Clinical clustering with prognostic implications in Japanese COVID-19 patients: report from Japan COVID-19 Task Force, a nation-wide consortium to investigate COVID-19 host genetics.

ABSTRACT

BACKGROUND: The clinical course of coronavirus disease (COVID-19) is diverse, and the usefulness of phenotyping in predicting the severity or prognosis of the disease has been demonstrated overseas. This study aimed to investigate clinically meaningful phenotypes in Japanese COVID-19 patients using cluster analysis. METHODS: From April 2020 to May 2021, data from inpatients aged ≥ 18 years diagnosed with COVID-19 and who agreed to participate in the study were collected. A total of 1322 Japanese patients were included. Hierarchical cluster analysis was performed using variables reported to be associated with COVID-19 severity or prognosis, namely, age, sex, obesity, smoking history, hypertension, diabetes mellitus, malignancy, chronic obstructive pulmonary disease, hyperuricemia, cardiovascular disease, chronic liver disease, and chronic kidney disease. RESULTS: Participants were divided into four clusters Cluster 1, young healthy (n = 266, 20.1%); Cluster 2, middle-aged (n = 245, 18.5%); Cluster 3, middle-aged obese (n = 435, 32.9%); and Cluster 4, elderly (n = 376, 28.4%). In Clusters 3 and 4, sore throat, dysosmia, and dysgeusia tended to be less frequent, while shortness of breath was more frequent. Serum lactate dehydrogenase, ferritin, KL-6, D-dimer, and C-reactive protein levels tended to be higher in Clusters 3 and 4. Although Cluster 3 had a similar age as Cluster 2, it tended to have poorer outcomes. Both Clusters 3 and 4 tended to exhibit higher rates of oxygen supplementation, intensive care unit admission, and mechanical ventilation, but the mortality rate tended to be lower in Cluster 3. CONCLUSIONS: We have successfully performed the first phenotyping of COVID-19 patients in Japan, which is clinically useful in predicting important outcomes, despite the simplicity of the cluster analysis method that does not use complex variables.