Thread by @holdspacefree

Disheartening how little regard there is for consent in the medical-health-data community. For example this new study looking into suicide "prediction". This is a #BigProblem #MentalHealth #BioInformatics #DigitalPhenotyping #Suicidology #DataEthics nature.com/articles/s4174…

2/ In this study, with MASSIVE personal health datasets provided by Kaiser Permanente @kpthrive, HealthPartners @_HealthPartners , Henry Ford Health System @HenryFordHealth the researchers applied models (including "neural" network) to try to predict suicidal risk.

3/ Spoiler alert it didn't really work (comment: maybe put resources on things we already know work, like human connection & actual care & support?). Sadly, this result will probably just cause more effort to "improve" modeling. This is about $, making it & cost-cutting it.

4/ Now consent, where my advocacy is centered. CONSENT WAS WAIVED BY THE IRBs. This data is from 11 year olds and up! I believe consent was waived because they knew they couldn't get it, but that wasn't going to stop them. That is wrong. And they are training models on us.

5/ And this is all under HIPPA which actually has some regulations,but there are loop holes in federal regs when they de-identify. I'd argue this case does NOT meet Belmont principles, this is a vulnerable population-you need consent or if you can't get it, don't do the research.

6/ Large corporate USA health providers must be having a free-for-all w/ all the data, & I suspect partner with insurance. I can't possibly take that problem on but want to highlight it for others. Yes there are ethical boards, etc but confidence falls when consent is waived.

7/ Data set: "All outpatient mental health visits made by individuals 11 years and older between January 1, 2009 and September 30, 2017 in 7 health care systems..."

8/ "...Our sample for estimating and validating suicide attempt (fatal and nonfatal) risk prediction models contained 15,249,031 (59%) mental health specialty visits made by 1,507,684 people and another 10,551,857 general medical visits made by 2,592,332 (84%) people. ..."

9/ "...The sample used for suicide death prediction models contained 13,981,418 (59%) mental health specialty visits made by 1,433,584 people and 9,714,817 (41%) general medical visits made by 2,470,576 people."

10/ IRB Consent Waivers: "Responsible institutional review boards for each participating health system approved waivers of consent for use of records data in this research: Henry Ford Health institutional review board ([IRB], #9998, Henry Ford Health System), Kaiser Permanente.."

11/ "...Colorado IRB (#00002931, Kaiser Permanente Colorado), and Kaiser Permanente Interregional IRB (#799744, Washington, HealthPartners, Hawaii, Northwest, and Southern California regions of Kaiser Permanente)."

12/ Dataset Availability: "...The datasets generated and analyzed during this study are not publicly available because they contain detailed information from the electronic health records in the health systems participating in this study and are governed by Health Insurance..."

13/ "...Portability and Accountability Act (HIPAA). Data are, however, available from the authors upon reasonable request, with permission of all health systems involved and a fully executed data use agreement."

14/ Government / Institutional Support: "We acknowledge funding from the National Institute of Mental Health ... and the Food and Drug Administration ..."

15/ Obviously this is tip of iceberg. The overwhelming hunger for data must be curbed to Belmont Principles of respect for persons. Otherwise,health "care" won't be believed as authentic or trustworthy. Researchers & academics need to pick this up. People needing care are tired.

16/ More:
There were 1,076,873 children (age 11-17) who had mental health specialty visits, whose personal information was used in this study for training suicide death risk prediction models.
459,214 children’s personal data from mental health specialty visits was used ...

17/ ... to validate the death risk prediction models.
All with consent waived. Children, vulnerable. What am I missing? #BioInformatics #DataEthics

Link to human subject info: static-content.springer.com/esm/art%3A10.1…

18/ I emailed the corresponding author Dr. Susan Shortreed on Mar 30 requesting the IRBs for this study. I have yet to receive a response. Follow up sent today.
Affiliations:
Kaiser Permanente WA Health Research Institute &
Department of Biostatistics, University of Washington