Physician Networks and End of Life Care in US Hospitals

 

MICHAEL BARNETT

DEPARTMENT OF HEALTH CARE POLICY, HARVARD MEDICAL

 

Decades of research have documented extensive regional variation in health care spending and resource use in the US, particularly in end-of-life care. To explore possible mechanisms of this variation, we examined the relationship between physician patient-sharing networks and health care resource use in 805 US hospitals. Using complete 2006 Medicare claims from all patients residing in 51 hospital referral regions in the US, we constructed a professional network of 86,104 physicians where two physicians were linked if they filed a claim in 2006 for one or more patients in common. After partitioning the large network into 584 separate hospital subnetworks, we examined clinically relevant aspects of hospital network structure and their relationship to hospital performance. After adjustment for several key hospital characteristics: o The 90th percentile of physician degree in a hospital was much more significantly associated with high costs and care intensity than the 50th percentile of degree. o Higher median number of hospital connections per physician was associated with higher costs, total hospital days, total physician visits in last 2 years of life . o Conversely, increased hospital density and primary care physician centrality in hospital subnetworks were associated with lower costs and care intensity. o Network characteristics were not associated with hospice use or hospital quality scores. These results indicate the network structure of patient-sharing in US hospitals appears to have a significant relationship to the total spending and aggressiveness in end-of-life care. The associations observed are consistent with the hypothesis that hospitals with more centralized, primary care focused care have lower cost and less aggressive end-of-life care. This has implications for possible policy levers to improve cost-efficiency in US hospitals.