Comment on Cai et al. Obesity and COVID-19 Severity in a Designated Hospital in Shenzhen, China. Diabetes Care 2020;43:1392–1398

We read with interest the analysis of the association of obesity with the severity of coronavirus disease 2019 (COVID-19) by Cai et al. (1) using data from a referral hospital in Shenzhen, China. The important finding is that obese patients had increased odds (3.40-fold) of progressing to severe COVID-19 compared with normal-weight patients. However, there are few available data on BMI for patients with COVID-19. We aimed to investigate whether BMI is a predictor of severity in COVID-19 patients by using a machine learning–based prognostic model.

Our study recruited 98 patients with confirmed COVID-19 (by PCR assay) admitted to Taizhou Enze Medical Center, Zhejiang. The study was approved by the local hospital ethics committee, and written informed consent was obtained from patients involved before enrollment when data were collected. Severe COVID-19 was defined based on the World Health Organization Interim Guidance Report criteria for severe pneumonia (2).

In the study population, 26 people (36.5%) were categorized as severe patients and 72 as nonsevere patients. Most severe patients were older (53.35 ± 12.67 vs. 45.00 ± 14.73 years, P = 0.01) and had higher BMI (24.89 ± 2.54 vs. 23.11 ± 2.54 kg/m², P = 0.003) compared with nonsevere patients. Using Pearson analysis, we discovered 17 key factors affecting the severity of COVID-19, including lymphocyte count, aspartate aminotransferase, creatine kinase, alanine aminotransferase, C-reactive protein, albumin, γ-glutamyl transpeptidase, sodium, platelet count, BMI, age, highest temperature, myalgia or fatigue, fever, cough, shortness of breath, and bilateral involvement on chest radiographs. The predictive models were built based on four machine learning classification algorithms—XGBoost classifier, K-nearest neighbors (KNN), decision tree, naive Bayes, and multilayer perceptron (MLP) classifier—by using Python programming software version 3.6.5. After comparing different machine learning methods, we choose the naive Bayes model as the optimal prediction model. The naive Bayes model exhibited the best accuracy rate, and the performance was signifificantly better than the other models. The accuracy rate, precision, recall rate, and standard deviation were 91.0%, 81.4%, 90.6%, and 0.037, respectively. Using the variables exhibiting the highest coefficients of permutation importance for the severity of COVID-19 in the naive Bayes model, the variable importance plot suggested that the lymphocyte count level was the most important predictor, followed by BMI.

These findings add to the growing literature highlighting that higher BMI is a critical factor associated with severity of COVID-19. There is limited scientific evidence regarding the mechanisms linking obesity and COVID-19, but some information can be extrapolated from studies conducted on subjects with influenza (3). ACE2 is the putative receptor for the entry of severe acute respiratory syndrome coronavirus 2 into host cells (4), and the level of ACE2 expression in adipose tissue is higher than that in lung tissue (5). Therefore, individuals with obesity, having more adipose tissue, may have an increased number of ACE2-expressing cells and consequently a larger amount of ACE. This is an important finding because obese individuals might also be more vulnerable to COVID-19.

In the context of the unprecedented health crisis due to the COVID-19 pandemic, these results may have major implications in public health strategy, especially in Western countries affected by a high prevalence of obesity.

• © 2020 by the American Diabetes Association