Prescription patterns of direct oral anticoagulants and concomitant use of interacting medications in the Netherlands

Our study involved a cross-sectional study for the calendar years 2014 up to and including 2018 of pharmacy dispensing data in patients who had their prescriptions for DOACs filled in a community pharmacy connected to the Dutch Foundation for Pharmaceutical Statistics (SFK). The SFK obtains data from > 95% of the community pharmacies in the Netherlands and covers about 15.8 million people (on a population of 17.3 million) (https://​www.​sfk.​nl). Information on patients’ sex and year of birth were available. Dispensing data were collected within pharmacies on a patient level, in which each dispensing was linked to the individual’s unique identification number. For each dispensation, we also obtained information about the drug supplied, and the postal code (two first numbers to indicate a postal region) of the dispensing pharmacy. Dispensations for one person could not be merged from different pharmacies. Drug dispensing medication was available as Anatomical Therapeutic Chemical (ATC) codes. Privacy of participating pharmacists, as well as prescribing doctors and patients, was guaranteed by providing the investigators with restricted access to a de-identified data file.

We only included community pharmacies that provided complete data over a 5-year time window and did not change software (which involved 1,665 out of 1,981 pharmacies). From these pharmacies we randomly selected a representative sample (approximately half of these pharmacies), as we deemed this number sufficient to address our research question. We obtained annual prescription data of patients who had at least one prescription for DOACs filled between 1 January, 2014, and 1 January, 2019. The DOACs we included were dabigatran, rivaroxaban, apixaban, and edoxaban. For each dispensation, we also obtained information about the DOAC and interacting drugs supplied. First dispensations of DOACs were defined as a DOAC dispensation without a dispensation of any DOAC in the previous 12 months and a dispensing fee for a first dispensing registered by the pharmacy.

The outcome of interest was concomitant use of a DOAC and one or more of the selected medications with a known drug-drug interaction profile as listed in the EHRA document [5] complemented with psychotropic agents known to cause drug-drug interactions [7]. A list of the medications, as well as the presumed interaction mechanism for each DOAC can be found in supplementary Table S1. We defined concomitant use when there was a 30-day overlap between a DOAC and another drug, except for medications that are typically used for shorter durations (i.e. macrolides), for which we used a 5-day overlap. We recorded the concomitant use of interacting medications for each calendar year.

Data were presented using the total number of prevalent DOAC users within a study year as well as new users of DOACs. We displayed categorical variables as median (interquartile range—IQR) or as percentage and number in parentheses, and compared these using the chi-square or Fisher’s exact test. For continuous variables we used mean ± standard deviation or median (IQR) quartiles depending on whether there was a normal distribution (based on skewness and kurtosis). We compared continuous data using the independent samples t‑test for normally distributed variables, and the Mann-Whitney U test for non-normally distributed variables.

In 831 community pharmacies there were 179,748 DOAC prescriptions to 99,211 unique patients within the separate pharmacies over the 5‑year study period. The total number of DOAC prescriptions rose from 8,293 in the calendar year 2014 to 74,461 in 2018. Over time, age as well as proportion of females among first-time DOAC users increased significantly, from mean age 70.6 ± 9.9 years and 39% female in 2014, to mean age 72.2 ± 11.1 years and 42% female in 2018 (see Fig. S1 in Electronic Supplementary Material [ESM]).

The initiation of DOAC treatment quadrupled over a 5-year time window. In our sample of 831 pharmacies, 8,293 patients were given their first DOAC prescription in 2014, while in 2018 a total of 35,415 were newly started on a DOAC. Temporal trends in the use of individual DOACs are displayed in Fig. 1. All DOACs showed an increase in use over time. This increase appeared to go in parallel for individual DOACs, except for dabigatran. Overall, rivaroxaban was most frequently prescribed (n = 12,672) among new DOAC users, followed by apixaban (n = 11,784), dabigatran (n = 7,473) and lastly edoxaban (3,486). The uptake of DOACs as a group was comparable between five geographical regions of the Netherlands over time, however there were notable differences in the preferred type of DOAC. As shown in Fig. 2, apixaban (52%) was most commonly prescribed to new DOAC users in the Eastern region of the Netherlands in 2018, whereas rivaroxaban was most frequently prescribed (32–48%) in the other regions.

The vast majority (99.3%) of included patients used concomitant interacting medications at time of first DOAC dispensation. Of those, most used either one (77.9%) or two (18.2%) interacting medications, with 4.8% using 3 or more. The most commonly prescribed interacting medications were digoxin (37.8%), atorvastatin (31.5%), verapamil (13.7%) and amiodarone (9.7%). As shown in Fig. 3, the distribution of the number of interacting drugs at first DOAC dispensation remained stable across the study years, as was the number of patients with 1, 2, or ≥ 3 interacting medications as percentage of total prescriptions. The median number of interacting medication remained 1 (IQR 1–1) across study years.

When assessing individual medications, we found a decrease in the use of antiarrhythmic medications between 2014 and 2018, most notably digoxin (41.7% versus 35.5%) and verapamil (20.5% versus 10.8%) (Tab. 1). On the other hand, there was a significant increase in this same period for concomitant use of many other interacting medications. Notable examples are atorvastatin (29.2% versus 31.9%), dexamethasone (0.9% versus 7.1%), amitriptyline (4.0% versus 5.3%), antiepileptic drugs (2.5% versus 3.8%), and haloperidol (0.5% versus 2.2%). As shown, in Tab. 2 these temporal changes can be observed for dabigatran, rivaroxaban as well as apixaban. In the latest DOAC prescription data (2018), differences were minimal between the individual DOACs, except for dexamethasone, which was more common among edoxaban users (8.8% versus 3.1–5.4% for the other DOACs).

Studies that assessed the impact of drug-drug interaction on clinical outcomes among DOAC users are limited. We have provided a summary of these studies in Tables S1 and S2 in Electronic Supplementary Material [ESM] [18‐20]. To date, the study with the most impact is a Taiwanese population-based analysis, in which the authors found substantially increased bleeding risk among DOAC users who also used amiodarone, fluconazole, rifampicin (rifampin), and phenytoin [19]. Of interest, both the use of fluconazole and antiepileptic drugs have increased over time in our analysis. The Taiwanese study also found that other drugs which shared key metabolic pathways with DOACs that would mechanistically result in an increased bleeding risk, did not do so. The most intriguing example is the concomitant use of atorvastatin, which was associated with a lower risk of bleeding. In our study almost one-third of the population was using this type of statin. In contrast to atorvastatin, other statins such as simvastatin and lovastatin were associated with higher bleeding risk when used concomitantly with dabigatran in a study by Antoniou et al. [18]. These findings underline that pharmacokinetic data may not always line up with associations observed in clinical outcome-based studies. Unfortunately, the Taiwanese study does not provide information on the impact on efficacy outcomes, such as stroke and mortality, nor on the relative efficacy and safety when compared with warfarin. A recent meta-analysis based on post hoc analyses from randomised controlled trials seems to provide reassuring answers in that regard, as the data clearly demonstrate that in the presence of interacting drugs, the use of DOACs remains more effective than warfarin [21].

So what are the clinical implications of our study? First, physicians should be aware that while DOACs may have fewer interactions than do VKAs, they do exist, and could lead to serious complications. Prior studies indicate that clinicians recognise only half of common drug-drug interactions as well as disease contraindication pairs [22]. Therefore, an improvement in the clinician’s ability to detect possible interactions could improve patients’ safety. We would therefore recommend that physicians who manage patients with AF use an online guidance tool, such as https://​www.​NOACforAF.​eu, when considering DOAC therapy or adding a different treatment among DOAC users. On the mentioned website, physicians can find practical information on prescribing DOACs, which is based on the EHRA practical guide, along with patient information in different languages (including Dutch). There are also other sources to help check for possible drug-drug interactions, such as Lexicomp® (integrated in UpToDate), or the tool by Medscape, which can be directly assessed via https://​reference.​medscape.​com/​drug-interactioncheck​er [23].

Our study relied on prescription data from a representative sample of community-based pharmacies in the Netherlands, in which we presume that these medications had been filled and had been used by the patients. The database did not collect in-depth data on patient characteristics, indication for DOAC use, nor on clinical outcomes while using the DOAC, which did not allow us to draw inferences on proper dosing, or on the impact of interacting drugs on the risks of bleeding and stroke/systemic embolism. Moreover, the SFK database did not allow for merging individual subjects’ data in the case of moving to a different SFK pharmacy. This may have led to overestimation of the number of first DOAC users in the observed period. However, this is unlikely to bias the comparison between the study years and overestimation may be limited as shopping behaviour is typically low in older patients.