We used data from the KORA (Cooperative health research in the region of Augsburg) myocardial infarction registry from the region of Augsburg, Germany (approx. 600,000 inhabitants from the city of Augsburg and the two adjacent districts Augsburg and Aichach-Friedberg) to identify people with first MI. Case finding, diagnostic classification of events and data quality control methods are described elsewhere [12,13,14]. The population-based Augsburg MI registry was implemented in 1984 as part of the WHO MONICA project [12, 13, 15]. It was part of the KORA platform from 1996 until 2019 and continues as Augsburg Myocardial Infarction Register from 2020. It continuously registered all cases of first MI (non-fatal MI and fatal MI) in the study population [13].

Data from four population-based surveys and two follow-up surveys conducted in the study region of the KORA MI registry are available. The first three surveys were part of the MONICA project (S1:1984/1985, S2:1989/1990, S3:1994/1995). The fourth (S4:1999–2001) and the follow-up surveys of S3 (F3: 2004/2005) and S4 (F4: 2006–2008) were conducted as KORA studies. These surveys were used to define the background population with and without diabetes. The baseline survey population of S1 was between 25 and 64 years old and between 25 and 74 years old in S2, S3 and S4. The survey was carried out using cluster sampling procedures with a response rate between 79% in S1 [16] and 68% in S4 [17]. In the F3 survey (survey population 35–84 years old) 76% of all eligible S3-participants responded [18] while in F4 (survey population 32–81 years old) 80% of all eligible S4 participants made their response [19].

Moreover, data from the Central Research Institute of Ambulatory Health Care (Zi-data) in Germany, available since 2009, were used to estimate the population with diabetes between 2009 and 2016 [20]. The Zi-data comprises all people with statutory health insurance having at least one medical contact in a calendar year. Over 80% of the total study region population were covered.

The present study excluded all people aged between 25 and 44 since no valid diabetes prevalence estimates were available for that age group for the study period. Thus, the remaining study population (45 to 74 years) ranged from 179,689 people in 1985 to 254,773 in 2016.

MI definition methods are described in detail elsewhere [11, 21]. Briefly, all cases of non-fatal MI and fatal MI between January 1, 1985 and December 31, 2016 were recorded according to the WHO MONICA protocol [12, 14].

For this study the clinical diagnosis of non-fatal MI was categorized and validated according to MONICA criteria [12], which included acute clinical symptoms (acute chest pain lasting 20 min or longer, not relieved by rest or nitrates), ECG diagnostic criteria (Q waves, non-Q waves in up to four electrocardiograms) as well as a subsequent increase in the serum activity of at least one of three enzymes (creatinine phosphokinase, aspartate aminotransferase, and lactate dehydrogenase) of more than twice the normal upper limit. Since January 1, 2001, non-fatal MI has been diagnosed according to the European Society of Cardiology and American College of Cardiology criteria [22]. An MI case was defined as non-fatal if a patient reached the hospital and survived for at least 24 h. Patients were additionally interviewed during hospitalization using a standardized questionnaire including factors such as previous myocardial infarction. Further data were gathered in a concluding chart review. We only analysed first cases of non-fatal MI in the current study.

Fatal MI was defined as cases of coronary death or early fatal MI occurring before or within 24 h of hospitalization if no previous non-fatal MI was recorded. Coronary deaths were identified via regional health offices by checking all death certificate diagnoses giving suspected coronary heart disease as main cause of death. Additionally, a written questionnaire enquiring about disease history (in particular previous myocardial infarction), prior medication, risk factors, and circumstances of death was routinely sent to the deceased person’s last treating physician and the coroner (mean response 85%).

First MI was defined as the sum of first non-fatal MI and fatal MI.

All patients with first MI were assessed by age, sex, date of first MI, and diabetes status.

Diabetes in people with first non-fatal MI was defined by a physician’s diagnosis assessed from chart review (99.7% of the cases) or by self-reporting in a personal interview (0.3%). Diabetes status in case of fatal MI was obtained from questionnaires sent to the last treating physicians or coroner. All people with unclear diabetes status were excluded from the analysis (n = 300).

Diabetes among the population at risk: KORA surveys classified a person as having diabetes based on self-reported physician’s diabetes diagnosis or reported use of glucose-lowering medication. The Zi-dataset classified someone as having diabetes if an outpatient diagnosis was made in at least two quarters of a calendar year.

All main analyses were performed for the entire population, stratified by sex, and for first MI as well as separately for, first non-fatal MI and fatal MI. For diabetes population definition, we estimated the age- and sex-specific prevalence of diabetes for each calendar year using the age classes 45–54, 55–64 and 65–74 years based on combined KORA surveys and Zi-data. Age-sex specific diabetes prevalence in the KORA survey in 2013/2014 was as follows: men: 2.0% (45–54 years), 8.7% (55–64 years), 19.8% (65–74 years); women: 2.9% (45–54 years), 6.4% (55–64 years), 11.9% (65–74 years). The mean of the age-sex specific diabetes prevalence in the Zi-data in 2013 and 2014 was as follows: men: 7.1% (45–54 years), 16.5% (55–64 years), 27.4% (65–74 years); women: 4.1% (45–54 years), 11.0% (55–64 years), 19.8% (65–74 years).

A correction factor was calculated for each age and sex stratum by dividing the mean prevalence from the Zi-data (years 2013 and 2014) by the 2013/2014 KORA prevalence. We then multiplied all age- and sex-specific KORA prevalence estimates by this correction factor. The corrected KORA prevalence estimates for the years of the surveys (1984/85, 1989/90, 1994/95, 1999–2001, 2004/2005, 2006–2008) were used to calculate diabetes prevalence between 1985 and 2008, assuming a linear time trend between the survey years. Age- and sex-specific diabetes prevalence was estimated directly from the Zi-data for the time period 2009 to 2016.

The population with diabetes was estimated for each stratum (age class, sex, and calendar year of first MI) by multiplying the population of the study areas by the age- and sex-specific prevalence of diabetes as described above. The population without diabetes was estimated by subtracting this population from the total population in each stratum.

In the sensitivity analysis using only the KORA prevalence, a linear trend for the age- and sex-specific diabetes prevalence was assumed between the surveys. Moreover, it was assumed that this prevalence remained constant in the years after the last survey (i.e., 2015 and 2016).

Calendar time periods were aggregated as follows to account for random fluctuations in the individual years of first MI: 1985–1988, 1989–1992, 1993–1996, 1997–2000, 2001–2004, 2005–2008, 2009–2012 and 2013–2016.

Stratum-specific and age- and sex-standardized MI IRs were calculated for each time period and for both populations with and without diabetes. We used the German population from the year 2000 as standard population.

Incidence rate ratios (population with vs. without diabetes) were estimated from the standardized IRs.

To test for time trends, we performed Poisson regression models using calendar period of MI IR, age (classes as described above) and sex as independent variables. We used the calendar period 1985–1988 as a reference to estimate the effect of calendar time. Furthermore, analogous Poisson models were fitted including a variable presence of diabetes (yes vs. no) stratified by calendar period of first MI. The change of the RR between time periods using 1985–1988 as baseline period was statistically tested including a variable presence of diabetes (yes vs. no) and an interaction term “diabetes*calendar period of first MI”.

The main analyses were repeated in a sensitivity analysis, whereby only KORA diabetes prevalence was used (detailed information see appendix). Another sensitivity analysis estimated linear time trend for the years 2009–2016 using only the Zi-data for each calendar year.

All analyses were performed using de-scale adjustment to account for over-dispersion of the dependent variable [23]. We performed all analyses using the statistical analysis systems SAS (SAS for Windows 10, Release 9.4 TS1M5, SAS Institute, Cary, NC, USA).

The KORA MI Registry and KORA survey data collection has been approved by the ethics committee of the Bavarian Medical Association (Bayerische Landesärztekammer, Germany). All survey participants and MI patients who actively participated in the registry gave written consent. The study was performed according to the principles of good epidemiological practice [24].