*3.3. Further Evidence of the Relationship between Land-Use Change and Environmental Health Risks*

Transformations in population, land use, and industry are essential features of urbanization. Thus, UR, CLA, and PHI were considered as proxy variables and were included as independent variables in the GAM model, to further verify the impact of urban expansion or urban land-use change on health risks. The results are shown in Table 2. UR and PHI were found to have a linear relationship with total mortality (*p* < 0.01), and CLA presented a nonlinear relationship with total mortality. The explained deviances and the

determination coefficients were low (Dev < 55%; R2 < 0.50). However, UR, CLA, and PHI showed significant nonlinear relationships with the number of cancer cases (*p* < 0.001), and the models had good explanatory power and goodness of fit (Dev > 95%; R<sup>2</sup> < 0.70). In addition, UR showed a linear relationship with the mortality of cancer, and CLA presented a nonlinear relationship with the mortality of cancer (*p* < 0.01). The models had good explanatory power and goodness of fit. PHI also showed a linear relationship with the mortality of cancer, but the model was non-significant and had poor explanatory power and goodness of fit (*<sup>p</sup>* = 0.379, adjusted R2 <sup>=</sup> −0.01).


**Table 2.** Results of the GAM model and its comparison with the SLR.

Note: eDF donates the estimated degree of freedom, Dev donates the explained deviance, and β is the regression coefficient of SLR.

The specific effects of these urbanization factors on health are shown in Figure 7. The total mortality grew linearly with increases in UR and PHI (Figure 7(a1,a3)), and generally showed nonlinear growth along with an increase in CLA (Figure 7(a2)). The number of cancer cases nonlinearly increased along with the three factors, but the nonlinear relationships were more significant between the number of cancer cases and UR and PHI (Figure 7(b1–b3)). The mortality of cancer exhibited linear growth and nonlinear growth with increased UR and CLA, respectively (Figure 7(b1,b2)). Figure 7(b3) shows that the mortality of cancer linearly decreased with an increase in PHI, but the explained deviance (15.7%) and adjusted R<sup>2</sup> (−0.01) proved that the relationship was non-significant and unreliable. In addition, the simple linear regression (SLR) was calculated for comparison with the GAM. The estimated coefficient generally confirmed the relationships identified by the GAM, but the SLR showed disadvantages when the relationship between dependent and independent variables was complicated or nonlinear, because they yielded a lower R<sup>2</sup> than that of the GAM (Table 2).

**Figure 7.** Effect of urbanization on health: (**a1**–**a3**), total mortality; (**b1**–**b3**), number of cancer cases; (**c1**–**c3**), mortality of cancer. Note: the vertical axis donates the fitted function value, the horizontal axis is the observations of the independent variable; solid line represents the fitted line (or curve) of the dependent variable, and the dotted line represents the confidence intervals.
