The diffusion of a medical innovation: is success in the stars?

• Author: Burke, Mary A.
• Position: Author abstract

1. Introduction

   Traditional economic theories of technology adoption and diffusion have naturally emphasized the role of economic incentives, predicting that new technologies will be adopted based on their expected costs and benefits. More recent research in economics has considered the social conditions under which diffusion takes place, with particular concern for the process by which information about new technologies spreads across individuals and becomes salient. The latter approach does not necessarily negate the former, but it introduces the possibility that perceived benefits and costs may depend on the social environment. For example, Munshi (2004) finds a significant role for social learning in the adoption of high-yield crop varieties in India, and Huckman (2003) finds that individuals with a high level of expertise or "technological status" exert a significant influence on the technology choices of others in the same professional community. While many economists viewed Griliches' early study of hybrid corn adoption as roundly invalidating the prior sociological theories of diffusion based on social ties, more recent research suggests that such ties should not be lightly dismissed (Griliches 1957; Skinner and Staiger 2005).

   With regard to medical technology adoption in particular, a number of studies identify a significant role for social interactions (see, for example, Young et al. 2003; Browman et al. 2005). Coleman, Katz, and Menzel (1966) showed that prescription of the then-new drug tetracycline occurred earlier among physicians with higher numbers of academic citations; adoption among less-cited doctors depended significantly on whether or not they interacted with more prominent peers. Schectman et al. (2003) and Bhandari et al. (2003), among others, show that treatment practices respond more strongly to in-person educational interventions and to changes in peer practices than to dissemination of practice guidelines or new clinical findings via peer-reviewed journals. Medical industry literature shows that pharmaceutical companies and medical device manufacturers target "opinion leaders" in the medical community in their marketing efforts, on the assumption that adoption by such individuals will serve as an efficient engine for more widespread adoption of the drug or device. (1) A number of studies indicate the effectiveness of such marketing efforts, for good or ill.

   Despite a recent surge in interest in social interactions within economics, rigorous study of the phenomenon of "opinion leadership" in technology diffusion remains relatively scarce outside of the medical literature (see, for example, Soumerai et al. 1998; Borbas et al. 2000; Berner et al. 2003). In the case of medicine, the susceptibility of diffusion to the contingencies of social influence may entail large welfare consequences; therefore, the issue warrants further study. A rich data set on medical treatment in Florida enables us to examine the diffusion process for coronary stents. A stent is a cylindrical-mesh metal device that is inserted into the coronary artery in the process of balloon angioplasty, in order to keep the artery open after surgery. The device reduced the incidence of post-surgical coronary artery collapse, thereby reducing the need for subsequent emergency coronary bypass surgery (CABG), and reducing the incidence of restenosis (re-narrowing of the coronary artery) in the year following angioplasty. Stent use spread to a large proportion (about 90%) of angioplasty patients within a relatively brief period (from 1995 to late 2001 in the Florida data), and contributed to an increase in angioplasty usage in general. While this might be viewed as a success story (for example, compared with the relatively slow and spotty diffusion of the use of beta-blockers shown by Skinner and Staiger 2005), we observe considerable variation in the adoption and utilization rates during the diffusion period, and the eventual convergence to widespread use was not inevitable.

   Defining "stars" on the basis of having completed medical residency at a hospital ranked in the top 30 nationally, and alternatively, at one ranked in the top 10, we study the influence of the number of stars working contemporaneously at the same hospital as a given non-star physician on the adoption hazard and utilization rates of non-stars. It may be that expert practitioners teach others the appropriate use of the procedure, or simply that experts, among others using a new procedure, are more likely to be imitated by their peers. Over the diffusion period, we find that the presence of stars increases the probability of adoption and utilization rates among non-stars, but we find no social influence in the opposite direction. Thus, the unequal distribution of prominent physicians across hospitals leads to treatment inequalities that may entail welfare losses.

   A number of recent papers have studied the properties of models with social interactions (Glaeser and Scheinkman 2001; Blume and Durlauf 2001). In the standard approach, preferences involve a penalty for departing from the average behavior of a reference group. In the case of physicians, there are justifications for an influence of average group behavior, such as when individuals seek to conform with local practice standards for fear of malpractice liability, or if productivity spillovers and social learning effects depend on average behavior. However, asymmetric influences have not been well-examined within this literature--influences we think should be particularly relevant within professional communities. Although the literature on herding and information cascades recognizes that early decisions by a small number of individuals can play a large role in determining technology choice (Bikhchandani, Hirshleifer, and Welch 1992), the order of decision-making in such models is often random. In social learning models (Conley and Udry 2001; Munshi 2004) agents are thought to be learning from each other simultaneously, and differences in the ability to learn independently are suppressed.

   Professional communities tend to give rise to hierarchies based on skill and expertise, in which those at the top are emulated by others. In their classic study of the forms of social power, French and Raven (1968) identify, as one of the bases of power, expert power. Under expert power, an individual exerts influence on others because of her superior knowledge or ability. The importance of a subset of key individuals was also outlined in Rogers' study of the diffusion of innovations, following which they are often referred to as key opinion leaders (Rogers 1995). Studies in the medical literature have shown that "best practices" may not be used even when clear practice guidelines exist. The failure to adopt best practices cannot be readily attributed to ignorance of practice guidelines or results of clinical trials. Rather, some physicians argue that success of a therapy in clinical trials does not necessarily imply its success "in the real world" (Collins, Hawks, and Davis 2000). As documented by Soumerai et al. (1998), the experience of influential peers working in similar environments may, rightly or wrongly, be perceived as more relevant or convincing than evidence from clinical trials. The current study provides further evidence of asymmetric, local influences in technology adoption, and also helps to identify the characteristics of influential individuals.

2. Medical Background

   The full technical name for coronary angioplasty is percutaneous transluminal coronary angioplasty, or PTCA. PTCA is one of three major classes of treatment for coronary artery disease (CAD), together with coronary artery bypass grafting (CABG) and non-invasive medical management. During PTCA, a balloon catheter is inserted through the skin in the patient's groin and threaded through to the coronary artery, where the balloon is inflated to expand the artery and restore blood flow. PTCA is far less invasive than CABG but comes with risk of complications. For example, the coronary artery can collapse immediately following the procedure, in which case emergency CABG is required. In addition, restenosis (a return of blockage of the coronary artery) can occur within a period of several months, necessitating additional intervention (either repeat PTCA or CABG) and possibly resulting in death.

   A study of California discharge data estimates that in 1993, prior to the introduction of stents, about 6% of acute myocardial infarction (AMI) patients and 3.7% of non-AMI patients required emergency CABG after PTCA, while in 1996, one year after their introduction, these figures had fallen significantly, to 5.2% and 2.6%, respectively. In addition, the use of stents resulted in significant reductions in 6-month restenosis rates from 30% to 25% based on one clinical trial (Kimura et al. 1993), and from 61% to 41% in another (Agostoni et al. 2006). Study results differ in whether they find significant declines in one-year mortality rates with use of stents, however, because revascularization (i.e., repeat PTCA or CABG) reduces mortality risk from restenosis. However, revascularization rates are significantly lower when stents are used (Agostoni et al. 2006).

   In 2003, a second generation of stents received Food and Drug Administration (FDA) approval in the U.S., called "drug-eluting" stents for their ability to release medications to prevent restenosis. The latter innovation further reduced restenosis rates to 8.9%, according to the larger of two clinical trials of Johnson and Johnson's CYPHER stent. (2) Over the period of our study, 1995-2001, we observe the diffusion of just the bare-metal stent. While stent outcomes may have improved during this period as a result of learning-by-doing, such improvements can be considered marginal relative to the discrete effect of the introduction of stents. Drug-eluting stents can be seen as a distinct...