Table 5.
Table summarizing the main characteristics of COVID-19 prediction models reported in the literature based on loss of smell and taste.
| Menni et al. [9] | Roland et al. [10] | Clemency et al. [14] | Kowall et al. [35] | Gerkin et al. [15] | Our Study | |
|---|---|---|---|---|---|---|
| Sample | (UK)6452+/9186− (US)726+/2037− |
145+/157− | 225+/736− | 296+/1641− | 4148+/546− | 421+/356− |
| Demographic Data | (UK) Positive group: Mean Age: 41.25 71.88% female Negative group: Mean Age: 43.2 76.40 % female |
Mean age: 39 Sex: 72% female |
N/A | Mean age: 53.5 years Sex: 61.3% females in the negative group and 57.8% in the positive group |
Positive group: Mean Age: 40.6 74% female Negative group: Mean Age: 43.2 78% female |
Positive group: Mean Age: 47.3 61% female Negative group: Mean Age: 45.2 78% female |
| Data collection | App-based symptom tracker | Public survey posted on social media | Nurse call center for healthcare workers (HCW) | Self-administered questionnaire | Online survey | Self-administered questionnaire |
| Variable Types | Categorical | Categorical | Categorical | Categorical | Categorical, continuous VAS | Categorical, Continuous VAS |
| Classification Methods | ● Stepwise (forward and backward) ● Logistic Regression ● Akaike Information Criterion (AIC) Classifier threshold at 0.5 |
● Stepwise Logistic Regression ● (p = 0.05 for entry and 0.10 for removal with maximum iterations set at 20) Classifier threshold at 0.5 |
Logistic regression with maximum positive likelihood ratio (PLR) criterion |
Stepwise backward logistic regression (p = 0.10 for entry and for removal) |
L1 regularized logistic regression (penalty α = 1) | ● Stepwise (forward and backward) ● Logistic Regression ● Bayesian Information Criterion (BIC) ● Random Forest (RF) ● Support Vector Machine (SVM) Classifier threshold at 0.5 |
| Predictors | Age, sex, loss of smell and taste, severe or significant persistent cough, severe fatigue, skipped meals | (1) Smell or taste change, fever, body ache, shortness of breath, sore throat (2) Smell or taste change, fever and/or myalgia |
(1) Fever, shortness of breath, dry cough (2) Fever, loss of taste or smell (3) Fever, shortness of breath, dry cough, loss of taste or smell |
Age, sex, age, return from abroad, close contact with a confirmed case, the presence of fever, cough, exhaustion, taste or smell disorder, current smoking, general health condition and number of comorbidities | (1) Loss of smell, time duration (2) Model with 70 features |
Five model datasets (see Table 4) including different variables among: age, sex, loss of smell, loss of taste, nasal obstruction, nasal discharge, facial pain, cough, dyspnea, fever and diarrhea |
| Validation method | ● Holdout 80:20% ● training/test ● 10-fold cross-validation in the UK sample ● US validation sample |
Holdout 75:25% training/test | N/A | Holdout 60:40% training-test | 100-fold cross-validation with 80:20% training-test | ● Holdout 75:25% training-test ● 50-fold cross-validation with ● 75:25% training-test |
| Accuracy Parameters | AUC = 0.76 SE = 0.66 SP = 0.83 PPV = 0.58 NPV = 0.87 |
(1) AUC = 0.82 SE = 0.56 (2) AUC = 0.75 SE = 0.70 SP = 0.73 |
(1) AUC = 0.63 SE = 0.93 SP = 0.09 (2) AUC = 0.75 SE = 0.89 SP = 0.48 (3) AUC = 0.77 SE = 0.98 SP = 0.08 |
AUC = 0.821 | AUC = 0.72 SE = 0.85 SP = 0.75 |
AUC = 0.80 SE = 0.82 SP = 0.78 PPV = 0.81 NPV = 0.78 |