Why Is There a Quality Chasm?

Joseph P. Newhouse


Health Affairs. 2002;21(4) 

In This Article

Why The Poor Performance?

I point to five possible causes of poor performance: consumers' ignorance; the rate of technological change; the role of administered prices; the difficulty in assessing the performance of given provider; and the role of the public sector and objectives other than efficiency.

Although consumers will not necessarily understand the technical details around production of other goods or services any better than they understand the technical details of medical care, they can judge how well they like the ride and performance of car, for example, or the quality of sound from sound system. By contrast, they may not be able to distinguish whether bad medical outcome is attributable to poor-quality care or to the underlying disease. Furthermore, for many acute medical problems there is little or no repeat buying, so consumers may have little experience with their specific problem or provider. For both reasons consumers may continue to use providers or delivery systems that give inferior results rather than gravitating toward those with better results and leaving others to fail.

Three types of evidence suggest that the Darwinian process found in most markets does not operate as ruthlessly in medical care. First, publishing the results of substantial variation in cardiac surgery mortality among New York and Pennsylvania hospitals did not provoke patients to change hospitals.[11] Thus, cardiac surgery patients did not behave like 1970s U.S. automobile buyers, who deserted GM, Ford, and Chrysler for Toyota, Honda, and Nissan. Poorly performing hospitals and their medical staffs often did respond constructively to the information, but their responses appeared more motivated by professional ethics than by any actual loss of business, contrasting sharply with the U.S. automobile industry's near-death experience in the late 1970s.

Second, malpractice rarely causes claim. Although the popular impression is nation wash in malpractice litigation, only small proportion, perhaps 1-2 percent, result in claim.[12] Only one-third of cases with much at stake -- patients under seventy years of age who either died or had disability lasting longer than six months as result of negligence brought -- claim.[13] These patients, or their heirs, were likely leaving substantial monies on the table. Although there could be several explanations, such as unwillingness to sue their physician, many patients may have simply been unaware that their care fell below professional standards and attributed the adverse outcome to the underlying disease.

Third, patients with higher cost sharing, including large deductibles, do not search for lower-price physicians to any greater degree than do patients with no cost sharing.[14] Because the patient with large deductible keeps any money he or she might save by using lower-price physician, this is contrary to the expected behavior in more standard market.

Consumers' ignorance, however, cannot be the sole explanation of poor performance. As Kenneth Arrow emphasized in his seminal paper four decades ago, because of their ignorance, patients rely on physicians to act as their agent, so the issue becomes why some physicians apparently do not carry out this role in exemplary fashion.[15] In fact, there are several reasons.

Technological change in medicine takes many forms, including the development of new devices, drugs, or procedures, as well as the adaptation of existing procedures or drugs for new patients. Many illustrations can be found. Coronary stents exemplify new device. They came onto the market in the 1990s to help prevent restenosis (occlusion) of the coronary artery after angioplasty. Although rarely used in 1994, by 1998 around half the angioplasties in seven of nine areas in various countries used stents.[16]

The pace of new drug development and new procedures is similarly swift. Between 1990 and 2000 nearly 1,000 new drugs were introduced into the U.S. market, and the number of new molecular entities introduced exceeded 300.[17] Just since 1990 the number of cancer drugs in the pipeline has increased from 28 to 402; in 1990 there were six cancer gents in Phase I trials, and today there are 150 200.[18] The use of catheterization to treat elderly heart attack patients in the U.S. Medicare population increased from 11 percent in 1984 to 41 percent in 1991.

New drugs, devices, and procedures are easily recognized as change. A less well recognized form is the learning that physicians, especially surgeons, acquire as they employ new procedures. Proficiency rises with familiarity, and physicians become more willing to perform the procedure on clinically riskier patients.[19] Discovering effective off-label uses of drugs represents analogous learning.

Another indicator of the increase in clinical knowledge (and its perceived value) is the increased resources devoted to clinical trials. In constant 2000 dollars, spending on trials by the National Institutes of Health (NIH) more than doubled between 1990 and 2000, from $875 million to $1.9 billion.[20]

Rapid change makes knowledge quickly obsolete and places heavy burden on mechanisms that enable physicians and other health professionals to keep up. The profession's main formal instrument for keeping current is continuing medical education (CME). However, the usual CME conference has little effect, and more-promising strategies are seldom used.[21] The IOM Quality Chasm report, which also emphasizes the rate of change in knowledge as cause of poor performance, points toward more systems-oriented approach and greater use of information technology to help practitioners cope and to make knowledge diffusion more rapid and more uniform.

Numerous well-known obstacles loom to the successful implementation of such strategy, obstacles that I do not dwell on here. Even ignoring those issues, the rate of change causes more fundamental problem, as emphasized by Barbara McNeil in her recent Shattuck lecture.[22] The problem is illustrated by work of Edward Guadagnoli and his colleagues, who measured compliance with the guidelines for angiography following an cute myocardial infarction (AMI). Their initial study simply gave more evidence of poor performance. When patients had characteristics for which the guidelines held angiography to be "useful and effective" (Class I), only 46 percent actually received the procedure in traditional fee-for-service (FFS) Medicare. At the other extreme, when patients h d characteristics for which the guidelines held angiography to be "ineffective" (Class III), 13 percent of cases received angiography.[23] Greater use of systems and information technology could bring these rates more in line with evidence-based medicine.

However, these remedies cannot address the difficulty posed by the high rate of change for evaluating medical capabilities at any point in time; evidence-based medicine labors under the onslaught of new knowledge. In subsequent study Guadagnoli and his colleagues divided the United States into ninety-five regions and examined the degree to which various categories of AMI patients accounted for variation cross the regions in angiography rates. Most of the variation was not in the Class I and Class III categories just described, but rather in two more ambiguous categories in which the guidelines judged angiography to be either "appropriate, but not necessary" or "uncertain."[24] But it is precisely for the patients in the latter two categories that clinical trial data on the efficacy of angiography are unavailable or unconvincing.

In short, it is not just that some health care providers fail to keep up; at any point in time, some procedures are sufficiently new that their efficacy has not been established for substantial numbers of patients. Larger trials, of course, would permit better measurement of efficacy in subclasses of patients; under-representation of the elderly, women, and children in existing trials are well-known examples. Larger trials are problematic solution, however, because accruing sufficient numbers of patients in various subclasses may delay knowledge for those classes of patients where results are more definitive.[25] Furthermore, by the time any trial is complete, better procedure or drug may have appeared. Or physicians may have become more proficient at the procedure being tested. Either way, the results of the larger, more expensive clinical trial would be obsolete. And any delay to accrue more patients simply increases the chance that the results from the trial will be out of date when they appear.

A more mundane problem that greater use of information technology does not address, although greater use of systems might, is the outmoded specialist. Knowledge accretion results in greater specialization, process that has gone on for centuries in all fields and disciplines. Specialists, however, cannot or do not always retrain to use new methods, in part because the new methods may be the province of another specialty. For example, in an earlier era some of general surgeon's business was gastric ulcer surgery; treatment is now generally by drugs and not by general surgeons. Although surgery for gastric ulcers has essentially disappeared, in other cases the specialist may simply continue to perform an outmoded procedure because it is effective and is what he or she does to earn living. In different settings this is termed featherbedding, reminder that inefficiency is not limited to medical care.

Although rapid rate of technological change surely has something to do with the poor performance of medical care, it cannot be the entire story either, since other industries with rapid technical change exhibit much different performance. As everyone knows, technological change in the semiconductor and computer industries has been rapid; between 1971 and 1999 the number of transistors per chip increased 10,000 times.[26] Between 1974 and 1996 the price of memory chips, adjusted for this phenomenal change in capabilities, decreased by factor of 27,270 times, staggering 41 percent per year.[27] Prices of logic chips, the data for which have been available only since 1985, fell an even greater 54 percent per year in the 1985-1996 period. Although these figures do not directly show that high-defect -- implying high-cost -- producers have not survived, that seems likely.[28] Furthermore, over this period dynamic random access memory (DRAM) and metal-oxide-silicon (MOS) logic chips became commoditized, implying that quality was nearly uniform.[29] Why do these two industries have performance that is so different from medical care?

A critical difference between medical care and many other sectors of the economy, including semiconductors and computers, is the wide-spread use of administered prices to pay medical providers.[30] The need for administered pricing arises because health insurers, whether public or private, cannot agree to reimburse any price provider names. If providers are not to be excluded from reimbursement on the basis of price, as is generally the case with public insurance and was de facto the case in U.S. private insurance for many years, administered prices must be used. In some countries the price may be implicit, as in the case of hospital with fixed budget, but in other countries an explicit price per unit of service is either set by the government or negotiated industry-wide. In the U.S. Medicare system, for example, Congress for the most part simply legislates prices; in Canada and Germany physician fees are negotiated between the profession and public or quasi-public entity.

Because pricing affects providers' behavior, getting the administered prices set at the right level is important.[31] Paying above marginal cost for defined service offers incentives for overly intensive care, and paying less does not elicit supply.[32] But getting the right price is difficult for many reasons. Marginal cost must be estimated from econometric or engineering (time and motion) studies, which are likely to be imprecise. Because it is easier to compute, actual price setting tends to aim at average rather than marginal cost. Estimating average cost requires knowing only total cost and number of services, something that normal accounting practice will reveal and that can be independently audited. And there is an even more basic problem. Costs, both average and marginal, adapt to what is paid.[33] As result, there is mutual causation between observed costs and reimbursement.

Not only the level but also the basis of price matters for quality. Consider hospitals that are paid on the basis of the hospital day or the hospital stay, as most U.S. hospitals are. Suppose that the hospital can purchase medical supply that reduces the incidence of adhesions after surgery and hence the need for readmission. If the hospital purchases the supply, not only will the cost of surgical procedures be higher, but the hospital will lose the revenue it would otherwise have received from the readmission. Similarly, it will lose revenue from the prolonged stays and readmissions caused by fewer medical errors if it invests in computerized drug order entry system.

This problem has led some to advocate paying health care providers partly on the basis of results achieved. Such methods, however, unless adequately standardized for differences in the patient mix of various providers, could promote adverse selection; the provider's incentives, for example, would be to shun the frail or noncompliant patient if reimbursement were tied to raw outcome measures. And techniques for carrying out such standardization are still underdeveloped, another illustration of the difficulty of getting administered prices right. Moreover, if only certain outcomes were rewarded, resources could simply be diverted from areas with unmeasured results, to no overall benefit.

The rate of technological change complicates all administered pricing methods. Costing studies of particular treatments may be quickly obsolete. Trial-and-error pricing, the usual approach, does not function well if techniques change frequently or there is learning by doing. In short, rapid technological change and administered pricing interact to produce poorer performance than either would individually.

Administered pricing also affects performance by affecting the rate of technological change. In standard market, the manufacturer of new product simply offers it at given price, and the market either accepts or rejects it. In the case of medical care, however, the manufacturer of new drug or device must obtain the approval of the relevant regulatory body to market the product at all. After obtaining approval, the manufacturer, to be paid, must convince insurers to cover the product, as must physician employing the new procedure. In FFS systems code for the product or procedure must sometimes be issued by another body to implement coverage decision. All of this introduces delay relative to other markets, which in turn reduces the expected reward for an investment in development. Potentially offsetting the effect of this delay, however, is the profit to be gained if the procedure is covered, because the moral hazard from insurance coverage will induce greater use than in standard market. The net effect on innovation is unclear, but these institutions differentiate health care from other markets.

Insurers, whether private or public, might seem at first blush to offer some help in improving performance. After all, they market directly to purchasers, whether employers or individuals. Why are they not more like hotel chains, which acquire reputation not always favorable for the quality of their lodgings? Part of the reason is reprise on the theme of consumers ignorance. If purchasers do not reward higher-quality health plans, we should not be surprised to see quality problems. (In the context of national health services, reward takes the form of more votes for politicians who support improved quality.) But like providers with an information problem in keeping current, health insurers have an information problem in identifying high-quality providers.

Many manufacturers can readily compute defect rates from alternative suppliers because they typically purchase large quantities of supply made to given specifications. Health plans, however, face the constraint that the sample of patients of any provider is often too small to draw reliable inferences about that provider's performance for particular disease.[34] Plans could, of course, measure the performance of an organized delivery system or aggregation of providers instead, using Health Plan Employer Data and Information Set (HEDIS) or Consumer Assessment of Health Plans (CAHPS)-type measures, but this leaves the possibility of within-system variation in performance.

Implicit in the Quality Chasm report's call for greater use of organized systems is that management of these systems could reduce within-system variance in performance to minimal levels. To do so requires that managers not only be able to measure the performance of providers but also have the incentive to reward the better performers. Both requirements are problematic for reasons already described. It is not obvious that the organized delivery system is much better placed than the plan is to overcome the problems of small samples and difficulty of risk adjustment. Nor is it clear that the marketplace, whether economic or political, will reward plans that restrict choice to better-performing organized systems.

The existing entities with the strongest incentives and tools to minimize cost for given quality -- or equivalently maximize quality for given cost -- are firms that integrate insurance and delivery, such as U.S. group- and staff-model health maintenance organizations (HMOs). But these entities do not appear to have any pronounced superiority (or inferiority) in their quality of care, which suggests that the barriers to good performance are more fundamental than simply the lack of organized systems.[35] This inference is consistent with group-and staff-model HMOs' failure to thrive in the marketplace.[36] In short, although the creation of organized and accountable delivery systems may be necessary condition for improvement, the existing evidence suggests that it is not sufficient.

Nonintegrated insurers in the United States, as well as in Canada, Germany, and Japan, generally defer to physician's judgment about which services should be reimbursed. As result, nonintegrated insurers implicitly delegate the responsibility for quality to physicians and other providers. From the nonintegrated insurer's perspective, deferring to the physician's judgment, although it may raise cost by covering inappropriate or overly intensive care, is consistent with the desire to maintain reputation for product that offers risk protection. As has become clear from the backlash against managed care, the insurer that seeks to inhibit inappropriate care will often incur the wrath of the patient, who generally believes that the physician, not the insurer, is his or her gent. After all, the insurer is corporation or public agency, not professional, and the insurer does not take the Hippocratic oath. And the consumer cannot be certain that the care the insurer seeks to inhibit is in fact inappropriate.[37]

The health care financing systems of all developed countries have an important public-sector component. Because efficiency is not the only goal of such systems, one should not expect the same performance as in standard markets.

Viewed from the local community, health care financed with federal or state taxes is an export good, as is care financed by premiums if premium payers are geographically dispersed. Local legislators will therefore seek to maximize funds coming to providers in their districts. In rural and inner-city areas health care spending may also serve community development purposes; the hospital may be among the largest employers in the local area. Although it may be possible to reduce costs and improve the quality of both medical and defense services by closing either hospital or defense base, both are notoriously difficult to close.

Furthermore, all developed countries regulate entry into the health care professions. The regulations necessarily specify who may perform certain tasks. Such regulations probably inhibit delegating tasks to allied health personnel in cases where delegation would improve performance.[38] Needless to say, the regulations are vigorously defended by those advantaged by them.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.