Using machine learning probabilities to identify effects of COVID-19.

Area of Science:

• Infectious Diseases
• Epidemiology
• Health Informatics

Background:

• Coronavirus disease 2019 (COVID-19) has caused significant global health and economic impacts.
• Retrospective analyses of electronic health records (EHRs) are crucial for understanding long-term COVID-19 effects.
• Incomplete EHRs pose challenges in accurately identifying COVID-19 cases for analysis.

Purpose of the Study:

• To develop a robust method for identifying COVID-19 diagnoses within patient electronic health records.
• To investigate the association between confirmed COVID-19 diagnoses and subsequent development of specific health conditions.

Main Methods:

• A random Forest classifier was trained to predict the probability of a COVID-19 diagnosis for each patient visit.
• The model utilized electronic health record data to assign these probabilities.
• Statistical analysis was performed to correlate COVID-19 probabilities with future diagnoses.

Main Results:

• The random Forest classifier effectively assigned probabilities for COVID-19 diagnoses during patient visits.
• Higher probabilities of COVID-19 diagnosis were significantly associated with increased future risk of myocardial infarction.
• Elevated COVID-19 probabilities also correlated with higher incidence of urinary tract infections, acute renal failure, and type 2 diabetes.

Conclusions:

• The developed classifier improves the accuracy of identifying COVID-19 cases in EHR data.
• COVID-19 diagnosis is associated with an increased risk of developing serious long-term health complications.
• This methodology supports ongoing research into the long-term sequelae of COVID-19.