Technology Diffusion in Health Care

10/06/2011
Featured in print Bulletin on Aging & Health

Health care spending in the U.S. accounts for 16 percent of GDP, a much larger share than in other developed countries. Is this money well spent? On the one hand, several studies have concluded that increases in health care spending over time have yielded dramatic increases in life expectancy. Yet other studies have found that there are marked differences in spending across hospitals and regions and that higher spending is not associated with better outcomes. These latter studies suggest that there are inefficiencies in the U.S. health care system and that the U.S. may be on the "flat of the curve" of the medical production function.

In Technology Diffusion and Productivity Growth in Health Care (NBER Working Paper 14865), researchers Jonathan Skinner and Douglas Staiger examine whether differential adoption of new technologies across hospitals can help to explain empirical patterns in health outcomes. The authors draw on macroeconomic models of technology diffusion and productivity, in which small differences in the propensity to adopt technology can lead to wide and persistent productivity differences across countries, or here, hospitals.

The authors focus on the treatment of heart attacks (acute myocardial infarction, or AMI). They examine three specific treatments: aspirin, beta blockers, and the restoration of blood flow to the heart muscles, which may be accomplished with "clot-busting" drugs or surgical angioplasty. All of these treatments are relatively inexpensive, have been shown to be effective in saving lives, and are administered based on the decision of the physician rather than the patient.

A key variable of interest is the diffusion rate, which measures the rate at which a new technology is applied to eligible patients. In the authors' model, the rate of diffusion of a given technology at a particular hospital will depend on the length of time the technology has been available and on a "common factor" capturing the hospital's intensity of search for new innovations in general. The authors use Medicare data on 2.8 million patients who experienced heart attacks from 1986 to 2004 to test the theoretical implications of their model.

The authors' first key finding is that the diffusion rate for each of the three technologies is strongly correlated with the common factor, suggesting that hospitals that adopt one innovation early are also more likely to adopt other innovations as well. Hospitals with quicker adoption tend to be major teaching hospitals, to have higher patient volume, and to be located in states with higher average income. These hospitals may find it easier to adopt new technologies and place more value on early adoption.

The authors also show that differences in the rate of technology adoption lead to meaningful differences in treatment patterns. The use of beta blockers varies from 65 percent among hospitals in the highest quintile of the common factor (fastest adopting hospitals) to only 31 percent among hospitals in the lowest quintile. Similarly, aspirin use varies from 90 percent in the highest quintile to 65 percent in the lowest quintile.

Critically, differences in technology adoption also lead to large differences in health outcomes. The authors project that survival rates in the fastest adopting hospitals are 3.3 percentage points higher than in the slowest adopting hospitals, an amount that is equivalent to one-third of the total improvement in survival rates over the past two decades. Interestingly, the authors' model predicts that there will not be convergence in hospitals' diffusion rates or survival rates over time, hypotheses that are supported by the data.

Finally, the authors compare the effectiveness of speeding the rate of technology adoption vs. adding more traditional health care inputs for improving survival. They find that a one-standard deviation increase in the diffusion rate has the same effect as doubling traditional inputs. They also use this framework to reassess the cost-effectiveness of health care for heart attack patients. When they fail to control for technology adoption, the authors estimate that it takes $355,000 of additional health care spending to generate one year of life saved, suggesting that the U.S. is on or near the "flat of the curve." However, when they include controls for each hospital's technology level, the cost per life-year saved falls to under $100,000, a more favorable cost-effectiveness ratio. The authors also show that hospitals with faster technology adoption should and do choose lower levels of other health care inputs, because the returns to traditional inputs are lower once technologies like aspirin and beta blockers have been adopted.

The authors' model of health care productivity can reconcile both the dramatic increases in life expectancy for AMI patients over time and the "flat of the curve" inefficiencies at a given point of time. The authors argue that the dramatic growth in survival over the past several decades was largely due to the diffusion of inexpensive and highly effective treatments, while the apparent point-in-time inefficiencies may result from a failure to control for hospital-specific rates of technology adoption. The authors caution, however, that their estimates are sensitive to the specification of the model and do not necessarily apply for the treatment of diseases other than heart attacks.

The authors conclude by considering the puzzle of why hospitals and physicians don't adopt new technologies more quickly, particularly a technology like aspirin that is so inexpensive and would appear to require little physician training to implement. They suggest that physicians historically have faced little pressure to change old habits due to the notoriously imperfect nature of health care markets. However, public reporting of technology use by hospitals, as was recently instituted in the case of beta blockers, may help to speed the pace of technology adoption and reduce inefficiencies in the provision of health care.


The authors thank the National Institute on Aging (P01-AG19783) and the Robert Wood Johnson foundation for financial support.