One premise of the CDC ATSDR National ALS Registry was that by passively mining data from Medicare, Medicaid, and VA files, they could identify around 85 percent of ALS cases in the United States.
People with ALS go onto Medicare much more quickly than people with other diseases. The premise seemed reasonable.
These administrative databases cover approximately 90 million Americans, and the algorithm identifies 80 to 85 percent of all true ALS cases when applied to these databases.We thought the files would only be missing around 15 percent of Americans with ALS. Those were the people who could be counted only if they self-enrolled in the online portal.
Question 1 --
When the first Registry report came out this week, I was anxious to see how many of the 15 percent of people with ALS who are missing from these administrative databases took the initiative to self-enroll at the Registry web portal.
And how many of that theoretical 1810 self-enrolled at the web portal? A remarkable 1926. Yes, every last one of the 15 percent and then some took the time to self-enroll at the web portal. Self-enrollments far exceeded what we expected in this "15 percent" group.
To add to the amazingness, only 17 percent of the people who were in the group that was in the administrative databases self-enrolled at the portal. What a difference.
Something is really odd here.
Question(s) 2 --
The expectation based on small pilot studies that the Registry data mining algorithm could identify 85 percent of US ALS cases from files was clear to all of us who worked hard to get the Registry approved and well-funded.
Question 1 now has me wondering if that expectation was simply wrong.
From the recent Registry report --
The algorithm was developed initially during the pilot
projects and categorized persons as either “definite ALS,”
“possible ALS,” or “not ALS,” with a sensitivity of 87% and specificity of 85%
Now I'm wondering if the premise should have been that if a case was identified in the passive data mining that 87 percent of the time it was really ALS (and that if somebody looked like a non-ALS person in the administrative data that 85 percent of the time that person did not have ALS) ?
To add to my head-scratching, if they found MND codes (rather than the more specific ALS codes) in the administrative files, that could well account for around 15 percent of the mined records being a MND other than ALS.
Does the 15 percent refer to something strictly within those in the administrative files (and not the Americans who aren't in the files)?
Was the idea that the Registry could identify 85 percent of ALS cases in the US by passive data mining wrong?
Do we really know anything about how many people with ALS are expected to be outside of these administrative databases?
It seems like the answer to Question(s) 2 will tell me if Question 1 is a silly question asked by an advocate who didn't read the fine print closely enough.