3 ms·

They should have been designed and planned with deidentification with the expectation that data would be released. If they are literally manually redacting pat

They should have been designed and planned with deidentification with the expectation that data would be released. If they are literally manually redacting patient data from pages and notes then that is a flaw in the study, I think. I would expect patient data to be pretty limited in the trial data since they are purposefully double blind. Notes are a different story, but I’m sure there are techniques that can be purchased for entity resolution that can automate the scrubbing for PHI and PII. Seems more like the FDA and Pfizer being silly than the judge.

lalaland1125 5y ago
> Notes are a different story, but I’m sure there are techniques that can be purchased for entity resolution that can automate the scrubbing for PHI and PII. You can't fully automate this because we require almost 100% accuracy. You need a manual review to ensure no data is leaked.
- prepend 5y ago
  I’m not sure why you think we require 100% accuracy. The actual level depends on many factors. For example, the expert determination method of deidentification in hipaa safe harbor [0] can allow up to 1% reidentification and still be considered anonymized. This is how medical providers share notes without manually reviewing them all. [0] https://www.hhs.gov/hipaa/for-professionals/privacy/special-topics/de-identification/index.html https://www.hhs.gov/hipaa/for-professionals/privacy/special-...
lesuorac 5y ago
Would you not use trial data to determine if Pfizer's vaccine should be approved? The FOIA request was for all of the documentation used to approve the vaccine. > I’m sure there are techniques that can be purchased for entity resolution that can automate the scrubbing for PHI and PII. Sounds like a good way to have "anonymized" data that isn't actually anonymized like most datasets ...
- prepend 5y ago
  That’s part of the downside of “anonymized” data as there’s not any clear legal definitions so it’s quite possible to identify some persons within anonymized data. I think this is a big problem when it comes to stuff like data brokering and selling. Less so with issues like this because vaccine data researchers are less likely to cause harm to someone than others who buy anonymized data.
robbedpeter 5y ago
>>If they are literally manually redacting... That's a feature, not a bug. Bureacracy is a game, and they are grandmasters.
anigbrowl 5y ago
To be fair they were in a bit of a rush, and opted to dump the problem on the bureaucrats rather than spend more time designing protocols. This isn't an example of wicked plotting, it's just an argument about who picks up the tab for a large redaction job.
- prepend 5y ago
  I think it’s also a situation of someone not wanting to release so coming up with a bone headed way of reviewing for privacy.

lalaland1125 5y ago

> Notes are a different story, but I’m sure there are techniques that can be purchased for entity resolution that can automate the scrubbing for PHI and PII. You can't fully automate this because we require almost 100% accuracy. You need a manual review to ensure no data is leaked.

prepend 5y ago
I’m not sure why you think we require 100% accuracy. The actual level depends on many factors. For example, the expert determination method of deidentification in hipaa safe harbor [0] can allow up to 1% reidentification and still be considered anonymized. This is how medical providers share notes without manually reviewing them all. [0] https://www.hhs.gov/hipaa/for-professionals/privacy/special-topics/de-identification/index.html https://www.hhs.gov/hipaa/for-professionals/privacy/special-...

lesuorac 5y ago

Would you not use trial data to determine if Pfizer's vaccine should be approved? The FOIA request was for all of the documentation used to approve the vaccine. > I’m sure there are techniques that can be purchased for entity resolution that can automate the scrubbing for PHI and PII. Sounds like a good way to have "anonymized" data that isn't actually anonymized like most datasets ...

prepend 5y ago
That’s part of the downside of “anonymized” data as there’s not any clear legal definitions so it’s quite possible to identify some persons within anonymized data. I think this is a big problem when it comes to stuff like data brokering and selling. Less so with issues like this because vaccine data researchers are less likely to cause harm to someone than others who buy anonymized data.

robbedpeter 5y ago

>>If they are literally manually redacting... That's a feature, not a bug. Bureacracy is a game, and they are grandmasters.

anigbrowl 5y ago

To be fair they were in a bit of a rush, and opted to dump the problem on the bureaucrats rather than spend more time designing protocols. This isn't an example of wicked plotting, it's just an argument about who picks up the tab for a large redaction job.

prepend 5y ago
I think it’s also a situation of someone not wanting to release so coming up with a bone headed way of reviewing for privacy.