Admission Predictors of In-Hospital Mortality and the Combined Outcome of Death or Invasive Mechanical Ventilation in Patients with COVID-19 During the Pre-Vaccination Era: A Retrospective Cohort Study

In a retrospective cohort study of 482 hospitalized COVID-19 patients in Brazil during the pre-vaccination era, five admission variables—age, arterial pH, neutrophil-to-lymphocyte ratio, comorbidity burden, and serum creatinine—were identified as independent predictors of in-hospital mortality, supporting the use of routinely available clinical data for early risk stratification.

The Gist

Imagine the early days of the COVID-19 pandemic (before vaccines existed) as a massive, sudden storm hitting a city. Doctors were like firefighters rushing into burning buildings, trying to figure out which houses were structurally sound and which were about to collapse. They needed a quick way to guess who would survive the fire and who might not, so they could prioritize their limited resources.

This paper is essentially a "Firefighter's Field Guide" written by a team of doctors in Central Brazil. They looked back at 482 patients who walked into their hospital with the virus between 2020 and 2021. Their goal was to find the specific "warning signs" present the moment a patient arrived that could predict if they would survive the hospital stay or if they would need a breathing machine (ventilator) or worse.

Here is the breakdown of their findings using simple analogies:

1. The Five "Red Flags" (Predictors of Death)

The doctors found that you don't need a crystal ball to predict a bad outcome. You just need to check five specific things when a patient first walks through the door. Think of these as the five dials on a dashboard that, if they are in the "danger zone," mean the car is likely to break down.

• The Age Dial (Time): Older patients had a much harder time. It's like an old, worn-out engine; it just doesn't have the same reserve power to handle the stress of a new, violent storm. For every 10 years older a patient was, their risk of not surviving went up significantly.
• The "Comorbidity" Backpack (Burden): Imagine every disease a person has (like diabetes, high blood pressure, or heart issues) as a heavy rock in a backpack. If you have one rock, you can still run. If you have a backpack full of rocks, you are already exhausted before the race even starts. The study found that the total number of health problems a person had was a huge predictor. It wasn't just one specific disease; it was the weight of the whole backpack.
• The "Inflammation Fire" (NLR): This stands for the Neutrophil-to-Lymphocyte Ratio. Imagine your immune system as an army. Neutrophils are the foot soldiers rushing in to fight, and lymphocytes are the generals organizing the defense. In severe cases, the foot soldiers go crazy (too many of them), but the generals disappear (too few of them). This ratio is like a smoke detector; a high number means the "fire" inside the body is raging out of control.
• The "Kidney Filter" (Creatinine): Your kidneys are like the water filters in a house. If the water coming out is dirty (high creatinine levels), it means the filters are clogged or broken. In this study, patients whose kidneys were struggling right when they arrived were much more likely to die. It's a sign that the virus was damaging the body's internal plumbing.
• The "Blood Acid" Meter (pH): This is the most unique finding. Think of your blood as a swimming pool. It needs to be at a perfect balance (pH) to keep the swimmers (your cells) happy. If the water gets too acidic (low pH), it's like adding vinegar to the pool—it starts corroding everything. The study found that even a tiny drop in this "acidity" level was a massive warning sign. It suggested the body was in a state of metabolic chaos, struggling to keep its internal environment stable.

2. The "X-Ray" (Chest CT Scans)

The doctors also looked at X-rays (CT scans) of the lungs. They found that while the pattern of the damage (like whether it looked like fog or solid blocks) didn't matter as much, the amount of damage did.

• The Analogy: Imagine a room covered in smoke. It doesn't matter if the smoke is in the corners or the center; if the smoke covers more than half the room (>50%), the situation is critical. Patients with more than half their lungs affected had nearly three times the risk of dying.

3. The "Composite" Outcome (Death OR Breathing Machine)

The researchers also asked: "Who will need a breathing machine?"

• The answer was almost the same as the "Who will die?" list, but with one twist: The kidney filter (Creatinine) wasn't as strong a predictor for just needing a machine. This makes sense because you can need a breathing machine just because your lungs are tired, but you usually only die if your whole body (including kidneys) starts shutting down.

The Big Takeaway

The most important lesson from this paper is that you don't need expensive, high-tech tests to know who is in trouble.

The doctors found that the five "Red Flags" listed above are all things that can be measured with a simple blood test and a quick physical exam the moment a patient arrives.

• Older age? Check.
• Heavy backpack of other diseases? Check.
• High inflammation? Check.
• Clogged kidneys? Check.
• Acidic blood? Check.

If a patient has these signs, the medical team knows immediately: "This is a critical situation; we need to pull out all the stops."

Why This Matters

This study was done in a specific region of Brazil before vaccines existed. It gives us a "baseline" of how the virus behaved when it was at its most dangerous and when the world had no immunity. It reminds us that even in a high-tech medical world, the most powerful tools are often the simplest ones: listening to the body's basic signals and understanding that a patient's history (age and other illnesses) is just as important as the virus itself.

In short: If you walk into a hospital with the virus, the doctors can look at your age, your medical history, and a few blood numbers to get a very clear picture of your odds. It's like checking the weather forecast before a storm; if the radar shows a hurricane, you don't wait to see if the roof blows off—you prepare immediately.

Technical Summary

1. Problem Statement

During the pre-vaccination era of the COVID-19 pandemic (early 2020–early 2021), healthcare systems faced unprecedented pressure to identify patients at high risk of adverse outcomes. While numerous studies investigated predictors of severity, many suffered from methodological limitations, including heterogeneous datasets, incomplete variable integration, and a lack of adjustment for confounding factors. Specifically, there was a scarcity of comprehensive studies in Brazil that integrated clinical presentation, arterial blood gas (ABG) analysis, and chest computed tomography (CT) findings within a single well-characterized cohort. Furthermore, the Central-West region of Brazil was significantly underrepresented in existing literature, which was dominated by data from the South, Southeast, and Northeast regions.

2. Methodology

• Study Design: Retrospective, single-center observational cohort study.
• Setting & Population: Conducted at Hospital do Coração Anis Rassi (a tertiary cardiac center) in Goiânia, Central-West Brazil. The study included 482 consecutive adults (median age 61 years; 64.3% male) with RT-PCR-confirmed SARS-CoV-2 infection hospitalized between March 13, 2020, and January 28, 2021 (pre-vaccination period).
• Data Collection:
  • Variables: 76 admission variables were analyzed within the first 48 hours, including demographics, comorbidities, vital signs, symptoms, laboratory parameters (complete blood count, biochemistry, coagulation, and arterial blood gas), and chest CT findings.
  • Exclusions: Patients leaving against medical advice, inter-hospital transfers before outcome ascertainment, and deaths not attributable to COVID-19 were excluded.
• Outcomes:
  • Primary: In-hospital mortality attributable to COVID-19.
  • Secondary: Composite outcome of in-hospital death or need for invasive mechanical ventilation (IMV).
• Statistical Analysis:
  • Univariable Analysis: Performed on all 76 variables.
  • Multivariable Modeling: Used binary logistic regression with an a priori variable selection strategy based on clinical relevance and representation of distinct pathophysiological domains (demographics, inflammation, metabolic status, organ dysfunction).
  • Constraints: Adhered to the "events-per-variable" principle (~10 events per predictor). Variables with >10% missing data (e.g., troponin, lactate) were excluded from multivariable models. Chest CT variables were analyzed descriptively and in univariable models only due to incomplete availability and potential collinearity.
  • Approach: Complete-case analysis was performed without imputation.

3. Key Contributions

• Comprehensive Integration: This study is one of the few to systematically integrate arterial blood gas (ABG) parameters (specifically pH) alongside inflammatory markers (NLR), renal function, and radiological burden in a single predictive model.
• Regional Representation: It provides critical data from the Central-West region of Brazil, an area previously underrepresented in pandemic literature.
• Methodological Rigor: The study avoided automated stepwise selection, instead using a clinically driven, pathophysiology-based variable selection approach to prevent overfitting and ensure interpretability.
• CT Analysis: While not included in the multivariable model, the study provided a detailed univariable analysis of CT findings, establishing that the extent of pulmonary involvement is a more significant prognostic marker than specific radiological patterns.

4. Key Results

• Outcomes:
  • In-hospital Mortality: 9.3% (45/482).
  • Invasive Mechanical Ventilation (IMV): Required in 15.4% (74/482); mortality among ventilated patients was 58.1%.
• Independent Predictors of Mortality (Multivariable Model):
  Five variables were independently associated with death (Nagelkerke $R^2$ = 0.43):
  1. Age: Adjusted OR 1.66 per 10-year increase (95% CI 1.19–2.32; $p=0.003$).
  2. Arterial pH: Adjusted OR 0.47 per 0.1-unit increase (95% CI 0.25–0.89; $p=0.021$). Lower pH significantly increased risk.
  3. Neutrophil-to-Lymphocyte Ratio (NLR): Adjusted OR 1.30 per unit increase (95% CI 1.18–1.44; $p<0.001$).
  4. Number of Comorbidities: Adjusted OR 1.59 per additional comorbidity (95% CI 1.25–2.02; $p<0.001$).
  5. Serum Creatinine: Adjusted OR 1.37 per unit increase (95% CI 1.05–1.77; $p=0.019$).
• Composite Outcome (Death or IMV):
  • Age, pH, NLR, and comorbidity burden remained significant.
  • Serum creatinine showed a non-significant trend ($p=0.069$), suggesting renal dysfunction is more strongly linked to fatal outcomes than respiratory failure alone.
• Chest CT Findings:
  • Pulmonary involvement >50% was associated with a nearly threefold increase in odds of death (OR 2.87; $p=0.003$) in univariable analysis. Specific patterns (e.g., ground-glass opacity) were common but not independently predictive of mortality compared to the extent of involvement.

5. Significance and Implications

• Early Risk Stratification: The study demonstrates that a model using five routinely available admission variables (Age, pH, NLR, Comorbidity count, Creatinine) can effectively stratify risk with moderate-to-high explanatory power.
• Prognostic Value of Acid-Base Status: A key finding is the independent prognostic value of arterial pH. Even after adjusting for renal function and inflammation, acidemia remained a strong predictor of mortality, suggesting it captures systemic metabolic decompensation and tissue hypoperfusion not reflected by other markers.
• Clinical Utility: These findings support the use of simple, low-cost admission data for early triage and resource allocation, particularly in settings where advanced imaging or biomarkers may be limited.
• Reference for Future Pandemics: By characterizing the "natural history" of the disease in a pre-vaccination cohort, this study provides a baseline for understanding severe respiratory infection pathophysiology and preparing for future outbreaks.

Limitations: The study is retrospective, single-centered, and limited by the number of events (45 deaths) and missing data for certain biomarkers. External validation is required before applying these findings to contemporary, vaccinated populations.