Mortality predictors in a cohort of patients with COVID-19 admitted to a large tertiary hospital in the city of São Paulo, Brazil: a retrospective study

Abstract BACKGROUND: There is discrepant information across countries regarding the natural history of patients admitted to hospitals with coronavirus disease (COVID-19), in addition to a lack of data on the scenario in Brazil. OBJECTIVE: To determine the mortality predictors in COVID-19 patients admitted to a tertiary hospital in São Paulo, Brazil. DESIGN AND SETTING: A retrospective analysis of medical records of COVID-19 patients admitted to the Hospital Central da Irmandade da Santa Casa de Misericórdia of São Paulo. METHODS: Overall, 316 patients with laboratory-confirmed COVID-19 between March 1, 2020, and July 31, 2020, were included. The analysis included the baseline characteristics, clinical progression, and outcomes. RESULTS: The mortality rate of the sample was 51.27%. Age ≥ 60 years was determined as a risk factor after multivariate logistic regression analysis. Patients with an oxygen (O2) saturation ≤ 94% upon admission accounted for 87% of the deaths (P < 0.001). Vasoactive drugs were used in 92% (P < 0.001) of patients who progressed to death, and mechanical ventilation was employed in 88% (P < 0.001) of such patients. However, patients who received corticosteroids concomitantly with mechanical ventilation had a better prognosis than those who did not. The progressive degree of pulmonary involvement observed on chest computed tomography was correlated with a worse prognosis. The presence of thrombocytopenia has been considered as a risk factor for mortality. CONCLUSION: The main predictors of in-hospital mortality after logistic regression analysis were age, O2 saturation ≤ 94% upon admission, use of vasoactive drugs, and presence of thrombocytopenia.

In this context, since most studies so far have focused either on a specific population subset, such as patients admitted to intensive care units or those with a specific disease, for instance or indiscriminately analyze data from public and private services altogether, there is still a gap in the literature on the mortality predictors in the general admitted population in a public hospital setting. Thus, this study aimed to characterize the predictive mortality factors in a cohort of patients admitted with COVID-19 to a tertiary public hospital in the city of São Paulo, Brazil, while also seeking to determine the clinical characteristics and prognostic factors of patients representing the most severe cases of the disease and hence requiring hospitalization. 8 Therefore, these findings may support the creation and design of scientifically supported clinical management protocol for inpatient admission and hospitalized patient care in large public facilities.

OBJECTIVE
This study aimed to determine the predictors of mortality in patients with COVID-19 admitted to a tertiary hospital in São Paulo, Brazil.

Study design
This retrospective study was conducted at the Hospital Central da Irmandade da Santa Casa de Misericordia of São Paulo (ISCMSP), a large tertiary public hospital in São Paulo, Brazil.
The Hospital Central has a total capacity of 559 beds, 550 of which are currently in use. At the time of this study, 152 of these beds were intended for COVID-19 patients, of which 72 beds were intended for intensive care.
Data were collected from the electronic medical records of patients with confirmed SARS-CoV-2 infection between March 1, 2020, and July 31, 2020.
The patients were selected from a list generated by the insti-   The vital signs considered were those upon admission to the COVID-19 department, either at the time of initial care in the emergency room or upon suspicion of nosocomial infection in patients who had previously been admitted to the hospital. A heart rate between 50 and 100 beats per minute was considered as normal, 11 as was a respiratory rate between 12 and 20 breaths per minute.
The patients were considered to have been subjected to hemodialysis in all cases where this procedure had been employed during their hospital stay, whether due to a prior need or not.
The CT scans included in the analysis corresponded to the first study carried out within the first 24 h of patient admission to the COVID-19 sector.
With regard to the laboratory tests, normal values, such as those used by the clinicians at the laboratory where the tests were performed, were used as the parameters. Thus, in relation to the arterial blood gas test, an acid-base disorder was considered when the pH was below 7.35 or above 7.45, whereas hypoxemia was considered when the partial pressure of oxygen (PaO 2 ) was lesser than 80 mmHg. With respect to the blood count, anemia was considered when the hemoglobin level was less than 12 g/dL, leukocytosis when the leukocyte count was greater than 10,000 cells/mm³, and thrombocytopenia when the platelet count was lesser than 140,000 platelets/mm³. In addition, prothrombin time (PT) was classified as normal between 11 and 12.5 s, normal activated partial thromboplastin time (aPTT) between 24 and 40 s, and altered D-dimer when values were above 0.50 μd/mL.
The use of any given medication was considered as positive if at least one dose thereof was administered.

Statistical analysis
Data analyses were performed using the SPSS version 25.0 software (IBM Corporation, Armonk, New York, United States).
The statistical analysis of the study population was based on whether a given variable could be characterized as a mortality predictor.
Thus, several variables related to the clinical characteristics, supplementary tests, and treatments were compared according to patient progression (discharge from hospital or death). The corresponding frequencies were established, and Pearson's chi-square test was used for the analysis, with a level of significance (P value) of 0.05.
Subsequently, to assess the risk factors associated with death, a multivariate logistic regression analysis was performed using the stepwise method, with a P value ≤ 0.05 requirement for data entry.
However, since the number of variables was quite extensive, pre-selection was carried out using bivariate analysis and setting a P value ≤ 0.20, according to Pearson's chi-square test or Fisher's exact test, along with excluding those with a data loss greater than 10%.

RESULTS
A total of 316 patients admitted to the hospital with COVID-19 were included in the analysis. The mortality rate in this population was 51.27% (162/316; 95% confidence interval [CI], 45.75-56.78%).
In addition to the overall mortality rate of the study sample, three other groups were considered (clinical, supplementary, and treatment characteristics) to better elucidate the potential predictors of mortality.
However, it is noteworthy that for each item, the number of patients may vary according to the availability of data.

Clinical characteristics that are mortality predictors
Of the 316 patients, 262 (82.91%) patients were communityacquired, and SARS-CoV-2 infection was the primary cause of hospitalization. The other 54 (17.09%) cases were nosocomial cases. Subsequently, the same patient samples were assessed for vital signs upon admission. Oxygen saturation, as measured by pulse oximetry, had a significance probability of < 0.001; thus, it was statistically significant. Considering the admission data, patients with saturation ≤ 94% appeared to have a worse prognosis ( Table 2).
The use of vasoactive drugs, mechanical ventilation, and hemodialysis as potential prognostic indicators was also analyzed, with results shown in Table 3. However, after a multivariate logistic regression analysis, only the use of vasoactive drugs could be considered as a predictor of mortality.
In eTable 2 in the Supplement (https://drive.google.com/ file/d/1WwAiHPfkpZjC1wvaKArI6FpAoEtR-8e9/view), symptoms are displayed according to their order of relevance as potential predictors. For this analysis, a report of a subgroup of 307 patients was considered, whereas those from which patient histories could not be retrieved were excluded. Of the symptoms listed, only the absence of fever was higher in the in-hospital mortality group.

Supplementary characteristics serving as mortality predictors
To assess the possible associations between supplementary tests and hospitalization outcomes, subgroups were considered according to the available tests. Altogether, a chest CT analysis was performed for 116 patients (detailed in eTable 3 in the Supplement: https://drive.google.com/file/d/1WwAiHPfkp ZjC1wvaKArI6FpAoEtR-8e9/view), who represented the sample with such tests accompanied by their respective reports.
In addition, entry oxygen saturation and the degree of impairment, as seen on chest CT, were correlated in a subgroup of 113 patients.
For chest CT analysis, pulmonary involvement was quantified in the imaging study, and the main findings were ground-glass opacity, consolidation, pleural thickening, and reticular opacities, with ground-glass opacity being the most common finding.
Three degrees of pulmonary involvement were considered (< 25%, between 25% and 50%, and > 50%); the greater the alterations, the greater the correlation with death, and therefore, the worse the prognosis. Additionally, when correlated with oxygen   saturation upon admission lower or equal to 94%, the progressive severity of the pulmonary imaging findings remained associated with a poorer prognosis; taken together, these two constitute an important set of predictive mortality factors. However, this variable could not be included in the multivariate logistic regression analysis because of excessive data loss. The laboratory characteristics were also evaluated, as shown in Table 4.
The altered levels of PT or aPTT, and anemia, leukocytosis, and thrombocytopenia, were more prevalent in the population that progressed to death. However, the in-hospital mortality was only associated with thrombocytopenia after a logistic regression analysis.

Treatment characteristics
Although the only treatment recommended by the WHO is the use of systemic corticosteroids in severe or critical cases of COVID-19, 12  None of the evaluated medications showed an association with better prognoses in the study population.
The correlation between patients on mechanical ventilation and corticosteroid use was also evaluated. Thus, patients on mechanical ventilation who received corticosteroids had a mortality rate of 72.73% (40/55), whereas for those who did not receive corticosteroids, the rate was 91.89% (102/111) with a P value of 0.006.
Finally, as shown in Table 5, some variables were chosen for the multivariate logistic regression. Therefore, age > 60 years, O 2 saturation ≤ 94%, use of mechanical ventilation without concomitant corticosteroid therapy, thrombocytopenia, and the use of vasoactive drugs were considered as the risk factors for mortality.

DISCUSSION
The COVID-19 pandemic is a serious threat to public health owing to the uncertainty it has brought worldwide. The panorama of a new virus with high transmissibility, high heterogeneity, and pathophysiology has not yet been fully elucidated, with cases ranging from an asymptomatic clinical picture to death, PaO 2 = partial pressure of oxygen; PT = prothrombin time; aPTT = activated partial thromboplastin time; NS = not significant; * statistically significant. achieving a reduction in mortality due to the disease has become a priority in the international scientific community.
Furthermore, reports issued by the WHO show that the mortality rate associated with COVID-19 varies greatly across different countries, indicating that particular geographic features likely play a role in the progression of the disease, and not every study can be extrapolated to individual realities.
In addition, our sample was composed solely of patients from the SUS, the nation's public healthcare services, on which approximately 70% of the population relies. Therefore, protocols designed to prioritize patients for hospital admission are essential to avoid overcapacity and the consequences of avoidable hospitalization.
Hence, the identification of predictive mortality factors that could aid in the establishment of protocols for the care of the studied population is essential, especially in the management of the most severe cases of the disease spectrum requiring hospitalization, and our study was designed with the aim of filling that gap. patients distributed across hospitals in northern Italy, 14 our study population was characterized by comprising mostly older adult males and patients with comorbidities; being older than 60 years of age can also be considered an independent risk factor for mortality.
Moreover, the mortality rate (51%) in our study was much higher than that reported by Richardson et al. 13  With regard to the characteristics upon admission, an oxygen saturation level ≤ 94%, as measured by pulse oximetry, had an odds ratio of 3.8, i.e, such patients were 3.8-fold more likely to progress to death than those with an oxygen saturation > 94%. The most prevalent symptoms in the sample were dyspnea, coughing, and fever; however, the presence of none may be associated with higher in-hospital mortality.
Although CT scans of the chest accompanied by their respective reports were available for only 37% of patients, they also showed that the greater the degree of pulmonary involvement, the worse the prognosis of the patient. The conclusion remained the same when associated with entry oxygen saturation, i.e., these parameters could be useful for the rapid stratification of cases being admitted to the hospital.
As already explored in other studies, 16 the hypercoagulable state seems to be associated with higher mortality via pathophysiological mechanisms that are yet to be fully elucidated. This phenomenon can also be seen in our study population, as translated by the association between higher mortality and altered PT and activated thromboplastin time, in addition to thrombocytopenia, which can be considered as a risk factor, with an odds ratio of 20.0, as found by logistic regression.
The use of vasoactive drugs and the need for mechanical ventilation were also highly prevalent in the death group, with the former being considered as a risk factor for such outcomes. However, the concomitant use of corticosteroids in mechanically ventilated patients, as recommended by the WHO, 12 could result in better prognoses.
This study has a few limitations, primarily due to its observational and retrospective nature; not all the data initially planned for analysis could be retrieved. Therefore, the total sample, comprising 316 patients, could not be analyzed for all the categories initially envisaged. In addition, variables, such as ethnicity and pulmonary involvement as seen on chest CT scans were not only described by a large number of different professionals, but also by dependent examiners. It is also worth noting that the hospital where this study was carried out is a referral facility located in the central region of the city and receives patients transferred from various other less complex minor hospitals/health care system units, which may indicate that the patient's condition could already be more advanced at the time of the initial care provided by our team and consequently result in worse outcomes and higher mortality.
Our study contributed to determining the phenotype of patients admitted to the hospital with COVID-19 at a higher risk of in-hospital mortality, with the aid of identifying those in need of hospital admission, priority care, case stratification, signs of clinical deterioration, and worse outcomes.

CONCLUSION
According to our findings, age ≥ 60 years, oxygen saturation ≤ 94% upon admission, use of vasoactive drugs, and thrombocytopenia were the main clinical predictors of in-hospital mortality.
In addition, mechanical ventilation, pulmonary involvement as seen on chest CT scans, altered values of PT, aPTT, leukocytosis, and anemia were more prevalent in the death group; however, they could not be considered as risk factors for mortality after adjustments.