Introduction
Drug–drug interactions (D–DIs) are responsible for many adverse patient outcomes. Different studies suggest that D–DIs may cause up to 3% of all hospital admissions [
1‐
4]. A D–DI is defined as a pharmacokinetic or pharmacodynamic influence of drugs on each other, which may result in desired effects, in reduced efficacy and effectiveness or in increased toxicity [
5]. Although many D–DIs exist, only a small part of these D–DIs is clinically relevant [
6‐
8]. The potential benefits of drug combinations should be weighed against the seriousness of the D–DI, taking into account the availability of alternatives. Only in cases that the risks associated with the D–DI are higher than the benefits or if a better alternative is available, the D–DI should be avoided.
In the Netherlands, one of the tasks of the pharmacist is to intervene in case of D–DIs, which involve a high risk for the patient. Hereto, the pharmacist uses patient characteristics and the medication history. All prescriptions, which are submitted to the pharmacy, are screened on potential interactions with the help of medication surveillance software. These D–DIs are evaluated by the pharmacist who intervenes if necessary. This task is important but cumbersome, and requires great attention from the pharmacist. The organisational aspects, such as the tuning of the medication surveillance software and instructions of technicians, should be managed by the pharmacist in such a way that in case of D–DIs with a high risk the pharmacist intervenes. This is important for the prevention of adverse patient outcomes [
9].
The objective of this study was to assess process and structure characteristics associated with the dispensing of interacting drug combinations, which carry a high risk of adverse patient outcomes.
Results
The database contained a total of 100,295,311 dispensings in the selected study period. One thousand one hundred and forty-two pharmacies were recorded in the database with 5,000 or more dispensings. The number of dispensings per pharmacy varied from 5,019 to 264,631. Because pharmacies receive reimbursements from several health care insurance companies and because not all health care insurance companies were included in the database, these numbers do not correspond with the total number of dispensings per pharmacy. The eleven potential D–DIs were dispensed 11,594 times. In 5%, more than one pharmacy was involved. As these D–DIs could not be assigned to a single pharmacy, they were excluded from further analyses. The number of dispensings and D–DIs are shown in Table
1. Disopyramide (D–DI number 6) and azapropazon (D–DI number 11) were not dispensed by 44% and 46% of the pharmacies, respectively. Therefore, a ratio could not be calculated for these pharmacies and these D–DIs were excluded from the analyses.
The number of times a ratio above one was found was calculated (Table
1) and pharmacies were selected as shown in Fig.
1. Two hundred and sixty-eight pharmacies were selected to receive a questionnaire and 74 pharmacies were selected for a visit by the IHC. For several reasons, such as recent visitations and duplications in the database, 12 pharmacies were excluded. Eventually, 256 pharmacies received a questionnaire and 62 pharmacies were selected for a visit. Two hundred and forty-six questionnaires were filled in (response rate 96.1%) and 58 (93.5%) pharmacies were visited after the questionnaire was completed. The judgements during the visits by the IHC were compared with the answers by the pharmacists. In 33 of the 37 verified questions, the IHCs judgement matched in more than 90% the answer of the pharmacist. Except four questions, the judgement by the IHC was equally more positive and more negative than the answers by the pharmacist.
In the univariate analysis, all combinations between the questions and D–DIs were searched for significant correlations. Two correlations were found with D–DI number 1 between macrolide antibiotics and digoxin (Table
3). Pharmacies, which are part of a health care centre dispensed this interacting drug combination more often than other pharmacies. A correlation with the type of medication surveillance system was also found. Pharmacies using the Euroned system dispensed this interacting drug combination more often, while pharmacies using the Pharmacom system dispensed this interacting drug combination less often.
Table 3
Significant univariate correlations between the questionnaire and the number of dispensings of the D–DIs between macrolide antibiotics and digoxin (number 1)
Is the pharmacy part of a health care centre? (1 yes, 2 no) (yes n = 18, no n = 228) | −0.165 | 0.009 |
Which medication surveillance system is used in the pharmacy?
|
Pharmacom (1 yes, 0 other) (n = 81) | −0.261 | 0.000 |
Aposys (1 yes, 0 other) (n = 62) | 0.088 | 0.170 |
Euroned (1 yes, 0 other) (n = 89) | 0.197 | 0.002 |
For the multivariate analysis, 32 variables were selected, representative of the whole range of questions. These variables were used in the analysis-set to compose models. The adjusted explained variance ranged from 2.6% to 28.9% (Table
4). The model explaining the D–DI between macrolide antibiotics and digoxin had by far the highest adjusted explained variance. The models were validated in the validation-set, calculating the unexplained variance (Table
4). The six variables in this model explaining the D–DI between macrolide antibiotics and digoxin are shown in Table
5.
Table 4
Predictability of the models composed in the multivariate analysis
1 | 28.9 | 0.61 |
2 | 12.8 | −0.22 |
3 | 17.3 | 31.5 |
4 | 7.0 | −0.18 |
5 | 14.4 | −0.41 |
7 | 6.5 | 6.4 |
8 | 16.1 | 0.68 |
9 | 14.0 | −0.43 |
10 | 2.6 | 0.90 |
Table 5
The questions in the multivariate model predicting the dispensing of the D–DI between macrolide antibiotics and digoxin (number 1)
Constant | | 3.3679 |
Is the pharmacy part of a health care centre? (yes n = 18, no n = 228) | Yes (0) versus no (1) | −2.2749 |
Co-trimoxazole—acenocoumarol: no appointments were made with the GPs. The drug will be dispensed. | Option 1 ‘with all GPs’ (1) versus other option (0) (n = 11) | Reference |
Eight options of choice option 1 ‘with all GPs’ and option | Option 2 (1) versus other option (0) (n = 10) | 1.0308 |
Eight ‘with no GPs’ | Option 3 (1) versus other option (0) (n = 4) | 0.3788 |
Option 4 (1) versus other option (0) (n = 4) | −0.4542 |
Option 5 (1) versus other option (0) (n = 3) | 0.9026 |
Option 6 (1) versus other option (0) (n = 2) | −0.5100 |
Option 7 (1) versus other option (0) (n = 4) | −0.1912 |
Option 8 ‘with no GPs’ (1) versus other option (1) (n = 202) | 0.0886 |
Are separate signal texts in the medication surveillance program adjusted to the situation in the pharmacy? (yes n = 72, no n = 165) | Yes (0) versus no (1) | 0.1793 |
Is the management of signals traceably recorded on the receipt? (yes n = 211, no n = 35) | Yes (0) on the receipt, no not on the receipt (1) | 0.2691 |
The supervision on management of signals takes place on the basis of signal lists (yes n = 158, no n = 86) | Yes (0) on the basis of signal lists, no (1) not on the basis of signal lists | 0.0723 |
How many receipts are dispensed per year divided by the number of fte technicians | | <10–4
|
Discussion
In this study, we investigated determinants for the dispensing of 11 undesirable interacting drug combinations. In general, our results are in line with the expectation that the medication surveillance system plays an important role in medication surveillance. Although the 11 potential D–DIs were counted 11,594 times which suggests that a considerable number of patients is exposed to potential and avoidable adverse patient outcomes, these results should be judged against a background of approximately 100 million dispensings. It is possible that in these cases due to particular circumstances any other option, such as substituting or not dispensing one of the drugs, is a less favourable choice than dispensing the D–DI. In 5% of the total number of D–DIs more than one pharmacy was involved, indicating the importance of communication. For the D–DI between macrolide antibiotics and digoxin, two determinants were found. Although the type of medication surveillance system was a determinant, this does not mean that the differences are determined by the quality of the system itself because they may also correlate with the attitude of the pharmacists using the systems. The three medications surveillance systems differ in the extent to which communication with other healthcare providers is possible and developments were made in recent years. The Pharmacom system has the most advanced communication possibilities and compared to the other systems, new developments to the Euroned system were modest. Unexpectedly, pharmacies part of a health care centre dispensed this D–DI more often than other pharmacies. In health care centres, the communication lines between pharmacists and general practitioners are much shorter, suggesting that intervening undesirable D–DIs will be easier. Possibly, pharmacies which are part of a health care centre oppose the opinions from the general practitioners less often, to avoid harming the cooperation within the health care centre but, of course, there may be several other reasons.
For the other eight assessed D–DIs no determinants were found in the univariate analysis, neither did the models in the multivariate analysis have a good predictability. A possible explanation is that the quality of medication surveillance in community pharmacies in the Netherlands is high. Therefore, the number of pharmacies dispensing high-risk D–DIs seems to be small.
Our study has some potential limitations. First, because we used strict inclusion criteria to prevent false-positive results, it is likely that the number of dispensings of undesirable interacting drug combinations in this study is an underestimation and it is possible that important determinants were not recognized or difficult to assess. In the univariate analyses, only those questions are given which had a significant (p < 0.01) correlation in two independent sets. Although we included 183 questions and nine D–DIs in the univariate analysis, the possibility of including a significant correlation by chance was small (on average 0.16 question). Second, the reimbursement data from eight health care insurance companies were used. In the Netherlands, these companies work mostly regionally. It is nevertheless not to be expected that the determinants of dispensing interacting drugs differ per region or that pharmacies differ in their management of D–DIs between patients of different health care insurance companies. Third, from all potential D–DIs, only 11 (but highly clinically relevant ones) were selected for this study. According to the Dutch guidelines, for all 11 combinations the dispensing of an alternative was strongly advised as a good alternative was available. Nevertheless, it is possible that these dispensings were not an error because any other option was not possible. For example, when a patient is hypersensitive to the alternative drug recommended in the guidelines or when the alternative drug is not effective. In these cases, the benefit of both drug therapies should be weighed against the potential risks of the D–DI. The potential risks can partly be avoided by taking appropriate measures such as monitoring of drug levels. In this study, we could not retrieve why the pharmacist had dispensed the interacting drug combination, and whether the dispensing was erroneous or not.
Fourth, the questionnaire was composed on the basis of a literature search and interviews with experts. It is possible that not all characteristics correlating with the dispensing of undesirable interacting drug combinations were disclosed, such as differences in population characteristics between pharmacies. For example, pharmacies with an elderly population using more drugs simultaneously have a higher risk of dispensing interacting drug combinations than pharmacies with a younger population. Also, it is possible that in areas with many general practitioners who use a medication surveillance system for prescribing, the background chance of a D–DI is much smaller. Fifth, it is possible that the differences between pharmacies were too small compared with the power of this study to distinguish determinants.
All associations found in this study were directly related to the management of signals. In our questionnaire, we also included other topics, such as pharmacy preparations and patient care. Future research should focus on the management of a larger variety of signals than the ones in our study and on how D–DI associated dispensing could be further reduced.