Last week I posted on some content from Alcoholism and Drug Abuse Weekly on the substance use treatment data issue, about which we’ve been writing for some time. Today’s issue includes a story by Alison Knopf focused on the concerns of researchers. A few quotes and editorial notes follow:
- “The Center for Medicare and Medicaid Services (CMS), at the request of the Substance Abuse and Mental Health Services Administration (SAMHSA), has quietly been scrubbing all Medicaid, Medicare and State Children’s Health Insurance
Program (SCHIP) files from any information that identified patients who had been treated for substance use disorders (SUDs). The scrubbing began in 2013.” - “Historically, CMS research files have not contained any patient-identifying information,” Aaron Albright is quoted. In fact, Research Identifiable Files have been available from CMS for many years. Here’s an example from over 20 years ago of use by researchers of Social Security number and other identifying information to match VA and Medicare claims records. I’m very worried that key people at CMS and/or SAMHSA think that researchers have never had access to identifiable SUD-related claims (the implication being that recent developments aren’t a change). That is false, and if they think otherwise they need to open up some communication with researchers and research organizations. Recent developments are a big change, which is why we’re upset. We’re not pushing for something new. We’re pushing for access to the type of claims data we’ve been able to use for decades.
- “We asked Albright if the SAMHSA request was documented — he said it was made ‘verbally.’”
- “The scrubbing was first brought to light in a blog, The Incidental Economist, [in posts] written by Austin Frakt and Nicholas Bagley.” Nicholas is quoted at length, expressing similar thoughts as he’s written here.
- According to the article, a notice of proposed rulemaking (NPRM) on 42 CFR Part 2 is expected out in 2015.
I want to emphasize that there is a difference between facilitating linking with patient identifiers and providing wide access to patient identifiers. Researchers require the former, not the latter, and the two are separable. Linking can be done by dedicated, credentialed staff and/or in defined, high security data processing environments to greatly reduce the potential for harm. (This has become relatively standard, in my experience.) Linked files can be subsequently stripped of identifiers for distribution to researchers for analysis. (Though I work with VA-Medicare linked files routinely, I neither obtain nor need (nor want!) patient identifiers in my analytic files, for instance.)
Moreover, access to claims data for research is not the same thing as access to electronic health records by providers (e.g., in a health information exchange or HIE). Both are important issues, but they should not be lumped together as they require different measures to reasonably ensure patient privacy. For example, for large-scale, observational studies there is absolutely no way to obtain consent from millions of patients over many years for research purposes. HIPAA waivers exist precisely for this reason, but may not be appropriate in the HIE context.
Full suppression of all substance-use related claims data to research organizations is overkill. But that’s what’s happening. I understand that’s what the regulations currently demand. My concern is that some people are talking about this as if this level of suppression is required for patient protection. That’s simply false. Moreover, it is not an expression of lack of concern for the privacy of such patients to want access to research claims data. It’s an expression of an interest in improving their care. The idea that privacy is the only dimension of potential harm to patients is a disservice to them. Harm can come from not being able to observe the quality of care they obtain and the nature of outcomes it produces, for instance.
If you can obtain access to it and are interested in this issue, the Alcoholism and Drug Abuse Weekly article is worth reading in full.
