This study found generally lower nonadherence rates compared to previous research for each of the three diseases [
2]. However, definitions of nonadherence varied strongly across previous studies. The reason why we found lower rates of nonadherence may be that that we focused on early dropout and refill adherence < 80%. Moreover, our pharmacy record data did not include a measure for primary nonadherence, where the patient does not redeem the prescription [
15]. In the dataset that was the source for this study, 7.6% of all prescriptions were not redeemed [
16]. In absolute numbers beta blockers and antidepressants were among the top of non-redeemed drugs. However, these drugs also have a high prescription volumes in the Netherlands [
17].
Correlates for serious nonadherence
One of the aims of our study was to find out whether risk profiles for nonadherence were comparable for patients who used drugs for the following three chronic diseases: depression (antidepressants), hypertension (antihypertensives) and type 2 diabetes (oral hypoglycemics). Our main conclusion is that no such common risk profile emerges even when using the same source population and measurements. For different diseases and its related medication risk profiles differ and, therefore, nonadherence should be studied and treated disease-specific.
For oral hypoglycemics few correlates for nonadherence were found. On the contrary, early dropout from antihypertensive use was correlated with many factors. Hence, the question is whether these are indeed risk factors. For example, younger and highly educated patients were more likely to stop using antihypertensives at an early stage. These patients may find other ways to lower their blood pressure, for example by losing weight or doing more exercises.
We did find some important correlates for nonadherence to antidepressants and refill nonadherence to antihypertensives. First of all, we found that non-western immigrants were more vulnerable for nonadherence to both antihypertensives and antidepressants. This can be related to their socio-economic status but also to a lack of understanding about their disease and its treatment since not all immigrants are able to communicate easily in Dutch or English. Both poor socio-economic status and poor understanding are found to be related to lack of adherence in antihypertensives in previous studies [
2,
18]. Our findings cannot be attributed to moving out of the country and, therefore, out of the registration databases because non-western immigrants in the Netherlands have low emigration rates. In 2001, the year of our study, only 1% of Dutch non-western immigrants left the countr [
19]. Our study also showed that in a primary care setting, type of medication is associated with nonadherence to both antidepressants and antihypertensives. Early discontinuation was higher among TCA-users. This may be due to the fact that TCAs are more often prescribed for other diagnoses such as pain, and ensuresis (in children). Once patients took antidepressants more continuously, users of TCAs proved to be most adherent users of antidepressants. This may be attributed to the fact that TCAs are more often subject to discontinuation because of side effects as was shown in a meta-analysis that compared treatment discontinuation of SSRIs and TCAs [
20,
21]. Once patients overcome the first prescriptions and either do not have side effects or accept them they may be more inclined to take the medication. For antihypertensives users of ace-inhibitors or/A2 antagonists were most adherent and type of medication proved to be the strongest correlate for adherence to antihypertensives. Nonadherence was highest among users of diuretics, which is in line with an earlier study in the Netherlands [
22]. Studies from other countries not always found higher nonadherence rates for diuretics [
23]. Lower adherence to diuretics is, for example, attributed to adverse effects and easiness of taking medication.
GP consultation is important for adherence to antihypertensives. Users of antihypertensives who visited their GP for hypertension and/or diabetes had higher adherence rates than patients who only had repeat medications (either from the GP or the medical specialist). This stresses the importance of communication about disease and treatment. Communication is facilitated by face-to-face consultation.
Somatic co-morbidity is associated with adherence to antidepressants: patients with somatic co-morbidity were more often seriously nonadherent. Since we controlled for complex medication regime (number of different type of drugs within a year) this effect cannot be attributed to difficulties encountered in taking multiple medications.
In sum
For each definition of nonadherence we found that the nonadherent patient population is hard to characterize by its sociodemographic characteristics, GP consultation, and medication related information on the patient, especially since correlates – partly – vary across diseases. A prescriber determine out characteristics such as sociodemographics ans GP consultation pattern rather easily. However, adherence is probably also influenced by patient characteristics that are less visible and more subtle, for example the patients' attitudes towards taking medication or the patients' trust in the health care system [
2]. These characteristics are more difficult to detect and need more time and attention from prescribers.
Differences between practices
We found that adherence rates vary across general practices, even though the number of general practices included in the analysis was low (n = 72). Previous studies found that clear instructions on the management of disease has a positive impact on patient adherence [
24] as has a good relationship between prescriber and patient [
25]. Communication styles are found to differ between doctors [
26,
27]. Future research should therefore further unravel what characteristics and mechanisms cause patients from one general practice to be more adherent compared to patients listed in another practice.
Strengths and limitations of the study
This study used a population-based dataset with a large sample, that enabled a multilevel analysis. Moreover, we combined registration information with data from a patient census, providing us with more information on the patient than most regular registration databases. However, our database also has some limitations. Dispensing general practices were excluded from the database, while about 10% of the patients in the Netherlands is listed in such practice. Moreover, not all patients could be linked. This was mainly due to the available linking keys [
10,
28]. The way the data were linked caused that the database included more patients with chronic medication. As our study included drugs that should be taken chronically, such bias towards our study is expected to be limited. We also had to exclude almost 30% of the (linked) patients for whom we had dispensing data and consultation data because these patients did not fill out the census form with patient characteristics. However, these patients did not differ in adherence rates from the patients who were included in the study.
A main advantage of refill data is that adherence rates can be estimated without the patient being aware of it. It increases the accuracy of the estimates by eliminating any Hawthorne effect [
29]. Moreover, we used the medication refill adherence measurement (MRA), which is, according to a study by Hess et al [
30] the preferred measure of adherence when using administrative data. The use of administrative pharmacy data also has some disadvantages. The first problem is that it is not possible to assess time of dosing [
31,
32] and that the data do not absolutely reflect patients' drug use. Roter et al (1998) suggested that prescription refills reflect patients' intention to comply rather than their actual drug consumption, i.e. patients fill their prescriptions more readily than they consume their medicine [
33]. However, some authors argue that patterns of ongoing prescription refilling probably provide the most accurate estimate of actual medication use in large populations [
34] , to assess drug exposure retrospectively or when direct measurement of medication is not feasible [
31]. A recent study on hypertension medication found that compliance measured by electronic monitoring revealed higher adherence rates compared to prescription refills [
34]. The researchers argued that the reason for this finding may be that electronic monitoring systems make patients aware of taking medication and as such influence adherence.
The data in our study refer mainly to the year 2001, because patient characteristics and morbidity data in general practice were only collected for that year. A one-year period may seem short to study adherence patterns. However, for refill adherence we only included patients with at least three prescriptions, which – given the fact that repeat prescriptions in the Netherlands often are prescribed for a three month period – covers about the whole year. In fact, this definition of refill adherence refers to patients who are inclined to used their medication but – in case they are nonadherent – who fail to do so.
Another problem is that the data do not tell what the reason for discontinuation is. Doctors may as well decide to stop the medication rather than the patient. As such we could not distinguish between a gap due to lack of adherence and a gap due to medical decisions. Since we expected this problem to be larger in the first stage of medication use, we separately analyzed early dropout from refill adherence. If a patient has three prescriptions over a one-year course (which was the minimal number of prescriptions for us to calculate refill adherence) we expect that there is an intention to continue the treatment. Moreover, our results showed that patients who consult their GP for complaints related to their medication (hypertension, anxiety, diabetes) are more compliant, which may be an indication that GPs are not very much inclined to stop the medication. Still, part of the discontinuation may be due to medical decisions and therefore, estimates for nonadherence in our study will be biased and – in real – levels of adherence will be higher. This may be especially true for antidepressants because this medication is not always chronic. However, over 80% of antidepressant users with three or more prescriptions in 2001 also had an antidepressant prescription in 2002, indicating that for the majority of patients there is an intention to continue treatment. A final disadvantage of the use of dispensing data is that such data do not reflect primary nonadherence.
Implications for clinical practice and future research
For health care professionals it is hard to recognize nonadherent patients by their socio-demographic background. Moreover, socio-demographics that are correlated with nonadherence vary across different types of drugs. There is an exception, though: patients' ethnic background. Non-western immigrants have lower levels of adherence. In order to get insight into motivations of these patients to be nonadherent it is important to check upon and discuss this issue in patient-prescriber consultations: did they not understand why taking their medication is important, did they not agree upon taking medication etc. Cultural aspects may influence the attitude towards taking medication, an aspect we did not include in our study [
35]. In fact, regular monitoring of and discussion on nonadherence is important for all patients.
We argue that guidelines for prescribers should include information on levels of nonadherence to certain types of medication. For example, the Dutch GP guideline for hypertension recommends diuretics as first choice medication in hypertension treatment. Our analyses showed that in Dutch general practice adherence to diuretics is lower than that to any other antihypertensive. For patients who are suspected by the prescriber to be nonadherent it may be more rational to prescribe a beta-blocker or an ace-inhibitor. At least prescribers should be aware of the nonadherence problem with this type of medication and closely monitor medication use. Such monitoring should include questions on adverse effects and or experiences in taking the medication during follow-up consultations.
Most of the implications mentioned refer to the importance of communication between professionals and patients. The fact that prescribers vary when it comes to their patients' adherence levels might, at least partly, be explained by differences in communication styles between prescribers. Further research into these differences among prescribers should provide more insight into this issue.