Prospective predictive performance comparison between clinical gestalt and validated COVID-19 mortality scores

Area of Science:

• Infectious Diseases
• Clinical Epidemiology
• Healthcare Outcomes

Background:

• Early COVID-19 mortality scores were developed with limited clinical experience and interventions.
• Advancements in clinical practice and evidence-based treatments may impact the performance of existing predictive models.

Purpose of the Study:

• To prospectively evaluate the current predictive accuracy of established COVID-19 mortality scores.
• To compare the performance of these scores against clinical gestalt predictions in a contemporary patient cohort.

Main Methods:

• Prospective evaluation of six COVID-19 mortality scores (LOW-HARM, qSOFA, MSL-COVID-19, NUTRI-CoV, NEWS2, neutrophil-to-lymphocyte ratio) and clinical gestalt.
• Area Under the Curve (AUC) analysis was used to determine predictive accuracy in 166 COVID-19 patients.

Main Results:

• All evaluated COVID-19 mortality scores demonstrated significantly lower AUC values compared to their originally reported performance.
• Clinical gestalt predictions showed non-inferiority to the mortality scores, with an AUC of 0.68.
• Score performance did not improve with local data adjustments, and some scores were outperformed by clinical gestalt when clinician confidence was below 80%.

Conclusions:

• The predictive performance of most COVID-19 mortality scores has decreased since their initial development.
• Clinical gestalt, despite its subjective nature, offers relevant advantages in predicting COVID-19 outcomes.
• Regular re-evaluation of the necessity and performance of COVID-19 mortality scores is crucial.