Study covariates

The 45 and Up Study questionnaire data included self-report information on key potential confounders and mediating factors including: baseline age; sex; marital status; born in Australia; Indigenous status; current housing; household income; education level; smoking history, intensity and age stopping; alcohol use; physical activity³³; time spent sitting; body mass index; psychological distress³⁴; level of limitation reported; social support³⁵; socioeconomic status³⁶; and Accessibility/Remoteness Index of Australia.³⁷ Covariate categories are provided in online supplementary file 1.

bmjopen-2018-027158supp001.pdf

Comorbidity was ascertained using the Multipurpose Australian Comorbidity Scoring System,³⁸ defined as the sum of comorbidities excluding diabetes at 1 and 5 years prior to study entry. We also used the PBS data to calculate the Rx-Risk index (ie, number of condition groups excluding diabetes for which medicines were dispensed) at 1 and 5 years prior to entry to the study period.³⁹

Health service use during and prior to the study period was captured in several ways. The number of GP and specialist physician visits 4 years before and during the study period was ascertained separately using MBS data. The frequency of contact was captured as: (1) a count of the number of days each person had a GP or specialist physician contact and (2) the SD of the annual count of days with a contact during the ascertainment period. Variables also captured if participants had a PBS records for any diabetes-related medicines, hospitalisations for dialysis, MBS diabetes cycle of care claim⁴⁰ and other chronic disease related MBS-funded primary care services or glycosylated haemoglobin (HbA1c) testing during or in the 4 years prior to the study period. Finally, the number of unplanned diabetes-related hospitalisations and ED presentations in the 3 years prior to study entry was also included as a potential confounding factor.

To reduce ascertainment bias, the method (ie, self-report only, APDC only, PBS only or combinations of the three) and first year the individual who identified as having diabetes was entered into the model. In addition, a binary variable was used to determine if the participant died during the 6 years of the study period. Person-time at risk of the outcome event was included in the count models.