Physician patient-sharing networks and the cost and intensity of care in US hospitals

Abstract

Background: There is substantial variation in the cost and intensity of care delivered by US hospitals. We assessed how the structure of patient-sharing networks of physicians affiliated with hospitals might contribute to this variation.

Methods: We constructed hospital-based professional networks based on patient-sharing ties among 61,461 physicians affiliated with 528 hospitals in 51 hospital referral regions in the US using Medicare data on clinical encounters during 2006. We estimated linear regression models to assess the relationship between measures of hospital network structure and hospital measures of spending and care intensity in the last 2 years of life.

Results: The typical physician in an average-sized urban hospital was connected to 187 other doctors for every 100 Medicare patients shared with other doctors. For the average-sized urban hospital an increase of 1 standard deviation (SD) in the median number of connections per physician was associated with a 17.8% increase in total spending, in addition to 17.4% more hospital days, and 23.8% more physician visits (all P<0.001). In addition, higher "centrality" of primary care providers within these hospital networks was associated with 14.7% fewer medical specialist visits (P<0.001) and lower spending on imaging and tests (-9.2% and -12.9% for 1 SD increase in centrality, P<0.001).

Conclusions: Hospital-based physician network structure has a significant relationship with an institution's care patterns for their patients. Hospitals with doctors who have higher numbers of connections have higher costs and more intensive care, and hospitals with primary care-centered networks have lower costs and care intensity.

Figure 1. Schematic of Network Identification and Measurement Methods

Figure 1A illustrates the observed connections between a hypothetical group of physicians and patients (denoted by solid lines connecting patients to physicians), the first step in creating a physician network. In Figure 1B, the group of patient-physician ties depicted in Figure 1A are transformed into a physician-physician network, where physicians are linked by a tie with strength equal to the number of patients they share. For instance, in Figs. 1A and 1B, physicians A and C share 1 patient, and physicians C and D share 2 patients. Also depicted in Fig. 1B is that physicians are assigned to hospitals, and their networks can extend within and outside their affiliated hospitals. Figure 1C shows a simple network to illustrate the network measures used in this study. Each circle represents a physician and is colored by its specialty. The size of each circle is proportional to its betweenness centrality within hospital X (which by definition is 0 for physicians in hospitals Y and Z). The betweenness centrality measures the centrality of a physician in her network by quantifying how often the physician functions as an intermediary in the shortest paths connecting each physician to every other physician (see Appendix for details). Degree: The adjusted degree measures how many colleagues physicians share patients with, adjusted for the number of patients they treat. In Fig. 1C, Doctor A has a degree of 4 because she is connected to 4 other physicians. Doctor B has a degree of 6. If Doctor A treated 10 Medicare patients in 2006, her adjusted degree would be 40 colleagues per 100 shared patients. Similarly, if Doctor B also treated 10 Medicare patients in 2006, her adjusted degree would be 60 colleagues per 100 shared patients. Relative Centrality: Relative centrality measures on average, how central one group of physicians are compared to another group within a hospital network by taking the ratio of the average betweenness centrality of, for example, PCPs, versus other physicians in a hospital. In Hospital X, the PCP relative centrality is approximately 2.0, meaning that the PCPs in this network are on average about twice as central as other physicians. This is visually apparent when looking at the betweenness centrality of physicians, represented by the size of physicians’ circles in Fig 1C, where PCPs appear to have a larger overall share of the centrality than other physicians. The concept of relative centrality is illustrated with larger, real hospital networks in Fig. 2

Figure 2. Example Hospital Networks: Illustrating Relative Centrality

Three example hospital networks from the dataset in this study are depicted in Figures 2A–C. Each point represents a physician, colored by the specialty of that physician (red = primary care, orange = medical specialist, green = surgical specialist [including general surgeons], blue = other specialist). Each tie between two physicians represents the sharing of 5 or more patients. This figure depicts both the complex organization of physicians in different hospitals, but also visually demonstrates how the concept of relative centrality reflects changes in physician networks. In Figure 2A, the PCP relative centrality (explained in Fig. 1C) is 0.35, so PCPs are about a third as central as other physicians in this network. This is reflected by the tight group of medical and surgical specialists at the center of this physician network, with the many PCPs in the network pushed to the periphery of this network. In Fig. 2B and 2C, PCPs move more toward the center of patient-sharing exchanges as the PCP relative centrality grows.

Figure 3. Adjusted Estimates of Hospital Network Structure vs. Cost and Utilization Outcomes

Each section represents the estimated effect of increasing a network measure (Fig 1A, median adjusted degree, Fig 1B, PCP relative centrality) by one standard deviation (SD) for the average-sized urban, non-profit, non-teaching hospital in our sample on three different cost, hospital day, and physician visit outcomes. All estimates are adjusted for several hospital characteristics described in the Methods, including hospital size, urban/rural location and case mix. Error bars show 95% confidence intervals for each estimate. *5 hospitals had missing data for the general medical/surgical and ICU hospital days outcomes, but did have data for the total hospital days outcome