Tuesday, March 11, 2014

#HealthIT 2014?

The CIO Perspective

Below, a THCB comment:
Jeff Goldsmith: ICD-10 is going to be the next healthcare.gov implementation fiasco. It’s actually going to be worse than healthcare.gov, because the way I read it, physician and hospital incomes basically stop dead if they haven’t converted.

And perhaps 15% of providers actually have this covered. It won’t be eager uninsured folk can’t get a website to work- it will the healthcare equivalent of an ischemic stroke...

I’d be diving for cover now, if I were CMS and the White House. This is a seriously stupid idea.
I guess we'll know before long. W6142XA, Struck by turkey, initial encounter.


Great post by Andy Oram on THCB:
HIMSS Unplugged

HIMSS has opened and closed in Florida and I’m in Boston with snow up to my rectus abdominis. After several years of watching keynote pageants and scarfing up the amenities at HIMSS conferences, I decided to stay home this year...

In general, I’ve found that my attendance at HIMSS leads moaning and carping about the state of health IT. So this year I figured I could sit in my office while moaning and carping about the state of health IT.

In particular, my theme this year is how health IT is outrunning the institutions that need it, and what will happen to those left behind...

Here are the trends in HIT
  • Meaningful Use requires the purchase of electronic health records, which run into the hundreds of thousands of dollars just for licensing fees. Training, maintenance, storage, security, and other costs add even more. The incentive payments from the federal government come nowhere near covering the costs. EHR providers who offer their record systems on the Web (Software as a Service) tend to be cheaper than the older wave of EHRs. Open source solutions also cost much less than proprietary ones, but have made little headway in the US.
  • Hot on the heals of Meaningful Use is ICD-10 compliance, a major upgrade to the diagnostic codes assigned to patient conditions. Training costs (and the inevitable loss of productivity caused by transitions) could be staggering. Some 80% of providers may miss the government’s deadline. The American Medical Association, citing estimated prices for a small practice of $56,639 to $226,105 (p. 2), recently urged the government to back off on requiring ICD-10. Their point of view seems to be that ICD-10 might have benefits, but far less than other things the providers need money for. Having already put off its deadline, the Department refuses to bend further.
  • All providers need sophisticated analytics. This is because both the Centers for Medicare & Medicaid Services and private insurers are moving to bundled payments and “pay for value.” In various configurations, these schemes pay doctors the amount that a payer thinks treatment should cost (in their estimation), not the cost of the actual procedures the doctors performed. Smart institutions that provide well-integrated care and track their patients to ensure adherence to treatment plans can suppress their costs and actually earn more money. But if a provider carries on the way most do–losing track of patients who fail to come for appointments, performing unnecessary tests because they can’t get the results of earlier tests, etc.–its bottom line will suffer. The change is certainly good for health care consumers, but smart treatment also costs money: to collect accurate data, to share it among providers, and to carry out the data crunching that turns up risks among patient populations.
  •  Patients and staff are coming to expect other amenities, such as web portals, that require investment in both technology and workflow changes. All these IT factors are compounded by other rising costs in the health care industry. Even the most powerful institutions are having trouble passing those on to payers and consumers, as they have done year after year. Aside from the “pay for value” programs, Medicare costs are being kept down for budgetary reasons, but the government’s intent is good: they expect providers to learn how to be more efficient...
With respect to his 3rd bullet point, he continues,
The problem is that doing analytics costs a lot of money, given all the steps involved:
  1. Someone must enter the data to be passed to the analysis. Natural language processing systems can look for important indications in the text entered by clinicians, but it’s very hard to track everything that way. Structured data takes a lot more time to enter, and staff must be trained to do so in consistent terms and formats.

  2. Analysts must detect the relationships in the data, a task that can trip them up in health care because it presents so many confounding factors. For example, an analyst might compare two hospitals without statistically adjusting for their relative mix of procedures, the types of patients that frequent each hospital, etc. Just graphing data without looking for such deeper patterns can be very risky. Organizations adopting predictive analytics must do additional work to generate the predictive models.

  3. Analysts must then create charts or reports showing important relationships. Many people have the training to do this, but not necessarily well. “It’s all too easy to generate graphs that look insightful but that do not communicate statistically sound insights,” says to Arijit Sengupta, CEO of BeyondCore.
  4. Business users must take time to review results. This calls for training in how to review data. Often the results presented aren’t exactly the data needed by the business user, who must now provide feedback to the analyst to help her redo the analysis.

  5. The hospital or clinic has to act on results–calling patients, scheduling visits and tests, etc.

"Analytics." Another of those fashionable buzzwords. I worry about the increasing ease with which statistical naifs can wreak time-wasting, aesthetically comforting yet worthless empirical havoc.

Read all of Andy's post, it is excellent.

apropos, my ASQ Quality Progress magazine arrived via snailmail today.

The cover story:
Probing Probabilities
A simple explanation of Rev. Thomas Bayes’ theorem and why it remains relevant
by William Hooper

Let's do a little experiment from an article published in the New England Journal of Medicine in 1978. Suppose you are tested for a rare disease that occurs in the population at about 1%. The test is 95% accurate. This means if you have the disease, the test result is 95% positive. If you get a positive test result, how likely is it that you have the disease?
  • 95%.
  • 85%.
  • 16%.
Sixty doctors at four Harvard Medical School teaching colleges were asked this question. Only 11 answered correctly, with almost half saying 95%. So, what is the correct answer?

Sixteen percent.

When teaching a course in beginning statistics, a student asked me why the answer wasn’t 95%. How could a test result that shows a 95% probability of being correct result in only 16% probability of having a disease?

Welcome to Bayesian statistics, made famous by the Rev. Thomas Bayes (1702-1761). After he died, researchers found the famous Presbyterian minister and part-time mathematician’s formula from a paper he had written. Bayes gets credit for the formula, but it’s connected to the much more famous mathematician, Pierre-Simon Laplace, who popularized it and demonstrated how to use it in the courtroom.

What is Bayes’ formula?
Let’s use the example from the opening question about being tested for a rare disease. Let P(A given B) be the probability you have the disease, given you have been tested positive:

P(A given B) = P(B given A) x P(A)/((P(B given A)XP(A) + P(B given not A) x P(not A)).
Plug some numbers into the formula and work through it.
  • P(B given A) = the more typical posterior statement. Given you have disease (A), what is the probability it will be found? In this case, it will be 95%.
  • P(A) = the probability you have the disease. In this case, the probability is only 1%. So, it is relatively rare (an important point emphasized later).
  • P(B given not A) = the probability you will be diagnosed with the disease, even though you do not have it. That is the 5% (100% − 95%). In quality worlds, this is a false positive.
  • P(not A) = the probability you will not have it. In this case, 100% − 1%, or 99%.
Plugging the numbers into the equation, it works out like this:
P(that you have the disease because you tested positive) = 0.95 x 0.01/(0.95 x 0.01 + 0.05 x 0.99) = 16%.
Strange, isn’t it? According to Bayes’ theorem, if you’ve tested positive for a rare disease, and it turns out positive, don’t panic. In this case, there is only a 16% chance you have the disease. Why?
Because Base Rates Matter.

I've been a Bayesian for a long time. "Frequentism" -- the otherwise eminent Mario Bunge aside (he inexplicably disses Bayes) -- is simply inadequate in many areas of analysis. See my post-9/11 screed on the anti-terror "Total Information Awareness" proposal.
What is "Bayesian Statistics"? 
Bayesian methods are used to refine a posterior probability estimate by using anterior probability knowledge. The table above is familiar to anyone who works in health care or epidemiological analysis. For example, we know both the approximate prevalence of a clinical condition (the proportion of people in the population with the condition) and the historical false positive and false negative rates of a relevant lab test. Using Bayes formula (below), we can better estimate the likelihood you in fact have a disease given that your test comes back positive, or the probability that you are actually disease-free given a negative lab test.
So, yeah, Andy, astute analytics will not come cheap. Both the academics and the in-the-seat experience are demanding.  Were I willing to whore myself back out into subprime (large pdf), I could easily stay warm and dry making $200k+ doing "distressed consumer debt risk modeling and management." More than the average PCP.

Not gonna happen. I'll play street music first, for BART fare and grocery money.

Just up on THCB:
The ACO Hypothesis: What We’re Learning

...It remains to be seen if these trends continue as more experience with the program accumulates, but the first year results from MSSP suggests Medicare ACOs are on the right track and with prudent evolution, can continue to move providers closer to greater accountability for health care costs and quality.

My comment:
OK, I re-read this post 3 times looking for data on / discussion of significant improvement of clinical outcomes and patient satisfaction.

Nyet, Zip, Zilch, Nada. Maybe those data were collected, but they are not even mentioned here. That WAS, recall, the entire distinction proffered to distinguish ACOs from merely being Suits-in-Charge HMO 2.0 zombies?

Sadly, it always ends up just being about the money. I suppose in the prevailing federal policy ADHD world, that’s just reality. Flit impatiently from one initiative to another every couple of years.

Now to be fair, the authors do note that “shifting to an accountable care model is a long-term, multi-year transition that requires major overhauls to care delivery processes, technology systems, operations, and governance, as well as coordinating efforts with new partners and payers.” But even THAT this-will-take-more-time caveat contains not one word about outcomes and patient satisfaction.

Freudian, much?
I'm not the first to ask "Accountable to whom?"

JAMA Forum: The Innovation vs Consumer Protection Tug-of-War in Health Policy

In the overheated political environment surrounding the Affordable Care Act (ACA), it’s easy to miss the fact that conservative and liberal health policy proposals exist along a coherent continuum: each strikes a different balance between the desires to promote innovation and protect consumers.

For convenience, let’s situate the ACA at the political center. (In doing so, however, I am not claiming it strikes a perfect balance between liberal and conservative ideas.) To promote innovation, the ACA establishes marketplaces in which many private health plans with different design characteristics can compete for consumers. This ACA’s promotion of market-based innovation is a conservative principle, even if it is not implemented in a fashion all conservatives find appealing.

To protect consumers, the ACA prohibits plans from offering deals that are too good to be true. As such, plans are not allowed to leave out basic protections that a reasonable consumer would presume would be included. Among other things, plans must cover 10 categories of “essential health benefits” and pay for at least 60% of the average costs of its policyholders’ medical care. These consumer protection provisions reflect liberal ideals, even if they’re not implemented in ways all liberals would prefer...
Good stuff. Read the entire piece. Austin is one of "The Incidental Economist" principals. I love their banner tagline: "Contemplating health care with a focus on research, an eye on reform."



More to come...

No comments:

Post a Comment