Try again later for healthcare

The much-anticipated switch-on of the Affordable Care Act ("Obamacare") health exchanges happened today, and as anticipated, it was not exactly a smooth ride:

And those going to the federal exchange site, which is handling enrollment for 36 states that didn't fully establish their own exchanges are greeted with this message: "Health Insurance Marketplace: Please wait. We have a lot of visitors on our site right now and we're working to make your experience here better. Please wait here until we send you to the login page. Thanks for your patience!"

California also had quite the surge of users:

The agency that runs the [California State] exchange, Covered California, said it received 1 million hits on the website during the first 90 minutes after the exchange opened. By 3 p.m., the site had received 5 million hits and the two service centers had received 17,000 phone calls.

There are 5400 seconds in 90 minutes, so that's about 200 hits per second. A respectable number, to be sure; many of those will be people just looking at the (static, small, hand-written, easily cached) front page, which is cheap to serve. It would be more interesting to see a rate of user sign-ups and quote serving, with the fraction of errors served in each case, but I guess that's too much to hope for. (For those interested, Covered California is using Microsoft IIS/7.5 and ASP for at least its front page.) I was amused to see the range of languages covered: Spanish, Tagalog, Vietnamese and Chinese as you'd expect, but no Indian languages I recognised. Arabic, Farsi, Russian and Armenian but not French, Portuguese or German. Korean, Hmong and Khmer but not Thai. It seems fairly clear which immigrants learn the language and which don't...

The "please try again later" screen is actually entirely the right approach. The current traffic spike is a short-lived phenomenon - the widespread publicity has led a tidal wave of real users to the exchange websites in a short time window, along with plenty of other "browsers" who are interested in seeing what kind of plan quotes are available. The right thing to do is to tell users to try again later. Real exchange users aren't going to go elsewhere for their plans, and they're not really time-limited - they will try again later in the evening, or tomorrow, or next weekend. The browsers will be put off and either not come back at all (preferred) or come back much later (manageable). In the meantime the exchange operators will have discovered the bottlenecks in their system, and can make tactical fixes for quick wins and start planning for the more involved capacity fixes that may take weeks or months to implement.

Based on the above I'd say that the ACA exchange rollout has gone roughly as well as could be expected. As the user load flattens out over the next few days and weeks the exchanges should work better, and downtime will be less frequent and shorter.

Information Week has a nice dissection from Oregon of why operating a health exchange is hard:

Presenting a product -- an insurance policy -- isn't the hard part. The hard part is figuring out which federal and state programs and tax credits a person or family is eligible for. Getting that part right takes creating an extremely complex rules engine.
About 1,700 individual rules affect eligibility for health insurance subsidies in Oregon.

For each user you have to gather their circumstances - family size, age, income, location - which is fairly straight forward. You then have to write this into a distributed data base, handling de-duplication where required. Then you hand off the data to the rules matching engine - probably the computational workhorse and main memory and disk hog of the system - and wait for it to come back with an offer. There will be separate links to the insurance providers which will receive feeds of what they are prepared to bid for customers, and publish notifications of what bids they made, won and lost.

The real technical challenge for the exchanges will come in the next few months, and it won't be the performance/load issue which has been making headlines today. There will be the inevitable security breaches, usually by insiders trying to sell on personal data. There will be discrepancies between the insurance bid record held by the exchanges and that recorded at the insurers' backends - and hot disputes about which is correct. There will be IT failures on both sides leading to users getting too-high or stupidly-low offers for insurance, and insurers being pressured to honour too-good-to-be-true offers which they (or the exchange on their behalf) made and a wave of consumers snapped up.

(Update: those security breaches seem to be coming earlier than I thought:

One exchange staffer’s simple mistake gave insurance broker Jim Koester access to an Excel document of Social Security numbers, names, addresses and other personal data for whole a list of insurance agents.

Oopsie.)

The policy problems for the exchanges and insurers will start to become more visible next year. Insurers aren't allowed to discriminate on the basis of pre-existing conditions; as such, the first wave of insurance buyers will be heavily seeded with sick or disabled people with no current insurance. The insurers will be hoping to pull in as many healthy people as possible (due to the "individual mandate" making people buy ACA-compliant insurance or face a fine. However, that fine is relatively small for the first year (the greater of $95 per adult and $47.50 per child, or 1 percent of family income), giving healthy adults a good reason to avoid the exchanges for at least a year. The fine will gradually rise, and it will be interesting to see the fine level required to make most people get onto the exchange. In the interim, the insurers are going to be stuck with a wave of very expensive people requiring treatment. There is a natural throttle though - the number of doctors and hospitals willing to accept the reimbursement rates of those insurers is going to go down steadily, so the insurers will have some limit on treatment rates. But then there are going to be widespread complaints that people with insurance from the exchanges can't get timely treatment - a complaint that the insurers can only relieve by raising their reimbursement rates, which isn't going to happen.

Hmm, healthcare rationed by medical provider availability and payer-authorised treatments. This is starting to sound awfully familiar... Think of the NHS but without its effective monopoly on health care purchases to keep down prices.

The $1bn question is whether the insurers are willing to stay on the exchange and absorb the costs of the treatments of their disproportionately sick initial customers, in anticipation of future revenue from young, healthy adults compelled to buy overpriced (for them) and needlessly comprehensive "insurance" - in effect, an inefficient payment plan for all medical costs. This is going to hinge on what they think the federal government is going to give them in subsidies (tax breaks for poorer payers) and how much purchasing leverage they have on the existing health system. Since bureaucracy is one of the biggest problems in the American health care system, and Washington politics is notoriously dysfunctional and lobby-driven, I don't expect this to work out too well.

I should add that there will be beneficiaries of this system - if you are poor and sick, but have enough money to cover your year's deductible, then it's a potentially life-saving bonanza. I fear, though, that help for these people - who undoubtedly need help - is going to come at a high price for everyone else.