PIN108 Using the Apriori Algorithm to Identify Risk Factors Associated with Survival and Mortality Among COVID-19 Patients

ABSTRACT

Objectives: The SARS-Cov-2 pandemic has resulted in more than 130,000 deaths. Factors like old age and presence of comorbid conditions have been hypothesized to increase likelihood of mortality. In this study, the Apriori data mining technique was used to identify combinations of patient variables associated with either mortality or survival. Understanding combinations of key risk factors may help practitioners classify patients as high or low risk for mortality. Methods: This is a retrospective data analysis using records of patients with COVID-19 diagnoses in the Premier Healthcare Database, from September 2019 to May 2020. Patients were characterized by comorbidities (Elixhauser and Charlson), demographics, provider characteristics, diagnoses and symptoms at time of admission and treatments during admission. Association rules were generated using Apriori algorithm keeping minimum support and minimum confidence as 0.004 and 0.65 respectively for outcome of mortality, and 0.10 and 0.95 for outcome of survival. Results: A total of 39 rules of factors predicted an increase in mortality by 5.87 times (average lift). Elderly patients (age greater than 65 years) who had a ventilator usage of more than 96 hours had the greatest mortality risk. Along with a COVID-19 diagnosis, they were also observed to have hypertension, fluid and electrolyte disorders, renal and congestive heart failure as comorbid conditions, with mostly Elixhauser Index of 5 or above. For survival, 132 rules were generated, all rules had a lift > 1 with an average of 1.08. Age groups less than 64 years, female and Elixhauser less than 3 were included in the algorithm. Diagnoses of “fever” and “cough” were also observed in the survival cohort. Conclusions: The Apriori algorithm, first developed to evaluate association rules in retail and other industry, may have significant applications in understanding concurrent risks for COVID-19 and tailor preventive measures and care specifically to patients with greater risk factors.