Via djinn and juice:
It’s all bike safety today! New studies and old on Healthcare Triage News:
Here’s some further reading:
- Effects of bicycle helmet laws on children’s injuries
- Safety effects of permanent running lights for bicycles: A controlled experiment
- Bicycle Injury Interventions: Bicycle Helmet Effectiveness
As we’ve highlighted the suppression of substance use related Medicare and Medicaid research data, people have asked how they can help rectify the situation. The answer, from AcademyHealth:
Please share stories of research findings and evidence-based, improvements in care that would not have been possible, or will not be possible in the future, without these identifiable data at email@example.com! We will use these examples as we make a compelling case to the administration and other policymakers for these data’s release.
We must gather compelling evidence that the suppressed data is essential. If you’re aware of key studies that led to important findings affecting policy or practice, send them to the address above. Please make this easy for AcademyHealth by summarizing the work and its import — the more specific the better.
Go read the rest of the post on AcademyHealth’s blog.
I was perusing this RWJF sponsored report, “Outbreaks: Protecting Americans from Infectious Diseases”:
The Ebola outbreak has been a major wake-up call to the United States — highlighting serious gaps in the country’s ability to manage severe disease outbreaks and contain their spread.
It is alarming that many of the most basic infection disease controls failed when tested. After more than a decade of focus on preparing for public health emergencies in the wake of the September 11 and anthrax tragedies, there have been troubling errors, lapses and scrambles to recreate practices and policies that were supposed to have been long considered and well established.
And that’s just the first paragraph. More than a third of seniors and most adults don’t get all recommended vaccinations. Thirty-six states couldn’t vaccinate even half their population against the flu. Fifteen states didn’t even get 90% of kids vaccinated to Hepatitis B. Still, one out of 25 hospitalizations results in a hospital-acquired infection. Forty states failed to reduce the number of central line-associated bloodstream infections in recent years. New HIV infections in certain populations are still increasing. Three million baby boomers have hepatitis C, and most of them don’t know it.
And about 48 million people get a foodborne illness each year.
I got all of that from just the key findings on the lower half of page 5. The document is 112 pages long. There are some decent charts, like this one:
We can do better.
The following was originally posted on the AcademyHealth blog earlier this year.
Long understood by some clinicians and policymakers, there is growing awareness of America’s opiate use crisis, and renewed efforts to do something about it.
A few statistics serve to make the case that use of opioids is a serious problem in the U.S.: Opioid use is now the leading cause of substance use-related ambulatory visits. The number of fatal drug overdoses doubled between 1999 and 2010; deaths from opioid pain relievers account for most of that increase, quadrupling between 1999 and 2010. Overdose deaths are the leading cause of accidental death in the U.S., exceeding those from motor vehicles. As if loss of life weren’t alarming enough, abuse of opioids costs a lot of money too: prescription opioid abuse costs were about $72.5 billion in 2007, and patients abusing opioid analgesics cost insurers about $14,000 more than average patients between 1998-2002.
(Source: Nora Volkow et al.)
Of course, my state of residence—Massachusetts—and its neighbors are not immune to the growing opioid abuse problem. According to the Alanna Durkin, reporting for the Associated Press, in Maine, deaths from heroin overdoses quadrupled from 2011 to 2012; in New Hampshire the number of such deaths doubled from 2012 to 2013; in Vermont they doubled recently as well. New England governors suggested approaches to addressing this growing epidemic. Proposed solutions often include enhancing treatment programs and “hunt[ing] down drug dealers.”
These numbers serve to shock, but they do not address the pervasive misunderstandings of drug use and abuse in general and the opioid problem in particular. An insightful conversation between Harold Pollack and Keith Humphreys illuminated some hard truths. A few highlights:
- Substance use disorders are not confined to Hollywood stars or the poor. Most people with drug problems have jobs; a huge proportion are “nice, middle class people.” They are your neighbors, friends, family members, coworkers, or you.
- Though deaths due to drugs now rival those from AIDS at the peak of its epidemic, unlike the early outbreak of AIDS, substance use is not focused within a “pre-existing community of people. […] It’s spread all over.”
- Detox that isn’t “followed up with careful interventions and monitoring” have relapse rates close to 100%. There are no quick fixes here.
- Most opiate overdose deaths and growth in them can be traced back to prescriptions: “4 in 5 of people today who start using heroin began their opiod addiction on prescription opioids.” Clamping down on the supply of heroin alone will not address the problem.
- U.S. clinicians “write more prescriptions for opiate painkillers each year than there are adults in the United States.” Think about that.
The health system itself plays a large role in the problem, and should be part of any solution. Recent work by Christopher Jones and colleagues, published in JAMA Internal Medicine, shows that while friends and relatives are the predominant, proximate source of opioid analgesics used for nonmedical purposes, about 20% are directly prescribed by physicians, the second most common source (others being stolen or bought from friends or relatives, bought from a drug dealer or stranger, and other). However, almost all opioid analgesics begin with a prescription, even if they’re later transferred by sale or theft.
What can be done? A full answer takes far more space than one post, and I’ll come back to it later (stay tuned). Ways to address the problem include alternative pain management, adjustments to prescribing practices, and expansion of access to a wider range of longer-term addiction treatment programs.
Responding to the overprescribing of opiate-based painkillers, in July 2012 Blue Cross Blue Shield of Massachusetts attempted to change prescribing patterns. It began requiring prior authorization for more than a 30-day supply within a two-month period. After 18 months, the insurer estimates that it cut prescriptions of opiate painkillers by 6.6 million pills.
In addition to attacking the problem on the front-end, there’s another way to save lives from the opioid epidemic. Better access to and protections for use of naloxone—an opioid agonist—could prevent death from overdose. Massachusetts Governor Deval Patrick recently made naloxone (brand name: Narcan) more available to first responders and to families and friends of drug abusers. Other states have already taken similar measures to promote life-saving naloxone use. Maine Governor Paul LePage is considering reversing his opposition to allowing health care providers prescribe naloxone to opioid addicts’ family members and allowing them to deliver it as required.
To better understand what’s known about options to manage opiate dependence, the New England Comparative Effectiveness Public Advisory Council (CEPAC)* will hold a public meeting to consider the relevant evidence and how it applies to New England patients, payers, and providers. I will summarize findings in a future post.
* I am a member of CEPAC. [This meeting has now been held. Details here.]
Here’s a spreadsheet that provides very detailed counts and percentages of suppressed records across Medicare claim types. It’s fairly consistent with what I posted previously, but with more detail. For instance about 4-5% of inpatient claims are suppressed, in general.
Here’s another that provides the same for Medicaid (MAX) files. As I suspected, the damage is greater than for Medicare. Up to 8% of inpatient claims are suppressed. I don’t know why the suppression proportion drops so much in 2011, but so do record counts. I’m guessing the analysis is based on incomplete files.
Remember, all the suppressed records are for claims related to substance use disorder treatment and diagnoses. So, this is far from a random sample. (It wouldn’t be that harmful if it were random.)
The spreadsheets come by way of ResDAC. Also, they use the term “suppression” in them and in the file names. I have mostly avoided that term, but I’m going to stop doing so. If this is what we’re calling it, then let’s all use the same terminology.
This tag pulls up all our posts on this issue. It’s the one you should share with colleagues to bring them fully up to speed.
The Senate Select Committee on Intelligence’s Study of the Central Intelligence Agency ‘s Detention and Interrogation Program showed in ghastly detail that physicians and psychologists participated in the torture of prisoners. In response, I wrote a post documenting how the ethical codes of physicians and psychologists unanimously and absolutely forbid participation in torture. Some readers responded with arguments for why torture is often or sometimes justified, which is the view held by a majority of Americans. In this post, I’ll try to explain why medicine and (my profession) psychology must take an absolute stand against participation in torture.
In response to the Senate Report, the President of the AMA, Dr. Robert Wah, wrote that
the AMA’s Code of Medical Ethics and the World Medical Association’s Declaration of Tokyo… forcefully state medicine’s opposition to torture or coercive interrogation and prohibit physician participation in such activities. The physician’s most important role is that of healer, and that role is seriously compromised in situations of torture and coercive interrogation.
The key to understanding why physicians and psychologists cannot torture is to understand what Dr. Wah said about the role of a healer. Being a clinician is more than simply having knowledge about treatments for disorders of the body or mind. Healing is also an interpersonal relationship. Francis Peabody wrote
There are moments, of course, in cases of serious illness when you will think solely of the disease and its treatment; but when the corner is turned and the immediate crisis is passed, you must give your attention to the patient… The good physician knows his patients through and through, and his knowledge is bought dearly. Time, sympathy and understanding must be lavishly dispensed, but the reward is to be found in that personal bond which forms the greatest satisfaction of the practice of medicine. One of the essential qualities of the clinician is interest in humanity, for the secret of the care of the patient is in caring for the patient.
As patients, we have strong expectations about the character and motivations of the doctors who treat us. We expect them to be devoted to relieving our suffering, that they will respect our rights to control our bodies, and that our healers will not exploit us to their own advantage. Finally, we want our healers to treat us in ways that accord with our values, even when the healer does not share our values.
To be a healer is not just to master a therapeutic technology. It’s also a matter of character, of your disposition to behave virtuously. Healers are compassionate. They respect patients’ autonomy and are zealous for their wellbeing. Becoming a healer should involve not just learning technical skills, but also cultivating your emotional being so that you care for patients in the proper way.
Torture is the antithesis of healing. To torture is to deliberately cause unendurable suffering to either extract information or – let’s be honest about it – to terrify others in the victim’s community. Rather than respecting the autonomy of the patient, torture is an assault on a defenseless prisoner. This assault may be inescapable. When the purpose of torture is to extract information, the prisoner who does not have that information has no recourse: how can you prove that you do not know what you do not know? Rather than valuing the patient as an end in himself, torture completely subordinates the wellbeing of the prisoner to the ends of the torturer.
We cultivate a virtuous character by practicing virtue. Conversely, practices like torture build a vicious character incompatible with the role of a healer. We see this in the histories of the Nazi doctors, the Stanford Prison Experiment, and everyone’s experience of the character of bullies: if you practice brutality and sadism, you may acquire a taste for it. Moreover, those who practice cruelty develop not only the habit but also a skein of rationalizations to justify it.
Torture should therefore be wholly repugnant to members of the healing professions. The public should be able to trust that its healers are, in their guts and at their cores, incapable of this behavior. Leaders in medicine and psychology recognize that torture is not just wrong but also morally corrupting. This is why medical and psychological societies absolutely ban participation in torture.
Richard Smith goes there at the BMJ. “Are some diets “mass murder”?”
Jean Mayer, one of the “greats” of nutrition science, said in 1965, in the colourful language that has characterised arguments over diet, that prescribing a diet restricted in carbohydrates to the public was “the equivalent of mass murder.” Having ploughed my way through five books on diet and some of the key studies to write this article, I’m left with the impression that the same accusation of “mass murder” could be directed at many players in the great diet game. In short, bold policies have been based on fragile science, and the long term results may be terrible.
Attributing disease or mortality to diet is scientifically difficult. Associations are first made through observational studies, but recording exactly what people eat is hard. We eat very varied diets, and maybe over time our diets change. Then converting our diet into components of fat, carbohydrate, protein, and the like is unreliable. So to make a link between diet recorded over a short period of time and diseases and deaths encountered perhaps decades later is inevitably difficult.
Then intervention trials are unreliable. Unlike with a drug trial, where there will be one variable (taking or not taking the drug), trials of diet include more than one variable: for example, a diet of less fat probably means more carbohydrate so as to supply enough energy. Adherence is an important problem in drug trials but a much bigger problem in trials of diets, as people may find it very difficult to follow an unfamiliar diet. Also, the trials are usually short term and rarely include hard outcomes such as cardiovascular events or deaths.
John Ioannidis, the scourge of poor biomedical science, has shown the great unreliability of most studies linking nutrition to disease and mortality, and perhaps we fail to recognise the complexity of relations between diet and disease when we pick out single components, whether it’s total fat, saturated fat, trans fats, sugar, or salt.
He goes on to praise and review Nina Teicholz’s The Big fat Surprise, which I also enjoyed. I will get to blogging about it one day. His piece is worth a read, too.
This post uses charts to explain some concepts underlying economic interpretation of health care geographic variation studies. I learned most of the content from Amitabh Chandra and Jon Skinner. (This paper by Amitabh offers one, insightful review of the basics.) I thanks Garret Johnson for drafts of some of the charts below. Any errors are mine.
Productive and Allocative Efficiency
The only way to begin is with productive and allocative efficiency. For convenience, let’s imagine the health system as composed of geographic regions. (This is not always ideal. At the end, I will consider organizations within regions.) A region is productively efficient if its resources (inputs, spending) are used to produce the most health possible. Allocative efficiency is obtained within the health system if no rearrangement of the same, aggregate resources across regions can increase health.
Different policies are appropriate depending on whether the health system is allocatively inefficient or whether some regions are productively inefficient. What do geographic variations tell us about policy? It depends.
Consider the chart below, from a paper by by Chandra and colleagues. (I’ve discussed this chart before, here and here.) It shows regional health care inputs (e.g., spending) and outputs (survival/quality of life) for regions A-F. A production function — which relates inputs to outputs — for some of the regions is plotted. You can think of the production function as a codification of all the expertise, processes, and technology of a system. The one shown is also labeled “production possibility frontier,” which designates the most health (survival/quality of life) a region can produce for any level of resources (factor inputs).
Regions associated with points A and F are not productively efficient since they’re below the frontier. The policy goal is to improve productivity of A and F to get them up to the frontier (or at least closer). It’s not clear that either adding or subtracting inputs to/from A or F would help in this regard. They need to be made more efficient with whatever level of inputs they are provided.
Regions associated with points other than A or F share the same production function and happen all to be productively efficient (because they’re on the frontier). For those, policy intervention that reallocates inputs is clearly helpful. For the same total inputs, we can increase allocative efficiency by, for example, moving point D up the curve (rightward) and B down the curve (leftward). Because of diminishing marginal returns, for every additional unit of input allocated to D, greater additional survival/quality is achieved than is lost by subtracting a unit from B.
We cannot fully assess and address allocative inefficiency until we’ve addressed variation in productive inefficiency. Job one is to get regions A and F up to the same production function as the other regions.
Now let’s explore what geographic variations can tell us about productive or allocative inefficiency. Suppose all we know is mean and variation (ΔS, as illustrated in the really dumb chart just below) in spending across regions. Does that reveal whether there’s productive vs. allocative inefficiency? (No, but keep reading to see why.)
Let’s add axes (see the chart just below). “Quality” on the vertical axis is anything from process quality measures to hard outcomes (mortality, QALYs). Spending ($) is the horizontal axis. The vertical, dashed lines indicate what we can infer from spending variation ΔS (and mean): regional spending is in the range of values between the vertical lines (or most of it, if we’re talking some number of standard deviations, for instance). For simplicity, let’s imagine there are only two regions, one with spending somewhere on the left dashed line, the other on the right.
What does this imply about production functions? (Keep reading.)
One possibility is that the two regions could be on the same production function, as shown below. In that case, there’s allocative inefficiency because we can improve quality, on average, by moving the left point rightward and the right point leftward by the same amount along the common production function until the two points meet. (Technicality: All production functions are concave down, i.e. with non-positive second derivatives.) There may also be productive inefficiency if the common production function is not the production possibility frontier.
Another possibility is that the two regions could be on different production functions, as shown below. In that case, there’s definitely productive inefficiency: one region’s production function is below another. From spending variation alone, we can neither distinguish the chart immediate below from the one immediately above nor infer whether either region is operating at the production possibility frontier. We cannot conclude whether there is productive or allocative inefficiency. Hence, it’s not clear what the appropriate policy intervention should be.
Now let’s add quality variation (ΔQ) information, as in the chart just below. This lets us locate the regions’ operating points. They’re at the intersection of the dashed, horizontal and vertical lines. Notice that, in general, we still cannot tell whether these points lie on one production function or two. (These are the same locations of points used in the prior two charts.) So, we still don’t know very much about policy.
However, if we imagine we know just a bit more, we can make some progress. Let’s suppose we’ve measured quality in two ways: we’ve measured rates at which the two regions provide what I’ll call “costly never” and “cheap always” services. Here’s what I mean by those:
- “Costly never” = high cost stuff that shouldn’t happen (or should be rare). Example: hospitalization for issues that could have been addressed with ambulatory care (aka, prevention quality indicators or ambulatory care sensitive conditions).
- “Cheap always” = low cost stuff that should always happen (to the right people). Example: annual blood glucose testing for diabetics.
If we find evidence that each of the two health systems provide some “costly never” services while not always providing “cheap always” services and at different rates, that suggests two different production functions.* And, in that case, a reasonable policy goal would be to push the lower production function up towards the higher one, as indicated below. Importantly, it is not evident that it would be a good policy idea to withdraw resources from one of these regions and reallocate them to the other one because even if we know there are two production functions, we don’t know their shapes (which I’ve drawn arbitrarily): we don’t know which is operating in a region of higher marginal quality returns to resources.
Notice that as the lower production function is pushed up (and changes shape, perhaps), variation in some types of quality and spending could each grow or shrink. Degree of variation alone isn’t necessarily informative. But, as noted above, variation in certain kinds of services (e.g., “costly never” and “cheap always”) may be.
Comparing Two Systems
The point that variation alone is not informative about efficiency is the principal takeaway from the recently published study led by Michael McWilliams (for which I was a coauthor). Let’s go through that point in charts.
Imagine two health care systems: system 1 and system 2. (System 1 could be the Medicare and system 2 the VA, as in our study, for example.) The amount of spending variation across regions served in each system, ΔS1 and ΔS2, is illustrated in the chart below. What does this geographic variation in spending tell us about the efficiency of each system? Which system is more efficient, system 1 or system 2? (Not much by itself and can’t tell, but read on.)
Consider also variation in quality produced by each system. We’ll call that ΔQ1 and ΔQ2, which are illustrated by the lines shown along the vertical axis in the chart above. Now which system is more efficient? (In general, we still cannot tell, actually.)
It’s tempting to say that a system with greater variation in spending and/or quality (in this case system 2) is less efficient. Is that right? (Nope.)
Let’s draw in some possible production functions, as in the chart below. Looks like it’s possible for system 2 to be more efficient than system 1, even though variation is larger in the former than the latter. And, even though that’s the case, both systems harbor some inefficiency. We were able to draw these conclusions in our paper based on the kind of quality we measured and relying on prior work. But, in general, variation all by itself isn’t informative about type of and relative efficiency.
Positive and Negative Associations
Are you puzzled by some findings that show a negative association between spending and health while others show a positive association? Do you wonder which results are telling us the truth about the efficiency of our health system? Turns out, those types of results, by themselves, may not be very revealing. This is illustrated in the following chart from a paper by Amitabh Chandra.
Panel A shows regions that are all operating on the positive marginal return parts of their production functions. Yet the spending-health relationship across them is negative. Panel B shows regions all operating on the negative marginal return parts of their production functions. Yet the aggregate spending-health relationship is positive.
The deeper point about this and all of the forgoing discussion is that without a thorough analysis of production functions, we don’t know very much. Some types of geographic variation studies can be a bit revealing, depending on what they analyze, but geographic variation in general (by itself) is not informative about efficiency.
Within vs. Across Regions
One final note: I’ve couched all the above in terms of regions, but studies have shown that variation across providers (e.g., hospitals) within regions is larger than variation across regions. Each region can harbor both more and less efficient providers. This implies that region based policy is too crude. Now go back to the top of the post and read it all again replacing “regions” with “hospitals.”
* If the analysis is done right, meaning controlling for risk-adjusted costs across the two service types. A lot of stuff in this post only makes sense controlling for risk and if “spending” is interpreted as “real resource use.” Fuzzing out some technical details like this has pedagogical value.
Previously, I posted a spreadsheet of codes that I believed CMS was using to select Medicare and Medicaid claims for removal from research data. Today, a colleague sent me more details, in another spreadsheet obtained from General Dynamics Information Technology, which, I’m told, they characterized as “a CMS approved list based off of conversations had with the SAMHSA legal staff.”
I have not digested the full spreadsheet. (Actually, it’s a workbook with seven tabs.) Here’s some of what’s on the first tab, with my bold added:
If no criteria met, then SAMHSA suppress code = 0.
If any treatment code criteria found or primary DGNS [diagnosis] found, then SAMHSA suppress code = 1.
If any treatment code criteria found or any DGNS found, then SAMHSA suppress code = 2.
Option1 (Conservative): Suppress records with SAMHSA suppress code in(1,2).
Option 2 (Less Suppression): Suppress records with SAMHSA suppress code = 1.
In general, we used the conservative option when identifying substance abuse claims for suppression. That is, we exclude the claim for any treatment code criteria found, or primary diagnosis code found, or any diagnosis code found in the extract files of interest.
So, now you know. I’ve opened up comments to this post (for one week) so folks who go over the spreadsheets can weigh in on what they’re seeing, what they think it means or implies, and ask relevant questions.