key: cord-1012499-befjeo9b authors: Mody, Aaloke; Lyons, Patrick G; Guillamet, Cristina Vazquez; Michelson, Andrew; Yu, Sean; Namwase, Angella Sandra; Sinha, Pratik; Powderly, William G; Woeltje, Keith; Geng, Elvin H title: The Clinical Course of COVID-19 Disease in a US Hospital System: a Multi-state Analysis date: 2020-12-22 journal: Am J Epidemiol DOI: 10.1093/aje/kwaa286 sha: f03b7bfe577d3d79978336e90609aa1e38b2a3f3 doc_id: 1012499 cord_uid: befjeo9b There is limited data on longitudinal outcomes for COVID-19 hospitalizations that account for transitions between clinical states over time. Using electronic health record data from a St. Louis-region hospital network, we performed multi-state analyses to examine longitudinal transitions and outcomes among hospitalized adults with laboratory-confirmed COVID-19 with respect to fifteen mutually-exclusive clinical states. Between March 15 and July 25, 2020, 1,577 patients were hospitalized with COVID-19 (49.9% male, median age 63 years [IQR 50, 75], 58.8% Black). Overall, 34.1% (95% confidence interval [CI] 26.4%, 41.8%) had an ICU admission and 12.3% (CI 8.5%, 16.1%) received invasive mechanical ventilation (IMV). The risk of decompensation peaked immediately after admission, discharges peaked around day 3 to 5, and deaths plateaued between days 7 and 16. At 28 days, 12.6% (CI 9.6%, 15.6%) of patients had died (4.2% [CI 3.2%, 5.2%] received IMV) and 80.8% (CI 75.4%, 86.1%) were discharged. Among those receiving IMV, 39.1% (CI 32.0%, 46.2%) remained intubated after 14 days; after 28 days, 37.6% (CI 30.4%, 44.7%) had died and only 37.7% (CI 30.6%, 44.7%) were discharged. Multi-state methods offer granular characterizations of the clinical course of COVID-19 and provide essential information for guiding both clinical decision-making and public health planning. A careful characterization of the clinical course of COVID-19 during hospitalization will offer important insights into patients' prognosis and the anticipated burden and duration of resources required for their care-basic clinical information which is still coming into focus for this novel pathogen. Hospitalized patients may take numerous pathways: some only require brief stays while others deteriorate and require admission to the intensive care unit (ICU) with or without invasive mechanical ventilation (IMV) (1) (2) (3) (4) (5) (6) . Even if these patients survive, many will experience protracted hospital courses prior to discharge. Deaths could occur immediately after admission or after decompensations later on in the hospitalization. An understanding of how patients transition through multiple clinical states over the course of their hospitalization-and the timing of these transitionswill offer situational awareness and information for clinical decision-making and public health planning as the epidemic continues to evolve. To date, published data on the hospital course for COVID- 19 do not yet provide a comprehensive descriptive picture indicative of the experience in the US. For example, while case series do describe the number or incidence of deaths (1) (2) (3) (4) (5) (6) , such analyses have not captured movement between multiple clinical states over the course of hospitalization. Additionally, the rapidly evolving nature of the pandemic means that in many reports a substantial proportion patients are still in the midst of their illness (7) . These analyses have either presented cross-sectional estimates that do not account for this unequal follow-up time, or have excluded patients with incomplete follow-up time, potentially creating bias in both scenarios (1) (2) (3) (4) (5) (6) (7) (8) (9) . Furthermore, much of the early data on hospitalizations focus only on critically-ill patients and comes from single-center studies earlier in the epidemic, largely from the worst-hit areas such as Wuhan, China (1-3), Lombardy, Italy (4), and New York City (5, 6) where outcomes may not be representative of outcomes elsewhere. Thus, more rigorous data from regions where the burden of COVID-19 did not exceed the capacity of healthcare systems is needed to inform COVID-19 planning in the US going forward. To address these needs, we used data from the BJC conditions (e.g., inpatient floor admission, ICU, death, discharge) in a cohort of patients who were admitted with COVID-19. We used multi-state methods to estimate the proportion of patients in various clinical conditions over time as well as the time spent in and rates of transition from each state. This analytic technique permits a more comprehensive examination of the cascade of outcomes (10) during COVID-19 hospitalizations for informing planning and policy. after IMV, (8) inpatient floor after ICU admission without IMV, (9) inpatient floor after IMV, (10) discharged without ICU admission, (11) discharged with history of ICU admission without IMV, (12) discharged with history of IMV, (13) died, (14) died with history of ICU admission without IMV, and Fifth, we assessed the cumulative incidence of ICU admission, NIV, intubation, and death by 28 days since inpatient admission stratifying by patient subgroups. We also performed Cox proportional-hazards models to identify patient characteristics independently associated with the time from inpatient admission to ICU admission, intubation, and death. We selected covariates using directed acyclic graphs based on a priori hypotheses of causal relationships between baseline sociodemographic and clinical characteristics and patient outcomes. We examined the proportional hazards assumption using Schoenfeld residuals (16) . Lastly, to assess the changes in patient outcomes over time and explore the potential impact of the introduction of evidence-based therapies (i.e., remdesivir (17) and dexamethasone (18) in moderate or severe disease), we obtained adjusted age-stratified estimates of patient outcomes based on the time period in which they were admitted (i.e., March 15 to May 3 [prior to remdesivir availability] or May 4 to July 25 [after remdesivir availability]). We report these as marginal estimates of age-stratified Poisson models adjusted for sex, race, comorbidities, and whether the patient lived in a long-term care facility. All analyses were conducted using R 3.2.4 using the mstate package (13, 14) and Stata MP 16.1. Patient characteristics. Between March 15 and July 25, 2020, 2,940 patients who presented to the ED were confirmed to have COVID-19, and 1,577 were admitted to the hospital (Web Figure 1 ). Among those hospitalized, 571 were subsequently admitted to the ICU, 343 received NIV, and 213 patients received IMV (Table 1) . Median age was 63 years old (IQR 50, 75) and 927 patients (58.8%) were Black ( Table 1) . As the pandemic progressed, patients admitted later on were younger, had fewer comorbidities, were less likely to be Black, and were less likely to reside in a long-term care facility. They were more likely to be treated with remdesivir and steroids and less likely to be treated with tocilizumab and hydroxychloroquine (Web Table 1 ). Overall, Black patients tended to be younger, less likely to be male, and to have more comorbidities (Web Table 2 ). Table 3 ). After admission, the rate of transfer to the ICU and intubation peaked on hospital day 1 and declined thereafter, whereas the rate of discharge peaked between hospital day 3 to 5, and the rate of death plateaued on days 7 through 16 ( Figure 4 ). Among patients admitted to the ICU and those who received NIV, 50.8% (95% CI: 35%, 66.6%) and 39.5% (95% CI: 26.6%, 52.4%) received IMV at some point, respectively (Figure 2, Web Table 3 ). The rate of noninvasive and invasive ventilation peaked immediately after ICU transfer; whereas the rate of death (without intubation) peaked around day 5, and the rate of transfer to the ward peaked at day 3 and again at day 12 ( Figure 6 ). At Table 3 ). The median duration of ICU admissions was 1.9 days (IQR 1.1%, 3.2 days) without NIV or IMV, 4.5 days (IQR 2.0, 9.2) with NIV only, and 10.3 days (IQR 4.6, 20.1) for those who received IMV ( Figure 5 , Web Table 5 ). Lastly, among patients who received IMV, the rate of extubation increased through day 14, while the hazard for death plateaued between days 5 and 12 ( Figure 6 ). At (7). Studies that did include censored observations only considered time to single outcome (i.e., inhospital death) (5), but did not consider intermediate events such as ICU transfers or intubation. Additionally, these estimates did not account for competing events (15) , such as hospital discharge, that would preclude the occurrence of an in-hospital mortality event, potentially also contributing to bias (8, 9) . This study adds to this existing literature in several ways. We use rigorous longitudinal methods to estimate the incidence and timing of events in the setting where both competing events are present and where the observation time between participants is not equal (8, 9) . Additionally, we use these multi-state methods to assess transitions between multiple clinical states-as opposed to a single one-over the course of one's hospitalization (13, 14) . Furthermore, most early reports were single-center studies conducted in regions that were the hardest hit by COVID-19, potentially limiting the generalizability of their experiences. In contrast, our data includes a diverse and representative population from a variety of settings (e.g., both academic and community hospitals, rural and urban settings, diverse patients from both affluent and marginalized communities) and across different phases of the pandemic (i.e., before and after the introduction of evidence-based therapies). Thus, it provides one the most comprehensive characterizations of the clinical course of COVID-19 hospitalizations to date. Our results offer an additional layer of nuance to characterizations of COVID-19-related hospitalizations, but are also consistent with what has been previously reported (1-6, 26, 27) . The majority of patients were admitted to the wards and discharged within 3 to 5 days, but an important subset present critically ill (or decompensate early in their hospitalization) and generally experience a protracted hospital course, often with prolonged periods of IMV and a high-risk for mortality. In our cohort, older age was most strongly associated with poor outcomes such as need for IMV and mortality, followed by male sex. Additionally, we found that patients admitted after May 4 th (i.e., after remdesivir was introduced in our hospital network) had reduced mortality rates, though patients admitted during this period were also substantially younger and healthier. Still, this association remained even after adjusting for age and comorbidities and may thus also be indicative-though not definitively so-of the positive impact of routine use of these evidence-based therapies for COVID-19 (i.e., remdesivir (17) and dexamethasone (18) estimates are also similar to those for influenza-associated and general ARDS (33, 34) , more work is needed to understand how COVID-19 clinical phenotypes relate to their underlying pathophysiology and how they differ from other disease states (35) (36) (37) (38) . Ultimately, further research extending these findings is needed to help us understand to whom, when, and what types of interventions and treatments are needed for optimizing our response to COVID-19, both at the individual patient and public health levels. There are several limitations to this study. First, we leveraged observational EHR data, which may have misclassified some patient outcomes, COVID-19 diagnoses, hospital events, or their timing. In particular, we did not have granular data on patients' disease severity (e.g., oxygenation levels), the exact timing of multiples events occurring within an hour of each other, or the history or circumstances leading up to admission at a BJC hospital (e.g., duration of symptoms, prior events if patient transferred from a different hospital). Second, we obtained adjusted age-stratified outcome estimates by time period prior to explore the potential impact of routine use of evidence-based COVID-19 therapies, but these analyses were not adjusted presenting disease severity and it is still possible that these estimates are affected by residual confounding. Third, our study only included hospitals from a large health system affiliated with an academic medical center where health care capacity was not exceeding and may not necessarily be reflective of outcomes in other regions of the country or the world, particularly places that experienced a COVID-19 epidemic surge that exceeded their health systems' capacity. Still, we did include patients from several hospitals ranging from an academic, quaternary-care medical center to smaller community hospitals located in both urban and rural settings. In conclusion, we used multi-state analytic methods to provided nuanced characterizations of (5) invasive mechanical ventilation in ICU, (6) noninvasive ventilation after invasive mechanical ventilation, (7) ICU after invasive mechanical ventilation, (8) inpatient floor after ICU admission but no invasive mechanical ventilation, (9) inpatient floor after invasive mechanical ventilation, (10) discharged without ICU admission, (11) discharged with history of ICU admission but no invasive mechanical ventilation, (12) discharged with history of invasive mechanical ventilation (IMV), (13) died, (14) died with history of ICU admission but no invasive mechanical ventilation, and (15) died with history of invasive mechanical ventilation. This figure depicts all the possible transitions patients could make from each state. Patients were not restricted to starting from State 1; those who were directly admitted to the hospital or transferred from another hospital started from the state in which they were first observed. Abbreviations: ICU=intensive-care unit; NIV=noninvasive ventilation; IMV=invasive mechanical ventilation. . We present overall estimates as well as estimates stratified by populations with specific clinical outcomes. Dots represent the median values, the surrounding box spans between the 25 th and 75 th percentiles, and the surrounding violin plot represents a kernel density plot spanning the full range of values. Of note, kernel density plots extend below one due to estimation algorithms, but no patients had length of stays less than 0 in any state. Abbreviations: ICU=intensive-care unit; NIV=noninvasive ventilation; IMV=invasive mechanical ventilation. Clinical course and risk factors for mortality of adult inpatients with COVID-19 in Wuhan, China: a retrospective cohort study Clinical course and outcomes of critically ill patients with SARS-CoV-2 pneumonia in Wuhan, China: a single-centered, retrospective, observational study Clinical Course and Outcomes of 344 Intensive Care Patients with COVID-19 Baseline Characteristics and Outcomes of 1591 Patients Infected With SARS-CoV-2 Admitted to ICUs of the Lombardy Region Epidemiology, clinical course, and outcomes of critically ill adults with COVID-19 in New York City: a prospective cohort study Presenting Characteristics, Comorbidities, and Outcomes Among 5700 Patients Hospitalized With COVID-19 in the New York City Area Clinical course and mortality risk of severe COVID-19 Reappraisal of Ventilator-Free Days in Critical Care Research Measuring and Analyzing Length of Stay in Critical Care Trials Longitudinal Care Cascade Outcomes among People Eligible for ART who are Newly Linking to Care in Zambia: A Multi-State Analysis Multi-state models for event history analysis The mstate package for estimation and prediction in nonand semi-parametric multi-state and competing risks models mstate: An R Package for the Analysis of Competing Risks and Multi-State Models An empirical transition matrix for nonhomogeneous Markov chains based on censored observations Regression Methods in Biostatistics: Linear, Logistic, Survival, and Repeated Measures Models Remdesivir for the Treatment of Covid-19 -Final Report Dexamethasone in Hospitalized Patients with Covid-19 -Preliminary Report Predictive Mathematical Models of the COVID-19 Pandemic: Underlying Principles and Value of Projections Forecasting the impact of the first wave of the COVID-19 pandemic on hospital demand and deaths for the USA and European Economic Area countries LEMMA: Local Epidemic Modeling for Management & Action Locally Informed Simulation to Predict Hospital Capacity Needs During the COVID-19 Pandemic Hospital Preparedness for COVID-19: A Practical Guide from a Critical Care Perspective Staffing With Disease-Based Epidemiologic Indices May Reduce Shortage of Intensive Care Unit Staff During the COVID-19 Pandemic Clinical Characteristics and Morbidity Associated With Coronavirus Disease 2019 in a Series of Patients in Metropolitan Detroit Risk Factors for Intensive Care Unit Admission and Inhospital Mortality among Hospitalized Adults Identified through the U.S. Coronavirus Disease 2019 (COVID-19)-Associated Hospitalization Surveillance Network (COVID-NET) Association of Race With Mortality Among Patients Hospitalized With Coronavirus Disease 2019 (COVID-19) at 92 US Hospitals This Time Must Be Different: Disparities During the COVID-19 Pandemic Hospitalization and Mortality among Black Patients and White Patients with Covid-19 Disparities In Outcomes Among COVID-19 Patients In A Large Health Care System In California The Disproportionate Impact of COVID-19 on Racial and Ethnic Minorities in the United States COVID-19 And Racial/Ethnic Disparities In Health Risk, Employment, And Household Composition Influenza virus and factors that are associated with ICU admission, pulmonary co-infections and ICU mortality Patterns of Care, and Mortality for Patients With Acute Respiratory Distress Syndrome in Intensive Care Units in 50 Countries Subphenotypes in critical care: translation into clinical practice Machine Learning Classifier Models Can Identify Acute Respiratory Distress Syndrome Phenotypes Using Readily Available Clinical Data Identifying Novel Sepsis Subphenotypes Using Temperature Trajectories The perils of premature phenotyping in COVID-19: a call for caution Acknowledgements: We want to thank the Institute for Informatics (I2) at Washington University in St. Louis for assistance with data extraction from the electronic health records.