Since the middle of December 2019, cases of pneumonia caused by coronavirus disease 2019 (COVID-19) were found in Wuhan, Hubei Province, and the virus began rapidly spreading throughout China (1-3). The percentage of patients with diabetes and/or hypertension contracting COVID-19-infected patients with diabetes or hypertension was approximately 10% more higher than that of those without these coexisting disorders (3). However, statistical evidence to suggest that patients with diabetes or hypertension were more predisposed to suffer more severe illness was insufficient so far. This study aimed to explore the effects of patient characteristics on COVID-19 severity including age, sex, and underlying diseases. We analyzed hospital-based data from January 23, 2020 to February 25, 2020 in Yichang, Hubei Province by applying ordinal and binary logistic regression. The results suggested elderly (≥65 years) patients with diabetes are more likely to suffer severe illnesses. Based on the fact that severity is directly related to patients’ survival, proper treatments for this group are particularly important. Furthermore, prevention must be emphasized.
The medical records of patients who were diagnosed with COVID-19 by nucleic acid tests were obtained from hospital-based data from January 23, 2020 to February 25, 2020 in Yichang, Hubei Province and were collected by a monitoring network for two hospitals including the Third People’s Hospital of Yichang and the Jiangnan branch of the Central People’s Hospital of Yichang based on the big data platform. This study sample consisted of all 882 symptomatic living patients, whose clinical data involved individual information (age and sex) and a personal history of underlying disease (diabetes and hypertension). The severity of disease was diagnosed by the reporting system of communicable disease. This severity was categorized into mild, moderate, and severe illnesses, which could be regarded as an ordinal response variable based on specialized medical knowledge. Age, sex, and personal history of underlying diseases were considered to be study factors. According to global recommendations on physical activity for health developed by WHO, age, which was originally a continuous variable, was dichotomized into the young and middle-aged group (age <65 years), and senior group (age ≥65 years). Personal history of underlying diseases was classified into the following categories: patients without diabetes and hypertension, patients with hypertension only, patients with diabetes only, and patients with diabetes and hypertension. Before we fitted the ordinal logistic regression (OLR) model with studies factors, all independent variables chosen from the studied study factors should be significantly associated with the severity of this disease, which were was ensured by applying Pearson chi-square test (4). First an OLR model-proportional odds model (POM) was adopted to explore the risk factors of the severity of this disease if the proportional odds assumption was not violated (4-5). Second, if the above stringent assumption was not satisfied, two efficient approaches were generally developed to fit the data. The One solution for analyzing the data is to apply an OLR partial proportional odds model (PPOM), in which the proportional odds assumption is not valid (5). The other method is to use multiple separate binary logistic regression (BLR) models Besides, the global goodness-of-fit is crucial to evaluating the adequacy and efficiency of statistical models. The likelihood-ratio test was commonly conducted to assess the overall goodness-of-fit of ordinal logistic regression POM and PPOM, while that of BLR models was verified by the Hosmer and Lemeshow test (4). All statistical analyses were conducted by R statistical software (version 3.5.1; The R Foundation for Statistical Computing).
The proportion of patients with mild, moderate, and severe illnesses in symptomatic living patients was 17.69%, 61.00%, and 21.31% respectively. The distribution of patients with different illnesses according to studied influential factors are depicted in Figure 1 and Table 1. The percentage of males (51.25%) was similar to that of females in patients, and the Pearson chi-square test revealed that there was no difference on gender for the severity of this disease of patients at the 0.05 significance level. The severity of this disease was significantly different on the two age groups and four statuses of underlying diseases according to the results of Pearson chi-square statistics and p-values. The severity of this disease as three ordinal categories was regarded as the dependent variable, while age group and status of underlying diseases selected to be the explanatory variables were employed to fit the OLR POM. However, the score test of the proportional odds assumption is significant (Chi-square=31.569, df=4, p-value<0.001), and the results of single score test of that assumption for each independent variable indicated that age violated that assumption (p-value<0.001). Therefore, we considered the estimate of coefficients obtained to be biased and applied an OLR PPOM as well as a BLR model to analyze the data. Results of the PPOM and BLR model are separately presented in Tables 2–3. The results of PPOM suggested that the risk of suffering severe illness was 4.71 (95% CI: 3.25–6.82) times higher among senior patients, when compared with young and middle-aged patients. The odds ratio for patients with diabetes as well as patients with diabetes and hypertension compared to patients without diabetes and hypertension were approximately 3.00 (95% CI: 1.47–6.09) and 2.89 (95% CI: 1.43–5.82), respectively, when the severe illness was compared with non-severe illnesses, or moderate and severe illnesses were compared with mild illness. Nevertheless, the risk of suffering from severe illness was found insignificant for patients only with hypertension compared to patients without hypertension and diabetes. The likelihood ratio statistic and p-value indicated PPOM built with adequate goodness-of-fit, while the Nagelkerke pseudo R2 of POM and PPOM were 0.103 and 0.140, respectively, further favoring PPOM as a better fit than POM. The results of the BLR model were relatively consistent with that of PPOM.
Independent variable Mild illness Moderate illness Severe illness Total Pearson chi-square (p-value) Age (years old) <65 125 457 91 673 104.41 (<0.001) ≥65 31 81 97 209 Sex Male 84 271 97 452 0.596 (0.742) Female 72 267 91 430 Underlying diseases Non-diabetes and non-hypertension 132 450 115 697 53.294 (<0.001) Only hypertension 18 57 39 114 Only diabetes 3 17 14 34 Diabetes and hypertension 3 14 20 37 Total 156 538 188 882 –
Table 1. Results of univariate analyses for the effect of the influential factors on the severity of illnesses.
Influential factor (Moderate & Severe illnesses) vs. Mild illness Severe illness vs. (Mild & Moderate illnesses) Coefficient OR (95% CI) p-value Coefficient OR (95% CI) p-value Intercept 1.416 − <0.001 −1.963 − <0.001 Age (years) (0–64 years as reference) ≥65 years 0.100 1.105 (0.711, 1.719) 0.657 1.549 4.706 (3.248, 6.816) <0.001 Underlying diseases (non-diabetes and non-hypertension as reference) Only hypertension 0.328 1.388 (0.915, 2.106) 0.123 0.328 1.388 (0.915, 2.106) 0.123 Only diabetes 1.095 2.990 (1.467, 6.086) 0.003 1.095 2.990 (1.467, 6.086) 0.003 Diabetes and hypertension 1.060 2.885 (1.430, 5.822) 0.003 1.060 2.885 (1.430, 5.822) 0.003 Note: Score test for the proportional odds assumption: Chi-square=0.607, df=3, p-value=0.895. Goodness-of-fit of overall model (Likelihood Ratio): Chi-square=111.47, df=5, p-value<0.001, Pseudo R2=0.1402.
Table 2. Results of the effect of the influential factors for the severity of illnesses obtained by OLR PPOM.
Influential factor Severe illness vs. Mild illness Severe illness vs. Moderate illness Moderate illness vs. Mild illness Coefficient OR (95% CI) p-value Coefficient OR (95% CI) p-value Coefficient OR (95% CI) p-value Intercept −0.4808 − 0.001 −1.738 − <0.001 1.273 − <0.001 Age (years) (0–64 years as reference) ≥65 years 1.260 3.524
<0.001 1.619 5.048
<0.001 −0.385 0.681
0.123 Underlying diseases (non-diabetes and non-hypertension as reference) Only hypertension 0.545 1.725
0.101 0.429 1.536
0.100 0.063 1.065
0.835 Only diabetes 1.623 5.069
0.014 1.077 2.937
0.007 0.547 1.727
0.390 Diabetes and hypertension 1.499 4.478
0.022 1.025 2.787
0.010 0.483 1.620
0.462 Hosmer-lemeshow goodness-of-fit test 0.614 0.380 0.980 Sample size 344 726 694
Table 3. Results of the effect of the influential factors for the severity of illnesses obtained by three separate BLR models.
|Independent variable||Mild illness||Moderate illness||Severe illness||Total||Pearson chi-square (p-value)|
|Age (years old)|
|Non-diabetes and non-hypertension||132||450||115||697||53.294 (<0.001)|
|Diabetes and hypertension||3||14||20||37|