On September 22, 2023, there was a joint fireside chat by Dr Lauterbach, Minister of Health, Germany, and Dr Tripathi, who heads the HHS Office of the National Coordinator for Information Technology. With my laptop on my lap, I typed 1400 words during the hour. Chat GPT, with no further assistance, generated a clean 680 word article from my fragmented notes.
My (manual) Blog on the event:
https://www.discoveriesinhealthpolicy.com/2023/09/lecture-notes-german-health-ministers.html
The Auto-GPT article on the same event (also right column, here):
https://brucedocumentblog.blogspot.com/2023/09/ai-corner-lauterbach-at-harvard-kennedy.html
I should mention there is a bit of "symbiosis" here, I took typed notes to write my own article later, but I also took them with the thought in mind, they would be given as a "dump" to Chat GPT to test its writing skills.
I have not tried to line up the two texts.
Rough Notes - Typed During Meeting |
Chat GPT Article |
1400 words German American Club Data and AI in Healthcare 2023 09 22 Dr Agrawal, Heidelberg and JHU. Ulm.
Boston, postdoc Dana Farber.
Yavando Health empowers cancer patients in clinical trials. Welcome, a translatlantic discussion, the German Health
Minister, Dr Lauterbach, and at HHS, Dr Tripathi, Office of the National
Coordinator (ONC-IT). Dr. Lauterbach holds a Harvard SPH PhD. He holds an adjunct professor there, and at
the Univ Cologne. Dr Tripathi holds an MS Public Policy Harvard, and PhD
Poli Sci MIT. He was a fellow at the
Dept of Defense. Twenty years of IT
leadership experience, including Acadia, and Mass EHealth Colloborative. In 2021, National Coordinator for Health
IT (Biden). Agrawal: Health data saves lives every day. Q: Where do you see
the most tangible benefits? Thanks for the opportunity. My remarks are brief. We are behind on health IT tech and must
catch up. #1 is free of fossil energy
and #2 is digital economy esp healthcare.
We are bringing in 4 new laws as priorities. We work also with Dr Lennerz and
colleagues here at Harvard. We have 4 laws that cover two circles of data. The outer circle is data for Care, and the
inner is data for Research.
Connected. We have 70M people
to connect to a EHR by 2025 (opt out is possible). 10M may opt out. Then we will have 70M people, 20,000
pharmacies, 1800 hospitals, 180,000 doctors offices. Also labs.
So we get patient reports immediately in EHR. The patient, too, can see the
results. Doctor may be given
permission to see all the data, hospital, radiology, etc. This is the outer sphere, which is encrypted but not
anonymized. The inner circle can pull the clinical data and cohort
study and genomic data into the FRS the Federal Research Space, a
confidential research space which is open (not encrypted) but
anonymized. Standard crawlers can
analyze the data, it is a federal data space. It will be a huge volume of data, by
European standards, and will grow rapidly.
The system can be a training ground for AI systems. Whether drug company or research questions, they get
access to the data. What are high level challenges? KL: If you want to
train AI, you want to be absolutely sure there is data protection from the
outside. Make the data accessible and
usable, but very confidential space.
The second issue, a fair amount of data is narrative data. This should be however somewhat
structured. Or in retrospect will
revise structure. Who will do this? “Lewins has partnered in the US” [?] Retrospective structuring of data. Q: Dr Tripathi, How
would you describe ONC priorities. MT: Thank you. I listen to the minister with envy!! For the top down authority he has. We are very fragmented. Our HC delivery system “isn’t a system” but
loosely coupled providers and payers.
We don’t have the ability to say we’ll do everything in a big single
database. Not in 1M years. We like to think the complexity makes us creative. We’d like open architecture systems that
free the data across the chain including the patient, payer,
provider-provider etc. Because we
live in fragmentation, it’s AND logic - A AND B AND C. In terms of priorities.
Spent $35B in incentives for EHR to lay the foundation. We let them purchase any system that met a
(low) bar. We certify systems, and
from 10% to 97% of hospitals and 86% of docs use a certified ONC system. Things like CDS, new regs about AI in
EHRs, predictive interventions. That’s the digital foundation. Now the next set? What do we build, ON the foundation. Resaerch, better quality, better
continuity, better measures (e measures).
Modern inter operability. Health equity by design.
Be sure these tech’s don’t amplify that. Safety by design, security by design,
equity by design. With AI, we have some draft regulations on venders to be transparent on embedded
tools. Origin and uses, not regulation of the tool. MGH has to decide, but give it 14 data
elements. Where the data came from, who paid for it, known risks,
etc. That’s in draft, final by EOY. Agrawal: Get your
questions ready. Agrawal: Q. There is excitement about new bills in the
German cabinet, can you clarify? KL: We are working
on 15 bills. 8 bills relevant
here. Digitalizing bills and hospital
bills. For Digital, passing cabinet,
the universal EHR (with opt outs) 2025.
Basis for the research ring.
A digital research bill, the federal research space. Some of the research is done by AI
(Prediction) and Inference is done by people.
A genetics bill (I think) all new digital formats are inter operable. IF it’s not inter operable, you can’t bill
(affects sickness funds). A fourth
bill on medical research of all kinds, if you want to do any medical trial,
in 21 working days, all the evaluation is done and in 5 days a decision. We want to speed up the first patient in,
day. All the data can be provided
digitally, not in journal but English
language, drugs can go direct to the sites. THere is a hospital law that has to do with avoiding
hospitalizations and better primary care. Dr Tripathi, on inter operability. We leverage 88% adoption only if we have
intraop (which is different than adoption.)
How can we improve inter op?
Can AI aide in closing some gaps? MT: YOu have
mutiple bills on the Hill. We think of
this as multi faceted. There are
miinimum data set standards. We
define the USCDI Core Data Inter Operability like vocab standards. It grows, a rising floor. This has spill over value, the APple Health
Record buids on USCDI. And FIRE API
approach. We don’t regulate Apple but
it behooves them to piggy back on us. We are doing extensions such as ISCDI-CANCER. What is a restful API?
Take an app, but in name and password, and it connects. 21CC asks us to make HC discover the
Internet - including API. We have December 2022 FIRE restful API standard. We can require this for payers as well, and state agencies
and public health at CDC. And data
must be available to patients.
Venders have to allow use of a different but compliant API
(competition for API UI). Apple, Google, Saleforce, all jumping into HC and
potentially via APIs. Prior athor via
API. The last priority.
How we have a safe network backbone.
LIke in banking.. There are
elaborate credit card or banking B2B rules and interfaces. TEFKA will be that backbone. As I opened by remarking, we have AND initiatives. AGRAWAL: Thank you
both. A softer question to KL.
Algorithms can be black boxes.
How do we ensure digital trust.
KL: These standards
are transparent. If AI is used for
recommendations or trials. The system
should be as transparent as possible.
Align #1, with all AI, you don’t know exactly how it works. Align #2, there is the alignment, what
data is used, and what was excluded and incuded. Here, we can be clear. We would work with a LLM that is as
transparent as possible, Aleph Alpha may be there. We have time for 2
brief audience questions. MPA
student. What is Germany’s approach
to accessibility to the health data?
Active ownership? What about
non digital society? People without an
ATM card? KL: EHR is opt out
and granulated opt out, you can hide data you want to hide, such as hide only
the psychiatric layer. You can see
all your data. You can block R&D
data. YOu own the data and can always
access it. You don’t need tech
expertise to use it. You may never ever see your EHR but your doctors can enjoy
using it. SIDE : How do you opt out if you can’t use an app?? …You can ask your doctor to hide your data. TRIPATHI: There’s a
lot of nuance. Patients don’t own
their US records. They have only a
“right of access” to a copy. And patient can choose the app on which they request the
copy or store it. KL The patient can
destroy his data which he owns. QUESTION LAST As a MD and Postdoc, I’m excited by the changes, but AI
must learn from data that is good, and some exams and docs are (not so
good!) Won’t that be part of the
AI? University or wealthy patients
may have more genomics, for example. KL: |
680 words On the 22nd of September, 2023, at the Harvard German
American Club, an enlightening discussion took place on the topic of
"Data and AI in Healthcare." The distinguished speakers were Dr.
Lauterbach, the Health Minister of Germany, and Dr. Tripathi, the head of the
Office of the National Coordinator (ONC) for US health data. The
conversation, held at the Kennedy School of Government, was skillfully
moderated by Dr. Agrawal of Harvard University. Dr. Agrawal, who boasts affiliations with Heidelberg,
Johns Hopkins University, and Dana Farber, opened the session by introducing
the central theme - the critical role of health data in saving lives every
day. In response to the question of where the most tangible
benefits of health data lie, Dr. Lauterbach underscored the need for Germany
to catch up in health IT technology. He emphasized two key priorities:
transitioning to fossil-free energy and fostering digital advancements,
particularly in healthcare. Dr. Lauterbach outlined four forthcoming laws
that would address these priorities, highlighting their collaboration with
Dr. Lennerz and colleagues at Harvard. The first set of laws pertains to two circles of data: the
outer circle for Care and the inner circle for Research, both interconnected.
By 2025, Germany aims to connect 70 million individuals to an Electronic
Health Record (EHR), allowing for opt-out options. Within this network,
20,000 pharmacies, 1,800 hospitals, 180,000 doctors' offices, and
laboratories would enable immediate EHR access to patient reports. This data
can also be accessible to patients, with doctors potentially granted
permission to access comprehensive medical information, including hospital
and radiology reports. The outer circle is encrypted but not anonymized, ensuring
data security. In contrast, the inner circle can access clinical, cohort,
and genomic data, residing in the Federal Research Space (FRS). The FRS is a
confidential, non-encrypted research space, open for analysis, and will serve
as a valuable resource for AI systems. Researchers and organizations,
including pharmaceutical companies, will gain access to this data for various
purposes. Addressing the challenges of data privacy and structure,
Dr. Lauterbach stressed the importance of making data accessible yet keeping
it confidential. He also recognized the need to structure narrative data for
retrospective analysis, possibly through partnerships like "Lewins
[?]." Dr. Tripathi, speaking on ONC priorities, expressed
admiration for the top-down authority held by Dr. Lauterbach. In contrast,
the United States faces fragmentation in its healthcare delivery system.
Instead of a unified database, the U.S. system comprises loosely connected
providers and payers, making interoperability a complex challenge. Dr. Tripathi highlighted the significant investments made
in the U.S. to lay the foundation for digital health, including incentives
for Electronic Health Record adoption. He discussed the need for modern
interoperability, emphasizing health equity, safety, and security.
Additionally, Dr. Tripathi mentioned draft regulations regarding transparency
for AI tools, ensuring data origin, uses, and potential risks are known. The discussion delved into audience questions, with Dr.
Agrawal posing inquiries to both speakers. Dr. Lauterbach detailed Germany's
legislative efforts, including digitalizing bills, universal EHR
implementation by 2025, and a strong focus on medical research acceleration.
[for example, a goal to approve research protocols in Germany in 30 days,
much faster than today.] Dr. Tripathi
addressed the challenges of interoperability and the role of AI in closing
these gaps. He also mentioned efforts to standardize data sets and promote
accessibility, including through Application Programming Interfaces (APIs).
[ONC rulemaking on AI here] In closing, Dr. Agrawal touched on the importance of
digital trust and transparency in AI algorithms. Dr. Lauterbach emphasized
the need for transparency in AI systems and data usage. The session concluded with brief audience questions,
including inquiries about Germany's approach to health data accessibility and
the challenges faced by patients in the U.S. regarding their medical records.
Dr. Lauterbach highlighted Germany's granulated opt-out system and patient
ownership of data, while Dr. Tripathi discussed the nuances of patient rights
and data access in the U.S. In summary, the discussion shed light on the evolving
landscape of healthcare data and AI, highlighting both the progress and
challenges faced by Germany and the United States in harnessing the potential
of these technologies to improve healthcare outcomes. |