3.1 Decision Tree Analyses of Clinical Data
3.1.1 Distinguishing Dengue Fever from other febrile Illnesses
3.1.1.3 Dengue Prediction based on Cytokine Data
We also addressed the question whether it would be possible to classify dengue patients by only using the cytokine data. The first constructed tree (DENPRE_CYTOA_291) (Figure 3.9) showed a sensitivity of 84% and a specificity of 94% which resulted in an overall error rate of 9.28% (27 cases). The average profit of the model was 0.814 and the positive AUC was 0.92 (95%CI: 0.88, 0.96). The negative AUC had also a value of 0.92 (95%CI: 0.89, 0.95) (Table 3.15; Figure 3.10).
The first split of the tree was represented by IFN_ALPHA_1>389.21 (OR: 79.54;
95%CI: 76.61, 82.47) for dengue positive cases. Patients showing lower levels of IFN_ALPHA_1 were further sub divided by IL_2_1<=2.72 (OR: 63.33; 95%CI: 58.63, 68.03) which represented a higher risk for patients to be dengue positive. Classification of cases having IL_2_1>2.72 additionally needed IL_10_1>9.14 (OR: 16.10; 95%CI:
13.34, 18.85) as well as TNF_1<=4.33 (OR: 32.5; 95%CI: 23.07, 41.93) as splitting criteria 6 (Table 3.13; Table 3.14).
ROOT 95 positives / 196 NEGATIVES
IFN_ALPHA_1 > 389.21 58.43 POSITIVES / 4 negatives
IFN_ALPHA_1 <= 389.21 36.57 positives / 192 NEGATIVES
IL_2_1 > 2.72 20.57 positives / 190 NEGATIVES
IL_2_1 <= 2.72 16 POSITIVES / 2 negatives
IL_10_1 > 9.14 14.57 positives / 21 NEGATIVES
IL_10_1 <= 9.14 6 positives / 169 NEGATIVES
TNF_1 <= 4.33 13.57 POSITIVES / 6 negatives
TNF_1 > 4.33 1 positive / 15 NEGATIVES ROOT
95 positives / 196 NEGATIVES
IFN_ALPHA_1 > 389.21 58.43 POSITIVES / 4 negatives
IFN_ALPHA_1 <= 389.21 36.57 positives / 192 NEGATIVES
IL_2_1 > 2.72 20.57 positives / 190 NEGATIVES
IL_2_1 <= 2.72 16 POSITIVES / 2 negatives
IL_10_1 > 9.14 14.57 positives / 21 NEGATIVES
IL_10_1 <= 9.14 6 positives / 169 NEGATIVES
TNF_1 <= 4.33 13.57 POSITIVES / 6 negatives
TNF_1 > 4.33 1 positive / 15 NEGATIVES
Figure 3.9: DENPRE_CYTOA_291: Decision tree for dengue prediction calculated on 291 patients only using cytokine data. IFN_ALPHA=interferon-α; IL_10=interleukin-10; IL_2=interleukin-2;
TNF=tumor necrosis factor α; 1=1st visit data.
Table 3.13: DENPRE_CYTOA_291: Decision tree for dengue prediction calculated on 291 patients only including cytokine data. Statistical analysis of splitting criteria performed on the whole dataset.
IFN_ALPHA=interferon-α; IL_10=interleukin-10; IL_2=interleukin-2; TNF=tumor necrosis factor α;
1=1st visit data; RR=relative risk; OR=odds ratio; CI=confidence interval.
Decision Node Feature RR OR 95% CI (OR) p value IFN_ALPHA_1 [pg/ml]
Cut-off value > 389.21 6.07 79.54 76.61, 82.47 < 0.001 IL_2_1 [pg/ml]
Cut-off value <= 2.72 3.69 42.62 38.32, 46.93 < 0.001 IL_10_1 [pg/ml]
Cut-off value > 9.14 2.80 6.33 4.49, 8.16 < 0.001
TNF_1 [pg/ml]
Cut-off value <= 4.33 11.49 23.93 21.53, 26.33 < 0.001
Table 3.14: DENPRE_CYTOA_291: Decision tree for dengue prediction calculated on 291 patients only including cytokine data. Statistical analysis of splitting criteria performed on each subgroup at the decision nodes. IFN_ALPHA=interferon-α; IL_10=interleukin-10; IL_2=interleukin-2; TNF=tumor necrosis factor α; 1=1st visit data; RR=relative risk; OR=odds ratio; CI=confidence interval.
Decision Node Feature RR OR 95% CI (OR) p value IFN_ALPHA_1 [pg/ml]
Cut-off value > 389.21 6.07 79.54 76.61, 82.47 < 0.001 IL_2_1 [pg/ml]
Cut-off value <= 2.72 8.79 63.33 58.63, 68.03 < 0.001 IL_10_1 [pg/ml]
Cut-off value > 9.14 10.06 16.10 13.34, 18.85 < 0.001 TNF_1
Cut-off value <= 4.33 10.95 32.50 23.07, 41.93 < 0.001
Table 3.15: DENPRE_CYTOA_291: Summary of K-fold (k=10) cross validation for dengue prediction based on 291 patients only including cytokine data.
Overall Evaluation Value (n=291) Confusion Matrix Total
misclassifications 27.0 Predicted Class
Overall error rate 9.276% neg pos
SE of error rate 5.637
Average profit 0.814 neg 184
(94%) 12 (6%) SE of profit 0.113
AUC negative 0.9193 95%CI: 0.89,
0.95 Actual Class pos 15
(16%) 80 (84%) AUC positive 0.9193 95%CI: 0.88,
0.96
Figure 3.10: DENPRE_CYTOA_291: Receiver operating characteristics (ROC) curve for dengue
Finally, we calculated a tree (minimum cases set to 14; pruning confidence set to 25%) based on only the cytokine data and excluding IFN_ALPHA_1 (DENPRE_CYTO_291) (Figure 3.11). This resulted in a correct classification of 264 (90.72%) of the cases with a sensitivity of 84% and a specificity 94% (Table 3.18;
Figure 3.12). The overall performance of the tree was slightly worse compared to the inclusion of IFN_ALPHA_1 with an AUC for positive (95%CI: 0.87, 0.95) and negative (95%CI: 0.88, 0.94) cases of 0.91 but had the same average profit of 0.814.
IL_2_1<=2.54 (OR: 88.21; 95%CI: 80.73, 95.68) was chosen as the first splitting criteria whereby cases below this threshold were more likely to have dengue. On the other hand, patients having higher levels of IL_2_1 were further classified into positive cases by taking IP_10_1>878.7 (OR: 29.12; 95%CI: 26.77, 31.46) and TNF_1<=4.29 (OR: 25.75; 95%CI: 22.68, 28.82) as characteristics for dengue infections 7 (Table 3.16; Table 3.17).
ROOT 95 positives / 196 NEGATIVES
IL_2_1 <= 2.54 29 POSITIVES / 0 negative
IL_2_1 > 2.54 66 positives / 196 NEGATIVES
IP_10_1 <= 878.7 7 positives / 152 NEGATIVES
IP_10_1 > 878.7 59 POSITIVES / 44 negatives
TNF_1 > 4.29 5 positives / 31 NEGATIVES
TNF_1 <= 4.29 54 POSITIVES / 13 negatives ROOT
95 positives / 196 NEGATIVES
IL_2_1 <= 2.54 29 POSITIVES / 0 negative
IL_2_1 > 2.54 66 positives / 196 NEGATIVES
IP_10_1 <= 878.7 7 positives / 152 NEGATIVES
IP_10_1 > 878.7 59 POSITIVES / 44 negatives
TNF_1 > 4.29 5 positives / 31 NEGATIVES
TNF_1 <= 4.29 54 POSITIVES / 13 negatives
Figure 3.11: DENPRE_CYTO_291: Decision tree for dengue prediction calculated on 291 patients only including cytokine data (excl. IFN_ALPHA_1). IL_2=interleukin-2; IP_10=interferon-inducible protein 10; TNF=tumor necrosis factor α; 1=1st visit data.
7 IL_2=interleukin-2; IP_10=interferon-inducible protein 10; TNF=tumor necrosis factor α; 1=1st visit
Table 3.16: DENPRE_CYTO_291: Decision tree for dengue prediction calculated on 291 patients only including cytokine data (excl. IFN_ALPHA_1). Statistical analysis of splitting criteria performed on the whole dataset. In case of 0 values in the original contingency table, OR calculations were adjusted by adding 1 to each table value +1. IL_2=interleukin-2; IP_10=interferon-inducible protein 10; TNF=tumor necrosis factor α; 1=1st visit data; RR=relative risk; OR=odds ratio; CI=confidence interval.
Decision Node Feature RR OR 95% CI (OR) p value IL_2_1 [pg/ml] +1
Cut-off value <= 2.54 3.81 88.21 80.73, 95.68 < 0.001 IP_10_1 [pg/ml]
Cut-off value > 878.7 10.67 29.36 27.28, 31.45 < 0.001 TNF_1 [pg/ml]
Cut-off value <= 4.29 11.65 24.45 22.05, 26.86 < 0.001
Table 3.17: DENPRE_CYTO_291: Decision tree for dengue prediction calculated on 291 patients only including cytokine data (excl. IFN_ALPHA_1). Statistical analysis of splitting criteria performed on each subgroup at the decision nodes. In case of 0 values in the original contingency table, OR calculations were adjusted by adding 1 to each table value+1. IL_2=interleukin-2; IP_10=interferon- inducible protein 10; TNF=tumor necrosis factor α; 1=1st visit data; RR=relative risk; OR=odds ratio;
CI=confidence interval.
Decision Node Feature RR OR 95% CI (OR) p value IL_2_1 [pg/ml] +1
Cut-off value <= 2.54 3.81 88.21 80.73, 95.68 < 0.001 IP_10_1 [pg/ml]
Cut-off value > 878.7 13.01 29.12 26.77, 31.46 < 0.001 TNF_1 [pg/ml]
Cut-off value <= 4.29 5.80 25.75 22.68, 28.82 < 0.001
Table 3.18: DENPRE_CYTO_291: Summary of K-fold (k=10) cross validation for dengue prediction based on 291 patients only including cytokine data (excl. IFN_ALPHA_1).
Overall Evaluation Value (n=291) Confusion Matrix Total
misclassifications 27.0 Predicted Class
Overall error rate 9.287% neg pos
SE of error rate 5.415
Average profit 0.814 neg 184
(94%) 12 (6%) SE of profit 0.108
AUC negative 0.9126 95%CI: 0.88,
0.94 Actual Class pos 15
(16%) 80 (84%) AUC positive 0.9132 95%CI: 0.87,
0.95
Figure 3.12: DENPRE_CYTO_291: Receiver operating characteristics (ROC) curve for dengue prediction calculated on 291 patients only including cytokine data (excl. IFN_ALPHA_1).