The Healthcare Diff | Hacker News Reader

The Healthcare Diff

kmeme.com

47 pointspbw27 comments

Neat use of development tools to illustrate a point. But, to be quite blunt, very naive in the best case and quite stupid in the worst case. We can "ship early, iterate often" in software because quite often it doesn't cost much (certainly a naive way of putting it) to do so. However that paradigm breaks down even in software when dealing with things like enterprise software or say medical device software or FDA approved software. Guess what would happen if the large scale infrastructure construction industry went with this model?

Think about what it costs for an organization to adapt to a new set of regulations. A boatload of new paperwork needs to be setup, ordered, printed and distributed, after a thorough analysis of the regulations and a whole long process to determine documentary requirements. If software or electronic records are used, those need to be updated. The new set of rules need to be communicated to a large number of participating health care providers with appropriate document and/or software updates. Each of those things needs to be audited and certified. Personnel have to be trained to understand and comply to the new set of regulations and potentially certified. There would be a period of enforcement where nobody will quite understand the limits and those limits would be tested and the legal system would then clarify those regulations.

And that is just the tip of the iceberg. In such cases the massive one-shot refactor gives you way more bang for the buck than not.

They should write the bill in a programming language instead of English. Then it won't be so tough to interpret it. :)

They do... what you see there in the "4,000 page" bill is legislative language - pretty clear code for a lawyer or legislative aide. On the other hand, the Congressional Research Service puts out some pretty good nonpartisan "plain-language" bill summaries (and for anyone who asks "well why don't they just use those the laws themselves", ask yourself "why don't you just write code in English" - they both have the potential for ambiguity).

Incidentally, I put "4,000 page" in quotes because it's a pet peeve... if you actually look at the text, it's in 24pt font and giant margins - makes it much easier to read and mark up around a table, but in no way approximates the amount of text people envision when you say "X-thousand pages".

Yea but if you compile it, wouldn't it be faster than being interpreted? And the interpretor is also a human brain, which is much more error prone than a computer.

The "trying multiple approaches" point he makes at the end exists in America, to an extent. It's called federalism. There can be substantial, though not absolute, differences between the laws of the states. Different states value different things, so some will try things before the nation as a whole is ready. If they work, the nation can look to their experiences and try to implement a nationwide plan that has all of those benefits and avoids any pitfalls they experienced.

Examples: women's suffrage, universal healthcare, medicinal marijuana even though it's federally illegal, physician-assisted suicide for terminal patients, probably tons more.

Agreed. The key is to improve the data collection process. To be more scientific, per se, in how we approach state-wide and national problem-solving. To treat other states as controls, etc. What we have now is people citing different stats from oft sketchy sources, to say the least, when they mean they're relying on "the numbers."

Massachusetts is a relevant example, they passed a state-wide health care reform bill in 2006.

One problem with this system in practice is that the federal government is already so expansive in terms of regulation and tax burden that there is not very much room left for states to experiment in.

California and Texas are very different states by some measures, but there are no states without Social Security, No Child Left Behind, the FDA, the War on Drugs or the Patriot Act, all things that I would prefer to do without.

i'm curious about your qualms with the FDA and what you'd like to replace it with.

FDA liberalization is not exactly a fringe position among economists. The basic argument is that the FDA may kill more people than it saves and may hamper medical innovation in the long run. It's hard to measure counterfactuals, but you can measure the costs of FDA regulation which are huge, about 80% of the cost of a new drug. So the benefits of the FDA would also have to be huge to justify its existence.

The arguments are out there if you are curious. You can find people that argue the position better than I can on a Hacker News comment thread.

We don't need to do this only within the states, we can also just do systematic experiments.

Once upon a time, we wanted to determine how much health insurance affected health. So we (and by "we", I mean the RAND corporation) created several different health insurance plans, and randomly assigned people to different plans. From this, we determined that very high deductible insurance lowers costs by 30% (relative to full coverage), but health is unaffected.

We have not performed a single experiment since then.

He tries to string together two different points:

1. The language used for describing a change in a legislative document is verbose and difficult to understand. A simple "diff" of the two documents would be a substantial improvement

2. Because a developer tool (diff) might be useful in one specific aspect of legislation, other development methodologies might also be useful. Specifically, the short time to release, fast iteration strategy should be considered for legislation.

The author has a good point in #1. In this century, we should be using tools that make our lives easier. Maybe not the exact diff tool, although that would be an improvement - but any move away from a prose describing the changes down to the last comma is probably a good move.

Point #2 fails to hold water in many respects. First, the particular utility of a tool like diff has little implication of the end methodology that should be used. Second, even if everyone decided that lawmaking should be more like software building, the "fast iteration" methodology is not necessarily the correct choice. Just because that may be a good way of building web-based software, if you are building software for a space-station, a medical device, or a nuclear reactor, this is rarely the accepted practice. Perhaps law-making is more like developing these forms of mission-critical software systems, rather than web search software.

Imagine the legislators running a distributed version-control system, exchanging diffs, pull requests, etc., with each contribution able to be logged and credited to an author. This whole thing is actually a merge of two branches in the health care bill source. Of course, our legislation is not run by intelligent people trying to get things done as efficiently as possible, so it's not done this way. In fact, if you can imagine the complete opposite of DVCS (or even regular monolithic VCS) that seems to be pretty close to what's going on.

Difficult to know who authored exactly what. Nebulous states of various parts of bills at various points in time. Opaque language designed to shift potential blame in the future if something goes wrong. Vestigial formatting rules, incomprehensible phrasing (to those who would equate law with plain English) and a high level of redundancy.

Question for lawyers: if law (as in legal writing) is written as a sort of code built up over time based on past rulings and with concrete (or semi-concrete) meanings in terms of what effects it's supposed to have, why do we not get any 'comments' in legislative law? For example, it makes sense for a contract between two corporations to not have 'comments' to go along with the 'legal writing code' -- you're using the contract against someone else, so it's their job to decipher what effects it will have (right?)

But legislation isn't supposed to be used as a competitive chokepoint between two entities, it's supposed to be something that people who are affected by it (in the case of medical insurance reform, everybody) can understand. Reading legal writing to me seems like reading source code without comments, and you just don't give source code without comments to someone asking to read your code unless you have a grudge against them.

This is something I spend a lot of time thinking about. Law and computer code are quite similar in many ways, yet innovations in software architecture and tools barely affect the way laws are written. No matter how much "transparency" we have, legal text itself remains incomprehensible. And though online distribution has managed to catch on, lawmakers don't take advantage of any of the affordances of the medium--diff being a prime example.

If only law could be written in Python...

Is there a reason you picked Python? Python seems like a bad language for comparison since so much of it is checked at runtime. It seems like Coq or Agda would be a superior choice for writing laws in.

Republican congressman can't get enough Coq.

The problem with the comparison is that lawmakers are purposely being duplicitous. It's a well known fact that congress drafts laws in a manner that's designed to turn away the casual reader because doing so gives them the ability to "spin" when re-election comes around. So all the tools in the world aren't going to compensate for someone whose trying to be opaque.

This assumes the goal was/is clarity. Unlike software, people have a lot to gain from being unclear in this case.

But that is gain at the cost of people suffering from lack of clarity.

You could say same things about lack of clarity as you can say about corruption. It's huge potential gain for some people but horrible waste of resources to every one.

Instead of arguing over the one true solution on the basis of hunches and intuition

We don't. Imperfect as much of proposed legislation is, it's also thoroughly researched, argued over, picked apart, dissected, analyzed and so on. Often with a partisan tilt. The system for coming up with legislation is quite adversarial. But it's not based on 'hunches and intuition'.

we should implement more than one approach

A little principle of 'equality before the law' might really get in the way here.

The process of writing and designing software is a relatively new discipline and one in which significant, high-impact insights are being made and will continue to be made. Given these bright flashes of success, the temptation to apply software methodology to entirely unrelated fields ranging from law to personal relationships is very high. It's best resisted, since it's usually stupid. At a minimum, before applying it, ask yourself 'how is this not like software' rather than starting with 'how is this like software'.

I like the diff idea for many things in government data.

But the real physical world has much higher costs, ask the hardware manufacturer that makes a chip that is not bought. Software has much easier iteration properties. Hardware and physical changes or slow changing society/process changes are much much harder to iterate. It is like writing code on punch cards or having to make the entire computer.

I think there is a fair amount of experimentation, trial and error and more in healthcare mainly private but also public. What we are seeing in congress is what we all deal with individually and as group, it is just freaking hard to get teams to agree on even a direction on touchy subjects and the larger they get the harder it is. Resistance to change and posturing prevents faster change, mainly because of that error side of trial and error and large doses of groupthink.

We can't easily roll back changes in law, process, society as you can in software. But I agree it is so bad right now we have to do some changes to see what happens.

Nice thing would be to have automatically checked assertions, constraints so help spot interactions between law to be introduced with existing laws.

All that would require to parse language into what it actually means. So we will probably have to wait for strong AI to see any significant improvement in lawmaking.

But as author of this post postulates using readily available tools for dealing with complex text documents today also would be nice.

Except, in this case, trying out approaches and alternatives means treating humans as the subjects of the experiment. Some may get degraded experience, others may complain unnecessarily.

Furthermore, the problem in this particular instance there is a lack of agreement on the objective metrics by which to measure and optimize. ($Cost, %Insured, Access, Availability of Advanced Treatments, Eligibility for Procedures, Choice of Providers, etc.)

The idea of experiment-driven law is an interesting one that should be explored. I'm reminded of a TED talk I watched recently about "charter cities" that have experimental legal and economic policies.

http://www.ted.com/talks/paul_romer.html

I was hoping he had hacked together some code to apply the diffs to the text to make it more readable.

My opinion (as someone who nearly died and has gotten a lot healthier):

The real problem is that study after study after study shows that diet and lifestyle are major factors in health outcomes for every single major deadly medical condition. But doctors are in the business of selling drugs and surgeries. Telling people to eat right and live right isn't "sexy" (in the advertising sense of that idea - basically exciting, attractive), it doesn't win Nobel prizes, and it is highly unlikely to get you headlined as some "hero" who is saving lives. (In fact, it seems to me that telling people to live right is more in line with what most religions try to do than what most doctors try to do. Of course, the problem with that is religion then has too much baggage about "sin" and the like -- ie you are told in essence "You are going to hell and morally bankrupt" rather than "That activity exposes you to disease" or "It will clog your arteries".)

So they frame the problem as "healthcare", then work on bills about "medical care" instead of actual healthcare, and basically miss the boat entirely. You aren't really going to make the nation healthier by getting them better designer drugs and manipulating who pays for what. You also aren't likely to bring down costs that way. Most bills of that sort shift costs rather than reducing them. Actually getting healthier is the only humane, effective means to genuinely reduce healthcare costs. And getting healthier is mostly about taking proper care of yourself to begin with rather than eating crap, not being active, and then running to the doctor for a new pill for every new symptom brought on by your bankrupt lifestyle.

Of course, that sh*t don't sell. (Yet. Maybe I will find a way around that problem. :0 )

  git diff-and-fix-hubris-and-corruption-but-oh-wait-not-really government/legislation/intentional_obfuscator.rb

?

Reply on news.ycombinator.com