Data source and study sample

The study followed a retrospective cohort design with staggered entry. Adults (aged ≥ 18 years) who visited a GP for any reason at one of the 39 MedicineInsight participating general practices in Western Australia were included in this study. As of the data extraction date of January 26, 2022, patients attending these clinics were categorised as follows: "active," defined as having had at least three encounters with the GP in the two years preceding data extraction; "inactive," defined as having had fewer than three such encounters during that period; or "deceased" (Fig. 1).

Fig. 1

Staggered entry sequence diagram

The de-identified electronic health records were extracted from the MedicineInsight database, an Australian national general practice data programme established by NPS MedicineWise, which included records from general practices that had consented to be part of the programme [21, 22]. The programme uses validated diagnostic algorithms to identify individuals with chronic diseases [23]. Approval for access to the data was given by the NPS MedicineWise Data Governance Committee (2020–003).

Information from MedicineInsight used in this study included demographics, diagnoses, reasons for consultations, laboratory and pathology investigation requests and their results, prescription data, patient screening, anthropometric measurements based on measured weight and height, smoking status and clinical measurements. Dates of diagnoses, tests, referrals, and medical treatments were also available. Socioeconomic disadvantage measure was based on the Socio-Economic Indexes for Areas – Index of Relative Socio-Economic Disadvantage (SEIFA-IRSD) [24], which is a residential postcode-based composite score that ranks geographic areas across Australia according to their relative socioeconomic advantage or disadvantage. All diagnoses were obtained from the “diagnosis”, “reason for encounter” and “reason for prescription” data fields using data extraction methods used by MedicineInsight, including standard clinical terminologies, misspellings, and abbreviations [21,22,23].

Ascertainment of diabetes mellitus (type 1, type 2, or unspecified type)

Diabetes mellitus (in this paper, referred to as “diabetes”) case identification was based on recorded diagnoses, prescription reasons, pathology results [25,26,27], and Medicare Benefits Schedule (MBS) item codes indicating presence of diabetes. MBS is an Australian government-funded list of medical services subsidised for Australian citizens, promoting accessible and affordable healthcare. To optimise the accuracy of diabetes detection, it was necessary to use two distinct records for the same individual to confirm the presence of diabetes [28]. To be defined as having diabetes, a patient needed to meet any of the following criteria:

1) Two separate diagnosis records indicating diabetes; 2) two separate HbA1c results ≥6.5% (48 mmol/mol); 3) two separate fasting plasma glucose tests ≥7.0 mmol/L; 4) two separate plasma glucose tests ≥11.1 mmol/L; 5) two separate recorded prescriptions of glucose lowering medications (Anatomical Therapeutic Chemical code: A10); and 6) two separate MBS item codes indicating management or diagnosis of diabetes (codes 66551, 66554, 66841, 73812, 73826, 73839, 73840, 81100, 81105, 81110, 81115, 81120, 81125, and historic codes 2517-2526, and 2620-2635).

The earliest recorded date of any of the above criteria was used as the diagnosis date.

Exclusion criteria

Without evidence of type 1, type 2, or unspecified type of diabetes, the following conditions were not counted as diabetes for the purposes of this study:

DefinitionsType of diabetes

Individuals identified as having diabetes were categorised as having type 2 if they had a recorded diagnosis indicating type 2 diabetes, non-insulin-dependent diabetes, or adult-onset diabetes. Individuals were recorded as having type 1 if they had a recorded diagnosis indicating type 1 diabetes or insulin-dependent diabetes. Those with a recorded diagnosis of diabetes with an unknown type (for example, “diabetes mellitus”) were classified as having unspecified diabetes. The majority rule was applied in cases where multiple types of diabetes were documented for a patient, determining the patient's classification based on the most frequently documented type. If different types were equally documented, the patient was classified as having "unspecified diabetes".

Study entry and exit

Patients entered the study on the initial date of their adult clinical encounter with the GP and exited either upon the patients’ death or upon their last clinical encounter in any of the 39 participating general practices. (Fig. 1).

Prevalence versus incidence

Cases diagnosed with diabetes over a period spanning 395 days from the first date of adult clinical encounter or before that were classified as prevalent (Fig. 1). To account for delays in patients’ electronic health recordings, “395 days” instead of the yearly “365 days” was selected. Similarly, patients diagnosed with diabetes based on abnormal HbA1c levels within 12 weeks after 395 days from the first patient-GP encounter were regarded as prevalent cases as HbA1c levels reflect average plasma glucose over the previous 8–12 weeks from the time of the test [29].

Cases diagnosed after 395 days from the first patient-GP encounter (or after 479 days for HbA1c criterion) were classified as incident cases. All prevalent cases were excluded from the incidence estimation. Women who had a history of gestational diabetes but did not show evidence of type 1, type 2, or unspecified diabetes were included among those at risk of developing diabetes mellitus as such women were at high risk of developing type 2 diabetes [30, 31].

Cases with unknown diagnosis date were classified as unknown prevalent or incident diabetes.

Patient-GP consultations

Multiple consultations occurring on the same day for the same patient were considered as single consultations.

BMI measurements

We used the BMI estimate recorded in the MedicineInsight database. If this estimate was not available, we computed BMI using the measured weight and height of the participants.

Health targets and management

The clinical management goals assessed in this study align with guidelines from the Royal Australian College of General Practitioners (RACGP) [27]. Individual management goals encompass smoking cessation and BMI, while treatment management goals include HbA1c, lipid levels, urine albumin creatinine ratio, vaccination, and blood pressure. Screening for potential diabetes-related conditions (ever recorded in patient health records) and the pharmacological approaches to managing blood pressure, dyslipidaemia, and diabetes were also investigated.

Statistical analysis

Diabetes incidence (all types) (measured as cases per 1,000 person-years of follow-up) and prevalence rates (measured as cases diagnosed over a period spanning 395 days from the first date of adult clinical encounter or before, divided by the number of people in the sample) were estimated with their 95% confidence intervals (CI). Characteristics of the overall cohort as well as individuals with and without diabetes were summarised using standard measures of central tendency and dispersion. Pearson’s χ2 test compared the frequencies in categorical variables, while a Mann–Whitney test compared the mean ranks of continuous variables. Prevalence and incidence of diabetes were each compared by the sexes stratified by age groups. Comorbidities and medical conditions were compared by sex stratified by diabetes type and duration.

Clinical and screening measures and health targets were compared by sex and type of diabetes. To be included in this analysis, patients with prevalent diabetes needed at least three years of follow-up while those with incident diabetes required three years of follow-up after their diabetes diagnosis.

Multivariable analysis: time to incident diabetes

The proportional hazards assumption was violated, rendering Cox regression unsuitable for analysis. Instead, time-to-incident diagnosis of diabetes (all types) was modelled using an accelerated time Weibull regression which provided the best fit with the lowest Akaike Information Criteria (AIC) compared to other parametric survival distributions. Study participants without evidence of prevalent diabetes were followed up from the first adult clinical encounter until they were diagnosed with incident diabetes or died or were right censored at the last clinical encounter. The analysis adjusted for age at first clinical encounter, sex, SEIFA-IRSD, smoking status and BMI (both at the first adult clinical encounter), and Indigenous ethnicity, while also accounting for clustering effects within the 39 participating general practices. Risk adjusted probability of remaining free of diabetes over time was plotted by sex.

Multivariable analysis: achievement of glycaemic control in the last year of clinical encounter among patients with incident diabetes

Sex differences in achieving glycaemic control [HbA1c ≤ 7.0% (≤ 53 mmol/mol)] (yes/no) over a period of up to 395 days, ending at the last clinical encounter, in patients with incident type 2 or unspecified diabetes who had at least three years of follow-up from diagnosis, were modelled using a multilevel mixed effects logistic regression. The multivariable model adjusted for 1) demographics (sex, age at last clinical encounter, SEIFA-IRSD, and Indigenous ethnicity); 2) BMI and smoking status as recorded over a period spanning 395 days up to the last clinical encounter); 3) years of follow-up; 4) active status of the patient; 5) baseline adult first recorded HbA1c level; and 5) clinical conditions that could have resulted in falsely high or falsely low HbA1c levels including anaemia, chronic kidney disease, chronic liver disease, hypertriglyceridaemia, and pregnancy [32,33,34,35]. The model also adjusted for cluster effects within the 39 participating general practices.

Multivariable analysis: sex differences in diabetes management in prevalent and incident cases

Sex disparities in the absence of tests or clinical assessments, lack of screening for diabetes-related conditions, and non-treatment with medications for diabetes-associated conditions over a period spanning 395 days up to the last clinical encounter were each analysed using multilevel mixed-effects logistic regression models. These models were adjusted for age at the last clinical encounter, sex, SEIFA-IRSD, smoking status, BMI, Indigenous ethnicity, rurality, duration of follow-up, and type of diabetes. Additionally, the models accounted for clustering effects within the 39 participating general practices.

The analyses were conducted separately for prevalent and incident cases, each requiring at least three years of follow-up from their initial adult clinical encounter or from their diabetes incident diagnosis, respectively.

Sensitivity analyses

In sensitivity analyses, sex differences in achieving glycaemic control were separately modelled after excluding pregnant women (over the period of 395 days up to the last clinical encounter) and/or after limiting the analyses to type 2 diabetes.

All analyses were performed using Stata/MP 17.0 (StataCorp, College Station, TX, USA).