Discussion
In this observational study including 1503 patients with COVID-19 admitted to the ICUs of the largest public hospital in Sao Paulo, we found that the 28-day mortality rate was 44% (95%CI 42–47) and 60-day mortality was 49%. Invasive mechanical ventilation was used for 79% of patients, vasopressors for 73%, and renal replacement therapy for 35%. Protective ventilation was used for 82% of patients receiving mechanical ventilation on the first 24 h of ICU stay and was independently associated with increased survival.
This is the first large cohort study of patients COVID-19 in a low- and middle-income country (LMIC) and describes the outcomes of patients treated in a large academic hospital in the context of a state preparedness plan. The hospital was the primary referral center for critically ill patients with COVID-19 and received patients from all regions of the metropolitan area of Sao Paulo, which has a total population of over 23 million people. The hospital is public, and patients were treated at no cost in accordance with the Brazilian universal health system. The preparedness plan involved cohorting COVID-19 patients in a building dedicated for the care of these patients, the creation of surge ICUs, and hiring or reallocation of healthcare professionals.
Hospital mortality at 28 days in our study was 44% (95%CI 42%–47%), and 60-day hospital mortality was 49% (95%CI 46%-51%), lower than the mortality found in a large epidemiological study based on a nationwide database with more than 250,000 cases across Brazil, which found 57% mortality for patients admitted to the ICU [
34]. These figures are comparable to previous reports, showing wide variability in mortality [
3‐
11], and reflecting differences between countries and health systems [
35]. Importantly, many previous studies reported the mortality rate, while a considerable proportion of patients were still in the hospital, therefore underestimating mortality. In our study, patients were followed for at least 28 days, only 6% were transferred before 28 days, and 3.9% were still in the hospital at the end of follow-up. A high mortality rate was expected, as studies show that the burden of critical illness is higher in LMICs [
36,
37] and large epidemiological studies performed in several ICUs across Brazil found high mortality for patients under mechanical ventilation [
38] and for patients with sepsis [
39]. In addition, our study was conducted during the first surge of cases, when no treatment was known to be effective, and mortality was higher [
40]. Corticosteroids were used for only 25% of patients, since the results of the large randomized-controlled trial that showed that dexamethasone reduced mortality in hospitalized patients were released in mid-June, close to the end of our study period [
41].
At ICU admission, the median duration of symptoms was 9 days, which is longer that most series [
3,
10,
13], 39% of patients were already receiving vasopressors, and 60% were under invasive mechanical ventilation. Organ dysfunction on the first 24 h of admission, measured by SOFA, was higher than in most reports [
3,
20,
21]. None of the previous studies reported SAPS 3, but a few studies report APACHE II of 13 to 16 [
3,
20,
21], corresponding to 25% expected hospital mortality. These findings show that admission to the ICU was delayed, which may have contributed to high severity of disease at admission and higher mortality, and reflect barriers to access to health care in LMICs.
Gas exchange was severely compromised, as shown by the median PaO
2/FIO
2 of 171, compatible with moderate Acute Respiratory Distress Syndrome (ARDS). Ventilatory parameters on day 1 were similar to other reports [
13,
20,
21], and within protective levels for 82% of patients. Adherence to a protective ventilation strategy, which is recommended by experts [
42] and by the institutional protocol, was not complete but was reasonably high, and consistent with what was observed for ARDS [
43] and for COVID-19 patients [
21,
22]. In contrast, using more liberal tidal volumes, under the assumption that COVID-19 patients may have near-normal compliance, as recently proposed [
17], has not been proven to confer protection and may have contributed to nonadherence to protective ventilation in our study.
We found associations of several ventilatory variables, including plateau pressure and driving pressure with mortality in COVID-19, similarly to what has been previously shown for ARDS due to other causes [
43]. Interestingly, when ventilatory parameters were assessed one at a time, plateau pressure and driving pressure, but not tidal volume, or PEEP, were associated with mortality. This finding is compatible with our observation that limitation of both tidal volume and plateau pressure conferred an advantage not only in terms of lower tidal volume and plateau pressures but also in terms of lower driving pressures.
Compliance on day 1 was moderately low and had a wide distribution. This pattern does not support recently proposed conceptual models of two phenotypes in ARDS caused by COVID-19 [
14‐
17]. Our findings are in line with most recently published studies in COVID-19 patients, which showed lower respiratory system compliances in COVID-19 patients [
20‐
22]. It is possible that findings of normal compliance in severe respiratory failure in COVID-19 were influenced by the small sample sizes in early studies and timing from disease onset until the compliance measurements.
In our study, compliance was associated with mortality and provided relevant information to describe the application of protective ventilation. We found that for patients with lower compliance, non-protective ventilation was most commonly due to higher plateau pressure and driving pressure, while for patients with higher compliance, non-protective ventilation was most commonly due to higher tidal volumes. PEEP levels were moderate, in contrast to some reports of need for high PEEP [
4,
10]. Prone position and PEEP titration were the most common advanced therapies used for respiratory failure on the first 24 h of ICU stay, similar to another large cohort of COVID-19 patients [
21]. As per the institutional protocol, prone was indicated for all patients with PF ratio < 150 mmHg unless they had a contraindication, and was used for 36% of patients during ICU stay. The relatively low use of prone, given the severity of patients, may have impacted survival and could be related to high burden of care during the surge of cases. Similar findings were reported in large cohorts from northern Italy [
4] and New York city [
10], which showed that prone was used in 17% and 27% of patients, respectively.
Most patients needed advanced life support, reflected by a high incidence of use of invasive mechanical ventilation and vasopressors. Noninvasive ventilation was used for only 21% of patients prior to intubation. This finding could be due to high severity of disease at admission, lack of resources and concerns with aerosolisation with noninvasive ventilatory methods. Our findings are in line with a large multicentric study across several ICUs in Brazil, showing that noninvasive ventilatory support use increased over 8 months after the first surge of cases and was associated with decreased mortality [
40]. Renal replacement therapy was used for 35% of patients, which is associated with high cost and higher burden of care, in addition to high mortality.
Our study has several important limitations: it was performed at a single-center, a large academic hospital with an institutional protocol that included ventilatory management, and therefore, the results may not be generalizable to other hospitals in Brazil. However, patients were referred from all regions of the metropolitan area, the 20 ICUs were staffed with physicians with diverse backgrounds, and some were staffed with health professionals from private hospitals in Sao Paulo who sent their teams to contribute with the state plan during the first surge of the pandemic; most patients were referred and transferred from other hospitals, possibly representing the most severe cases in Sao Paulo; part of the data were collected retrospectively, since no data were collected until we obtained study approval in our ethical committee. However, we believe that the impact on data accuracy was minimal, since we had specific electronic forms for COVID-19 symptoms in our electronic medical record, which were filled out at hospital admission for all patients, and structured ICU forms which include detailed ventilatory parameters and ICU support measures such as use of vasoactive drugs and sedation; we only collected ventilatory parameters on the first day of the mechanical ventilation, and therefore, the adherence to protective ventilation over the following days and its association with survival is unknown; we also recognize that many other practices may have impacted outcomes over the course of the study, for which we could not account; and finally, it was an observational study, and therefore, the relationship between protective ventilation and survival may be influenced by residual confounding and causality cannot be assumed. The study also has strengths: all patients admitted during the study period were included, avoiding selection bias; the sample size was large, allowing for the identification of risk factors and precise estimation of outcomes; we recorded detailed ventilatory parameters on the first 24 h of ICU stay, which allowed us to estimate the association between ventilatory strategies and survival; follow-up was long enough and complete, providing an accurate estimation of ICU survival; missing data for clinical data were minimal and quality measures provided accurate estimation of outcomes.
Acknowledgements
We would like to acknowledge the outstanding work performed by healthcare workers and staff in our hospital during the COVID-19 crisis. We would also like to thank the Hospital das Clinicas COVID-19 crisis committee and the informatics department (NETI) for their support for this project.
We would like the names of the individual members of the Group to be searchable through their individual PubMed records
EPICCoV study group: Adriana Hirota, Alberto Kendy Kanasiro, Alessandra Crescenzi, Amanda Coelho Fernandes, Anna Miethke-Morais, Arthur Petrillo Bellintani, Artur Ribeiro Canasiro, Bárbara Vieira Carneiro, Beatriz Keiko Zanbon, Bernardo Pinheiro Senna Nogueira Batista, Bianca Ruiz Nicolao, Bruno Adler Maccagnan Pinheiro Besen, Bruno Biselli, Bruno Rocha De Macedo, Caio Machado Gomes De Toledo, Carlos Roberto Ribeiro De Carvalho, Caroline Gomes Mol, Cassio Stipanich, Caue Gasparotto Bueno, Cibele Garzillo, Clarice Tanaka, Daniel Neves Forte, Daniel Joelsons, Daniele Robira, Eduardo Leite Vieira Costa, Elson Mendes Da Silva Júnior, Fabiane Aliotti Regalio, Gabriela Cardoso Segura, Giulia Sefrin Louro, Gustavo Brasil Marcelino, Yeh-Li Ho, Isabela Argollo Ferreira, Jeison Oliveira Gois, Joao Manoel Da Silva-Jr, Jose Otto Reusing Junior, Julia Fray Ribeiro, Juliana Carvalho Ferreira, Karine Vusberg Galleti, Katia Regina Silva, Larissa Padrao Isensee, Larissa Santos Oliveira, Leandro Utino Taniguchi, Leila Suemi Letaif, Lígia Trombetta Lima, Lucas Yongsoo Park, Lucas Chaves Netto, Luciana Cassimiro Nobrega, Luciana Bertocco Paiva Haddad, Ludhmila Abrahao Hajjar, Luiz Marcelo Sa Malbouisson, Manuela Cristina Adsuara Pandolfi, Marcelo Park, Maria José Carvalho Carmona, Maria Castilho Prandini H. Andrade, Mariana Moreira Santos, Matheus Pereira Bateloche, Mayra Akimi Suiama, Mayron Faria de Oliveira, Mayson Laercio Sousa, Michelle Louvaes Garcia, Natassja Huemer, Pedro Vitale Mendes, Paulo Ricardo Gessolo Lins, Pedro Gaspar Dos Santos, Pedro Ferreira Paiva Moreira, Renata Mello Guazzelli, Renato Batista Dos Reis, Renato Daltro-Oliveira, Roberta Muriel Longo Roepke, Rodolpho Augusto Moura Pedro, Rodrigo Kondo, Samia Zahi Rached, Sergio Roberto Silveira Da Fonseca, Thais Sousa Borges, Thalissa Ferreira, Vilson Cobello Junior, Vivian Vieira Tenório Sales, and Willaby Serafim Cassa Ferreira.
All investigators above are from Hospital das Clinicas HCFMUSP, Faculdade de Medicina, Universidade de Sao Paulo, SP, BR
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.