The simultaneous effects of pharmaceutical policies from payers’ and patients’ perspectives: Italy as a case study

Since 1992, pharmaceuticals have been a preferred target of cost-containment actions [17]. Centralised price cutting, discounts on list prices and drug delisting prevailed in the first 10 years. Cost-containment approach was strengthened with the introduction of a spending cap on drugs (set as a percentage of public health funds) that was enforced by law in 2001. Initially, general price cuts were applied to cover the deficit. Since 2007, the industry has been partially in charge of covering the deficit: each pharmaceutical company is given a budget based on the national drug budget for the current year and market shares in the previous year; if the actual drug spending is over the budget, each company will contribute to the payback in proportion to its actual revenue (compared with its budget). Reimbursement and ex-factory prices are simultaneously negotiated by the national drug agency and the relevant company. The main criteria used in negotiation are disease burden, place in therapy and availability of alternative treatments, risk–benefit profile, therapeutic added value and impact on the drug budget. For most new drugs, managed market (financial-based and outcome-based) contracts are agreed on; for some drugs, more than one contract is negotiated [19]. Finally, reference pricing for genericated molecules (molecules with at least one generic version available) was introduced nationwide in 2001 and applied to the same molecule–package pair. TRP, instead, was introduced as a policy option implementable by regional governments in 2006.

Despite price and reimbursement being managed at the central level, since 2002, regions have implemented diverse actions to face financial constraints. In fact, regional governments have become accountable for their health-care-spending deficits. As a consequence, they strengthened their cost-containment actions on pharmaceuticals [17]. Copayment, actions on prescribing behaviour, including prescription quotas, and TRP were introduced by various regions to curb the retail market, whereas drugs used in hospital settings have been affected by regional formularies and procurement policies [4, 20].

Copayment in Italy is active in two forms: as a prescription fee and as the spread on the reference price. The latter has been active since 2001 (and then optionally increased in level with TRP) and is more the effect of a policy than a policy itself. Copayment as prescription fee, instead, was first introduced by regional governments in 2002 and is the form considered in this study as “copayment” policy; in 2014, drugs were subject to copayment in 16 ut of 21 regions. The introduction of copayment produces a payment shift from third-party payers to patients and is expected to lower drug consumption, at least in the short term. Therefore, we expect a short-term decline in public expenditures and increase in private expenditures. In absolute terms, private expenditures are expected to be lower than public expenditures because of the drop in consumption. In other words, we expect consumption to mediate the effects of copayment on public expenditures. In the long term, patients and prescribers may adjust their initial choices and increase consumption, leading to a positive impact on both public and private expenditures.

Prescription quotas were first introduced in 2005. They refer to binding prescription targets that address general practitioners prescribing more genericated molecules within a certain therapeutic class (e.g. x Percentage of genericated statins over the total prescription of statin). These quotas are usually enforced by (regional) law and linked with sanctions/incentives for prescribers. For the most important retail therapeutic classes [e.g. hypertension drugs, statins, selective serotonin reuptake inhibitors (SSRIs) and proton pump inhibitors (PPIs)], quotas were applied in 13 regions in 2014. Prescription quotas are expected to shift prescriptions within a therapeutic class of drugs from expensive to less expensive, with an overall reduction in public expenditures. No effect is expected on volume unless cheaper drugs lead to increased consumption, and no effect is expected on private expenditures unless generic products are associated with generic molecules that involve copayment due to reference pricing. We expect that prescription quotas require behavioural adjustments by patients and prescribers, and we expect these adjustments to occur over the long term.

TRP was intended to reduce the expected impact of generic reference pricing, which enhances a prescription shift from genericated molecules to patent-protected drugs in the same therapeutic area, avoiding the application of generic reference pricing. As of 2014, TRP was applied only to PPIs in nine regions. In October 2007, around 1 year after its introduction, this policy option was abolished at the regional level for equity reasons, but the new regulation has not had any retroactive effect, such that regions that had already activated TRP were allowed to maintain it. In principle, TRP is expected to reduce public and increase private expenditures. However, in the absence of copayment and prescription quotas, TRP could have adverse effects on public expenditures. In fact, TRP reduces the perceived minimum price for both private and public payers and could stimulate a higher demand. Because the demand mechanism is behavioural, we expect the effect to occur in the long term.

An interaction of policies is expected to occur to the extent to which the mechanisms behind them are not independent. Depending on how their mechanisms interact, one policy can reinforce or hinder the effects of the other. We expect that the three policies reinforce each other in decreasing public expenditures. The simultaneous presence of copayment and prescription quotas should lead prescribers to reduce inappropriate prescriptions and to favour cheaper drugs. Similarly, copayment associated with TRP should orientate prescribing behaviour toward less expensive molecules. Finally, prescription quotas should mitigate the expected positive impact of TRP on volumes, and TRP may enhance a shift towards cheaper drugs that are produced by prescription quotas. However, interactive mechanisms may exceed the intended impact, generating less equity and possible undertreatment. This effect could be signalled by a shift from reimbursed to nonreimbursed drugs.

Following the aforementioned gaps in the literature, this paper aims to assess (1) the individual and interactive effects of the three pharmaceutical policies on public and private retail drug spending and volumes in the short and long term, and (2) the causal relationship among policies, prescription/consumption behaviours and both public and private expenditures (i.e. the extent to which the long-term effect on expenditures is mediated by behaviours). Several models have been used to evaluate policy impact [21, 22]. We employed an enriched difference-in-differences (DD) model that allows simultaneous estimation of the effects of three policies and their interactions. We first estimated the separate effects of policies on public expenditures, private expenditures and volumes. Then, we tested the hypothesis that the effects of pharmaceutical policies on public expenditures are mediated by prescription/consumption behaviours (i.e. a transmission mechanism). To test for robustness, we also tested for possible reverse-causality and feedback mechanisms by switching the mediator and the independent variable, which allowed us to rule out the alternative hypothesis of ambiguous causality on behaviours.

Policies considered and their period of activation in every region are reported in Table 1. Figure 1, on the other hand, summarised the number of regions in which each policy was active in every period. Included variables, their measurement and relevant sources are summarised in Table 2. Dependent variables included monthly per capita public and private expenditures and volumes for partially and fully reimbursed as well as nonreimbursed retail drugs. Volumes were also used as mediators when testing for a behavioural transmission mechanism. The independent variable matrix included policies in the short term (i.e. introduced within 6 months) and long term (i.e. introduced at least 6 months earlier). We also included a set of control variables. Public and private expenditures are influenced by several factors other than policies. For example, in an investigation on the impact of copayment on drugs in Italy, Fiorio and Siciliani [5] included control variables and used a fixed-effects model (using the first-difference approach, equivalent to a fixed-effects model, since the number of periods was 2); the authors included a dummy variable for the regional government (regions may be governed by a left- or right-leaning coalition), per capita gross domestic product (GDP), proportion of people >65 years, number of pharmacists and number of general practitioners. We included the same variables (with the exception of pharmacists and physicians, as the relevant data were incomplete). Because of possible age-related differences in drug use, we added the paediatric population; we also included the political cycle (in pre-election periods, cost-containment actions may be relaxed to increase consensus) and added a double control for time: first, we included a time dummy for every single month (162 variables) in the model; and second, we controlled for repeated seasonal effects (Fig. 2) through dummies identifying the quarter (three variables). These two controls have different meanings: the second ne captures seasonal trends in the pharmaceutical market that are repeated every year, particularly evident from Fig. 2, while the first aims at reducing endogeneity by capturing any contingent effect happening in a particular month that is not explicitly accounted for in our model. In fact, despite the important seasonal effects, every period can carry some peculiarity that is outside our model and can influence our dependent and explanatory variables. However, including monthly dummies alone would have been enough to explicitly capture seasonal effects. Moreover, both monthly (unique) dummies and seasonal (repeated every year) ones are significant and do not present collinearity issues. We further considered two other possible sources of endogeneity: turnaround plans and spillover effects. Since 2007, regions with important health-care deficits have been asked to implement a turnaround plan, which increases the pressure to adopt cost-containment measures. The effects of pharmaceutical policies would likely be artificially inflated if turnaround plans were not considered. Accordingly, we included three dummy variables related to turnaround plans: (1) whether the region was treated with a turnaround (R turnaround = 1 if the region has introduced a turnaround); (2) whether the turnaround was active (T turnaround = 1 after turnaround plan approval); and (3) a DD estimator of the turnaround plan’s effect (Turnaround = 1 if the region was treated and the policy was active). Second, regions may experience imitative pressure to adopt drug copayment. This tendency was captured by a variable that considers, for each month, the overall number of regions with drug copayments (Fig. 1). Our hypothesis is that this imitative pressure may strengthen negative and positive impacts of copayment on public and private expenditures, respectively, by reducing reaction time from prescribers and patients who have observed other regions. We hypothesise that this effect will only influence copayment, as copayment is the policy most discussed and best known by the population; the other two policies are primarily known by technical staff, making spillover effects less likely. We also tested a model with spillover effects on the other two policies, but their coefficients were never significant, so we removed them to create the most parsimonious models.

A DD model was adopted and enriched to better address the parallel trend assumption (i.e. in the absence of the policy, dependent variables would vary similarly over time in both the treatment and control groups). In addition to standard procedures for estimating a DD effect, we adopted two strategies to better cope with the assumption. First, we controlled for the simultaneous presence of more than one policy at any time in every region, with the inclusion of interactions. Second, we included the difference between short- and long-term effects. This way of addressing the parallel trend assumption does not remove it but allows a more precise estimation of such a trend. The consequence of this strategy can be seen in the different R and T coefficients (see below) for copayment after inclusion of the other two policies. The model is described by the following equation:where R _j  = 1 if the region belongs to the treatment group, i.e. to the group of regions that have activated the policy for at least 1 month over the observed range; T _t  = 1 in the treatment period (T _t  = 0 before t,, the month when first the region introduces the focal policy); D _jt indicates that region j is on treatment at time t, i.e. region j belongs to the R = 1 group, t ≥ t and the policy has been introduced in region j; β ₃ is the main parameter of interest (DD); Run _jt is a pair of dummy variables that indicates whether the focal policy has been applied for <6 months or for at least 6 months (we tested different definitions of long term, finding consistent results and an indication that most differences between short- and long-term effects are visible using the 6-months split tables available from the authors); \({\text{Interactions}}_{jt}\) is the set of interactions among difference-in-differences in the three policies; and X _jt is the control variable matrix, including both quarterly (seasonal) dummies repeated every year and 162 monthly dummies.

The behavioural mediation hypothesis was tested with the Sobel–Goodman test and completed by a bootstrap estimation. The Sobel test, also known as the delta method, was introduced in 1982 [23] as a test of the significance of the indirect (mediated) effect. If a is the path from the independent variable and the mediator, and b is the path from the mediator to the dependent variable, then a × b is the indirect effect. After dividing the indirect effect by the square root of the variance b ² × standard error of a + a ² × standard error of b, a Z test is made on this ratio. When its value is >1.96, the mediation hypothesis is supported. However, MacKinnon et al. [24] provided evidence of the conservative nature of the Sobel test, which is, therefore, not very powerful. The main reason for the test being conservative is that the sampling distribution of the indirect effect is highly skewed. When the patter is positive, there is positive skew with many small estimates of the indirect effect and few very large ones. Since the Sobel test uses a normal approximation that presumes a symmetric distribution, it falsely presumes symmetry, which leads to a conservative test. For this reason, the Sobel test is generally associated with bootstrapping [25, 26], a nonparametric method based on resampling with replacement, which is done many times (1000 in this study), where the indirect effect is estimated at every replication. We estimated the indirect effect (and direct and total effects) using both the standard Sobel–Goodman test and the bootstrap method.

To consider possible simultaneous effects on private expenditures and volumes, we tested two mediation mechanisms. In other words, we allowed only volumes or private expenditures to react to policies in each model so we could establish their individual impacts. In particular, in the first model, we used public expenditures as the dependent variable, policies as independent variables and volumes as a mediator that was conditioned on all other control variables and private expenditures. In the second model, we used private expenditures as a mediator that was conditioned on all other control variables and volumes.

Descriptive statistics are summarised in Table 3. Box 1 illustrates the model matrix. Results are illustrated in Table 4a (expenditures models) and Table 4b (volumes models). Box 1 illustrates the model matrix (Table 3).

Copayment is the only policy in which its relationship with public expenditures is mediated by both volume and private expenditure. Prescription quotas and TRP may involve transmission mechanisms other than volume, such as change in mix of prescribed drugs, but these data were not available. Results of mediation tests are presented in Table 5, and details are provided in Table 6.

The first copayment mediation model , with drug volumes as the mediator, supports the mediation hypothesis. By holding private expenditures constant, copayment reduces volumes by 5.4 %. The elasticity of public expenditures to volumes is 0.81, such that a 5.4 % reduction in volumes produces a 4.4 % decline in public spending. Therefore, the direct impact of copayment (−3 %) is reinforced by the indirect effect, and the total effect of −7.4 % is partly mediated (59.7 % of the total effect) by a volume effect. We also assessed the amount of burden transferred to patients by holding volumes constant, thus observing the mediation of private expenditures. The total effect of copayment on public expenditures was reduced from 7.4 to 4.1 % when volumes were held constant, whereas the direct effect did not change (−2.9 %). In this case, copayment produced a 7.5 % increase in private expenditures, and the elasticity of substitution between private and public expenditures was −0.14. Therefore, the indirect effect of copayment via private expenditures was −1.1 %, mediating 26.5 % of the total effect on public expenditures. No mediation was observed in the long term, so we removed the long-term effect from Table 5. Comparing the two tests on copayment, we provide evidence that a behavioural volume effect attenuated the burden of a shift from public to private coverage. The effects of prescription quotas were not mediated by private expenditures. However, the effects of prescription quotas were fully mediated by volumes, creating an increase of 2.5 %. Such an increase led to a +2.0 % indirect effect of prescription quotas on public expenditures. Finally, the effect of TRP was only mediated by volumes in the long run. Differently from what emerged from regression analysis, when holding private expenditures constant, the impact of TRP on volumes was negative rather than positive. In fact, controlling for private expenditures, TRP produced a decrease in volumes of 2.7 %, which translates into a 2.2 % reduction in public expenditures. The direct effect, instead, remained positive (3.7 %), and therefore the total effect was not significantly different from 0. This situation is known as inconsistent mediation, because the direct effect has an opposite sign compared with the indirect effect, resulting in a null total effect. In other words, the mediated effect compensates for the direct one. Bootstrap estimations were consistent and confirmed the Sobel–Goodman test results (Table 6). All reverse causality tests rejected mediation at the 95 % confidence level.