A machine learning model exploring the relationship between chronic medication and COVID-19 clinical outcomes

Area of Science:

• Clinical Medicine
• Pharmacology
• Data Science

Background:

• The influence of chronic medications on COVID-19 prognosis remains debated.
• Understanding these associations is crucial for patient care and treatment optimization.

Purpose of the Study:

• To investigate the relationship between chronic medication use and COVID-19 outcomes.
• To identify key medication-related factors influencing COVID-19 severity using machine learning.

Main Methods:

• Analysis of 137,835 COVID-19 patients in Catalonia (February-September 2020).
• Utilized eXtreme Gradient Boosting for predicting hospitalization, ICU admission, and mortality.
• Complemented by logistic regression and sensitivity analyses for specific conditions (diabetes, hypertension, lipid disorders).

Main Results:

• Machine learning models accurately predicted COVID-19 mortality risk (AUCROC 0.89 for 18-65 age group).
• Key predictors included number of drugs, systemic corticoids, HMG-CoA reductase inhibitors, and hypertension drugs.
• Angiotensin-converting enzyme (ACE) inhibitors or angiotensin II receptor blockers (ARBs) were associated with lower mortality in hypertensive patients over 65.

Conclusions:

• Machine learning effectively identified COVID-19 outcomes based on patient data.
• Patients on ACE inhibitors, ARBs, or biguanides should continue treatment due to potential protective effects.
• Specific medications like metformin showed protective benefits, while DPP-4 inhibitors were linked to increased mortality in certain age groups.