I'm always a bit skeptical of presentations from tech company CEOs on
how their product areas are necessary in the mental health field.
That said, this article has a few good points:
/"Umar Nizamani, CEO, International, at NiceDay, emphasised that AI will
inevitably become an essential tool in mental health care: 'I am very
confident AI will not replace therapists – but therapists using AI will
replace therapists not using AI.'"//
/
I am beginning to think this also -- for better or worse. I took a VERY
fast 60 second look at NiceDay and it appears to be another
all-encompassing EHR, but with a strong emphasis on data. Lots of tools
and questionnaires and attractive graphs for therapists to monitor
symptoms. (I need to take a longer look later.) So data-driven could
be very good, if it does not crowd out the human touch.
/"Nizamani said there had been suicides caused by AI, citing the case of
a person in Belgium who died by suicide after downloading an anxiety
app. The individual was anxious about climate change. The app suggested
'if you did not exist' it would help the planet, said Nizamani."//
/
YIKES... So, yes, his point that care in implementation is needed is
critical. I worry at the speed of the gold-rush.
/"He [//Nizamni] //called on the industry to come together to ensure
that mental health systems using AI and data are 'explainable’,
'transparent', and 'accountable'." //
/
This has been my biggest focus so far, coming from an Internet security
background when I was younger.
/"Arden Tomison, CEO and founder of Thalamos"/ spoke on how his company
automates and streamlines complex bureaucracy and paperwork to both
speed patients getting help and extract the useful data from the forms
for clinicians to use. More at: https://www.thalamos.co.uk/
/"Dr Stefano Goria, co-founder and CTO at Thymia, gave an example of
'frontier AI': 'mental health biomarkers' which are 'driving towards
precision medicine' in mental health. Goria said thymia’s biomarkers
(e.g. how someone sounds, or how they appear in a video) could help
clinicians be aware of symptoms and diagnose conditions that are often
missed."//
/
Now THIS is how I'd like to receive my AI augmentation. Give me
improved diagnostic tools rather than replacing me with chatbots or
over-crowding the therapy process with too much automated tool data
collection (some is good). I just want this to remain in the hands of
the solo practitioner rather than being a performance monitor on us by
insurance companies. I want to see empowered clinicians.
TITLE: AWS rolls out generative AI service for healthcare documentation
software
Yeah... If it's going to be worth using it would have to listen to the
whole visit... But this needs more thought. In my past quick
experiments, it took 90% of the effort typing directions to get an AI to
generate a halfway okay note. So the AI (I think) would have to listen
in and then do the note to be worth it.
Would we want this? Can we trust this?
--Michael
+++++++++
"Amazon Web Services announced Wednesday a new AI-powered service for
healthcare software providers that will help clinicians with paperwork."
"AWS HealthScribe uses generative AI and speech recognition to help
doctors transcribe and analyze their conversations with patients and
drafts clinical notes, the company announced Wednesday at its AWS Summit
New York."
This is for a new product called AutoNotes.ai that will create progress notes for you for about $14+ per month.
I am tentatively interested in this and signed up for a free trial.
I have concerns (below) but am hopeful this continues to improve.
The system has no BAA agreement for HIPAA because they claim not to collect any PHI. This may be true.
Their terms of service and privacy policy may be amended at any time and the only notification provided will be on their website. I am weary of this now that I have been burned by such stealth changes by one of my own BAA subcontractors.
Information for each client will have to be entered from scratch each time as they do not store data about clients. For the free demo, it takes about 60 seconds to generate each note. You then cut and paste it into your EHR.
While they claim to collect no PHI, they do send SOME data of SOME sort to several tracking systems and data aggregators. At best, they are tracking YOU -- the clinician. At worst -- they may be sending some of the data you enter to data aggregators which -- hopefully -- do not have data from other sources to be able to pin down the individual clients you are describing.
DATA IS SEND TO THE FOLLOWING OUTSIDE TRACKING COMPANIES ON BOTH THEIR DATA INPUT PAGE & THE GENERATED NOTE PAGE:
PRIVACY BADGER BLOCKED:
connect.facebook.net
www.googletagmanager.com
s.gravatar.com
snap.licdn.com
i0.wp.com
GHOSTERY BLOCKED:
Google Tag Manager
LinkedIn Analytics
Facebook Connect
Gravatar
The following is an example of the note generated for a fake client with Acute Stress Disorder:
%GREEN = IN WHAT I REPORTED%
&RED = INVENTED BY AI, NOT REPORTED BY CLINICIAN&
BLACK = I HAVE NO STRONG OPINION
Subjective:
Client &reported increased levels of stress, particularly in the& %workplace. Described experiencing anxious mood and labile mood, but was able to demonstrate appropriate behavior and normal speech throughout the session. Client was self-reflective% and &engaged& %during the 45-minute video session%, %which allowed for a comprehensive exploration of the current challenges they face&. The client recognized the need to address these issues in order to improve their overall well-being.
Objective:
During the session, several areas were discussed. &Client was open and honest& about %feeling overwhelmed and stressed at work, further precipitating their anxiety. They were able to identify specific work situations that were particularly difficult for them to navigate. It was noted that the client was able to maintain self-reflective behavior during discussions,% which will be beneficial for making progress in therapy.
Assessment:
%The client is showing signs of progress in therapy, taking active steps to address and work through their challenges with stress and anxiety.% Their willingness to participate and self-reflect in session indicates a strong commitment to achieving their therapeutic goals. %Current therapeutic interventions, which are directed at helping the client build skills and coping strategies, are having a positive impact on the client's progress.%
Plan:
%In order to lower workplace stress and continue advancing in therapy, the client agreed to set specific objectives for themselves. They plan on talking to their manager about ways to manage their workload or address any outstanding concerns. Additionally, they will begin practicing meditation four times a week during their lunch breaks, as a means of managing stress and promoting relaxation.% &Continued exploration of these& and other stress reduction &strategies will be a focus in future sessions.&
Hmmm... My take-away is that this needs more work (that's fine); I want to know why they have to report to LinkedIn, Facebook, Gravatar, and Word Press while I'm logged in and what they report; and the system IS inventing minor elements that I did not tell it to add. For example, while I reported the client was overwhelmed and stressed, I did not say the client was open and honest about it. I told the system the client was "progressing", but never said that increased levels of stress were reported in this session.
#AI #CollaborativeHumanAISystems #HumanAwareAI #chatbotgpt #chatgpt #artificialintelligence #psychology #counseling #socialwork #psychotherapy #EHR #EPIC #medicalnotes #progressnotes #Microsoft
@psychotherapist@a.gup.pe @psychotherapists@a.gup.pe @psychology@a.gup.pe @socialpsych@a.gup.pe @socialwork@a.gup.pe @psychiatry@a.gup.pe #mentalhealth #technology #psychiatry #healthcare #patientportal #autonotesai
If AutoNotes.ai just plugged in a free AI available from elsewhere with a front-end they created, I wonder what legalese governs use of the AI on the backend? I am not a lawyer and none of us know the licensing agreement or ownership of the AI which AutoNotes.ai is using.
And, well hey -- while I'm just busy making up wild speculations -- let's play with this a bit:
So I know AutoNotes.ai is sending SOME kind of information to Google tracking services because my browser plug-ins are preventing this data from being sent to Google and telling me so.
Let's just suppose for a moment that they are using Google's Bard AI on the back-end. Because -- why not -- there is no PHI being collected anyway...
Meanwhile, both the therapist and the client are using the Google Chrome web browser for televideo. Or maybe they are using Gmail and the Gmail text is being mined. Or the data input for the note is sent along to Google datamining regardless of whether or not the Bard AI is used...
Let's go further out on our hypothetical limb and say that the therapist sees only three clients that day. The therapist creates three notes in AutoNotes.ai that day...
It's now a more than fair chance that one of those three unnamed clients has Acute Stress Disorder (like in my example above). If Google has gone to the bother to devote the computer tracking power to it, they might know from Gmail, or Bard, or data aggregation the names of the clients the therapist saw that day.
Of course, I really am making this all up -- we are just not given enough data to know what's real and false anymore.
Here is a paragraph from the welcome message they emailed me:
"Here are a couple of simple suggestions, first, complete as thorough a Mental Status Exam (MSE) as possible, submit a few sentences related to the session and theme, and include treatment plan goals, objectives, and strategies; this will ensure the best possible clinical note. Please revise and submit your revised version inside the app! This will assist all of us in building the greatest tool on earth for the field!"
Well, okay -- I do want the AI to get better at its job...
But this DOES mean they are keeping a version of what you provide, doesn't it?
It occurred to me this morning that by the time I fill out the form and write two or three sentences, I've already done all the work that is needed for an official note (after adding start and end times, diagnosis, name, client age, and a few other elements to the form). There is no need to convert it all to narrative -- it can stay in form factor mostly.
So -- while I want an AI I can trust to help with notes (and this one may grow into such) -- right now the effort of getting it to create a note is about exactly equal to the effort of just writing it myself anyway.