Risk factors on admission and condition at discharge of 529 consecutive COVID-19 patients at a tertiary care center in Santiago, Chile


 Background: The first case of COVID-19 was reported in Chile on March 3, 2020. Public and private hospitals were managed in a centralized manner. On May 30, Chile had 99,668 cases, 1054 deaths, 1383 ICU patients, 1174 patients on invasive mechanical ventilation (IMV), and 51 patients on non-invasive ventilation (NIMV). Research question: What are the variables associated with condition at discharge?Method: We performed a retrospective cohort study of 529 patients with a positive RT-PCR for SARS CoV-2who were consecutively discharged between March 14 and June 4, 2020, at Clínica Dávila, Santiago. Patients were analyzed according to laboratory variables on admission, Quality-Adjusted Life Year (QALY) score, health insurance, and type of respiratory support. Condition at discharge was survivor, non-survivor, or transfer to another center. Differences were evaluated by Chi-square test, Student’s t test, or Mann–Whitney U test. Logistic regression analysis was performed to identify variables that were predictive of condition at discharge.Results: Median (interquartile range, IQR) age was 49 (37–62) years, and the median (IQR) stay in the hospital was 6 (3–10) days. A total of 352 patients (66.5%) had respiratory symptoms, 177 (33.4%) had other symptoms or diagnoses on admission, and 116 required ventilatory support; 448 (84.7%) were survivors, 54 (10.2%) were non-survivors, and 27 (5.1%) were transferred. The median ages of the survivors and non-survivors were 46 (36–59) and 75.5 (66–84), respectively.Having state health insurance increased the risk of death by 2.8-fold (OR, 2.825; 95% CI: 1.383–5.772; P = 0.004). Multivariate analysis revealed the following predictive variables: age ≥ 60 years (OR, 15.3; 95% CI: 7.25–32.2; P = .001); PaO2/FiO2 on admission ≤ 200 vs > 200 (OR, 5,205; CI 95%: 1,942–13,94); high-sensitivity troponin, ≥ 15 vs <15 ng /L (OR, 5,163; 95% CI: 1.95–13,64; P = .001); and QALY ≤ 15 vs > 15 points (OR, 14,011; 95% CI: 4,826–40,679; P=.001).Interpretation: The variables analyzed and patient’s clinical evolution may allow assignment of ICU beds to patients with the greatest chance of survival, especially in countries or regions where this resource is limited.


Introduction
On March 3, 2020, the rst case of COVID-19 was reported in Chile. Over the following days, a growing number of cases were reported in the central and southern zones of the country, reaching the Metropolitan Region in mid-March 2020. The Chilean Ministry of Health assumed centralized control of all beds in public and private hospitals. By May 30, 2020, in the country as a whole, 99668 patients were infected and 1054 had died. A total of 1383 patients were in intensive care units (ICUs); 1174 were on invasive mechanical ventilation (IMV) and 51 on non-invasive mechanical ventilation (NIMV). At the same time, the Chilean Society of Intensive Medicine reported that 253 patients were on IMV outside critically ill adult units. Clínica Dávila, Hospital San José, and Hospital Clínico de la Universidad de Chile are the main level III hospitals in the northern area of Santiago.
Before the pandemic Clínica Dávila had 647 beds, including beds for adult and pediatric cases of varying degrees of complexity.
By June 4, 2020, Clínica Dávila had 334 COVID-19 beds with an occupancy of 95.8%. The number of ICU beds for adult COVID-19 patients increased by 500%, from 12 to 62, including 15 ICU beds in surgical rooms with all patients on IMV. The number of intermediate care unit (IMCU) beds increased by 300%, from 24 to 72, with 68 hospitalized patients, of whom 30 were connected to NIMV, 16 were connected to high-ow nasal cannula (HFNC), and four were ventilated by tracheostomy (TQT) through adapted noninvasive ventilators.
Overall, 81% of COVID-19 infections are mild, 14% severe, and 5% require intensive care 1 . The mortality rates published by China, Italy, and the United States range from 1.4% among hospitalized patients 2 to 61.5% among critically ill patients 3 .
Objectives Primary objective: To analyze demographic, clinical, and laboratory characteristics at admission that may have prognostic value regarding the condition of patients at discharge. Secondary objective: To analyze the type of ventilatory support used and condition at discharge.

Patients and method
This study was conducted in accordance with the Declaration of Helsinki 4 and approved by the Ethical and Scienti c Committee of Clínica Dávila.

Patients and data collection
All patients with positive laboratory tests for SARS CoV-2, hospitalized and discharged between March 14, 2020 and June 04, 2020, were included in this study. The patients were discharged in the following sequence: in March, six patients; in April, 17 patients; in May, 452 patients (85.4%); and in the rst 4 days of June, 54 patients.
COVID-19 disease was diagnosed based on guidelines from the World Health Organization (WHO). Con rmed cases were patients with positive results from real-time polymerase chain reaction (RT-PCR) test for SARS CoV-2, performed on samples of the upper respiratory tract harvested by nasopharyngeal swab 5 . Positive patients were entered into the mandatory noti cation system, Epivigila (https://epivigila.org.cl) 6 , created by the Ministry of Health. The most frequent respiratory symptoms were dyspnea, odynophagia, cough, and chest pain. Other predominant symptoms were vomiting, diarrhea, abdominal pain, and myalgias.
Gender, age, health insurance, duration of symptoms before admission, and comorbidities (T2DM, HT, cancer, HIV, immunosuppression from other causes, heart failure, kidney failure, obesity, coronary heart disease, bronchial asthma, active smoking, and chronic obstructive pulmonary disease) were recorded. For all patients, the Quality-Adjusted Life Year (QALY) score was calculated. QALY is a generic measure of disease burden that is used to evaluate the impact of therapeutic measures on the quality of survival expected with or without an intervention. It considers the life expectancy of a country or region from which the patient's age number is subtracted; the resultant value is multiplied by 1 in the absence of comorbidities, or in the case of existing comorbidities, 0.1 is subtracted from factor 1 for each compensated comorbidity, 0.2 for each decompensated comorbidity (or in the case of a semi-dependent patient), and 0.3 if the patient was previously bedridden 7 . Also, duration of hospitalization and noninvasive and invasive ventilatory support (expressed in days) were recorded. The type of bed used by the patient before discharge or transfer to another center was also recorded.
Sample collection, ethological agent, and laboratory tests Laboratory examinations on admission were taken in the emergency room, or within the rst 3 hours of admission, including RT-PCR for SARS CoV-2, arterial blood sample for arterial gases, relationship between the arterial partial pressure of O 2 (expressed in mmHg) and the fraction of O 2 that the patient was breathing at the time the sample was obtained (PaO 2 /FiO 2 ), ferritin, D Dimer, Creactive protein (PCR), procalcitonin, platelet count, white blood cell count, viral panel by real-time PCR (identifying 17 viruses), bacterial panel by real-time PCR (identifying seven bacteria), urinary antigen for Pneumococcus and Legionella, blood cultures, and positive expectoration cultures.

Methods of administration of oxygen and ventilatory support used
Administration of O 2 was carried out as follows: 1. Through the nose at a ow rate of up to 4 L/min. 2. High-ow multi-vent mask with FiO 2 from 40% to 50%. 3. Non-rebreather mask that delivered an FiO 2 between 50% and 90%.
Administration of O 2 through a HFNC was performed using AIRVO 2 (Fisher & Paykel, New Zealand).
Discharge conditions: survivor, non-survivor, or transfer to another institution.

Statistical analysis
A retrospective cohort study was performed on 529 patients with a positive RT-PCR for SARS CoV-2 consecutively discharged between March 14 and June 4, 2020 at Clínica Dávila, Santiago. Clinical information was obtained from the electronic medical record of Clínica Dávila and collected in a database designed to ensure that the identities of the patients was protected.
Categorical variables were described using absolute and relative frequencies, and quantitative variables were described using means and standard deviation for those with normal distributions and with median and interquartile range (Q1, Q3) for those without normal distributions. For the categorical variables, association with condition at discharge was evaluated using the Chisquare test, whereas for quantitative variable, association was evaluated using Student´s test (normal distribution) or Mann-Whitney U test (non-normal distributions). To assess risk factors for discharge status, we used univariate and multivariate logistic regression models. First, the variables were analyzed individually; those with a p-value less than 0.1 were incorporated into a stepwise model with "forward-selection", and variables with a Pearson correlation greater than 0.8 were excluded to avoid collinearity and choose variables with greater predictive capacity. Finally, univariate models were used with variables categorized according to clinical criteria, and variables with a p-value less than 0.1 were again incorporated into a stepwise model. Signi cance level (α) less than or equal to 0.05. All analyses were performed using STATA v14.2 IC software (StataCorp. LLC, USA).
Of the 529 patients, 352 (66.5%) reported respiratory symptoms in the emergency department and had a chest CT scan with a COVID-19 pattern; on the other hand, in the remaining 177 patients (33.4%), the grounds for hospitalization were non-respiratory symptoms. Of those, 84 patients (15.9%) had digestive symptoms (nausea, vomiting, diarrhea, or di cult-to-control abdominal pain) or intense myalgia and headache that had not responded to outpatient management. Ninety-three patients (17.6%) had other grounds for admission, but the presence of SARS CoV-2 was con rmed due to the obligatory testing of all patients who were hospitalized during that period. Thirty-ve were pregnant adult women hospitalized for pregnancy complications or in labor. Only two puerperal patients presented respiratory symptoms and required IMV. Both had pulmonary compromise on chest CT, and neither died. Thirty-two patients were admitted for other non-infectious causes (acute coronary syndrome, deep vein embolism thrombus, and others), and 23 were admitted for other concomitant infections (acute cholecystitis, acute pyelonephritis, cholangitis, and others) ( Table 1).
Of 529 patients, 177 did not receive oxygen or ventilatory support, 236 received oxygen at variable rates or HFNC, and 116 received ventilatory support with NIMV or IMV. None of the patients underwent extracorporeal membrane oxygenation (Table 3).
In our cohort, obesity was not identi ed as a poor prognostic factor: 15.8% of those who survived and 16.7% of those who died were obese (p = 0.437) ( Table 2).
Univariate and multivariate logistic analysis of demographics, comorbidities, and laboratory variables.
In the univariate analysis, the clinical variables on admission that differed signi cantly between survivors and non-survivors were age, hypertension, and diabetes. Laboratory variables that differed signi cantly were procalcitonin, ferritin, PaO 2 /FiO 2 at admission, leukocytes, double dimer, and creatinine (Table 5).

Discussion
In a previous series of 393 consecutive cases examined in New York city 10 , the median age was 62.2 years (48.6-73.7) and 40 patients died, corresponding to an overall lethality of 10.2%. In our series, the lethality was the same (10.2%), but our patients were younger with a median age of 49 years (37-62). In the previously described series, 130 patients needed IMV (33%), and the lethality of that group was 14.6% (19 patients). In our series, 84 patients (15.9%) needed IMV, of whom 46 (54.7%) survived, 18 died (21.4%), and 20 (23.8%) were transferred to other centers ( Table 2).
The strategy used in the New York group involved early IMV essentially without the use of HFNC or NIMV. In our series, patients who had no indication for intubation upon arrival at the emergency room received O 2 at increasing ow rates, with or without HFNC and NIMV. Lack of response led to IMV (Table 2).
In an analysis of a consecutive series of 78 patients, 11 patients (14.1%) presented with progression of respiratory failure and 62 improved or stabilized. Those who deteriorated were older, tended to be smokers, and had more comorbidities 11 . In our cohort, patients with state health insurance were 2.8 times more likely to die than patients with private health insurance. In part, this may have been because the mean age of patients with state insurance was 53.2 ± 17,8 years, whereas that of patients with private insurance was 45 ± 17.3 years (p = 0.001). In the group with state health insurance, 38% were ≥ 60 years old. Meanwhile, among the patients with private health insurance, only 19.4% of patients were ≥ 60 years old (Tables 5 and 6).
In patients with severe ARDS admitted to the ICU, PaO 2 /FiO 2 has prognostic value 8 . In our clinical practice during the pandemic, we used PaO 2 /FiO 2 ≤ 200 on admission to identify patients who needed to enter the intermediate care unit for HFNC or NIMV. On the other hand, patients with PaO 2 /FiO 2 > 200 were hospitalized in the general ward with supplementary oxygen . We retrospectively analyzed our series and found that this cut-off point had prognostic value.
In our patients, we categorized patients as QALY ≤ 15 or QALY > 15 points. In the multivariate analysis, this cut-off point showed a strong statistical signi cance for risk of death at discharge. Regardless of their clinical condition, we calculated the QALY score for all patients who were enrolled in our study. The purpose of this calculation was to identify patients who would bene t from an ICU bed in the clinical scenario of severe ARDS. This was important due to the limited number of these units and the need to transfer to another center with free ICU beds. Transfer was under the jurisdiction of the Centralized Bed Management Unit (UGCC) of the Chilean Ministry of Health. In our cohort, 27 patients were transferred to other centers with an available ICU bed.
One review of support modalities in acute respiratory failure in patients with COVID-19 recommend the use of HFNC in cases with mild respiratory insu ciency (PaO 2 /FiO 2 between 200 and 300). In cases with moderate respiratory insu ciency (Pa/FiO 2 between 100 and 200), NIMV alone could be useful. The authors of that review also suggested that rotation between the two modes (HFNC and NIMV) may be a bene cial strategy when PaO 2 /FiO 2 increases and respiratory rate or volume/minute improves 12 . However, the timely availability of IMV when this mixture fails depends on the center, as the prolonged use of NIMV without an evident clinical response and persistence of increased respiratory work can generate more damage 13 .

Conclusions
Our analysis of 529 consecutive patients included patients who were younger than the cases reported in North America and Europe. Lethality was 10.2%. A third of patients who were hospitalized had no respiratory symptoms. The remaining two-thirds had varying degrees of respiratory failure. Of those, 116 (21.9%) required ventilatory support.
Patients over age of 60 or with QALY scores <15 were at higher risk of dying. From the admission laboratory, patients with PaO 2 /FiO2 < 200, high-sensitivity troponin ≥ 15 ng/ml, or creatinine > 1.4 mg% also had a higher risk of dying.
Due to the reduced availability of ICU beds in the pandemic context, it was important to determine variables upon admission that allow clinicians to assign those beds to patients with the greatest chance of survival, especially in countries or regions where this resource is limited.
The differences in the prognosis that we observed between patients with state health insurance and private health insurance were related to the older age and higher burden of disease, expressed by the QALY score, in the former group.

Research limitations
We did not register the severity of the lung involvement in the lung CT SCAN of patients and follow-up laboratory variables. The only variable that we registered post-admission was the determination of PaO 2 /FiO 2 before connection to NIMV or IMV. We consider that the tobacco habit was under registered. Information about obesity was absent in some patients.

Con icts of interest
The authors have no con icts of interest related to this study.