Effects of pay-for-performance on prescription of hypertension drugs among public and private primary care providers in Sweden

Pay-for-performance (P4P), incentives tied to performance targets, has been a popular strategy to improve care quality in many countries. The research literature indicates that incentives tied to process measures (e.g., guideline compliance or screening rates) are often associated with small improvements of the performance measure, while incentives tied to health outcomes tend to be ineffective. However, the quality of the evidence is limited and there are comparatively few studies from outside the United States or United Kingdom (Eijkenaar et al. 2013; Herck et al. 2010; Ogundeji et al. 2016). Moreover, although the importance of contextual factors has been pointed out (McDonald et al. 2009), there have been few attempts to examine how providers with different characteristics differ in their reaction to P4P. In particular, although public provision is a key feature of many healthcare systems, there is to the best of our knowledge no study of whether private and public providers respond differently to P4P. Theoretically, one may expect the power of monetary incentives to be weaker for public care providers for at least two reasons. First, public providers lack a profit-maximization motive and often face soft budget constraints (Kornai et al. 2003). Second, public employees may be more driven by intrinsic motivation for their work than private employees, suggesting that external incentives (such as P4P) may be less effective in public organizations (Frey et al. 2013; Georgellis et al. 2011). Nevertheless, there are reasons why publicly employed physicians may not be completely insensitive to P4P. The components of the reimbursement scheme for providers clearly gives a signal of what is prioritized by leading policy-makers (Bénabou and Tirole 2003; Ellingsen and Johannesson 2008). Publicly employed physicians with career concerns may therefore respond to the scheme, as it allows them to signal high ability (Holmström 1999).

This study examines whether private and public primary care providers in Sweden respond similarly to P4P for compliance with hypertension drug guidelines. The Swedish primary care setting provides an excellent ground for studying such heterogeneity, as almost half of all providers are publicly owned. A study of publicly employed primary care physicians in a Swedish region indicated that they react to financial incentives, but did not examine P4P (Dackehag and Ellegård 2019). Of the three existing studies focusing on P4P in Sweden, none has addressed the potentially different responses across public and private providers. The previous Swedish studies indicated that P4P is associated with improvements of process measures such as antibiotics guideline compliance (Ellegård et al. 2018), registrations in a diabetes quality register (Ödesjö et al. 2015), and medication reviews (Ödesjö et al. 2017), though not with intermediate outcomes connected to elderly or diabetics patients (Ödesjö et al. 2015, 2017). The similarity with the international evidence on P4P indicates that the Swedish experience may be of interest outside the study context.

A second contribution of the study is to extend the limited evidence base regarding P4P for hypertension treatments. This literature, which has focused exclusively on the US and UK, has found zero or temporary effects of P4P on hypertension-related process measures and on intermediate or final health outcomes (Doran and Fullwood 2007; Lee et al. 2011; Petersen et al. 2013; Serumaga et al. 2011; Simpson et al. 2011). A major weakness of this literature is that all studies except Petersen et al. (2013) were conducted in the UK, where all providers became subject to P4P simultaneously and there was thus no control group. The decentralized healthcare system in Sweden allows for a stronger research design, as there are 21 independent county councils each designing its own reimbursement scheme for care providers. The influence of many common confounding factors can therefore be mitigated in a difference-in-differences analysis, comparing providers in counties that introduced P4P to providers in counties that did not.

Although the P4P and control counties had different average ACE shares (50% vs. 62% in the early period), Fig. 2 shows that the development in the groups was similar until 2010, the year when generic ARB was introduced. For providers in the control counties, the ACE share decreased between 2009 and 2010. In the P4P counties, the ACE share did not start to decrease until the year thereafter.

Table 3 shows the DID estimates of the P4P effect. Disregarding private/public ownership (column 1), the estimated effect of P4P on the ACE share was 3 percentage points (p < 0.01), a 5% increase relative to the mean ACE share (56.4%). The interaction specification indicates that the P4P effect was significantly stronger—twice as large—for private providers (column 2). The changing ACE share of private providers reflected a substitution of ARB for ACE, while for public providers it was more a case of less growth of the ARB prescriptions (columns 3–5). For both provider types, the effect on the ACE share became insignificant once the incentive was removed, although one reason for the insignificance may be that only 103 observations identify these estimates.

The estimates on PrivOwn are also imprecise, as they are identified by a very small number of providers that have switched from public to private ownership. Giving providers responsibility for a drug budget did not affect the ACE share, although it was associated with increased prescription of both ACE and ARB. The county-level number of GP visits was not associated with the provider-level ACE share, whereas the prescription of both ACE and ARB generally decreased following a choice reform.

The supplementary material displays the sensitivity tests. Notably, the parallel trends assumption appeared reasonable and the main results were robust to including differential trends for P4P counties. The positive effect for private providers was robust to the sensitivity tests, although the difference between public and private was not significant when using the wild cluster bootstrap. The effect for public providers was less stable and appeared to be driven by the two largest counties (Västra Götaland and Stockholm), in which the P4P incentive was indirect (see Table 1). Notably, the positive effect on private providers remained when excluding one P4P county at a time from the analysis, which is interesting given that the targets and incentive sizes differed across counties. The results were also similar when providers that failed to meet inclusion criterion 2 were included in the sample, although the magnitude of the public–private interaction was attenuated. The attenuation was expected, given that the more comprehensive sample included a large number of providers that were not subject to P4P [i.e., outpatient secondary care clinics (Anell 2015)].

The results strongly suggest that private providers reacted to P4P, while public providers reacted only in a few counties. The findings are consistent with the idea that monetary incentives are of higher importance for profit-maximizing providers, and also with the idea that individuals attracted to the private sector are relatively more motivated by extrinsic rewards (Georgellis et al. 2011). At the same time, it seems premature to rule out that public providers are completely insensitive to P4P; rather, the results indicate that their reaction depend on the circumstances.

The ACE share was lower for providers in the P4P counties than for providers in the control counties, suggesting that the decision to implement the P4P policy was driven by policy-makers wanting to approach the levels of other counties. While such selectivity in the decision to adopt P4P may limit the generalizability of the findings, it is not clear in what direction the bias would go. On the one hand, it may be easier to increase the ACE share if it initially is low. On the other hand, providers in the P4P counties may have firmly grounded aversion towards ACE, making it harder to influence their prescribing decisions. Notably, because policies like this are rarely randomized, a treatment-effect-on-the-treated like the one estimated here is a relevant policy parameter.

Many of the counties using P4P for the ACE share chose to remove the indicator after a few years. One obvious reason may be that the introduction of generic ARB made the issue less important. Previous research suggests that there are many reasons why counties often choose to replace their P4P indicators: satisfaction or disappointment with achieved outcomes, experiences that the performance measure is too volatile at the health care unit level, or simply a desire to prioritize something new (Ellegård et al. 2018; Johansson Krafve 2015). In Sweden, there has also been a growing critique against the use of monetary incentives in the health care sector (SOU 2017).

The analysis has several limitations. As already discussed, it was necessary to make assumptions about which providers in the prescription registers were primary care providers. It is thus possible that some providers were misclassified, although the sensitivity analysis indicated that the consequences of the sample delimitation were small.

Another weakness of the study is that the register does not include prescriptions that the patient did not redeem. Notably though, unless patients’ redemption decisions have changed differentially over time in P4P and non-P4P counties, the difference-in-differences analysis accounts for systematic differences between P4P and non-P4P counties.

A remaining caveat is that the empirical strategy only accounts for changes over time that affected providers in all county councils similarly. For instance, if the introduction of generic ARB had different impact in different counties, the estimated “P4P” effect also includes the differential effect of price changes. However, the model mitigates this concern by controlling for whether providers were responsible for a drug budget, as such responsibility would be the only reason (apart from P4P) why providers would care about the cost of prescribed drugs. Of note, the share of control providers with a drug budget increased a lot over the sample period (Table 1), but most of the increase took place before 2010, the year when many of the treated providers became affected by P4P. Therefore, the implementation of drug budgets in control counties should by itself not be a big confounder, although we cannot completely dismiss such concerns. Likewise, in both the P4P group and the control group, there are examples of counties that introduced a choice reform in 2010 and others that had already done so. Therefore, the choice reform variable limits the potential disturbance from these reforms at least to some degree.

The observational study setting implies some complications. Almost all county councils experimented with different P4P indicators during the study period, and it is not possible to rule out that the effect to some degree reflects that the ACE share received less attention by providers in counties that just had adopted incentives for other, unrelated goals (Holmström and Milgrom 1991). In one of the P4P counties (SLL), the total number of patients prescribed ACE or ARB was also incentivized, meaning that the impact in that county might reflect a changing patient mix. Notably though, the result that private providers react to P4P was unchanged even when providers from SLL was left out of the analysis.

Notwithstanding the limitations of the study, it is interesting to note that the estimated effect of 5% was similar in magnitude to the typical estimate for process measures found in previous P4P studies (Herck et al. 2010; Ogundeji et al. 2016). The estimate was also similar in magnitude to that of a previous Swedish study that concerned incentive for compliance with antibiotics guidelines (Ellegård et al. 2018) and qualitatively similar to other findings from Sweden (Ödesjö et al. 2015, 2017). In difference to previous evidence from Sweden (Ellegård et al. 2018), but similar to findings from other contexts (Constantinou et al. 2017), the significant effect only lasted as long as the incentive was in place. However, this result is tentative, as the point estimate for private providers indicated a lasting effect and the number of private providers in this category might have been too small to detect statistical significance.

Of studies on P4P for hypertension-related performance measures, this is the first to consider the ACE share. Observational time-series studies from England and Scotland found no trend breaks around the time of P4P implementation with respect to the number of prescribed hypertension drugs or recordings of blood pressure (Lee et al. 2011; Serumaga et al. 2011; Simpson et al. 2011). A randomized control trial in the US found a temporary increase in the percentage of patients on recommended drugs when incentives targeted individual physicians, but no effect of group incentives (Petersen et al. 2013). It has been argued that the target levels in the English P4P scheme were too low to spur further improvements (Serumaga et al. 2011); by contrast, the typical target levels in Sweden (Table 1) ought to have presented a challenge for the Swedish care providers. In relation to the results from the US study, it is interesting that the Swedish incentives appear to have been effective despite that they were directed to the health center rather than to individual physicians. However, given the vast differences in terms of institutional settings and study designs, it is difficult to pinpoint the reasons why the results differ.

P4P increased compliance with drug treatment guidelines for hypertension in Sweden. The effect was particularly strong for private providers, while the effect on public providers was present only in a few counties. Future research should seek to investigate the conditions under which public providers do react to monetary incentives (Dackehag and Ellegård 2019). Policymakers and researchers ought to acknowledge that the different underlying incentive structures of public and private providers may affect the effectiveness of P4P, even when little money is at stake.