My health data errors exploded in the newspaper. Why?
Part 5 in a series on the importance — and potential power — of access to our health data.
Yesterday’s post described what happened when I poked the button my hospital offered, to send my data to Google Health: a train wreck. I tried to get answers from my hospital about the bogus medication warning, the conditions they reported that I’d never had, but they had no answers. What’s up with that??
So I blogged about it—3500 words long including screen captures.
That post was shocking enough to me (“my hospital’s records about me are a mess?? The hospital that saved my life??”), and it produced a firestorm of social media (nearly 100 comments), but what I didn’t expect was when the Boston Globe called and asked if they could write about it.
And I really didn’t expect they’d put it on page 1. But they did.
My world blew up. Suddenly the phone was ringing off the hook, and Twitter had a second round of explosion. And now my hospital expressed interest: once my problem was on the front page, that very morning the CIO’s blog at last responded, saying “We’ll hold a conference call with e-Patient Dave, his doctor, Google, and me…” Funny how media attention works. :-)
The cause of the madness: they sent Google my “claims data” (insurance codes), not my clinical records
In his excellent book The Digital Doctor Bob Wachter MD refers to insurance records as a “waste product of administrative functions.” Well, guess what my hospital had sent Google? My insurance billing history, not my clinical records.
As just one example of how stupid that was, consider that the insurance billing code for a pregnancy test is the same, no matter what the result was.
I can’t tell you how many gasps and eye-rolls I witnessed when people learned they’d sent claims data. Here are just a few of the violations of good data practices that created this mess:
1. Fundamental data principle: it’s never good to take data created for one purpose and use it for another.
A CPA friend told me this at the time. All data involves some level of compromise when it’s created, and trade-offs that are legitimate in one context may create a really off-base picture in another context.
2. Claims data is not granular enough.
Granularity is the issue of how much detail a digital file conveys. We’ve all seen photos that are low quality, or pixelated, because there’s too little detail in the data. And even though CSI: Los Angeles might show the wizards miraculously inventing missing detail in a blurry photo, that ain’t how it works.
It’s not just pregnancy tests. In my case, I was tested for metastases to the brain or spine, so the billing code was appropriate — “Metastases to the brain or spine” — but there’s no separate code for “tested and didn’t find it,” so when my hospital’s IT people sent Google my billing codes, Google innocently told me what it said. (Right??)
In short:
There was no way the “vocabulary” of claims data could possibly express the clinical information needed for a meaningful health record.
But they used that vocabulary anyway, because it was convenient.
3. Upcoding
The other reason claims data is catastrophically the wrong choice for this task is upcoding: the routine (and legal) practice of inflating the billing charges to the most expensive charge possible, based solely on words that occur in the medical record, whether or not the resulting charge makes sense.
I’m not making this up. It sure made my jaw drop. But have a look at this screen capture from my CT scan report, 12/06/07:
Note where it says “dilatation of the aorta”? True — I’d had a stressful immune treatment. I did not have an aneurysm, but they billed for it anyway, legally, because the report included the words “dilatation” and “aorta.”
People in the industry told me this happens all the time. Apparently rather than assessing every charge individually, they just look for keywords that supposedly justify a charge. (See why Wachter calls it a “waste product of administrative functions”?)
4. Fraud.
Yes, fraud. Well, I can’t assess the intent in someone’s state of mind when they entered this data, but consider that these billing codes among others showed up in my billing record, with no resemblance to reality:
- Nonrheumatoid tricuspid valve disease (this heart diagnosis was billed during a femur treatment!)
- Volvulus (a lethal kink in the intestine, billed on a chest CT; no volvulus was reported on the radiology report)
Among other things, consider that my insurance claims history now says I have an aneurysm and other nasty things that I don’t. Fortunately I’m on Medicare, but if this happens in your family, you may have no way of knowing, and it could be used against you in assessing your risk.
Now, who knew that all this malarkey would be revealed when one naive patient took his hospital’s offer to send his “medical records” to Google? That’s what amazes me to this day. I mean, didn’t they know this would happen?
Turns out, the answer is no, because:
5. They released the software without testing it on real patient data.
That’s right. Back then people in the industry had marveled, “How did they get that done so fast???” Turns out they didn’t actually do the required work — they just announced it was done, declared victory and accepted applause.
I want to emphasize, when I clicked that button, I wasn’t looking for trouble. I had no idea it would blow up in my face.
But why would this cause an explosion in DC policy circles?
As I noted before, the ARRA/HITECH had just been passed — just a few weeks before I poked the Google Health button — and it included $40 billion to fund new EMRs (electronic medical records). This meant lots of people were studying “How will we get patients’ histories into the computers?”
It’s a real problem because your history is mostly piles of text (including handwritten numbers for blood pressure etc), not neatly organized into the“structured data” that computers like. But some genius was telling everyone, “We do have one kind of structured data — everyone’s claims data.” Yes, whoever they were, they were advocating that we should just send the EMRs everyone’s claims data.
Savvy people were yelling, “Noooo!! You can’t do that — that data is crap!” but I’m told the claims-data advocates were winning. Then along comes naive me and triggers this big discovery.
And the next punch line is …
Remember the start of this blog series, about healthcare’s Gordian knot, a thousand points of fiscal pain? Consider this: after the commotion had died down, I called my health insurance company (Harvard Pilgrim at the time) and asked if they wanted to do a thorough audit, to see if they’d been billed for things falsely. And you know what they said? “Don’t worry about it.”
I’ll leave you to ponder, because I don’t know: how is it that any other kind of insurance (fire, auto theft, etc) is vigilant about fraud, and health insurance seems not to be?
Whatever the answer, I bet it’s a factor in the incessant growth curves in my malignant tumor post. (2019 saw coverage in The BMJ of the second annual Shkreli Awards for the greediest companies. Not making this up. Go look.)
The real bottom line: Do you know what’s in your data?
Then, aside from the administrative / claims / billing malarkey, there’s the 2014 Wall Street Journal article [left] citing a Geisinger study that 80% of charts contain errors. Does yours? Your mom’s? A subsequent letter urged us to check for false diagnoses in our MIB records, which can affect life insurance premiums.
All of this can be crazy-making, not least because this was a shock to the whole industry. But let’s not lose our bearing: aside from the greedhogs, the only reason any of us care about the health system is because sooner or later, we’re all going to seriously need good care for a serious problem. And if the data in your chart isn’t accurate, how can they possibly treat you accurately?
Think it doesn’t happen? Think again. Some appetizers:
- When my mom got to rehab after surgery, her chart somehow had her thyroid condition backwards. (She’s hyper; it said hypo.) The wrong medication could have done serious harm.
- My colleague in the Society for Participatory Medicine, Peter Elias MD, recounts that during his wife’s treatment for lymphoma, her chart included a chest x-ray report from a different patient. (How could any doctor assess the wife’s status, given the wrong patient’s data?)
- It also said, two years after her chemo finished, that those drugs were still current.
Do you know?
I’ll have more examples in tomorrow’s post: A movement is born: Gimme My Damn Data
