Analyses of correlates of SARS-CoV-2 infection or mortality have usually assessed individual predictors. This study aimed to determine if patterns of combined predictors may better identify risk of infection and mortality. This is a retrospective cohort study of 106 hospitalized patients.
For the first 9 days of the pandemic in Indonesia, we selected all 18 confirmed cases, all 60 suspected cases, and 28 putatively negative patients with pneumonia and no travel history. Hierarchical cluster analyses (HCA) and principal component analyses (PCA) identified cluster and covariance patterns for symptoms or hematology which were analyzed with other predictors of infection or mortality using logistic regression.
For univariate analyses, no significant association with infection was seen for fever, cough, dyspnea, headache, runny nose, sore throat, gastrointestinal complaints (GIC), or hematology. A PCA symptom component for fever, cough, and GI symptoms tended to associate with increased risk of infection (OR 3.41; 95% CI 1.06-14; p=0.06), and a hematology component with elevated monocytes had decreased risk (OR 0.26; 0.07-0.79; 0.027).
Multivariate analysis revealed that an HCA cluster of 3-5 symptoms, typically fever, cough, headache, runny nose, sore throat but little dyspnea and no GI symptoms tended to reduce risk (aOR 0.048; <0.001–0.52; 0.056).
In univariate analyses for death, an HCA cluster of cough, fever and dyspnea had increased risk (OR 5.75; 1.06 − 31.3, 0.043). Other significant predictors of infection were age ≥ 45, international travel, contact with a Covid-19 patient, and pneumonia.
Diabetes and history of contact were associated with higher mortality. Cluster groups and co-variance patterns may be stronger correlates of SARS-CoV-2 infection than individual predictors.