dimer, SOFA score, coronary heart disease, and age as the five variables for our multivariable logistic regression model.
We excluded variables from the univariable analysis if their between-group differences were not significant, if their accuracy was unconfirmed (eg, exposure, which was self-reported), if the number of events was too small to calculate odds ratios, and if they had colinearity with the SOFA score.
We compared patient characteristics between the two hospitals and used a generalised linear model to adjust for possible differences in patients’ characteristics and treatment between the two study centres.
A two-sided α of less than 0·05 was considered statistically significant. Statistical analyses were done using the SAS software (version 9.4), unless otherwise indicated.