-Original-

Indicators of Acute Kidney Injury as Biomarkers to Differentiate Heatstroke from Coronavirus Disease 2019: A Retrospective Multicenter Analysis

Hirofumi Obinata^1,2, Shoji Yokobori^1,3, Kei Ogawa⁴, Yasuhiro Takayama⁵, Shuichi Kawano², Toshimitsu Ito², Toru Takiguchi¹, Yutaka Igarashi¹, Ryuta Nakae¹, Tomohiko Masuno¹ and Hayato Ohwada⁴

¹Department of Emergency and Critical Care Medicine, Nippon Medical School, Tokyo, Japan
²Self-Defense Forces Central Hospital, Tokyo, Japan
³Japan Association of Acute Medicine Heatstroke and Hypothermia Surveillance Committee, Tokyo, Japan
⁴Department of Industrial Administration, Tokyo University of Science, Tokyo, Japan
⁵Department of Emergency Medicine, Flowers & Forest Tokyo Hospital, Tokyo, Japan

Background: Coronavirus disease 2019 (COVID-19) and heat-related illness are systemic febrile diseases. These illnesses must be differentiated during a COVID-19 pandemic in summer. However, no studies have compared and distinguished heat-related illness and COVID-19. We compared data from patients with early heat-related illness and those with COVID-19.
Methods: This retrospective observational study included 90 patients with early heat-related illness selected from the Heatstroke STUDY 2017-2019 (nationwide registries of heat-related illness in Japan) and 86 patients with laboratory-confirmed COVID-19 who had fever or fatigue and were admitted to one of two hospitals in Tokyo, Japan.
Results: Among vital signs, systolic blood pressure (119 vs. 125 mm Hg, p = 0.02), oxygen saturation (98% vs. 97%, p < 0.001), and body temperature (36.6°C vs. 37.6°C, p<0.001) showed significant between-group differences in the heatstroke and COVID-19 groups, respectively. The numerous intergroup differences in laboratory findings included disparities in white blood cell count (10.8 × 10³/μL vs. 5.2 × 10³/μL, p<0.001), creatinine (2.2 vs. 0.85 mg/dL, p<0.001), and C-reactive protein (0.2 vs. 2.8 mg/dL, p<0.001), although a logistic regression model achieved an area under the curve (AUC) of 0.966 using these three factors. A Random Forest machine learning model achieved an accuracy, precision, recall, and AUC of 0.908, 0.976, 0.842, and 0.978, respectively. Creatinine was the most important feature of this model.
Conclusions: Acute kidney injury was associated with heat-related illness, which could be essential in distinguishing or evaluating patients with fever in the summer during a COVID-19 pandemic.

J Nippon Med Sch 2021; 88: 80-86

Keywords
heat-related illness, coronavirus disease, COVID-19, heatstroke, machine learning

Correspondence to
Hirofumi Obinata, Department of Emergency and Critical Care Medicine, Nippon Medical School, 1-1-5 Sendagi, Bunkyo-ku, Tokyo 113-8603, Japan
h-obinata@nms.ac.jp

Received, July 6, 2020
Accepted, July 31, 2020