The multivariate analysis included all parameters that were significantly (defined as a
P value of <0.10) associated with the dependent variable of ASL in the univariate analysis. In a first step, we dropped the parameters of anterior chamber depth and lens thickness from the list of independent variables, since their sum was the ASL. For reasons of collinearity, we then deleted the parameters of body weight (variance inflation factor, 107). We dropped body height (
P = 0.69), central corneal thickness (
P = 0.63), and prevalence of arterial hypertension (
P = 0.13), since these parameters were no longer significantly associated with ASL. We arrived at a model (overall correlation coefficient
r, 0.26), in which longer ASL was associated with male sex (
P < 0.001), longer axial length (
P < 0.001), higher degree of nuclear cataract (
P = 0.001), higher body mass index (
P = 0.02), and older age (
P = 0.0,
Table 3). For each year increase in age, ASL increased by 0.002 mm. If refractive error was added to the model, it was not significantly (
P = 0.10; β, 0.04) associated with ASL. If body mass index was replaced by body height in the model, body height was not significantly (
P = 0.58) associated with ASL. If body mass index was dropped, the association between longer ASL and older age increased only slightly (
P = 0.01; standardized correlation coefficient β, 0.05; correlation coefficient B, 0.003). If the study population was stratified by sex, men showed a significant association between longer ASL and older age (
P = 0.02; β, 0.07; B, 0.004), while the association was not significant (
P = 0.48) in women. If sex was dropped from the multivariate analysis and body length was added, longer ASL was associated with taller body height (
P < 0.001, β, 0.11; B, 0.005; 95% CI, 0.003, 0.008) after adjusting for longer axial length (
P < 0.001), higher degree of nuclear cataract (
P = 0.001), higher body mass index (
P = 0.01), and older age (
P = 0.005).