August 6, 2020

Can We Predict Which COVID-19 Patients Will Decompensate?

Background: As the COVID-19 pandemic continues a number of challenges have arisen. Amongst these is the ability of clinicians to predict which patients will suffer from early decompensation. It is well established that there are patients that will rapidly decline while others, who initially present similarly, will continue without disease progression. A clinical decision instrument (CDI) to guide clinicians can be useful placing patients requiring hospital admission at the correct level of care without over-utilizing ICUs or, putting patients on the floors who will suffer from early decompensation.

Article: Haimovich A et al. Development and validation of the quick COVID-19 severity index (qCSI): a prognostic tool for early decompensation. Ann Emerg Med 2020. Link

Clinical Question: Can an Emergency Department (ED) risk stratification tool be created that predicts respiratory failure within 24 hours of admission in patients with COVID-19?

Population: Admitted adult COVID-19 (between March 1 and April 27, 2020) positive patients in a single healthcare system composed of nine hospitals: six suburban community hospitals, two urban community hospital and one urban academic hospital.

Primary Outcome: Respiratory failure within 24 hours of admission as defined by oxygen requirement of > 10 L/min, any high flow device use, non-invasive or invasive ventilation or death.

Design:

  • Retrospective observational cohort study
  • Patients from eight hospitals were used in model generation and internal validation and, a ninth hospitals data was used for further validation

Excluded: Patients < 18 years of age and those who required > 6 L/min or met any other critical illness criteria (oxygen requirement of > 10 L/min, any high flow device use, non-invasive or invasive ventilation or death.) within four hours of hospital presentation

Primary Results

  • COVID-19 admissions: n = 1792
  • Critical respiratory illness w/in 4 hours: n = 620 (35%)
  • Study population for instrument
    • n = 1172
    • Admitted to ICU between 4-24 hours: n = 59 (5%)
    • Respiratory decompensation w/in 24 hours: n = 144 (12.3%)
      • Requiring > 10 L/min: n = 101 (8.6%)
      • Requiring high-flow device: n = 112 (9.6%)
      • Non-invasive ventilation: n = 4 (0.3%)
      • Invasive ventilation: n = 10 (0.8%)
      • Death: n = 1 (0.01%)
  • Full data set included 713 patient variables that were investigated

Critical Findings

  • Three variables identified as having strong predictive capabilities and used in the quick COVID Severity Index (qCSI)
    • Nasal cannula O2 flow rate
    • Minimum recorded pulse oximetry
    • Respiratory rate

  • qCSI < 3 Performance Characteristics
    • Sensitivity: 79% (65-93)
    • Specificity: 78% (72-83)
    • LR (+): 3.55 (3.51 – 3.59)
    • LR (-): 0.27 (0.26 – 0.28)
  • On-line calculator: covidseverityindex.org
  • MDCalc: Quick COVID-19 Severity Index (qCSI)
  • COVID Severity Index (CSI) included 12 additional variables

Strengths:

  • Addresses a clinically relevant question
  • Instrument derived from a large cohort of patients across multiple hospitals
  • Decision aid is simple and requires inputs that can be determined at bedside (i.e. no lab tests or imaging required)
  • Researchers created an on-line calculator to support clinicians
  • Focus on ED prognostication by evaluating patient outcomes within 24 hours of admission
  • Emphasizes O2 requirements and mortality rather than ICU placement as the latter involves numerous inputs and subjective decisions

Limitations:

  • Derivation and validation all performed on retrospectively collected data from a single health system
  • Data missing from the chart was filled in either with assumptions (e.g. if GCS not noted, was assumed to be 15) or from prior charts (i.e. for comorbidities)
  • Sensitivity has wide confidence interval
  • Composite outcome with non-equally weighted outcomes.  Death and invasive mechanical ventilation not the same as HFNC or NIV
  • There’s no data on where in the disease time course the patient is which may influence disposition decisions

Discussion:

  • The sensitivity and specificity of the qCSI are both on the lower side.
    • The target score (I.e. qCSI < 3) could be lowered to achieve a higher sensitivity but this would come at the expense of specificity
    • The wide confidence interval for sensitivity further jeopardizes the usefulness
  • The qCSI may be over-simplified by not taking into account lab data that has been previously shown to be associated with illness severity or radiographic information
  • There is no comparison to clinician gestalt
    • In order for a CDI to be useful, it should have superior performance to what is currently being done
    • This is a common issue seen in CDI creation
  • Despite the above issues, there may still be a place for a simplified CDI like this. As the pandemic overwhelms hospitals and hospital systems, non-emergency clinicians may be called upon to help in ED care. These clinicians may have more limited experience in risk stratification and lack the necessary gestalt for risk stratification. This group may benefit from a codified system for disposition.
  • Younger patients in this cohort had more risk of decompensation than older patients.  The cause of this is unclear

Authors Conclusions:

“A significant proportion of admitted COVID-19 patients progress to respiratory failure within 24 hours of admission. These events are accurately predicted using bedside respiratory exam findings within a simple scoring system.”

Our Conclusions: The qCSI is a good initial attempt to derive a CDI to support clinicians in disposition decisions for COVID-19 patients who require hospital admission. However, prospective as well as external validation is necessary prior to widespread use can be recommended.

Potential to Impact Current Practice: While we do not recommend using this tool in lieu of experienced clinician gestalt, there may be a role for its use in less experienced clinicians or, for non-emergency clinicians that support EDs during the pandemic.

Post Peer Reviewed By: Salim R. Rezaie, MD (Twitter: @srrezaie)

Cite this article as: Anand Swaminathan, "Can We Predict Which COVID-19 Patients Will Decompensate?", REBEL EM blog, August 6, 2020. Available at: https://rebelem.com/can-we-predict-which-covid-19-patients-will-decompensate/.
The following two tabs change content below.

Anand Swaminathan

Clinical Assistant Professor of Emergency Medicine at St. Joe's Regional Medical Center (Paterson, NJ)
REBEL EM Associate Editor and Author

Latest posts by Anand Swaminathan (see all)

No Comments

Post A Comment

Time limit is exhausted. Please reload CAPTCHA.

0