Skip to main content

Table 3 Multi-level logistic regression models predicting the odds of clients being found at risk in unstructured assessments

From: More than three times as many Indigenous Australian clients at risk from drinking could be supported if clinicians used AUDIT-C instead of unstructured assessments

 

Fixed effects

  

Predictors

OR [95% CI]

lnOR

\(SE\)

\(p\)

\(ICC\)

Likelihood Ratio Test

Model 1

–

–

–

–

45.56%

–

Intercept

0.01 [0.00, 0.02]

− 4.91

0.41

 < 0.001

–

 

AUDIT-C

1.87 [1.77, 1.98]

0.63

0.03

 < 0.001

–

 

Model 2

–

–

–

–

45.59%

\({\chi }^{2}\)(1) = 0.01, \(p\) = 0.90

Intercept

0.01 [0.00, 0.02]

− 4.90

0.42

 < 0.001

–

 

AUDIT-C

1.87 [1.77, 1.98]

0.63

0.03

 < 0.001

–

 

Same occasion

0.98 [0.72, 1.33]

− 0.02

0.15

0.90

–

 

Model 3

–

–

–

–

45.56%

\({\chi }^{2}\)(3) = 9.14, \(p\) = 0.028

Intercept

0.01 [0.00, 0.03]

− 5.06

0.85

 < 0.001

–

 

AUDIT-C

1.88 [1.78, 1.99]

0.63

0.03

 < 0.001

–

 

Age (decade)a

1.11 [1.03, 1.19]

0.10

0.04

0.005

–

 

Remoteness

1.10 [0.47, 2.55]

0.10

0.43

0.83

–

 

Male

0.87 [0.71, 1.06]

− 0.14

0.10

0.17

–

 

Same occasion

1.08 [0.79, 1.47]

0.08

0.16

0.63

–

 

Model 4

–

–

–

–

45.08%

\({\chi }^{2}\)(1) = 8.86, \(p\) = 0.003

Intercept

0.01 [0.00, 0.03]

− 5.06

0.84

 < 0.001

–

 

AUDIT-C

1.87 [1.77, 1.98]

0.63

0.03

 < 0.001

–

 

Age (decade)a

0.88 [0.74, 1.04]

− 0.13

0.09

0.14

–

 

Remoteness

1.11 [0.48, 2.56]

0.10

0.43

0.81

–

 

Male

0.86 [0.70, 1.06]

− 0.15

0.10

0.16

–

 

Same occasion

1.07 [0.79, 1.46]

0.07

0.16

0.65

–

 

AUDIT-C * Age (decade)a

1.04 [1.01, 1.07]

0.04

0.01

0.003

–

 
  1. OR odds ratio, lnOR natural logarithm of the odds ratio (logit), SE standard error (of lnOR). ICC Intraclass-correlation coefficient—the percentage of variance attributable to the random effects
  2. aClient age (a continuous variable) was divided by ten to represent decades. The age (decade) of each client was centered such that 0 represents 40 years. Likelihood ratio tests indicate whether a model significantly improves upon the fit of a simpler model. Model 2 did not significantly improve upon the fit of Model 1. Model 3 significantly improved upon the fit of Model 2. Model 4 significantly improved upon the fit of Model 3. This table presents the results of a multi-level regression. These models include both fixed (indicated by the column span) and random effects which enables clustering within the data to be modelled. The random effects include intercepts for each client (n = 4225) and service (k = 18). The thousands of random effect coefficients are not tabulated, but the percentage of variance explained by the random effects are described by the ICC statistics
Back to article page

Addiction Science & Clinical Practice

ISSN: 1940-0640