Information is frequently advocated as a tool to improve public services. But some suppliers’ responses to information shocks may be objectively better for consumers.1 In this paper we examine heterogeneity in supplier responses to a frequent type of information shock and explore how this affects their consumers.

Our focus is on responses to information in the context of an innovation in healthcare. Information shocks are particularly pertinent in this context. Innovations are generally introduced through clinical trials and successful trials are often based on a relatively homogeneous set of patients. Rolling out the innovation to a wider group of patients often leads to less positive (or negative) effects on patient health than in the initial trial. This can even lead to abandonment of the innovation, known as “medical reversal” (see, e.g., Prasad and Cifu, 2019).2 In this setting, acting upon or disregarding new information can have a large impact on patient welfare. Since even specialist medical societies appear to be resistant to making medical reversals (suggesting considerable inertia in changing ineffective clinical practice (see, e.g., Wang et al., 2015)), there is a need to investigate physician responses to such common information shocks, whether there is heterogeneity across physicians, and the impact of this heterogeneity on patient outcomes.

To do this we exploit an important innovation in the production technology that interventional cardiologists use in their work: the drug-eluting stent (DES). Starting in 2002 DES was widely heralded as the solution to a key problem in coronary catheterization. As a result it captured over half the market from the older technology (bare-metal stents, henceforth BMS) in only four years after regulatory approval. But in 2006 information was released showing DES caused potentially life threatening side-effects. This second information shock, widely reported internationally, drastically reversed the trend of increasing DES use worldwide. Within just one month, DES lost half its global market share to BMS and the decline continued until extensive re-evaluations of the safety of DES reaffirmed its superiority for certain patient types. This led many national regulatory bodies to issue guidance as to the appropriate use of DES and BMS a year after the negative information shock.3 After this, DES use rose again.

Fig. 1 shows the effects of these information shocks on the use of DES in our “test-bed” setting, Sweden. Between the date of DES approval in Europe in early 2002 and the release of adverse information in 2006, DES had become the dominant choice of Swedish cardiologists, accounting for approximately 60 percent of the Swedish stent market. Only one year later this share had plummeted to less than 10 percent.4 This led to a regulatory response with the introduction of new national guidelines in late 2007 providing guidance as to when DES should be used, which led to renewed uptake, albeit at a slower rate than the initial increase.

We use this context to examine whether physicians who rapidly embrace a new innovation have better or worse outcomes for their patients. We use the universe of cardiologists in Sweden and data on all their patients from 2002 to 2011. Our setting has several advantages for the study of the impact of information on physician behavior and patient outcomes. First, the initial positive information shock and the subsequent medical reversal were salient, exogenous and the behavioral responses large. Second, the setting is one that affects a relatively large patient population for which the consequences of medical errors can be important.5 Third, the issuance of national guidelines after the news shocks allows us to use this “guideline period” to define appropriate treatment and to benchmark cardiologist responses and patient outcomes prior to these national guidelines to responses and outcomes in this period.6 Fourth, the treatment alternatives we study (DES versus BMS) are in all relevant aspects equivalent in how they are clinically administered. Thus we can exclude potential explanations for heterogeneity in behavior and patient outcomes arising from differences in cardiologist motor skills or visual acuity. Furthermore, the introduction of DES did not affect the appropriateness of other treatment options (for example, coronary artery bypass grafting) so the relevant patient population of interest can be considered as fixed over time. Finally, by examining these issues in the context of the Swedish healthcare system, we are able to rule out market mechanisms that may drive decisions about treatment in many healthcare markets. These include patient selection (the treating cardiologist does not generally decide whether to stent a patient but only selects the type of stent, and patients have virtually no choice of selecting physicians in the Swedish inpatient sector; we explore this empirically below), competition (Swedish hospitals are publicly owned and managed and physicians are salaried) or costs of treatment (the expected price differential between the use of BMS and DES in PCI treatments was relatively small in Sweden).7 Thus, we may interpret variation in responsiveness to the initial positive news across cardiologists as arising from individual discretion in response to information.

We exploit the three information regimes of “initial positive news” (period one: the period between regulatory approval of DES in 2002 and 2006), “medical reversal” (period two: the period between 2006 and 2007) and the guideline period (period three: the period between 2007 and 2011) for our analysis. To define our measure of responsiveness to positive news about an innovation we construct a cardiologist-specific measure of responsiveness in period one, defined as the speed with which each cardiologist adopted DES relative to the period-specific average across all cardiologists. We use this to categorize whether the cardiologist’s response to the initial positive news was slower than their peers and test whether this is associated with their patient outcomes. We then explore whether the differences that we find are due to patient selection or treatment selection.

We find the following. First, there is less heterogeneity in the speed with which cardiologists take up DES after the guidelines were published compared to the two earlier periods. Hence the guidelines restricted (as they were intended to) individual discretion. They also resulted in improved patient outcomes. We use these findings to justify the use of the guideline period as a period of appropriate practice which we can use as a benchmark for earlier treatment behavior of the cardiologists. We also use the clinical information in the guidelines to define inappropriate (“off-label”) use of stents. Second, we show that cardiologists’ speed of response to the initial positive news is associated with their patients’ outcomes, with patients treated by slower responders to the positive news having a lower risk of adverse clinical events compared to those treated by their peers. We find no evidence that these differences in patient outcomes are driven by patient–physician sorting. Instead, we find that slower-responding cardiologists make, on average, better choices of which patients to give the new stents to, and they are less likely to use stents inappropriately (i.e., off-label), with such off-label use of stents being strongly associated with an increase in adverse cardiac events. However, despite these differences in the choice of stents and off-label use between slow and other cardiologists and their impacts on patient outcomes, we do not find that this can fully explain the observed differences in outcomes of the two groups of cardiologists. Thus we cannot fully identify the mechanism through which slow responders perform better. Finally, we find that those who respond least to the positive news are more likely to work in teaching hospitals. It is possible that this gives these cardiologists greater access to private information about the innovation, which they use when making clinical decisions.

Our work contributes to several strands of literature. The first explores causes and consequences of physician practice styles (see, e.g., Chandra and Staiger, 2007, Epstein and Nicholson, 2009). Chandra et al. (2012) provide an overview of potential causes for variations in provider treatment decisions across similar patients. These include (i) defensive medicine, where providers perform unnecessary procedures to avoid complaints, bad reputation and possible lawsuits from patients; (ii) financial incentives associated with fee-for-service reimbursement models (McClellan, 2011); (iii) patient preferences and demand for specific procedures (Cutler et al., 2019); and (iv) unobserved heterogeneity across providers (Doyle et al., 2010). Our institutional setting allows us to focus on variation in the behavior of providers, abstracting from the first three potential drivers of variation. In particular, we contribute to a small set of recent studies which analyze the relationship between provider practice styles and costs and quality of care. Currie et al. (2016) study whether more aggressive (defined as the use of more invasive treatments) or responsive (the tailoring of treatment to patient characteristics) practice styles matter for costs and health outcomes of patients with acute myocardial infarction. Currie and MacLeod (2020) explore whether physician experimentation with anti-depressant drugs is associated with better patient health outcomes. Molitor (2018) examines how cardiologists’ practice styles are affected by their environment by assessing how their behavior changes when they move across healthcare regions. He finds that migrating physicians are highly malleable and largely change their treatment behavior in line with the prevailing environment, suggesting that hospital characteristics may play a substantial role in shaping practice styles. Cutler et al. (2019) examine physician behavior using responses to vignettes (hypothetical medical cases) and identify types of behavior from these responses. Although they do not study the association with patient outcomes, they find that these types of behavior explain a relatively large share of variance in medical expenditures.

While information is likely to play a role in these decisions, none of these papers focus on responses to information. There is a large literature on prescribing responses to warnings by regulatory authorities concerning risk which shows that drug safety warnings may not be optimally received. A review of 20 years of US Federal Drug Administration communications of risk concluded that although some communications had immediate and strong impacts, many had either delayed or had no impact on health care utilization or health behaviors (Dusetzina et al., 2012). In a healthcare context closer to the Swedish setting, a meta analysis of 25 UK medicine risk communications relating to prescribing concluded that risk communications were associated with significant changes in both intended and unintended (detrimental) changes in health outcomes (Weatherburn et al., 2020). In the explicit context of a medical reversal, Everhart et al. (2021) examine physician-level deadoption of a combination of diabetes medications following evidence that they were not effective when taken together. They find on average physicians decreased prescribing following the release of public information but many physicians also increased prescribing.

Closest to our research, Staats et al. (2017) study the negative news shock for DES. They examine how physician experience (defined as volume of activity) affects the speed of response to this news in a US context. They find that more experienced cardiologists respond more sluggishly to the negative news shock. Our focus is on heterogeneity in response to the early good news about the innovation and, importantly, and in contrast to Staats et al. (2017) (and most other studies in this field), we link the speed of response to patient outcomes. Our Swedish setting also allows us to close down avenues relating to behavior that is a response to the financial incentives present in the US context.

We also contribute to the literature on responses to information and their impacts outside the medical setting. This literature shows that individuals may over- or under-react to news (see, e.g., Daniel et al., 1998), and that individuals respond differently to positive and negative news (e.g., in psychology (Baumeister et al., 2001), empirical finance (De Bondt and Thaler, 1985, De Bondt and Thaler, 1987, Veronesi, 1999, Hong et al., 2000, Hong and Stein, 2007, Kacperczyk et al., 2015), and politics (Soroka, 2006)). There is a growing interest in ideas of differential responses to common information driven by salience and limited attention, whereby cognitively overloaded individuals rationally pay attention to only a subset of information (see, e.g., Maćkowiak et al., 2018). Our study shows that heterogeneity in responses to common information shocks also affect physician behavior and, importantly, the health of their patients.

The paper proceeds as follows. The next section provides an overview of the Swedish healthcare system and the clinical context. Section 3 explains our empirical approach, how we estimate cardiologist responsiveness to news shocks, the effect on patient outcomes, and explores potential mechanisms that may be driving these differences. Section 4 presents the data, Section 5 the results and Section 6 concludes.