Definition of outcome measure and risk factors
The mortality status based on master data of health claims. In analysis, we set the time of death to the middle of the month of death.
We included BMS coded by OPS as ‘8–837.k’ and DES coded as ‘8–837.m’. For CABG we considered OPS-codes ‘5–361*’ and ‘5–362*’. We categorized the variable of intervention type into “BMS”, “DES”, “CABG”, “mixed”, and “none” (no procedure). The categories BMS, DES and CABG implied initial interventions, whereas patients in the mixed-group underwent at least two different revascularization techniques in one quarter or received different intervention types as a re-intervention. Patients remained in ‘none’ as long as they did not receive BMS, DES, or CABG. Intergroup changes from none to BMS, DES, CABG or mixed and from BMS, DES or CABG to mixed were possible.
Control variables and validation strategy
Analysis controlled for sex, and age at first valid CAD diagnosis or at first coronary intervention (50–54, 55–59, 60–64, 65–69, 70–74, 75–79, 80–84, 85+). To identify modifying effects of other cardiac diseases on mortality, we included the following ICD-10-diagnoses ever received between 2004 and 2015: acute coronary syndrome (I21 and I20.0), congestive heart failure (I09.9, I11.0, I13.0, I13.2, I25.5, I42.0, I42.5-I42.9, I43, I50), cardiogenic shock (R57.0), arrhythmia (I44.1-I44.3, I45.6, I45.9, I46, I47, I48, I49, R00.0, R00.1, R00.8, T82.1, Z45.0, Z95.0), multivessel coronary artery disease (I25.13).
We also considered non-cardiovascular diseases with high impact on mortality. We measured the risk factor “multimorbidity” by an additive score as the number of the following acute and chronic diseases ever diagnosed between 2004 and 2015: atherosclerosis (I70), Alzheimer’s disease (F00), diabetes mellitus (I10–I14), cancer (C00–C97), liver diseases (B18, K70–K72, K76, Z94.4), lung diseases (J44, J47), neurological diseases (G00–G09, G23, G24–G26, G31.8, G32), kidney diseases (N17–N19, Z94.2, T82.4, Z99.2), Parkinson’s disease (G20–G22), cerebrovascular diseases (I60–I69, G45, G46, H34), paralysis (G80–G83, G04.1, G11.4) and peptic ulcer (K25–K28). The multimorbidity score consisted three categories: 0–1, 2–6, or 7+ of the selected diseases.
In addition, we considered external injuries (S00–S03, T00, T01.3, T02.4, T02.6–T02.9, T05.8, T05.9, T14–T98) as non-degenerative risk factor of death. All diseases were defined as irreversible since the first observed diagnosis.
Another control variable concerned the duration since coronary intervention measured in years. We compared CAD patients in 1st to 2nd year since intervention with a third (residual) group that covered both, CAD patients in the 3rd year and CAD patients who never received BMS, DES, or CABG.
We reduced the problem of false-positive CAD diagnoses by applying a validation strategy: to define the first CAD diagnosis as valid, the patient required the co-occurrence of CAD diagnosis in another quarter over the whole observation period [
9]. We considered all covariates, with the exception of sex and age at incident CAD diagnosis, to be time-varying variables with value 1 since first valid diagnosis and 0 otherwise.
Statistical analysis
We estimated sex and age-standardized 3-year mortality rates from 2005 to 2015 by using the German average population of 2005 and computed the mortality rate for each intervention type by dividing the number of deaths (
\(D_{2005 - 15,x,a}\)) by the population under risk (Eq.
1). The Person-years under risk (
\(PY_{Risk}\)) from 2005 to 2015 defined the population of risk in the denominator.
$$Mortality_{2005 - 15,x,a} = \frac{{D_{2005 - 15,x,a} }}{{PY_{Risk2005 - 15,x,a} }}*100$$
(1)
To compare 3-year survival after coronary intervention, we applied Kaplan–Meier estimators. We also computed Cox proportional-hazard models, which aimed to examine disparities in the 3-year mortality among CAD patients with adjustment for demographic characteristics, cardiovascular and non-cardiac diseases and the years passed since intervention. In sensitivity analysis, we calculated an additional model excluding CAD patients who never underwent coronary intervention in the observation period, as well as the observation time prior to intervention. All analyses were conducted with the use of Stata (version 16.1).
We measured analysis time as the years since first valid CAD diagnosis as of Q1 2005 at the earliest (Additional file
1: Fig. S1, persons 1–5). Following the principles of Hernán et al. (2016) and Emilson et al. (2018), we emulated a target trial [
10,
11]. We maximally tripled all eligible observations (measured in person-times) and assigned each copy to the corresponding coronary intervention group (none, BMS, DES, CABG or mixed). In case of a group change we defined a time zero of analysis time (Additional file
1: Fig. S1, persons 1 & 2). Coronary intervention simultaneous with the first valid CAD diagnosis (study entry) was also possible (Additional file
1: Fig. S1, persons 3 & 5). To face the immortal time bias, we censored CAD patients after a follow-up of 3 years spend in the category of BMS, DES or CABG (Additional file
1: Fig. S1, persons 2, 3 & 5). Person-times spend in the categories of none and mixed had no observational time limit (Additional file
1: Fig. S1, person 4).