In this study we analyzed the route of a prescription, from the moment of a treatment decision by the GP to the drug delivery by the pharmacist, with a specific interest in generic and branded substitution. In particular, we analyzed the behaviour of the pharmacist with regard to generic and branded substitution by using information related to patients' and pharmacists' characteristics and the specific timing of the prescriptions and deliveries. For this purpose, prescription-level information from both a GPs' and a pharmacies' database was merged.
We found that two of every three prescriptions were generically prescribed by the GP, while for 32% of the prescriptions the GPs prescribed the branded drug. The relatively high proportion of prescriptions written using the brand name could possibly indicate the effect of pharmaceutical detailing and marketing efforts and routine prescribing of branded drugs by the GPs . We are currently directing one of the further researches into this aspect.
Following the prescription until delivery of the drug by the pharmacist, we noticed that generic substitution occurs in 8% of the total prescriptions (25% of the branded prescriptions), only marginally more often than branded substitution (6.7% of the total prescriptions, 10% of the generic prescriptions). We estimated generalized linear mixed models (GLMMs) for both types of substitution in an effort to identify the characteristics that may lead the pharmacist towards generic or branded substitution. Application of the GLMMs yielded results regarding duration from patent expiry that were in accordance with prior studies [7, 25].
In particular, the probability of generic substitution for a prescription was found to be significantly and positively related to the experience of the patient on the use of drugs of the same drug class but negatively related to the proportion of prior branded use in this drug class. Obviously, it is less likely for a patient who has routinely been delivered the branded drug in the past, and is again prescribed a branded, to have his prescription generically substituted by the pharmacist. The opposite holds true for the case of branded substitution where prior branded use seems to positively affect the probability of branded substitution. In the case of the first prescription for any user, pharmacists seem to be more reluctant regarding generic substitution. The pharmacist, possibly following the patients' perception that branded drugs are generally safer and more effective, prefers to either deliver a branded prescription or even substitute a generic one with the branded.
The higher likelihood for both generic and branded substitution close to the time of patent expiry can possibly be explained by the dynamic process of generic diffusion. After the patent expiry date, GPs may still prescribe the branded drug but the stock of the pharmacist might already be replaced with generic alternatives, enhancing the impetus for generic substitution. Oppositely, the generic might be prescribed while there is still branded stock available and/or lack of information in the pharmacy regarding the existence, efficacy and safety of generic alternatives. In general, we notice that substitution by the pharmacist is very much concentrated in the period where the generic drug starts diffusing in the market.
The lower probability for branded substitution, in combination with the higher probability of generic substitution of prescriptions dispensed by pharmacies belonging to a chain or a partnership might be related to the fact that these organizations have a stricter policy as well as additional incentives towards generic use . Also, bulk orders of generic drugs by chain pharmacies may result in higher discounts from the generic manufacturers and, hence, into lower generic acquisition costs .
Finally, differences were identified between substitution patterns of different drug groups. Antidepressants and PPIs were found to be less often generically substituted and more frequently substituted from generic alternative to branded, compared to ACE inhibitors and statins. The reduced probability for the class of antidepressants could possibly be attributed to concerns of patients on the effectiveness of the generic alternative. Furthermore, the patient expiry of the branded version of paroxetine was followed by specific patent extensions and legal battles from the producing pharmaceutical company in order to maintain the branded market share with a potentially substantial effect on substitution patterns . A similar cause might be attributed to the lower probability of generic substitution on PPIs. The branded manufacturer of omeprazole introduced a new formulation of the drug that, in combination with extensive marketing efforts, maintained a large share of the omeprazole market . Furthermore, concerns have been raised on the in vitro dissolution profile and kinetic properties of generic omeprazole compared to the branded drug .
Through the application of a GLMM approach, we detected a significant amount of heterogeneity in the behaviour of pharmacists towards, mainly generic, substitution. Except from the analysed characteristic of the pharmacy status, which was available to us, specific location and competition factors may be expected to additionally influence the decision of the pharmacist on substitution. A pharmacy in a more remote district which might enjoy a certain monopoly on drug distribution over the area will possibly be more inclined to substitution if this is financial-economically attractive. Oppositely, pharmacy competition within a single city might oblige pharmacists to be more elastic towards specific patient requests, for example, regarding preferences for staying on the branded product with which potentially a lot of experience already exists. Also, no information on potential discounts achieved by pharmacies and other financial benefits that might co-determine pharmacists' behaviour were available to us. There was also considerable evidence found regarding the unobserved variation in the substitution probability among different patients. This was expected since patients behave differently when confronted with the decision of substitution, depending on characteristics that are impossible to be identified through our dataset such as tolerability of the generic alternative, existence of comorbidities and attitudes towards drug use. Such heterogeneity in behaviour of pharmacists and patients, potentially related to above mentioned factors unobserved in our database, provides a strong rationale to use the random effects model as we did here.
To our knowledge, this is the first paper that describes the extent and nature of generic and branded substitution by pharmacists in the Netherlands. Our efforts to explicitly identify factors that might affect the likelihood of these substitutions may help in designing specific strategies to achieve general goals in this area, often concerning the increase of generic and discouragement of branded substitution. In respect, our findings indicate that policies enhancing generic substitution could be targeted to increase generic use at first prescription, for those patients with a long history on the branded drug, the relatively young patients and in individually-owned pharmacies. As these aspects are mirrored in our analysis on branded substitution, the same types of prescriptions, patient groups and pharmacies should be targeted for discouraging branded substitution. Explicit information on safety and efficacy of the generic PPIs and antidepressants could help raising levels of generic substitution (decreasing levels of branded substitution) to levels achieved for the other drug groups investigated namely of statins and ACE-inhibitors.
One caveat in our analysis is the lack of information concerning the capability of the pharmacist to dispense the generic drug instead of the branded, as some of the prescriptions explicitly indicate that generic substitution should not be performed. Another important factor of influence on which our databases offer no information is the monetary incentives that pharmacists are offered, most often in the form of discounts, from the pharmaceutical industry. Given that generic manufacturers have been evidently found to offer greater discounts than branded manufacturers [5, 31] it could however be hypothesized that the presence of discounts results into higher generic substitution levels, yet the data to actually analyse this currently lack us.
Furthermore, despite the use of a mixed model to correct for unobserved heterogeneity, we might still lack important information on other potential explanatory variables such as pharmacy stock sizes, regional insurance companies' rules and incentives for substitution (e.g. preference policy) and the intensity of detailing. Further research is directed towards providing more insight regarding these yet "unobserved" variables.
Another interesting extension of the current research would be an analysis on the effect of patent expiry on therapeutic substitution. Although Klok et.al  have already identified a positive effect of the patent expiry of branded omeprazole on therapeutic substitution within the PPI group, investigation in a larger number of drug classes would give a more general impression of the effect of patent expiries on therapeutic substitution.