Table 3 Multi-level negative binomial regression models showing the association of patient-level characteristics with the total number of referrals in the CPCSSN database between 1 July 2007–30 June 2012 (n = 28 practices and n = 87258 patients)
Model 1Model 2Model 3Model 4Model 5
VariableIRR (95%CI)IRR (95%CI)IRR (95%CI)IRR (95%CI)IRR (95%CI)
Malea0.937 (0.919–0.954)0.962 (0.945–0.979)
Age in Yearsa1.014 (1.013–1.014)1.010 (1.009–1.010)
#Chronic Conditionsa1.323 (1.311–1.335)1.049 (1.038–1.061)
#Visitsa1.029 (1.028–1.030)1.022 (1.022–1.023)
#Months of Exposurea1.030 (1.023–1.030)1.024 (1.023–1.024)1.022 (1.022–1.023)1.012 (1.011–1.013)1.012 (1.011–1.013)
Intercept0.234 (0.226–0.242)0.240 (0.232–0.248)0.227 (0.219–0.235)0.199 (0.193–0.205)0.238 (0.231–0.246)
/ln alphaa−0.724 (−0.759–0.689)−0.918 (−0.957–0.879)−0.900 (−0.938–0.861)−0.800 (−0.835–0.765)−1.179 (−1.233–1.134)
Intercept for Practice-Level Variance0.076 (0.074–0.079)0.162 (0.157–0.167)0.068 (0.066–0.070)0.813 (0.789–0.837)0.114 (0.111–0.118)

Note. IRR = incidence rate ratio; CI = confidence interval.

/ln alpha is an added parameter to model over-dispersion of the outcome variable, and its significance indicates the need for negative binomial modelling.

Intercept for practice-level variance is a variance parameter for nesting of patients within practices.

  • a P < 0.001 for this parameter across all models.