The results of this study show that dispensation data (a feasible proxy of doctor prescriptions) are compatible with our hypothesis on physicians' behaviour. The more expensive a drug is, the higher the price differential is between dispensations to pensioner and non-pensioner patients if non pensioners pay a 40% copayment, but this differential is inexistent in drugs with a reduced copayment, suggesting that Spanish NHS physicians are sensitive to the price that their patients have to pay.
There are three pharmacological FC groups for which the effect is clearer, because they include drugs of high price and drugs of low price: antiplatelet agents (clopidogrel vs. acetylsalicylic acid), statins (atorvastatin vs. simvastatin and other statins) and anti-dementia drugs (anticholinesterases and memantine vs. ginkgo biloba). In these groups, doctors seem to differentiate prices and predominantly prescribe low price drugs to non-pensioners and high price drugs to pensioners. Therefore, these groups have a positive price differential for pensioners in almost all healthcare areas. However, doctors do not seem to distinguish prices for non-expensive or low-price drugs. For them, the differential effect is negligible. One possible explanation for this behaviour is that physicians only have an approximate knowledge of drug prices. This explanation is consistent with a review of physicians' awareness of drug prices, showing a low cost accuracy (31% of estimates were within 20% or 25% of the true cost, and fewer than 50% were accurate by any definition of cost accuracy) . This result is also consistent with a previous study for Spain. Spanish family physicians were asked about the price of well-known drugs. They estimated correctly (with an interval of 25% around the real price) in 41% of cases. This study also suggests that physicians tend to neglect price differences between products of identical composition .
In contrast with the growth in the literature about (co)payment effects on healthcare service utilization (quantities), empirical studies about the effects on prices are scarce. Our study is consistent with those scarce antecedents, confirming that the selection of the prescribed drug is influenced by the price that the patient pays (or co-pays). Even in Japan, where physicians sell medicines to the patients and have incentives to obtain higher margins with more expensive drugs, one study in antihypertensive drugs found that "physicians are willing to give up one dollar of their profit in order to reduce the co-payment of non-elderly patients by 28 cents" . In Sweden, another study concluded that physicians prescribed less expensive drugs (generics) to patients that had to pay for them . On the other hand, several studies using qualitative methods  or surveys [14–18] have reported that physicians claim to consider out-of-pocket costs a more important issue when prescribing than the cost for the organization or for society (although physicians -occasionally in the same studies- declare their awareness of drug costs and that discussing the cost of treatment with patients is very uncommon) [15, 17–19]. Notably, the most likely strategy used to assist patients burdened by their out-of-pocket costs is to switch the patient from a brand name to a generic drug [20, 21].
Going beyond the agency relationship, doctors' price sensitivity to their patients' copayment scheme may be the result of different causal mechanisms. First, pensioners are older and probably sicker than non-pensioners, and occasionally some expensive drugs could be more appropriate for patients at higher risk (i.e., clopidogrel has a lower risk of gastrointestinal bleeding than aspirin and could be a better alternative for elderly people). Second, the pharmaceutical industry exerts strong promotional pressures on doctors to prescribe new, more expensive drugs with a patent in force. It is possible that marketing strategies focus -at least in some cases- on older patients, contributing to higher prices (although these hypotheses cannot explain the absence of differences in drugs with reduced copayment). Third, in Spain general practitioners maintain the prescription of medicines that have been indicated by specialists (so-called "induced prescription"). Specialists have different prescription patterns (with more innovative and expensive drugs) and, also, treat more complex -and probably, older- patients. Pensioners could be more exposed to the "induced prescription" phenomenon than non-pensioners and therefore receive more expensive prescriptions. Finally, if patients with co-payment do not pick up (selectively) the most expensive prescriptions from pharmacies, we would be facing a problem of patients' price sensitivity instead of physicians' sensitivity to the price that can be afforded by patients.
Apart from contributing to the scarce literature on this issue, our study has certain other strengths. We work with population data and we include all the dispensations for selected therapeutic groups for two common conditions. The study also has several limitations. First, pensioners are very different from non-pensioners in terms of age, disease patterns and their severity, and these differences could justify differences in the choice of drugs and, therefore, in the average price for pensioners and non-pensioners. The ecological nature of the data does not allow consideration of all the factors that influence medical prescriptions (disease and its severity, other accompanying health conditions, possible contraindications or interactions with other drugs that the person is taking alongside, and so on). But in our study, average price discrepancy is measured within specific and relatively homogeneous therapeutic groups. In most of these therapeutic groups, evidence of the superiority of one drug over others in terms of higher price rarely exists (i.e. atorvastatin vs. simvastatin, ARBs vs. ACEIs, brand name vs. generic drugs, one atypical antipsychotic vs. another atypical antipsychotics, and so on) and, with some exceptions, there are no clinical reasons for the systematic use of high price drugs in pensioners and low price drugs or generics in non-pensioners. Nevertheless, in some cases the therapeutic groups are more heterogeneous, including medicines with different indication profiles (i.e. antiplatelet drugs or atypical antipsychotics). For those groups including some medicines aimed at young people and others aimed at older patients, the price differential could be a compositional effect not related with doctors' sensitivity to patient costs (i.e. requirements for the prior authorization of clopidogrel consider age over 65 as a criterion; because people over 65 are mainly pensioners with no copayment, we could find a compositional effect in this therapeutic class).
Second, more severe patients may use stronger doses of the same drug. Although strong-dose packages have a higher price, the DDD metric oscillates between flat pricing (equal for all presentations without considering the number of units or their strength) and monotonic pricing (the price of the DDD decreases with increasing units or doses of the presentation). Because pensioners usually consume presentations with higher doses and more units per presentation, the DDD price is artificially lower in this group, underestimating the copayment effect and undervaluing the intensity of doctors' sensitivity to patient costs.
Third, the Spanish regulation of prices and copayments does not consider a reduced contribution for fixed-dose combinations, even if both (or more) drugs of the combination separately have this consideration. As disaggregate data of prescriptions within each group were not available for this study, groups with fixed-dose combinations suffer a miss-classification bias (i.e. a third of the dispensations of inhibitors of the renin-angiotensin system group were fixed-dose combinations subject to a general copayment of 40%, but they were classified as RC drugs). This bias also underestimates the copayment effect and doctors' sensitivity to patient costs. Nonetheless, Model 3 does not show differences among pharmacological groups, suggesting that doctors' sensitivity to patient costs is independent of the medicines used.
Finally, and probably the most important limitation in our study, our conceptual framework attributes decisions on prescriptions to physicians but uses dispensations (prescriptions filled out) as a proxy of prescriptions issued (which also includes unfilled prescriptions). The patients' ability to influence prescription decisions to cheaper drugs is irrelevant to the agency theory (in fact, if the agency relationship was complete, the decision would always reflect the patients' preferences), but unfilled prescriptions overestimate doctors' sensitivity to patient cost effects, especially if patients do not pick up the most costly medicines from the pharmacy. Some studies in the United States have shown that the drug abandonment rate increases as the out-of-pocket expenses increase . Although the generalization of these studies to the Spanish setting is uncertain, probably both behaviours (patients' price sensitivity and physicians' sensitivity to the price that can be afforded by their patients) occur at the same time and both contribute to the price differences between co-payment schemes detected in our study. The nature of our data (dispensation, not prescription) does not permit the estimation of the contribution of each factor to the price differences found.
The policy implications of our findings for cost-containment are diverse. First of all, specific measures addressed to patients (i.e. the use of reference prices as avoidable copayments for pensioner and non-pensioner patients) could be effective measures for increasing doctors' prescription of cheaper and generic drugs. Second, cost-containment policies could benefit from a better knowledge of drug prices among physicians. Also, if our results are related with the agency relationship, physicians' incentives to switch expensive medicines to cheaper equivalents or generics should considerer physicians' beliefs on the clinical value of the cheaper ones relative to more expensive drugs. Nevertheless, we need broader and deeper studies on cost-containment pharmaceutical policies, and specifically on the agency relationship between physicians and their patients . And, we evidently need better data about costs and reasons for prescription [23, 24].