A total of 1435 (823 males) patients were included in this study, where 1094 patients tested positive (HFMD group) for enterovirus RNA while 341 patients (non-HFMD group) tested negative (Table 1). Among 1094 confirmed cases of HFMD, 215 were EV710-positive, 283 were CA16-positive, and 596 were positive when tested with a set of non-EV71, non-CA16 universal primers for enterovirus (the pathogens include CA6, CA10, CA4, and other types of enterovirus). While no difference in gender composition was found between the two groups, HFMD patients were older and had longer illness duration when compared with non-HFMD patients (Table 1).
A subset of children in both groups (442 in the HFMD group and 39 in the non-HFMD group) endorsed a history of close contact with patients with HFMD or herpangina. The proportion of patients with clear exposure history was higher in the HFMD group than in the non-HFMD group (Table 1).
Since HFMD rashes are often concentrated in specific locations of the body, we quantified the rash severity by dividing the body into discrete regions. The oral cavity was divided into the hard palate, soft palate, tongue, buccal mucosa, lip mucosa, and gums. The remainder of the body was divided into the face, chest, back, buttocks, upper limbs, lower limbs, palms, back of the hands, fingers, feet, dorsum of the feet, plantar surface of the feet, and toes. In each patient, the number of rash spots/ulcers/sores in each body region was counted. We observed significant differences in rash densities in the upper jaw, soft palate, tongue, buccal mucosa, gums, chest, back, buttocks, and toes (Table 2).
Additionally, we analyzed additional clinical information such as fever severity, length of fever, fever-to-rash interval, presence of cough, gastrointestinal symptoms, WBC count, and neutrophil percentage. Between-group differences were found in fever frequency, WBC count, and neutrophil percentage (Table 3).
Of the 1004 cases in the training set, 764 were HFMD patients and 240 were non-HFMD patients. Multivariate logistic regression was then used to establish a scoring model based on the training set data. Among 15 tested clinical variables, 7 statistically significant clinical variables were identified and subsequently included in the scoring model. These included the following: (1) age; (2) exposure history; the number of ulcers on the (3) hard palate, (4) soft palate, (5) buccal mucosa; and cutaneous rash distributed on the (6) back and (7) buttocks (Table 4). To test the predictive accuracy of this scoring system, we applied this model on data from patients in the training set. The median score of the HFMD group was 10 (7, 15). The median score of the non-HFMD group was 4 (2, 8), which was significantly lower than that in the HFMD group (Wilcoxon rank sum test, Z=14.24, P < 0.001). The final scores ranged from −5 to 24 points with predictive accuracies of 0.12 to 0.99. The AUC was 0.804 (95% confidence interval [CI]: 0.773–0.835), which was significantly higher than 0.5 (defined as having a predictive ability at chance; P < 0.001). The sensitivity was 0.76 and specificity was 0.68. Additionally, we found the positive predictive value to be 0.88 and the negative predictive value to be 0.47 (Fig. 1). We found the optimal cut-off point to be seven; hence, a score of seven or greater suggested a positive HFMD diagnosis, while a score of less than seven could be diagnosed as non-HFMD.
The test set was then used to validate and evaluate the scoring model. The test set consisted of 431 patients (330 HFMD and 101 non-HFMD). Within the test set, the median score of the HFMD group was 10 (7, 14) and the median score of the non-HFMD group was 5 (2, 8) (Z=7.93, P < 0.001). The final scores ranged from −< 0.001 points with predictive accuracies of 0.12 to 0.99. The AUC was 0.76 (95% CI: 0.710–0.810) ( P < 0.001) ,with a sensitivity of 0.76 and specificity of 0.62. The positive predictive value was 0.87 and the negative predictive value was 0.44 (Fig. 2).
The comparison of general data and exposure history data between the training set and the test set is shown in Table 5. In the test set, the population is divided into three groups according to the tertiles of the predicted probability, and the actual prevalence rate of each group is relatively consistent with the predicted probability of disease (Table 6).