Background and Purpose: Association rule mining can discover hidden patterns and relationships between variables that may not be apparent through other data analysis techniques. We aimed to find practical patterns in COVID-19 data and predict patient survivor status using association rules.
Materials and Methods: In this cross-sectional study, clinical data of 51460 hospitalized patients tested by polymerase chain reaction (PCR) were collected from February 20, 2020, to September 12, 2021, in Khorasan Razavi Province, Iran. An Apriori algorithm was used to extract association rules or patterns in data.
Results: Most participants (51.0%) were male; their Mean±SD age was 54.55±22.15 years. Fever (37%), cough (38.4%), respiratory distress (56%), PO2 level less than 93% (52.9%), muscular pain (19.1%) and decreased consciousness (8.9%) were common symptoms. Based on the association rules, if a patient was older than 75 years, had respiratory distress, reduced consciousness and PO2 level <93%, then this patient is who has died. The PCR test result of a male who used drugs was positive. Vomit and diarrhea lead to positive PCR test results, too. The most common symptom seen in men was respiratory distress, while the most common symptom in women was hypertension. Muscular pain due to COVID-19 is more common in women than men. Furthermore, the accuracy and area under the receiver operating characteristics curve were obtained as 92.28 and 86.80 on the testing dataset, respectively.
Conclusion: Simple methods such as association rules mining and complex methods could be helpful and give valuable results, and predicting death using association rules provides high accuracy.