shows different levels of efficiency (as measured in QALYs gained per expenditure) in two hypothetical health care delivery systems.18
The purple curve represents an efficient system; along this curve, higher expenditure leads to better outcomes.19
Each segment of the curve represents either the addition of an intervention for a particular clinical condition or the substitution of an intervention that is more effective but more expensive for an intervention that is less expensive but less effective. Each slope represents value for money — the incremental gain in health per dollar spent (steeper is better) — and the length of the segment reflects the degree to which the service is utilized. The incremental value of an addition or substitution to the mix of services often depends on what other services are already in the mix; for example, the added value of coronary catheterization may be lower when the hospital is already providing appropriate medical management for patients with acute myocardial infarction.20
Levels of Efficiency in Allocating Health Care Resources
The blue curve in represents an inefficient health delivery system, in which some cost-effective health services are underutilized. For example, the health service that is represented by segment L′ in the inefficient system is shorter than the segment L in the efficient system. As examples, antihypertensive treatment, screening for colorectal cancer, and counseling for smoking cessation are all underutilized in the United States.21,22
In both the more efficient system and the less efficient system, it is possible to increase expenditures to the point where they yield little or no health improvement, which is sometimes referred to as “flat-of-the-curve” medicine.23
In , the efficient system spends less but obtains better health outcomes at point X than the inefficient system at point Y, even though the latter is doing nothing that is overtly harmful. The inefficient system does worse because it falls short in utilizing beneficial health practices — errors of omission, not commission. At point X on the curve, the efficient system forgoes some potentially valuable spending (between X and C) but still ends up at a higher level of health than at Y.
At a point in time, any statistical comparison between X and Y will show a negative association between spending and health outcomes, as shown by the dashed green line between the two curves. Indeed, one might conclude from cross-sectional data that spending more harms individual patients. In the case shown in , however, cutting spending in health system Y back to the level of system X would result in worse outcomes. However, system Y could save money without sacrificing health by shifting up to the higher curve — substituting more cost-effective health services for less cost-effective health services.
displays the relationship between expenditures and outcomes among various providers (shown by the circles) in managing a hypothetical condition. This picture, which is consistent with empirical data from a variety of studies, shows little or no association between spending and health outcomes. One might find in such a seemingly random scatter plot either a positive or negative (or no) association, but the wide variations in both expenditures and outcomes suggest that factors other than spending itself must be affecting outcomes. In particular, this kind of pattern can be generated by wide variations in the use of interventions that are cost-effective (good value for the money) and those that are cost-ineffective (flat-of-the-curve interventions). Differences in the skill levels of providers or administrative efficiency might also contribute to the variation.
Expenditures and Health Outcomes According to Various Production Functions
To provide some structure to the scatter plot, we can distinguish between two types of providers: those who are close to the more cost-effective curve, F1 (blue circles), and others who are close to the less cost-effective curve, F2 (green circles). Some of the differences in outcomes, at any level of spending, arise because some providers utilize a more cost-effective mix of services than others.
Empirical support for comes from a recent study24
that used chart-review data from the 1994–1995 Cooperative Cardiovascular Project to categorize hospitals as either high-adopting facilities or low-adopting facilities, according to their rates of use of aspirin, beta-blockers, and coronary reperfusion in the treatment of acute myocardial infarction. The researchers found that the high-adopting hospitals had consistently better rates of risk-adjusted survival, at no additional cost to Medicare. But after stratification according to the hospitals' adoption rates, there was a positive but diminishing effect of spending on the health outcome (12-month survival), similar to curves F1
in . The cost-effectiveness ratios at the margin were $95,000 per life-year or more but with slightly better returns for the hospitals that were slower to adopt cost-effective practices (curve F2
Another study showed that regions that had high rates of revascularization for patients with acute myocardial infarction received good health value for the expenditure on the intervention.25
Despite this, there was essentially a zero association between spending and outcomes across regions. The explanation is that the high-revascularization areas were also less likely to use beta-blockers and aspirin for their patients. Thus, like the region represented by point Y in they started out with poorer outcomes because of their failure to adopt cost-effective interventions, but they caught up to the other regions by devoting more resources to invasive interventions that were more expensive yet reasonably cost-effective.
A third example comes from a study of colon-cancer treatment, in which once again there was no association between overall spending and outcomes.26
The authors found greater use of inappropriate chemotherapy in high-spending areas, where adverse effects shifted the providers down from outputs that were more efficient (F1
) to those that were less efficient (F2
). However, these providers were also more likely to use highly appropriate (yet expensive) adjuvant chemotherapy, which would move them along and up the F2
curve, resulting in no difference in overall outcomes from the lower-intensity areas, despite the higher cost.
As illustrates, the apparent lack of correlation at the aggregate level between health spending and outcomes does not disprove the existence of a positive association within a hospital or other health care delivery organization. Moreover, a negative association does not imply that more spending is harmful. That said, there are undoubtedly treatments — such as testing of prostate-specific antigen levels in older men with limited life expectancy,27
arthroscopic surgery for osteoarthritis of the knee,28
and ventricular reconstruction surgery29
— that offer no measurable value and could be scaled back without compromising outcomes. And there are many more therapies for which the comparative effectiveness is unknown.