Editor’s Note: In October, Health Affairs published two papers on factors driving imaging utilization. One paper, by Jacqueline Baras and Laurence Baker, analyzes the relationship between MRI supply and care for fee-for-service Medicare patients with low back pain. It finds that increases in MRI supply are related to higher use of both low back MRI and surgery. The other paper, by Joseph Ladapo and coauthors, focuses on the adoption of 64-slice computed tomography, which can be used to image coronary arteries in search of blockages. It finds that early adoption is related to cardiac patient volume but also to hospital operating margins.

We asked two experts on the use of imaging technology to review and comment on these papers. Jonathan Sunshine’s commentary appears below, and a commentary by Jean Mitchell will follow later. Sunshine is Senior Director for Research at the American College of Radiology-American Roentgen Ray Society and Professor (Adjunct) at the Yale University School of Medicine; however, the opinions expressed below are soley his own. Mitchell is an economist and Professor in the Georgetown Public Policy Institute

In international comparisons, the most prominent facts about U.S. health care are that we spend almost twice as much as other developed countries, but our life expectancy is in the bottom quartile for developed countries.  High-tech imaging often receives much of the blame for the cost side of this unattractive value proposition.

Because I’m commenting on two papers that ultimately grow out of this sort of concern about imaging, and that deal with adoption of imaging technology, it’s useful briefly to consider why U.S. total costs are so high, emphasizing reasons relevant to imaging and to technology adoption.

There are many reasons our health costs are particularly high, some mutually reinforcing:

  • We’re a nation of big spenders.  A paradigm example: before the Great Recession hit, half the “cars” in the U.S. weren’t cars, but behemoths—SUVs, pickups, and minivans officially categorized as “light trucks.”
  • U.S. physicians are trained–particularly in diagnosis, which includes imaging–to do everything that might possibly be useful, rather than to parsimoniously identify the minimum actions reasonably likely to produce the desired result.
  • Resources, including costly high-tech resources, are abundant and widely distributed, facilitating this “do everything” practice style not only in training, but also in community practice.
  • Fear of malpractice lawsuits creates defensive medicine.  While malpractice awards and premiums total only a few cents of the health care dollar, estimates are that as much as 20% of health care consists of defensive medicine.
  • Almost the entire U.S. healthcare system works on fee-for-service (FFS), meaning providers obtain more income if they do more.  In other countries, sizable parts of the system are not FFS.  In particular, big-ticket equipment and facilities are often financed in other ways.
  • Physicians are increasingly going beyond conventional FFS and acquiring financial self-interest in a widening variety of health services, such as ambulatory surgi-centers and imaging.  For example, in Medicare over the past two decades, imaging by cardiologists increased from 6% of total imaging to 25%.  All three forms of cardiac imaging (echocardiography, coronary angiography, and cardiac nuclear medicine) increased at least ninefold, which suggests that diagnostic parsimoniousness received no attention.  However, before we say “gotcha”  to cardiologists, it’s important to add that the death rate from heart disease—the nation’s number one killer–dropped by more than 25% over this period.
  • Medical research, the engine that generates new technology, is more than $40 billion-a-year enterprise.  In contrast, for sorting out the value of research’s creations, we will have only $1 billion annually for comparative effectiveness research—if what’s now talk becomes a reality.
  • Moreover, comparative effectiveness research is likely to have very little clout, as its most (in)famous episode illustrates:  In the mid-1990s, AHRQ conducted a comparative effectiveness study of treatment for back pain.  The most notable finding was that surgery (which is mostly done by orthopedists) and physical therapy (mostly done by nonorthopedists, less expensive, and less invasive) are about equally effective.  The orthopedics community was up in arms and almost succeeded in having AHRQ abolished.

The list could continue, but it’s best to stop here and emphasize that despite popular impressions, inappropriate imaging is not a major factor in the cost story.  Consider the numbers:  imaging costs are 6-8% of total U.S. health care costs (my estimate).  Radiologists, when asked, say about 30% of imaging is inappropriate. That, however, includes situations in which imaging is desirable but an inappropriate imaging exam is ordered.  Assuming, as an upper bound, that 25% of imaging takes place when none should, eliminating inappropriate imaging would cut 25% x 8% = 2% from health care spending.  That’s trivial in explaining why we spend nearly twice as much as other countries.  Similarly, although Medicare has already planned or carried out multiple large cuts in payments for high-tech imaging, and private payers might do likewise, there’s not more than 2-3% of total health care costs to be eliminated.

Imaging—a view from 30,000 feet

Why, then, is imaging often identified as a major villain in high U.S. health costs?  Partly, I suspect, because it captures the public imagination with large, sci-fi-looking machines that produce amazing pictures of the inside of the body.  As well, for those who are a bit more sophisticated in health policy, there are the facts that the price of such a machine is high (though lower, by an order of magnitude or two, than the price of a major piece of military hardware), and imaging has received much publicity as a particularly rapidly growing category of health spending.

In terms of rapidly growing spending, it is important to note that the usual data (mostly found annually in Chapter 2B of MedPAC’s March Report to Congress) include only Medicare physician fee schedule spending, and that generates a major upward bias, as follows:  When Medicare imaging takes place in a non-hospital setting, the whole of it is paid for under the physician fee schedule.  At the hospital, however, the physician fee schedule pays only for the physician’s interpretation of the images (the “professional component”).  Payment for the imaging equipment, technicians, hospital space used, etc. are in Part A or in the hospital outpatient payment system and typically constitute roughly three-quarters of the total.  Thus, an imaging procedure in a physician’s office (or nonhospital imaging center) gets counted for approximately four times as much as the same procedure at a hospital.  (This is true in both dollars and relative value units (RVUs);  MedPAC’s data are in RVUs.)  Imaging, like health care generally, is moving away from the hospital, and each time an imaging procedure moves out of the hospital, the tabulations count it as four times as much imaging.  With approximately 1% of Medicare imaging moving from the hospital each year, the reported rates of growth are some three percentage points a year above the true figure.

Also important to note is that, in recent years, the rate of growth of imaging has much decreased.  Our preliminary estimate is that from 2007 to 2008, per capita imaging growth in Medicare, properly measured, was approximately 3%.

While these caveats are important, it is true that over the past two decades, imaging has grown quite substantially as a percentage of total U.S. health spending.

In evaluating this growth, it should be recognized that, at any time, some aspects of health care are growing faster than others, and this probably indicates their increasing value.  For imaging, a good part of the growth in value consists of displacing other more invasive and more costly forms of diagnosis.  For example, exploratory abdominal surgery has declined by more than half since the mid-1990s, being largely replaced by abdominal CT.  (See Exhibit 1) Similarly, the large majority of biopsies of internal organs are now performed with a percutaneous needle approach, predominantly under imaging guidance. The more invasive and costly open surgical and laparoscopic routes to obtaining biopsy tissue have largely been displaced.

Imaging has grown not only by replacing other forms of health care, but also by expanding what health care does for patients.  Mammography is a prime example.  For at least forty years preceding 1990, the breast cancer death rate was constant or slowly rising. Between the late 1980s and 2000, the mammography use rate among women age forty and older increased from 29% to 70%, largely due to an education and advocacy campaign.  The result:  a falling breast cancer death rate, which has decreased by 30% since 1990.  (See Exhibit 2) Statistical modeling attributes a substantial share of the reduced mortality to increased use of mammography.  (The role of improved treatment, of course, is also large—almost certainly larger.)

Difficulties of conducting and using technology assessment of imaging

Although imaging does not constitute a large part of the total U.S. health cost problem, for imaging, as for all health care, it’s desirable that care alternatives be soundly evaluated and that actual practice be governed by the resulting scientific evidence.  For imaging, there are special difficulties in achieving these objectives:

  • Imaging technology usually comes in the form of equipment—for example 64-slice CT, which can visualize smaller structures and is faster than the predecessor, 16-slice CT.  New imaging equipment typically has potentially dozens, if not hundreds, of applications.  Consequently, the assessment task consists of appraising new equipment not for one patient problem, but for a great many.
  • Imaging technology progresses particularly rapidly.  Thus, in the time from initial planning of a careful multi-institutional trial to publication of its results (say 4-5 years), technology typically has moved a generation ahead, the trial’s results apply to “yesterday’s technology,” and practitioners who are inclined to be up-to-date will already have next-generation equipment and tend to ignore the results.
  • Multiple imaging examinations can be performed on one patient.  Thus, for practice guidelines to be really useful, they can’t merely indicate how appropriate each of a set of examinations is for a given clinical presentation.  They need to indicate which examination to use first and what the decision tree should be for subsequent imaging, if any.
  • Especially with technology progressing rapidly, the radiologist is the expert on what imaging is most appropriate.  However, it’s the treating physician who orders the imaging, and there’s rarely a consultation between the two on what to choose.  The American College of Radiology has formulated appropriateness criteria covering over 800 clinical presentations, and updates them biannually.  But getting treating physicians to use them… “aye, there’s the rub.”
  • Systems that provide electronic order entry for the physician accompanied by decision support (that is, guidelines and other information that pop up in real time) are a potential solution.  But like most activities that might “go electronic” in health care, the practicing physician faces the problem that each hospital—and possibly each payer—is building its own system.  The main result may be that the practicing physician is driven crazy by having to work with a dozen different systems, each with its own software and its own (at least slightly) different practice guidelines.

MRI for low back pain

The foregoing “big picture” observations provide perspectives for thinking about the two papers that are the occasion for this blog.  Comments on the specific content of each paper are also important.

The Baras and Baker paper on MRI for low back pain is a fine example of studies that address the question, “Does supply of health care resources (in this case, high-tech imaging equipment) engender demand?”  This is perpetually a priority question, given the U.S.’s high-cost health system.  And it naturally applies to high tech imaging equipment because of the equipment’s visibility and the substantial revenues that have to be generated if the equipment is to be paid for.

Radiologists’ research on MRI for lower back pain is a delightful classic in technology evaluation:  Lower back MRIs on 98 asymptomatic volunteers showed what are usually regarded as “abnormalities” in two-thirds of the volunteers (MC Jensen et al., New England Journal, 1994). The conclusion:  MRI is of no use in determining if a back pain patient needs surgery; however, it may be useful in indicating where to perform surgery if the need for surgery is otherwise reliably established.

In short, as Baras and Baker note, they are studying an imaging examination that’s very much inappropriate, particularly in the first month of a back pain episode.  (After that, surgery might—or might not—be more reasonable.)  And,  interestingly, they find the proportional effect of equipment availability on utilization is greatest in the first month, when the examination is most clearly inappropriate, and decreases thereafter.

In general, one would expect that the effect of equipment availability on utilization is less, the more appropriate the examination.  In other words, their findings on MRI for low back pain may well represent something of a worst case in terms of supply creating demand.

Moreover, although they use a careful research design, I think there’s a major question as to whether causality underlies the statistical association they find.  The problem is as follows:  Their methodology assumes that demand and supply of MRIs were approximately in balance in each of the 300+ metropolitan areas at the beginning of their study and that changes in demand were only generated by (i) patient aging, changes in the supply of hospital beds and surgeons, etc. and (ii) an increased number of MRI machines.  But overall per capita utilization of MRI by Medicare beneficiaries increased by approximately 75% during their study period, as did the number of MRI machines per million Americans.  With such large increases, it’s reasonable to think that other things were happening—for example, physicians learning more about what MRI can do.  These changes could then generate demand, and the increase in the number of MRI machines could have been a response to this change in demand, not its cause.  In terms of what’s desirable, that sounds better than supply creating demand.  .

Hospital acquisition of 64-slice CT

Ladapo and colleagues address the always important question of how good–or bad—are the reasons for which providers adopt a technology.  Their study of hospitals’ acquisition of 64-slice CT is impressively wide-ranging in the reasons they consider, as well as thoughtful in its methodology.

However, for multiple reasons, it seems to me their paper paints 64-slice CT, and how it is being acquired, in too negative a light. For one, they focus on one application of 64-slice CT–admittedly the most ballyhooed one, namely, coronary CT angiography.  But, as for most equipment-embodied imaging technology, there are dozens of potential applications for 64-slice CT.  Thus, a disappointing performance in this one application is not adequate reason to be generally negative about the equipment.  Moreover, the recent evaluation literature on coronary CT angiography—including their own cost effectiveness study (JA Ladapo et al. American Journal of Roentgenology, 2008) and an extensive meta-analysis (G Mowatt et al, Health Technology Assessment, 2008)—is fairly strongly positive.   Finally, the total number of CT machines per million Americans increased, if anything, more slowly between 2004 (when 64-slice CT first appeared) and 2007 (latest data available) than in 2001-04 (OECD, Health Data 2009).  Thus, hospitals seem to be acquiring 64-slice CT when they were, in any case, going to replace old CT units or add one(s).  This is very different—and much less alarming—behavior than if the advent of 64-slice had set off an expansion boom.

Perhaps more importantly, I have concerns about their two most prominent conclusions.  The paper describes operating margins as a key determinant of whether hospitals acquire 64-slice CT.  Certainly, that sounds highly undesirable from a policy standpoint.  However, a close examination of their regression results shows the finding actually is that the probability of acquisition is similar over most of the spectrum of operating margins with, however, hospitals in the lowest quartile less likely to acquire 64-slice CT than others.  I’d read these results to mean that a bad financial situation keeps a hospital from obtaining the technology, not that the better off a hospital is, the more likely it is to acquire 64-slice CT.  The former is a more reassuring behavior pattern, by a good deal.  That’s especially true given that, as noted above, the literature is fairly positive about the technology.

The other perhaps questionable conclusion is that if hospitals have more cardiac patients per year, they’re more likely to acquire 64-slice CT.  This certainly seems a desirable pattern—at least if cardiac imaging is indeed the main field in which 64-slice CT provides a major advance over 16-slice CT.  However, there may be a problem because the regressions include as variables both the number of heart disease patients and the total number of patients.  Heart disease is a common ailment, in rich neighborhoods as well as poor, so the percentage of a hospital’s patients who have heart disease is probably fairly similar across hospitals (especially because specialty hospitals are excluded).  That means the two variables are probably strongly collinear.  And if the variables are, indeed, highly collinear, the coefficients on either–or both–may be severely misleading.  I’d have liked to see a small analysis dealing with this colinearity possibility.

Finally, the regression shows teaching hospitals are particularly likely to acquire 64-slice CT.  That’s desirable, because they’re where the researchers are who will evaluate its usefulness. However, this desirable finding receives relatively little mention.


In summary, close examination of the two papers suggests that adoption and use of high-tech imaging is probably more rational than the papers describe it to be, particularly for acquisition of 64-slice CT.  On the other hand, one wonders about MRI, for the U.S. has 26 MRI machines per million population and keeps these units fairly busy, while the typical OECD country has 5-10 MRIs per million (OECD, Health Data 2009).