Table 3.
Fitted GAM models to examine predictors linearity and interaction.
Model | Coefficient (SE) | p -value | EDF | Adjusted R2 | Deviance | p χ2 | AIC |
---|---|---|---|---|---|---|---|
Single term model | |||||||
HT = β1CMB + β2NIHSS + β3 Size | β1= 0.24 (0.03) | < 0.001 | 0.33 | Reference | 239.4 | ||
β2= 0.12 (0.03) | < 0.001 | ||||||
β3= 0.48 (0.09) | < 0.001 | ||||||
Interaction term model | |||||||
HT = CMB + NIHSS + Size + NIHSS*Size + CMB*NIHSS + CMB*Size + CMB*NIHSS*Size | β1= 0.98 (0.19) | < 0.001 | 0.40 | 29.6± | < 0.001± | 217.9 | |
β2= 0.47 (0.10) | < 0.001 | ||||||
β3= 1.84 (0.40) | < 0.001 | ||||||
β4= −0.04 (0.01) | < 0.001 | ||||||
β5= −0.17 (0.06) | < 0.01 | ||||||
β6= −0.06 (0.02) | < 0.01 | ||||||
β7 = 0.007 (0.003) | < 0.05 | ||||||
Spline fitted model | |||||||
HT = β1CMB + β2NIHSS + β3Size+ | β1 = 0.47 (0.2) | < 0.05 | 1.6 | 0.50 | 56.8± | < 0.001± | 201.5 |
s(CMB) + s(NIHSS) + s(Size) | β2 = −0.52 (0.14) | < 0.001 | 6.8 | 27.3§ | < 0.001§ | ||
β3 = 1.14 (0.65) | 0.08 | 2.0 | |||||
s(CMB) | < 0.05 | ||||||
s(NIHSS) | < 0.001 | ||||||
s(Size) | 0.09 | ||||||
Tensor product model | |||||||
HT= s(CMB)+ s(NIHSS)+ s(Size)+ | s(CMB) | < 0.001 | 2.6 | 0.52 | 66.6± | < 0.001± | 197.9 |
ti(CMB, NIHSS)+ ti(CMB, Size)+ | s(NIHSS) | < 0.001 | 1.0 | 37.0§ | < 0.001§ | ||
ti(NIHSS, Size) | s(Size) | <0.001 | 1.0 | 9.7¥ | <0.05¥ | ||
ti(CMB,NIHSS) | <0.01 | 4.6 | |||||
ti(CMB,Size) | 0.19 | 1.0 | |||||
ti(NIHSS,Size) | < 0.001 | 3.1 |
Compared to single-term model.
Compared to interaction term model.
Compared to tensor product model.
Size in the model refers to infarction size. Effective degrees of freedom (EDF) > 1 indicates non-linearity. In the final model, the smooth CMB, the product tensor of CMB and NIHSS, and the product tensor of NIHSS and infarction size were significantly no-nlinear. Intercept was not included in the table.