Skip to main content
. 2004 Sep;88(9):1125–1130. doi: 10.1136/bjo.2003.032383

Table 8.

 Model of logistic regression

A Multinomial regression
Self reported level of care Explanatory variables Exp(B) Wald Significance 95% CI
Level 2 Age 1.008 0.05 0.82 0.942 to 1.078
PF 0.982 1.85 0.17 0.956 to 1.008
P 1.013 1.05 0.31 0.988 to 1.039
MH 0.984 0.82 0.37 0.951 to 1.019
EV 1.012 0.28 0.60 0.968 to 1.057
GHP 0.985 0.80 0.37 0.952 to 1.018
DLTV subscale 1 0.962 4.08 0.044 0.927 to 0.999
DLTV subscale 2 1.002 0.01 0.91 0.967 to 1.039
DLTV subscale 3 0.991 0.16 0.69 0.950 to 1.034
DLTV subscale 4 1.020 3.42 0.064 0.999 to 1.041
DVA better eye 1.63 0.31 0.58 0.29 to 9.09
Level 1 Age 0.890 7.96 0.005 0.821 to 0.965
PF 0.940 12.58 0.000 0.909 to 0.973
P 0.999 0.01 0.941 0.970 to 1.028
MH 0.960 3.37 0.066 0.919 to 1.003
EV 1.062 4.13 0.04 1.002 to 1.125
GHP 0.992 0.14 0.703 0.973 to 1.025
DLTV subscale 1 0.985 8.40 0.004 0.929 to 1.044
DLTV subscale 2 0.962 2.66 0.10 0.918 to 1.008
DLTV subscale 3 1.009 0.09 0.76 0.954 to 1.067
DLTV subscale 4 0.998 0.01 0.91 0.973 to 1.025
DVA better eye 6.65 2.80 0.09 0.72 to 61.1
B Binomial logistic regression (base category level 3)
Level 2 DLTV subscale 1 1.039 17.09 0.000 1.020 to 1.058
SF 36 PF 1.017 3.89 0.049 1.000 to 1.033
C Binomial logistic regression (base category level 3)
The multinomial regression model (A) was run with the three levels of care as the dependent variable and age, sex, SF36 dimensions, and DLTV subscales as explanatory variables. The model shows that the decrease in scores in DLTV subscale 1 was most strongly influenced by the transition from care level 3 to 2. The decrease in PF scores and age most strongly influenced the transition from care level 2 to level 1 with a contribution from DLTV subscale 2. To progress to a more parsimonious model variables that were counterintuitive were first removed followed by those explanatory variables whose p values exceeded 0.1. (SPSS default for stepwise backwards and also equivalent to a 5% test in one direction only.)
The binomial regression model (B) confirmed that DLTV subscale 1 was most important in the transition from care level 3 to 2 while DLTV subscale 2 was most important in the transition to care level 1 (C).
Level 1 DLTV subscale 2 1.041 9.54 0.002 1.015 to 1.068
SF 36 PF 1.031 8.37 0.004 1.010 to 1.053
Min DVA 0.24 3.15 0.076 0.05 to 1.16