|Home | About | Journals | Submit | Contact Us | Français|
The passage of the Affordable Care Act of 2010 has created a real opportunity to advance delivery and payment system reform in the United States. The need for reform is widely recognized, not only because of the now well-established consensus that uneven quality and poorly coordinated care are endemic (IOM 2001) but also because of the growing concern that continued health care spending growth will exacerbate the federal deficit, potentially reducing U.S. Treasuries to junk bond status.
The article by Jeffrey Silber et al. (2010) in this issue of Health Services Research can contribute to a better understanding of how to improve quality and reduce costs in U.S. health care. They use data on Medicare beneficiaries undergoing major vascular, orthopedic, and general surgical procedures at U.S. acute care hospitals. Their primary results are that when a Dartmouth measure of end-of-life hospital resource use—or as they call it, “aggressiveness”—is higher, 30-day surgical mortality rates and the relative risk of failure to rescue (having a complication and dying) were lower. In other words, hospitals spending more on their chronically ill patients near death also experienced better outcomes among surgical patients.
There is much to admire in the study. The quality of the statistical analysis is very high and they do a commendable job of risk adjustment at the individual level. The failure-to-rescue quality measure is clinically meaningful and well validated. And we certainly approve of their using the Dartmouth end-of-life hospital measures.
Our major concern is that the study's results will be interpreted incorrectly. For example, their finding of a modest positive correlation between end-of-life health care intensity and surgical outcomes could be viewed—wrongly—as meaning that any cuts in health care costs must inevitably lead to worse outcomes, or that these results are somehow inconsistent with those found in earlier Dartmouth research. Instead, the conclusions we draw from this and other studies is (a) it is not how much you spend that is important, but how you spend it; (b) overall spending alone explains remarkably little in health outcomes, implying that most hospitals can cut costs without any loss in quality of care; and (c) there is a tremendous degree of inefficiency, in terms of lost lives and wasted dollars, in the U.S. health care system. We believe that these three messages in turn hold important lessons for designing and implementing real health care reform in the United States.
For many years, the widely held assumption in U.S. health care was that more medical care would lead inexorably to better outcomes. The benefits might be smaller and smaller for each increment of care, but benefits there would be. The public and many policy makers concluded therefore that reductions in utilization would inevitably lead to harm. This belief presented a serious barrier to health care reform and to efforts to slow spending growth. The marked variations in price and illness adjusted per capita spending across U.S. regions (Skinner and Fisher 1997)—and the twofold differences in long-term rates of readmission for patients with conditions such as hip fractures or heart attacks across hospitals (Fisher et al. 1994), provided motivation for research that began to look at the relationship between the quantity of medical care provided overall to similar patients, the quality of care they received, and their health outcomes.
Several early studies, based on data from the mid-1990s, tested a simple null hypothesis—is more always better?—in the Medicare population. The studies rejected this null hypothesis by demonstrating that neither quality nor long-term outcomes were better in higher intensity U.S. regions (Fisher et al. 2003a,b; Skinner, Fisher, and Wennberg 2005;) or in high-intensity compared with low-intensity academic medical centers (Fisher et al. 2004). Nor did regions with greater growth in spending for patients with acute myocardial infarction experience better improvement in survival than in those with slower growth in spending (Skinner, Staiger, and Fischer 2006).
Higher end-of-life intensity was largely explained by differences in the use of the hospital as a site of care (those in the highest intensity regions spent over 60 percent more time in the hospital than those in the lowest spending regions, after adjusting for case mix) and in the use of specialists, imaging services, and diagnostic tests. These studies suggested that there was at least the potential for reducing costs within the U.S. health care system—largely through reductions in the use of the hospital as a site of care.
While important at the time, this hypothesis is too simplistic. Studies comparing costs and outcomes of health care systems cannot capture the causal pathways observed in clinical trials that examine the (sometimes) beneficial effects of specific interventions. Not surprisingly, other observational studies have found settings where spending for specific clinical treatments appears to be associated with better outcomes—whether β-blockers, aspirin, and reperfusion for heart attack patients (Skinner and Staiger 2009), higher intensity care for ICU patients in Pennsylvania (Barnato et al. 2010), a greater supply of neonatologists in regions with low supply (Goodman et al. 2002), or intensive care for tourists admitted to emergency rooms in Florida (Doyle 2010). These studies point to plausible clinical mechanisms by which more spending could yield benefit—for example, the greater use of ICUs, mechanical ventilation, hemodialysis, tracheostomy, and feeding tubes in the Barnato study, or the value of diagnostic and treatment intensity for tourists about whom little is known, as in the Doyle study.
How can these studies be reconciled? It is not simply whether more spending is worse (or less spending is better). Nor (as Silber and colleagues appear to suggest) will cutting back on costs necessarily lead to worse outcomes. Instead, overall health outcomes depend on how the money is spent (Weinstein and Skinner 2010).
To illustrate this point, consider Figure 1, with each dot representing the spending and outcome measures for a specific hypothetical hospital. The hollow dots represent hospitals that have invested most heavily in cost-effective treatments (or organized themselves in a way that reduces ineffective care); they outperform the other (solid) hospitals for any given level of spending. It may be the case that by spending more, inefficient hospitals can move from point A (low cost, low quality) to point D (high cost, low quality) with slightly better outcomes (Skinner and Staiger 2009). But the impact of total spending on outcomes is dwarfed by the importance of spending the money right—by reducing fragmentation and minimizing ineffective care—as shown by points B and C corresponding to hospitals (one low cost, the other high cost) that have devoted more resources to effective care.
Figure 1 also illustrates other features of spending and outcomes across U.S. hospitals—wide variation in costs (Fisher et al. 2003a,b; Wennberg, Fisher et al. 2008; Barnato et al. 2010;), large variations in risk-adjusted outcomes (Fisher et al. 2004; Rogoski, Staiber, and Horbar 2004; Thabut et al. 2010;), and a weak correlation between spending and outcomes.
For example, the study by Silber and colleagues implies that increasing end-of-life spending by U.S.$10,000 per decedent is associated with a reduction in 30-day mortality rates of roughly 2.7 per 1,000 patients (95 percent CI, 1.9–3.5). Dr. Silber and colleagues kindly provided us with the 1-year average mortality rate (13.9 percent), which allowed us to calculate the mortality reduction of 1.2 per 1,000 (CI, 0.1–2.4). (An alternative approach using the 30-day and 31–365 day estimates yields 1.7 per 1,000.) These imply that increasing end-of-life spending by U.S.$10,000 per decedent yields an increase in life expectancy of between one-half and 1 day per surgical patient during the year following surgery.
Of course, this estimate does not account for benefits that persist beyond the first year, nor any other benefits associated with end-of-life spending. Furthermore, end-of-life spending is unlikely to represent the most productive use of health care dollars (Temel et al. 2010), and it would therefore likely understate the benefits of targeted spending to reduce the risk of failure-to-rescue for surgical patients.
While Figure 1 is drawn to have a weak positive correlation, as in the Silber and colleagues study, other studies from the Dartmouth group have often found weak negative associations. But the implications for policy are identical whether for a positive or negative correlation: Simply spending more money is a singularly ineffective and wasteful approach to improving quality—and might not even do so. Conversely, as is evidenced by the presence of low-cost but high-quality hospitals (as in Hospital B), cutting costs does not necessarily mean having to cut quality. What is needed is an understanding of mechanisms: It matters what you spend the money on. And these and other studies provide some insights.
The earlier research on regional variations in outcomes suggested that additional spending devoted to treating patients with chronic disease in the hospital as opposed to the outpatient setting does not offer benefits. A recent study finds that hospitals with intensive end-of-life treatment patterns are also far more likely to use feeding tubes in patients with advanced dementia—a practice that is both expensive and clinically inappropriate (Teno et al. 2010). And the Silber and colleagues study, along with another (Ghaferi, Birkmeyer, and Dimick 2009), suggests that effective treatment of patients with complications following surgery is likely to be beneficial. These potential mechanisms can—and indeed should—be understood as hypotheses that deserve to be tested in comparative effectiveness studies intended to help us understand what services yield the greatest benefits and which provide no benefit at all.
Return one last time to Figure 1, which as noted above shows a faint positive correlation, which might even pass a sufficiently lax cost-effectiveness test. Would this mean that the U.S. health care system is deemed to be “efficient”? The answer is a resounding no. That there exist hospitals like point B with consistently better-than-expected performance points to enormous levels of inefficiency, whether in institutions with similar quality but higher costs (point C) or in institutions with equal or greater costs, but worse outcomes (points A and D).
Inefficiency properly measured therefore includes both traditional measures of “excess” spending not justifiable by improved health outcomes but also the less-traditional measure of lost lives—the excess number of patients who are dying whether at low-cost low-quality hospitals like point A, or at high-cost low-quality hospitals like point D. If we were to convert these worse outcomes into dollar terms, relative to a best-case scenario, the combined inefficiency costs would vastly exceed conventional estimates of waste equal to 30 percent of health care spending (McKinsey Global Institute 2008; NEHI 2008; Kelly 2009;).
Several other issues arise in considering recent research on the relationship between spending, quality, and health outcomes—and deserve some comment.
Silber and colleagues have chosen the Dartmouth hospital-level measures of end-of-life spending as their exposure measure. We think this is a good choice, as it appears to accurately reflect health care intensity, and this “look back” measure of spending is very strongly correlated with other measures of intensity such as “look forward” expenditures for congestive heart failure and heart attack patients (Ong et al. 2009; Skinner, Staiger, and Fisher 2010;).
The only downside of using this exposure measure is that it never fails to confuse the lay press. For example, in one newspaper column, a former Lieutenant-Governor of New York unveiled the Achilles heel of Dartmouth research: that of course spending more on end-of-life patients yields no health benefits, because they are all dead by the end of the year (McCaughey 2009). We hope that the press is better able to understand this study showing that spending more money on decedents is associated with improved survival.
A shift toward bundled payment, whether for specific episodes or total per-capita costs, will require some form of risk adjustment to ensure that those caring for sicker populations are not penalized and to minimize the incentive to avoid such patients. But there is a tension in Medicare claims data between sins of omission—of not adequately risk adjusting for sicker or more complicated patients—and sins of commission, where aggressive hospitals are more likely to diagnosis disease in otherwise similar patients (Song et al. 2010). In the research and in the public reporting context, the use of diagnosis-based risk adjusters in regression models can lead to substantial biases in risk estimates: If higher-intensity hospitals diagnose more diseases in the months leading up to the surgery—as the Song and colleagues study suggests—they will appear to provide better risk-adjusted outcomes and at lower risk-adjusted costs. Use of these risk adjusters could bias research papers toward finding less regional variation in spending or intensity (MedPAC 2009), and that greater intensity leads to better outcomes. If these risk adjusters are adopted for public reporting, they could further lead patients to choose hospitals with worse outcomes than are actually the case. Further work on risk adjustment will clearly be needed.
For policy makers, the media, and even for some investigators, there is a temptation to overinterpret studies relating spending and outcomes. Some might hope to conclude from some of our studies that to get better quality, we need only cut spending or reimbursement rates in high-cost regions or health systems. Others interpret positive associations between intensity and mortality as causal and conclude that lower intensity would cause higher mortality. We disagree with these interpretations. As noted above, the factors that are causing higher spending are typically different from the factors that are causing better outcomes. Thus, a negative association between spending and quality does not necessarily mean that all spending yields worse outcomes (Weinstein and Skinner 2010). But a negative correlation, or even a weak positive correlation as was found by the Silber and colleagues study, is the proverbial canary in the coal mine: a strong warning of pervasive inefficiency with regard both to lost lives and wasted dollars.
The Affordable Care Act sets in motion a broad array of new initiatives, with expanded support comparative effective research and a major new program to develop and test new models of care and payment. Remarkably, the Secretary of Health and Human Services is authorized to adopt as national policy those innovations that demonstrably improve quality (while holding costs constant), reduce costs (while maintaining quality), or both. The opportunity for the health services research community to contribute to the development of national health policy is perhaps unprecedented. The difficult challenge is to establish and measure what works so well for highly efficient health care institutions and what does not work in less efficient systems.
Joint Acknowledgment/Disclosure Statement: This work was primarily supported through a grant from the National Institute of Aging (PO1-AG19783). Drs. Fisher and Skinner are also associated with the Dartmouth Atlas Project, which has received support from a coalition of funders, including the Robert Wood Johnson Foundation (lead funder), the California Health Care Foundation, Wellpoint Foundation, Aetna Foundation, and the United Health Foundation.
Additional supporting information may be found in the online version of this article:
Appendix SA1: Author Matrix.
Please note: Wiley-Blackwell is not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.