In Tuesday’s Wall Street Journal Op-Ed pages, physicians from Harvard and University Pennsylvania Medical Schools criticize subsidies for expanding the use of health information technology (HIT). The physicians cite a recent review article that failed to find consistent evidence of cost savings associated with HIT adoption. If true, this is bad news for the health economy, as supporters claim that HIT could cut health spending by as much as $1 trillion over the next decade.
How can something that is so avidly supported by most health policy analysts have such a poor track record in practice? In a new NBER working paper by myself, Avi Goldfarb, Chris Forman, and Shane Greenstein, we label this the “Trillion Dollar Conundrum.” One explanation may be that most HIT studies examine basic technologies such as clinical data repositories, while most of the buzz about HIT focuses on advanced technologies such as Computerized Physician Order Entry. In our paper, we offer a rather different explanation for the conundrum, one that would have eluded physicians and other health services researchers who failed to consider the management side of HIT.
My coauthors on this paper are experts on business information technology. They are not health services researchers. When I approached them to work on this topic, they insisted on viewing HIT much as one would view any business process innovation. As I have learned, this is by far the best way to study most any issue in healthcare management. Those who advocate that “healthcare is unique” – usually by ignoring broadly applicable theories and methodologies—often strain to explain data that are easily understood using more general frameworks. Such is the case with HIT.
Health services researchers have analyzed HIT much as they would analyze a new medical intervention. Some patients receive the treatment, others receive a placebo, and the treatment is deemed “successful” if the treatment group fares better than the control group and the difference passes statistical muster. While this methodology inspires a certain level of confidence in medicine, it has a critical shortcoming that has only recently been addressed through “personalized medicine.” The intervention might be effective for only some of the treatment group, and might be harmful to others. The typical research design masks these heterogeneous effects.
Our study articulates why we would expect heterogeneous effects of HIT and finds strong supporting evidence in the data. The key is to view HIT as a business process innovation. Like other such innovations, successful implementation requires complementary human capital. In other words, HIT does not operate itself. Skilled individuals must install it and train hospital personnel on how to use it. Hospital personnel must learn how to use the software and how to adapt it to their idiosyncratic needs. Not surprisingly, some individuals are better at this than others.
We argue that complementary human capital is most abundant in areas where there is a strong general IT presence – think the Bay Area or Seattle. Thus, El Camino Hospital near Palo Alto was an early and successful adopter of HIT. At the same time, hospitals located near the headquarters of major HIT firms, are more likely to get better vendor support. Thus, hospitals in Milwaukee have been very successful with the Epic system. (Epic is located near Madison.) Finally, hospitals with experience with primitive HIT are likely better prepared to take advantage of advanced HIT.
We find strong evidence that human capital is vitally important to the success of HIT. We find that hospitals adopting advanced HIT experience, on average a 1-2 percent increase in costs (including amortized HIT costs.) But this masks heterogeneous effects. Hospitals lacking complementary human capital see their costs increase by 2-4 percent, while those with complementary human capital enjoy cost savings of 2-4 percent. All of these findings are statistically significant.
The most exciting thing about these findings is that complementary human capital is not static. All of us are improving our general IT capabilities just by using our smart phones and the like. Hospital staff will, over time, improve their HIT-specific human capital. The benefits of HIT enjoyed by hospitals fortunate enough to have complementary human capital will almost surely spread to most hospitals. We would be foolhardy to promise $1 trillion in savings, but we do expect substantial savings. It is far more foolhardy to claim that the tepid average performance to date is the end of the story.
Postscript: Next week I will attend the Annual Health Economics Conference at Stanford. I have the privilege of discussing a new paper by Jeff McCullough, Steve Parente, and Bob Town that studies the impact of HIT on outcomes. Most studies have failed to find any benefits for the average patient. After interviewing many providers, these authors conclude that the benefits are likely concentrated on patients whose care requires substantial coordination and information transfer. And they find this — HIT improves outcomes (measured as mortality in this paper) for the minority of patients likely to enjoy the benefits of HIT but are otherwise unchanged. As with my own paper, this study shows that HIT is complex and the benefits are likely to be heterogeneous. I do not believe it is a coincidence that both of these studies were conducted by management professors.