Abstract
Importance: There is an urgent need to understand patient characteristics of COVID-19 and evaluate markers of critical illness and mortality. Objective: To assess the association of clinical features with patient outcomes. Design, Setting, and Participants: In this observational case series, patient-level data were extracted from electronic medical records (EMR) for 28,336 patients tested for SARS-CoV-2 at the Mount Sinai Health System from February 24 to April 15, 2020, including 6,158 laboratory-confirmed cases. Main Outcomes and Measures: Effects of race on positive test rates and mortality were assessed. Among hospitalized cases (N = 3,273), the effects of demographics, hospital site, vital signs, lab results, and comorbidities on discharge and death were estimated. Results: Hispanics (29%) and African Americans (25%) had disproportionately high positive case rates; however, no differences in mortality rates were observed in the hospital. Significant risk factors for mortality included age (HR=1.05), oxygen saturation (HR=0.985), ICU care (HR=1.58), and elevated creatinine (HR=1.75). Asthma was associated with increased length of stay but not mortality. Conclusions: While race was associated with higher infection risk, inpatient mortality was comparable across races. Clinical features identified here can help triage high-risk patients.
Introduction
During the initial peak of the COVID-19 pandemic in New York City, clinicians faced an unprecedented challenge: a novel virus with a highly variable clinical course. Identifying which patients were likely to recover and which were at risk of rapid deterioration was critical for hospital resource management. This study leverages a massive Electronic Medical Record (EMR) dataset from the Mount Sinai Health System (MSHS) to identify the clinical "red flags" that predict COVID-19 severity and to investigate early reports of racial disparities in outcomes.
Methods
Data Source and Cohort
The study utilized de-identified EMR data from the MSHS, encompassing 28,336 patients tested for SARS-CoV-2 between late February and mid-April 2020. From this group, 6,158 cases were confirmed via RT-PCR. The primary analysis focused on a hospitalized cohort of 3,273 patients admitted for at least 24 hours. To establish demographic baselines, a reference set of 1.6 million patients from the MSHS EMR (dating back to 2016) was used to represent the NYC population.
Statistical Analysis
The researchers employed a competing risks survival analysis. Unlike traditional survival models that treat discharge as a censored event, this study recognized that death and discharge are mutually exclusive, competing outcomes. Using the cmprsk R package, the team estimated Cumulative Incidence Functions (CIFs) to model these paths. Two types of multivariate models were developed:
- Etiologic Models: Used all data available throughout hospitalization to identify factors associated with death.
- Prognostic Models: Used only baseline data available at the time of admission to evaluate predictive power for eventual outcomes.
Key Insights and Results
The Disparity Gap: Infection vs. Mortality
A primary finding of the study was that while Hispanic (29%) and African American (25%) populations were significantly over-represented in infection rates relative to the general population, their in-hospital mortality rates were comparable to other groups once adjusted for comorbidities. This suggests that the higher death tolls observed in these communities were likely driven by higher exposure and infection rates—potentially due to housing density or occupation—rather than intrinsic biological differences or disparities in the quality of hospital care.
Clinical Predictors of Mortality
The study identified several high-confidence risk factors for mortality (Hazard Ratios > 1):
- Age: HR = 1.05 per year (p=1.15e-32), the strongest predictor of mortality.
- Oxygen Saturation: HR = 0.985; lower saturation levels at admission were strongly linked to poor outcomes.
- Body Mass Index (BMI): HR = 1.02 (p=1.09e-2), confirming obesity as a significant risk factor.
- Organ Stress Markers: Beyond respiratory failure, the strongest predictors of death included markers of kidney stress (elevated creatinine, HR = 1.75) and liver stress (elevated ALT, HR = 1.002). This highlights the multi-organ impact of the disease.
- Inflammatory Markers: In subsets where testing was available, elevated LDH, CRP, and D-Dimer were strongly correlated with mortality.
The Asthma Paradox
Contrary to early clinical fears that chronic respiratory conditions would lead to higher mortality, asthma was not a significant predictor of death. However, it was significantly associated with a longer hospital stay (HR=0.78 for discharge), suggesting that while these patients required more care, they did not necessarily face a higher risk of fatality.
Hospital Site Variation
Outcomes varied significantly across different hospital sites within the Mount Sinai system (e.g., Brooklyn vs. Manhattan). However, this variation was largely explained by the "older" age profile and higher comorbidity burden of patients at specific locations, rather than differences in hospital performance.
Discussion and Limitations
This study provides a robust prognostic framework for triaging COVID-19 patients based on EMR data. By identifying that the disparity in outcomes is primarily an issue of infection risk rather than inpatient care, the findings point toward public health interventions as the primary tool for addressing racial inequities.
Limitations: * Retrospective Nature: As a study of "real-world" clinical care, not all patients received the same laboratory tests (e.g., Ferritin or D-Dimer), limiting the inclusion of these markers in the primary multivariate models. * Temporal Snapshot: The data concludes in April 2020. The subsequent introduction of treatments like dexamethasone, remdesivir, and standardized anticoagulation protocols would likely alter the hazard ratios observed in this early cohort. * Baseline Focus: The prognostic model focused on admission data. Future research should incorporate the trajectory of vital signs over the first 48 hours of admission to improve predictive accuracy.
Conclusion
By analyzing one of the largest early-pandemic EMR datasets, this study identifies age, oxygen saturation, and markers of kidney and liver stress as critical prognostic indicators. These findings assist in identifying high-risk patients early in their hospital stay and suggest that inpatient outcomes for COVID-19 are consistent across racial and ethnic groups when adjusted for baseline health factors.
Summary updated