Background
Methods
Datasets
Evaluation
Training
Outbreak detection algorithms
Decision fusion methods (DFMs)
Taxonomy and choice
-
The measurement level: A classifier attributes a probability value to each label
-
The rank level: A classifier ranks all labels in a queue and chooses the top label
-
The abstract level (or single class label): A classifier only generates a single-label output (in our case, outbreak yes or no).
Voting methods
Logistic regression
Classification and regression trees (CART)
Bayesian networks (BNs)
Evaluation metrics
Results
Accuracy and quality of prediction assessment
Sensitivity per outbreak | Sensitivity per day | Specificity | PPV | NPV | AUC | |||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
Mean | STD | Mean | STD | Mean | STD | Mean | STD | Mean | STD | Mean | STD | |
CUSUM | 0.83 | 0.28 | 0.45 | 0.29 | 0.87 | 0.17 | 0.49 | 0.35 | 0.94 | 0.03 | 0.73 | 0.14 |
C1 | 0.72 | 0.34 | 0.10 | 0.07 | 0.99 | 0.00 | 0.38 | 0.21 | 0.92 | 0.01 | 0.53 | 0.02 |
C2 | 0.74 | 0.33 | 0.16 | 0.11 | 0.99 | 0.00 | 0.45 | 0.23 | 0.92 | 0.01 | 0.57 | 0.04 |
C3 | 0.82 | 0.25 | 0.25 | 0.14 | 0.96 | 0.00 | 0.36 | 0.16 | 0.93 | 0.01 | 0.62 | 0.07 |
Farrington | 0.86 | 0.20 | 0.20 | 0.11 | 0.97 | 0.02 | 0.51 | 0.33 | 0.92 | 0.01 | 0.66 | 0.10 |
EWMA | 0.89 | 0.20 | 0.29 | 0.17 | 0.95 | 0.02 | 0.37 | 0.20 | 0.93 | 0.02 | 0.64 | 0.09 |
Majority voting | 0.82 | 0.26 | 0.24 | 0.17 | 0.99 | 0.01 | 0.61 | 0.32 | 0.93 | 0.02 | 0.60 | 0.09 |
Weighted majority voting | 0.78 | 0.31 | 0.23 | 0.17 | 0.99 | 0.01 | 0.66 | 0.33 | 0.93 | 0.02 | 0.61 | 0.08 |
Logistic regression | 0.65 | 0.44 | 0.27 | 0.25 | 1.00 | 0.00 | 0.90 | 0.06 | 0.93 | 0.02 | 0.70 | 0.12 |
CARTa | 0.65 | 0.44 | 0.26 | 0.24 | 1.00 | 0.00 | 0.91 | 0.07 | 0.93 | 0.02 | 0.69 | 0.12 |
Bayesian Networks | 0.66 | 0.43 | 0.26 | 0.24 | 1.00 | 0.00 | 0.90 | 0.09 | 0.93 | 0.02 | 0.70 | 0.12 |
Timeliness assessment
Cases required | Proportion of delay | Time to detection | AMOC | AUWROC | ||||||
---|---|---|---|---|---|---|---|---|---|---|
Mean | STD | Mean | STD | Mean | STD | Mean | STD | Mean | STD | |
CUSUM | 0.47 | 0.24 | 0.41 | 0.20 | 5.10 | 2.83 | 0.83 | 0.05 | 0.66 | 0.11 |
C1 | 0.54 | 0.27 | 0.49 | 0.23 | 6.50 | 3.44 | 0.87 | 0.03 | 0.50 | 0.03 |
C2 | 0.52 | 0.27 | 0.48 | 0.23 | 5.90 | 3.09 | 0.86 | 0.03 | 0.54 | 0.05 |
C3 | 0.56 | 0.18 | 0.46 | 0.16 | 6.30 | 2.64 | 0.82 | 0.04 | 0.56 | 0.07 |
Farrington | 0.46 | 0.17 | 0.41 | 0.14 | 5.23 | 2.26 | 0.87 | 0.04 | 0.61 | 0.10 |
EWMA | 0.41 | 0.19 | 0.38 | 0.14 | 5.28 | 2.45 | 0.87 | 0.03 | 0.59 | 0.08 |
Majority voting | 0.49 | 0.22 | 0.44 | 0.18 | 5.30 | 2.56 | 0.75 | 0.11 | 0.57 | 0.07 |
Weighted majority voting | 0.53 | 0.24 | 0.47 | 0.20 | 5.43 | 2.54 | 0.75 | 0.11 | 0.57 | 0.07 |
Logistic regression | 0.59 | 0.31 | 0.56 | 0.30 | 7.15 | 3.82 | 0.82 | 0.07 | 0.63 | 0.10 |
CARTa | 0.60 | 0.30 | 0.57 | 0.30 | 7.10 | 3.85 | 0.77 | 0.12 | 0.62 | 0.09 |
Bayesian networks | 0.60 | 0.30 | 0.56 | 0.29 | 6.75 | 3.55 | 0.81 | 0.09 | 0.63 | 0.11 |
The influence of signal-to-noise difference on outbreak detection performance
Sensitivity per outbreak | Sensitivity per day | Specificity | PPV | NPV | Cases required | Proportion of delay | Time to detection | AUC | AMOC | AUWROC | |
---|---|---|---|---|---|---|---|---|---|---|---|
Positive SND: scenario with a SND = 65.4 | |||||||||||
CUSUM | 1 | 0.74 | 0.83 | 0.29 | 0.97 | 0.17 | 0.25 | 4 | 0.89 | 0.90 | 0.81 |
C1 | 1 | 0.25 | 0.99 | 0.69 | 0.93 | 0.08 | 0.15 | 2 | 0.59 | 0.92 | 0.57 |
C2 | 1 | 0.38 | 0.99 | 0.77 | 0.94 | 0.08 | 0.14 | 2 | 0.66 | 0.92 | 0.64 |
C3 | 1 | 0.54 | 0.96 | 0.61 | 0.95 | 0.09 | 0.19 | 2 | 0.77 | 0.90 | 0.72 |
Farrington | 1 | 0.42 | 1.00 | 0.99 | 0.95 | 0.18 | 0.21 | 2 | 0.84 | 0.93 | 0.79 |
EWMA | 1 | 0.58 | 0.97 | 0.67 | 0.96 | 0.14 | 0.22 | 2 | 0.76 | 0.90 | 0.70 |
Majority voting | 1 | 0.56 | 1.00 | 0.98 | 0.96 | 0.10 | 0.17 | 2 | 0.78 | 0.91 | 0.73 |
Weighted majority voting | 1 | 0.53 | 1.00 | 0.99 | 0.96 | 0.13 | 0.22 | 2 | 0.77 | 0.89 | 0.71 |
Logistic regression | 1 | 0.59 | 0.99 | 0.92 | 0.96 | 0.09 | 0.16 | 2 | 0.84 | 0.94 | 0.80 |
CARTa | 1 | 0.56 | 1.00 | 0.99 | 0.96 | 0.12 | 0.19 | 2 | 0.83 | 0.92 | 0.78 |
Bayesian Networks | 1 | 0.56 | 1.00 | 1.00 | 0.96 | 0.12 | 0.19 | 2 | 0.90 | 0.93 | 0.84 |
Quasi-null SND: scenario with a SND = −1.4 | |||||||||||
CUSUM | 1 | 0.61 | 1.00 | 0.93 | 0.96 | 0.49 | 0.38 | 5 | 0.86 | 0.88 | 0.77 |
C1 | 1 | 0.17 | 0.99 | 0.64 | 0.92 | 0.24 | 0.27 | 4 | 0.55 | 0.89 | 0.53 |
C2 | 1 | 0.28 | 0.99 | 0.75 | 0.93 | 0.24 | 0.26 | 4 | 0.61 | 0.89 | 0.58 |
C3 | 1 | 0.39 | 0.97 | 0.56 | 0.94 | 0.36 | 0.31 | 5 | 0.72 | 0.86 | 0.66 |
Farrington | 1 | 0.27 | 1.00 | 1.00 | 0.93 | 0.35 | 0.34 | 4 | 0.80 | 0.91 | 0.74 |
EWMA | 1 | 0.51 | 0.94 | 0.46 | 0.95 | 0.20 | 0.24 | 4 | 0.76 | 0.90 | 0.70 |
Majority voting | 1 | 0.42 | 1.00 | 0.99 | 0.94 | 0.25 | 0.28 | 4 | 0.71 | 0.86 | 0.65 |
Weighted majority voting | 1 | 0.38 | 1.00 | 1.00 | 0.94 | 0.34 | 0.33 | 4 | 0.50 | 0.50 | 0.50 |
Logistic regression | 1 | 0.70 | 0.99 | 0.93 | 0.97 | 0.22 | 0.27 | 4 | 0.86 | 0.88 | 0.77 |
CARTa | 1 | 0.68 | 1.00 | 0.93 | 0.97 | 0.25 | 0.27 | 4 | 0.84 | 0.86 | 0.75 |
Bayesian Networks | 1 | 0.70 | 0.99 | 0.94 | 0.97 | 0.23 | 0.27 | 4 | 0.86 | 0.88 | 0.77 |
Negative SND: scenario with a SND = −89.2 | |||||||||||
CUSUM | 0.29 | 0.03 | 1.00 | 0.96 | 0.91 | 0.87 | 0.77 | 11 | 0.65 | 0.82 | 0.59 |
C1 | 0.51 | 0.05 | 0.99 | 0.25 | 0.91 | 0.73 | 0.64 | 5 | 0.52 | 0.86 | 0.49 |
C2 | 0.60 | 0.07 | 0.98 | 0.30 | 0.91 | 0.70 | 0.60 | 5 | 0.55 | 0.86 | 0.51 |
C3 | 0.78 | 0.16 | 0.96 | 0.27 | 0.92 | 0.62 | 0.50 | 6 | 0.59 | 0.82 | 0.54 |
Farrington | 0.67 | 0.09 | 0.99 | 0.46 | 0.92 | 0.64 | 0.55 | 5 | 0.60 | 0.87 | 0.56 |
EWMA | 0.98 | 0.18 | 0.95 | 0.25 | 0.92 | 0.47 | 0.37 | 5 | 0.59 | 0.87 | 0.54 |
Majority voting | 0.60 | 0.07 | 0.99 | 0.45 | 0.92 | 0.71 | 0.61 | 5 | 0.53 | 0.69 | 0.51 |
Weighted majority voting | 0.53 | 0.06 | 1.00 | 0.69 | 0.91 | 0.75 | 0.65 | 5 | 0.55 | 0.72 | 0.52 |
Logistic regression | 0.29 | 0.03 | 1.00 | 0.96 | 0.91 | 0.87 | 0.77 | 11 | 0.60 | 0.81 | 0.55 |
CARTa | 0.29 | 0.03 | 1.00 | 0.96 | 0.91 | 0.87 | 0.77 | 11 | 0.59 | 0.77 | 0.54 |
Bayesian Networks | 0.51 | 0.06 | 1.00 | 0.96 | 0.91 | 0.80 | 0.68 | 7 | 0.60 | 0.81 | 0.55 |