Methods
Data sources
The unique Swedish personal identity number was used to link individual-based data from six national registers. Filled prescriptions were collected from the Swedish Prescribed Drug Register (SPDR), which contains information about all prescriptions filled at Swedish pharmacies since July 2005.20 Swedish prescriptions include information about the medication, patient, prescriber and the prescribed daily dosage. Clinical and health-related data about risk factors and complications of diabetes, as well as healthcare center characteristics, were collected from the Swedish National Diabetes Register (NDR), which contains patient information reported by physicians and nurses at hospitals and primary care clinics nationwide.21 Healthcare providers were identified at the center level, not by individual practitioners. Data concerning CV events were collected from the Swedish National Patient Register, which contains information about inpatient and outpatient care in Sweden.22 The date and cause of death were collected from the Cause of Death Register.23 Information about primary tumors was collected from the Swedish Cancer Registry.24 Individual data about socioeconomic status were collected from the Longitudinal Integration Database for Health Insurance and Labour Market Studies (LISA).25
Study population and period
Patients age 18 years or older with a clinical T2DM diagnosis26 who had filled at least one prescription for lipid-lowering medications between July 1, 2006 and December 31, 2012 were eligible for inclusion (figure 1). Prescriptions for bile acid sequestrants were excluded since the indication is typically not hyperlipidemia.27 The date of the first filled prescription was defined as the index date. To establish a new-user design, patients who had filled prescriptions for lipid-lowering medications within 1 year prior to the index date were regarded as prevalent users and excluded. Patients who filled prescriptions for a lipid-lowering combination therapy or lipid-lowering extemporaneous preparation that lacked information about package size were also excluded. Combination therapy has been described elsewhere.6 28 Patients who experienced CVD prior to or on the index date were classified as prescribed lipid-lowering secondary prevention; all other patients were defined as primary prevention.
Figure 1Inclusion and exclusion of the study population. CVD, cardiovascular disease; NDR, National Diabetes Register.
Patients were followed from cessation of the first filled supply (baseline date) until the first filled prescription for multi-dose dispensed medications, migration, CV event, death or December 31, 2016. The study period was broken down into subsequent intervals of 122 days until December 31, 2014, followed by annual intervals until December 31, 2016.
The healthcare center at baseline was assigned by selecting the entry in the NDR closest to the baseline date and remained the same throughout the study period unless the NDR indicated otherwise. A comparison between the assigned healthcare center from the NDR with the healthcare provider of the first filled lipid-lowering prescription in the SPDR showed agreement of 98% for county council and 87% for type of care.
Refill adherence
Patients’ (refill) adherence to lipid-lowering medications was assessed with data from the SPDR. We assumed that patients initiated medication use on the index date. We measured refill adherence with the medication possession ratio (MPR), representing the proportion of days with medications available. The supply of each prescription was calculated by dividing the number of tablets filled by the daily dosage stated as a free-text variable by the prescriber. Interpretation of the free-text variable to obtain the daily dosage has been described elsewhere.6 28 Overlapping supplies for the same substance and strength were added to the latter supply. Overlapping supplies for different substances or strengths were deleted. MPR was assessed for each subsequent interval and categorized as high or low based on the common cut-off of 80%.29 30
Guideline adherence
Based on data from the NDR, we assessed healthcare center adherence to national lipid-lowering prescription guidelines. Guideline adherence was defined as the prescription prevalence of lipid-lowering medications among patients with T2DM and LDL cholesterol above the recommended target levels existing at the time for the study (≥2.5 mmol/L for primary prevention31 and ≥1.8 mmol/L for secondary prevention1 10). Between 2007 and 2014, guideline adherence was assessed for each healthcare center and year, for primary and secondary prevention patients separately.
Guideline adherence was linked to patient intervals based on the year in which the interval started. For intervals starting in 2006, guideline adherence for 2007 was used. For healthcare centers that lacked information about adherence, we imputed the preceding year’s figure or the mean annual adherence for the county council and type of care. Guideline adherence was categorized as high or low based on a cut-off that represented the median for primary (48%) and secondary prevention (78%).
Outcomes
The outcomes of interest were CV events and mortality. A CV event was defined as a diagnosis of unstable angina pectoris, myocardial infarction (including percutaneous coronary intervention, coronary artery bypass grafting), stroke or ischemic heart disease. All-cause mortality was defined as death from any cause. CV mortality was defined as death from a cause of CVD or a CV event entered in the National Patient Register within 28 days prior to death. Starting from the second interval, the risk of CV events and mortality was analyzed for each interval until migration, CV event, death or December 31, 2016.
Covariates
Covariates regarded as potential confounders included sex, age, socioeconomic status (country of birth, marital status, education level, employment status, profession, income), concurrent medications (filled prescriptions for diabetes medications, anticoagulants and antihypertensives), and clinical and health-related characteristics (diabetes duration, hemoglobin A1c [HbA1c], estimated glomerular filtration rate [eGFR], blood pressure, cholesterol values, microalbuminuria, macroalbuminuria, kidney disease, cancer, body mass index [BMI], physical activity and smoking). These covariates have previously been shown to be important factors when analyzing adherence, as well as the risk of CV events and mortality.6 28
Sex, age and socioeconomic characteristics were collected from the LISA database. Sex and country of birth were regarded as constant, and age was based on the year of birth. Information about marital status, education level, employment status, profession and income was collected prior to or during the baseline year. Income was regarded as a continuous variable. The remaining socioeconomic variables included the following categories: country of birth: Sweden, other Nordic country, other European Union (EU)-15 country or the Soviet Union, rest of Europe, the Americas, Asia or Oceania, or unknown; marital status: unmarried, married or registered partner, divorced, or widow/widower; education level: compulsory school or lower, upper secondary school, or postsecondary; employment status: unemployed, employed or retired; profession: upper white-collar, lower white-collar, blue-collar, or other.
At baseline, cancer and kidney disease were defined as diagnosis within 5 years prior to the baseline date and were collected from the Swedish Cancer Registry and the National Patient Register, respectively. Cancer diagnosis included primary tumors, while kidney disease included acute or chronic kidney failure, as well as glomerular or renal complication due to T2DM.
Filled prescriptions for diabetes medications (anatomical therapeutic chemical (ATC): A10), anticoagulants (ATC: B01), and antihypertensives (ATC: C02, C03, C04, C05, C07, C08, and C09) were collected from the SPDR. Filled prescriptions within 12 months prior to the baseline date were considered concurrent use.
The remaining clinical and health-related characteristics were collected from the NDR. At baseline, data were collected between 24 months prior to and 14 days after the baseline date by selecting the value closest to baseline. Diabetes duration was based on the year of birth and diabetes diagnosis. Microalbuminuria and macroalbuminuria were dichotomously categorized. BMI and eGFR were categorized according to recommended references values.32 33 HbA1c, blood pressure and cholesterol levels were categorized as high or low according to recommended target values at the time of the study.31 Physical activity was defined as a 30 min walk or equivalent, categorized as less than once a week, 1–2 times a week, 3–5 times a week or daily. Smoking was dichotomized and defined as at least one cigarette/pipe daily or having quit within the past 3 months.
A total of 23% of patient characteristics at baseline were missing: 4% of socioeconomic status and 43% of clinical and health-related characteristics. No information was missing about age, sex or concurrent medications. Missing information at baseline was replaced with multiple imputations. Potential confounders (except for cancer) were assessed for each interval during the study period by assuming the imputed baseline value until the information had been updated in the registers.
Sensitivity analyses
To evaluate the cut-offs used to categorize refill and guideline adherence as high or low, new ones were set. For refill adherence, cut-offs of 60%, 70%, and 90% were applied to evaluate the 80% level that had been used to define patients as low-adherent or high-adherent. For guideline adherence, new cut-offs were set at 30% and 60% (from 48%) for primary prevention and 50% and 90% (from 78%) for secondary prevention. The results of the statistical analyses were compared between the cut-offs.
To estimate the impact of multiple imputations, the risk of CVD and mortality was assessed and compared between complete cases and imputed data.
Statistical analyses
The association between refill (MPR) and guideline adherence was analyzed by means of general linear regression. Multivariate imputations by chained equations (MICE) were used to replace missing information among baseline variables; 10 imputed data sets with 10 iterations each were generated. The risk of CV events and mortality was analyzed for each interval based on the 10 imputed data sets with Cox proportional hazard regression and Kaplan-Meier, adjusted for potential confounders. Covariates and guideline adherence for one interval were regarded as potential confounders for the subsequent interval of MPR measures. Similarly, MPR for one interval was regarded as the exposure for the subsequent interval of outcome measures (online supplementary figure S1). HRs generated for each imputed data set were pooled and one final set was established. The same procedure was performed to assess adjusted and pooled survival estimates and obtain Kaplan-Meier survival curves for CV events and mortality. All hypothesis tests were evaluated at a 5% significance level.
To present baseline characteristics for the study population based on imputed data, the mean values of continuous variables and the most frequent category of categorical variables were obtained from the imputed data sets. These generated baseline values were used to descriptively present the study population but were not used in the statistical analyses.
Multiple imputations were performed in R V.3.3.234 using the MICE package.35 All other data management and statistical analyses were performed with SAS V.9.4 software.