| Autor: |
French, Nigel, Jones, Geoff, Heuer, Cord, Hope, Virginia, Jefferies, Sarah, Muellner, Petra, McNeill, Andrea, Haslett, Stephen, Priest, Patricia |
| Rok vydání: |
2021 |
| DOI: |
10.6084/m9.figshare.16907278 |
| Popis: |
Additional file 1. Additional material. Figure S1. Distribution of the combination of 15 symptoms reported by confirmed, probable and ���not a case��� individuals. Only symptoms reported in >100 people and combinations that occurred in 5 or more individuals are included. Figure S2. Multiple correspondence analysis plots showing the location of each individual in the first two dimensions, coloured by their PCR status (plot A), and symptoms (plots B-L) and 90% prediction ellipses . Plots B and C are symptoms strongly correlated with the second dimension and determine the 3 distinct groups. The upper group are individuals with ageusia and anosmia , the middle group are more likely to be those with anosmia but no ageusia, and the lower group are those without ageusia or anosmia. Plots D-L are symptoms more strongly correlated with the first dimension (positive individuals are more likely to be to the right of all three clusters). Plots K and L are examples of two respiratory symptoms that are not correlated with either dimension. Figure S3. Scatterplot of sensitivity and 1-specificity for combinations of symptom variables using ���status��� as a gold standard. A A rule based on combinations of at least one of 5 symptoms. The red dot in the left plot identifies the current case definition based on respiratory symptoms (A|Z|C|B|T). The green and blue dots are sets of largely non-respiratory symptoms used for comparison. B A rule based on one or more of respiratory symptoms (coryza, cough, sore throat or shortness of breath, Z|C|B|T) AND one or more of 5 non-respiratory symptoms. The table shows the estimated values for the example sets highlighted in the figure. Figure S4. Posterior distributions of the sensitivity (upper row) and specificity (lower row) of both the symptoms (headache OR general weakness OR anosmia OR muscle pain OR joint pain) and the PCR assay, estimated using Bayesian Latent Class Analysis, based on three chains of 20,000 iterations after a burn-in of 5,000. Figure S5. Scatterplot of sensitivity and 1-specificity for combinations of symptom variables estimated using Bayesian Latent Class Analysis. A A rule based on combinations of at least one of 5 symptoms. The red dot in the left plot identifies the current case definition based on respiratory symptoms (A|Z|C|S|B). The green and blue dots are sets of largely non-respiratory symptoms used for comparison. B A rule based on one or more of respiratory symptoms (cough, sore throat, coryza or shortness of breath, Z|C|S|B) AND one or more of 5 non-respiratory symptoms. Figure S6. Decision tree built using a machine learning algorithm that minimises the misclassification of cases and non-cases. The thickness of the lines denotes the proportion of the population heading down each branch. Darker shading indicates greater node purity. |
| Databáze: |
OpenAIRE |
| Externí odkaz: |
|
