Trends in Incidence of Intracerebral Hemorrhage and Association With Antithrombotic Drug Use in Denmark, 2005-2018

Key Points Question Is use of antithrombotics associated with intracerebral hemorrhage (ICH), and is increased use of these drugs in recent years associated with increasing incidence of ICH in Denmark? Findings In this case-control study of 16 765 ICH cases, ICH was statistically significantly associated with antithrombotic drug use, including vitamin K antagonists, direct oral anticoagulants, clopidogrel, and low-dose aspirin. Whereas antithrombotic drug use (mainly direct oral anticoagulants) increased during 2005 to 2018, the incidence of ICH did not. Meaning In Denmark, increased use of antithrombotics was not associated with an increase in the incidence of ICH.


Assessment of antithrombotic drug exposure
Information on prescriptions dispensed at community pharmacies in Denmark have been prospectively recorded in the Danish National Prescription Registry (Prescription Registry) since 1995. For each prescription, the date of dispensing and a full account of the dispensed product, including the anatomical therapeutic chemical (ATC) code, are recorded. The indication and prescribed dose are not available in the Prescription Registry.
To calculate date of end of supply of each prescription, we set each prescription to last the number of days that corresponded to the number of pills dispensed (divided by 2 for dipyridamole, dabigatran, apixaban, and edoxaban and 1.5 for warfarin 1 ). Between pairs of consecutive prescriptions, we allowed a grace period of 60 days between the end of supply of the first prescription of that pair and the date of the next prescription. An episode of treatment comprised consecutive pairs of prescriptions presented without a gap. Based on the most recent treatment episode before the index date, we divided exposure by recency of use as described in the main text. However, as we observed increased ORs associated with past use of certain antithrombotics, we also performed analyses where past use (91-365 days before index date in main analysis) was subclassified into two categories (91-183 days vs 184-365 days before index date).
For analyses of duration of use, we subdivided current use according to the duration of the treatment episode into <1 month; ≥1 month to ≤3 months; >3 months to ≤12 months; >1 year to ≤3 years; and >3 years.
We calculated the risk of intracerebral hemorrhage (ICH) among current users with concurrent multiple use of antithrombotics (dual therapy with antiplatelets, or antiplatelets combined with anticoagulant). In these analyses, as explained above, use of dipyridamole was disregarded (except in the analysis concerning dual therapy with low-dose aspirin (ASA) and dipyridamole). In the analysis of current dual therapy, individuals who had used any other antithrombotic than the two in question within the 12-month period preceding the index date were not retained as dual current users. The same strategy was adopted for current triple therapy (i.e., only users of the three drugs in question within the 12-month period were included in that particular analysis). Some subjects switch between antithrombotic drugs. To analyse the effect of current use of antithrombotic drugs in monotherapy, we therefore repeated analyses comparing current users who had only used a single antithrombotic drug (or class of drugs, depending on analysis) within 12 months before index date with non-users. Therefore, we excluded users with use of more than one antithrombotic drug in the last 12 months from these analyses. We did not analyze the effect of single dipyridamole use (with no concurrent use of ASA), since, in accordance with Danish guidelines on stroke prevention, this drug is only recommended in combination with ASA. 2,3 Accordingly, we only studied the combination ASA/dipyridamole. We also performed analyses performed restricted to "new users" (also known as new initiators) of antithrombotics and "naive users" of oral anticoagulants. We defined "new users" as patients with only a single episode of use of the drug class (e.g. vitamin K antagonist (VKA)), or specific drug (e.g. warfarin) in question, who, furthermore, had no recorded use of the drug class/specific drug in the years 1995 to 1996; the latter criterion was used to minimize misclassification of prevalent use within the first 2 years after the Prescription Registry became operational in 1995. We defined "naive use" of oral anticoagulants (i.e. no recorded previous use of other oral anticoagulants) at group level (e.g., direct oral anticoagulant (DOAC) use by patient with no previous use of VKA) and at drug-specific level (e.g., dabigatran use by patient with no previous use of any other DOAC or VKA).
For DOACs, we calculated the effect of dose of the drug on risk of ICH. Based on capsule or tablet strength of the most recent prescription presented before the index date, we classified daily dose into standard (dabigatran 300 mg, apixaban 10 mg, rivaroxaban 20 mg, edoxaban 60 mg), or reduced (dabigatran ≤220 mg, apixaban ≤5 mg, rivaroxaban ≤15 mg, edoxaban≤30mg).
With few exceptions (see below), we used a common reference group of exposure. This reference group comprised subjects with non-use of antithrombotic drugs (end of supply of most recent prescription of any antithrombotic drug before index date stopped more than 365 days before index date, or no recorded use of any antithrombotic drug since 1995).
We wished to capture recent developments in use of oral anticoagulants and therefore, we performed analyses with data limited to 2014-2018. In these analyses we also calculated risk estimates using current use of VKA (warfarin in drugspecific analyses) to better mirror the choices of comparator in previous clinical trials and observational research.

Potential confounders
We classified disorders that we regarded as potential confounders based on diagnosis codes from in-patient (1977 to 1 day prior to index date) and out-patient (1995 to 1 day prior to index date) contacts at hospitals in Denmark, prescriptions (1995 to 1 day prior to index date), or a combination of diagnosis codes and prescriptions, see eTable 1. We also classified exposure to certain drugs (in separate variables: NSAIDs, SSRIs, statin, hormone replacement therapy, oral corticosteroids) we regarded as potential confounders based on prescription data for the period 1 year to 1 day prior to index date.

Fatal ICH
Date of death of patients with ICH was established through linkage with the Civil Registration System 4 which is continuously updated with regard to migration and vital status of all citizens in Denmark. We classified patients who died within 30 days of ICH onset as cases of fatal ICH.

Analyses of associations of antithrombotic drug use with risk of ICH
For main analyses of use of antithrombotic drugs with risk of ICH, see manuscript.

Supplementary analyses of influence of socioeconomic status as potential confounder
We performed several supplementary analyses that are mainly presented in the manuscript. Here, we focus on the evaluation of influence of socioeconomic status as potential confounder The main analyses are based on data up to 2018 held at the Danish Health Data Authority (DHDA) (in Danish, Sundhedsdatastyrelsen). However, the registries at DHDA do not hold information on socioeconomic status (SES), a potentially confounding factor. Therefore, we performed separate analyses of a dataset we built with data held by another vendor (Statistics Denmark) that included information on SES at the level of the individual but was only updated to 2016 (i.e. lacked data for 2017 and 2018). Importantly, all other data used in the analysis of this additional dataset were retrieved from copies of the same registries as held at DHDA and using the same codes as in the main analyses. We used this additional dataset to perform key analyses where we compared ORs for ICH associated with antithrombotic drug use with and without addition of two separate covariates concerning education level and income (ascertained the year before ICH onset) to the full model (for description of full model, see main text). This allowed us to gauge the direction and magnitude of confounding caused by SES.

Supplementary analysis of negative exposure controls
Use of negative exposure controls has been advocated as a tool for detecting confounding and bias in observational research. 5 For this purpose, we chose current use of proton pump inhibitors (PPI), as we had no a priori biologically founded hypothesis of a causal association between use of PPI and risk of ICH. In Denmark, in the study period, 97%-99% of the total quantity of PPI sold was dispensed on prescription (www.medstat.dk/en). We calculated recency and duration of PPI use as described above using dispensed defined daily doses (DDD) 6 to calculate the supply of individual prescriptions of PPI. We calculated the OR of current use of PPI after adjusting for all variables included in the main model.

Descriptive analyses
Annual incidence rates (IRs) of ICH and fatal ICH (overall and by sex) per 100,000 person-years were calculated using the number of ICH events as numerator and person-years at risk as denominator. Confidence intervals (CIs) were derived assuming a Poisson distribution. Standardized IRs (sIRs) were age-(5-year bands) and sex-standardized to the Danish population in 2011 using census data from Statistics Denmark. Using Poisson regression, the trend was calculated in the incidence rate of ICH of nationwide data (2005 to 2018). Trends in the 30-day case fatality rate of ICH patients were also tested.

Supplementary descriptive analyses
We repeated all descriptive analyses using the Patient Registry as the source, as no temporal trends in the high sensitivity of this registry for ICH diagnosis were observed in a recent validation study. 7 Further, as the introduction of a new electronic health record system in 2016 in hospitals in Eastern Denmark 8 could influence the completeness of the data, we also conducted analyses after excluding data from this region (i.e. with data from Western Denmark only).

Analyses of regional subset with verified diagnoses
We had access to data on all first-ever cases of spontaneous ICH in the Region of Southern Denmark (RSD; population 1.2 million) in 2009 through 2017. 5 In brief, we traced all adults (aged 20-99 years) in RSD with a first-ever diagnosis of ICH according to multiple sources (i.e., recorded admission in the Stroke Registry or any type of hospital contact recorded in the Patient Registry (admission, outpatient or emergency room)) in 2009-2017. For all thus identified potential cases, we retrieved medical record information (primarily discharge summaries and brain imaging study reports), based on which study physicians verified diagnoses of spontaneous ICH. Using this regional verified subset, we calculated sIRs (standardized to year 2011 as described above) of spontaneous ICH for 3-year time-bands (i.e., 2009-2011; 2012-2014; 2015-2017) and incidence rate ratios (IRRs;2012-2014vs 2009-20112015-2017vs 2009-2011. We could verify spontaneous ICH diagnosis in 2,556 patients. The positive predictive value (PPV) of a diagnosis of ICH varied by source and type of contact. 5 A small number of untraced records (n=188) were primarily due to paper medical records not being archived at some hospitals in the region. The distribution of untraced records was heavily skewed over time (2009-2011: 115; 2012-2014: 54; 2013-2015: 19). Therefore, we calculated sIRs by applying multiple imputation, i.e., 100 imputed data sets for the 188 patients with untraced status in the regional subset. We based imputation on published PPVs by calendar period. 5 We likewise calculated sIRs of fatal ICH based on verified cases after establishing 30-day mortality based on Civil Registration System data, as described above (see Fatal ICH).
We also computed the prevalence of current use of antithrombotic drugs among general controls that resided in RSD in 2009-2017 and could be matched to verified ICH cases on index date, age and sex. Based on these controls, we calculated percentage of users (age-and sex-standardized to year 2011) of antiplatelet and anticoagulant drugs within each 3-year time-band and prevalence ratios (i.e., 2012-2014 vs 2009-2011 and 2015-2017 vs 2009-2011 *Non-use of any antithrombotic drug is defined as no use of any antiplatelet or anticoagulant in the 12 months preceding index-date 1 Adjusted for age, sex, and calendar period (year) by design 2 Adjusted for age, sex, and calendar period (by design) and the following, based on register data: hypertension, previous ischemic stroke, diabetes, chronic renal insufficiency, chronic hepatic disease, coagulopathy, heart failure, ischemic heart disease, peripheral artery disease, cancer, high alcohol consumption, chronic obstructive pulmonary disease, use of oral anticoagulants, low-dose aspirin, clopidogrel, other adenosine diphosphate inhibitors (ticagrelor or prasugrel), statins, nonsteroidal anti-inflammatory drugs, selective serotonin reuptake inhibitors, hormone replacement therapy, or oral corticosteroid drugs. In drug-specific analyses of anticoagulants use of oral anticoagulants we adjusted with individual variables for use of dabigatran, rivaroxaban, apixaban, edoxaban, or warfarin, respectively. *Non-use of any antithrombotic drug is defined as no use of any antiplatelet or anticoagulant in the 12 months preceding index-date 1 Adjusted for age, sex, and calendar period (year) by design 2 Adjusted for age, sex, and calendar period (by design) and the following, based on register data: hypertension, previous ischemic stroke, diabetes, chronic renal insufficiency, chronic hepatic disease, coagulopathy, heart failure, ischemic heart disease, peripheral artery disease, cancer, high alcohol consumption, chronic obstructive pulmonary disease, use of oral anticoagulants, low-dose aspirin, clopidogrel, other adenosine diphosphate inhibitors (ticagrelor or prasugrel), statins, nonsteroidal anti-inflammatory drugs, selective serotonin reuptake inhibitors, hormone replacement therapy, or oral corticosteroid drugs.