Value of Laboratory Tests in COVID-19 Hospitalized Patients for Clinical Decision-makers: A Predictive Model, Using Data Mining Approach

Reports, mostly from high-income countries, have shown a wide range of symptoms, clinical profile, and outcomes for patients diagnosed with COVID-19. However, little is known about these issues in developing countries. This research used medical records in 15 hospitals in Tehran, the capital city of Iran, to assess predictors of in-hospital mortality in patients diagnosed with COVID-19.

The required information was extracted from patients' medical records, including age, gender, laboratory data (complete blood count, serum electrolytes, and liver, renal, and muscle injury tests) at admission, and the outcome of in-hospital mortality (yes/no) of 4,542 adult patients with confirmed COVID-19. This research used logistic regression to assess the predictors for mortality (measured as adjusted odds ratio [aOR]) and Chi-square automatic interaction detector to classify high-risk patients in different age groups as a decision tree model. Two models were developed through a machine learning approach.

Overall, 822 (18.09%) cases passed away in the hospital. Mortality risk was increased from 4.33% in patients aged 18-40 years old to 40.96% in those aged 80+ years old. After adjusting for covariates, age (aOR 1.62 to 7.05 vs. those aged 18-40 years old), high aspartate transaminase (aOR 1.64 to 3.21), high alkaline phosphatase (aOR=2.17), low sodium (aOR=1.31), high sodium level (aOR=5.05), high potassium (aOR=2.41), low calcium (aOR=2.31), high creatine phosphokinase (aOR=2.21 to 2.24), and high creatinine (aOR=3.43) were significantly associated with in-hospital mortality.

Based on the results of our study, the mortality rate was high among in-hospital patients, particularly among older age and those who had liver and renal dysfunctions, muscle injury, and electrolyte imbalance at admission. Triage and special care for these high-risk patients can improve in-hospital outcomes.