Misattribution Bias of COVID-19 Hospitalizations in Alberta Using an Admission Algorithm.

Area of Science:

• Public Health
• Infectious Diseases
• Health Informatics

Background:

• Early COVID-19 responses often presumed all hospitalizations were COVID-19-related.
• Alberta Health Services implemented an algorithm to classify COVID-19 admissions.
• The algorithm aimed to differentiate direct, contributing, and unrelated causes.

Purpose of the Study:

• To assess the precision of the COVID-19 admission algorithm.
• To compare algorithm-adjudicated causes against expert physician review.
• To identify and rectify misattribution bias in hospitalization data.

Main Methods:

• A quality improvement initiative was conducted.
• 261 COVID-19 hospitalizations from January-February 2022 were sampled.
• Algorithm results were compared to a panel of infectious disease physicians.

Main Results:

• Physician review found 39.9% direct, 17.2% contributing, and 37.6% unrelated COVID-19 causes.
• The algorithm identified 42.9% direct, 24.5% contributing, and 30.3% unrelated causes.
• Moderate agreement (Cohen's kappa=0.475) was observed, with algorithm over-attributing unrelated cases.

Conclusions:

• The admission algorithm exhibited misattribution bias, particularly overestimating contributing causes.
• Findings led to algorithm improvements for enhanced precision.
• Public health agencies should validate COVID-19 hospitalization reporting methods.