Algorithms and artificial intelligence in telemedicine: Making #TelemedNow Work for Us

#TelemedNow Twitter Chat Wednesday, September 9 – 8 PM ET Introduction & Welcome: Ami Bhatt (@AmiBhattMD) - Moderator Algorithms and artificial intelligence in telemedicine: Making #TelemedNow Work for Us Q1 What clinical input is essential to making AI work for #TelemedNow ?

A1 Notable Responses: Shereesepubhlth (@ShereesePubHlth) Clean data, a great idea in theory, but can we get this right? I'm not sure. #TelemedNow Salim Saiyed (@SalimCMIO) I think it has to start with the clinical question of what we are trying to solve. And then what would #augment a #physician workflow or mind to help diagnose, treat and manage a patient or population via #telemednow Janice Tufte (@Hassanah2017) Documentation on Data collection of #PatientHealthLiteracyLevel #DevicePreferences #PatientAccess #RPMData #MedicationManagement #ICDCodes #PatientPHBH #PatientTeleComfortLevel with #TelemedNow Aditi Joshi (@draditijoshi) What problem/issue you are trying to solve. in EM, we would want to use to triage patients at home to level of care, requiring chief complaints, branching for specific sx, hx, and red flag/triggers that could be virtual care, urgent care or emergent. Andrew Watson (@arwmd) And Aditi, maybe #AI could identify the highest risk patients, and the ones you need to monitor at home, or do a video visit with the next day? #TelemedNow Andrew Watson (@arwmd) It has to be a hybrid of #EHR data, #PGHD, #RPM data, and hopefully #HIE data. One big question is not just the data, but also the time intervals between data points. Salim Saiyed (@SalimCMIO) more #data does not always mean #good data in #healthcare or #telemednow. we have to carefully select which data points are included, monitor the #outcomes of #AI before widespread use Helen Burstin (@HelenBurstin) #AI should be designed to meet #telehealth needs of patients & clinicians. Clinical input can identify systemic issues where #AI integrated into workflow can improve care. For example, use #AI to find "signal" in #RPM data "noise” for chronic illness management. Ritu Thamman (@iamritu) EKGs: a rational, appropriate application of sequential ML methods has developed a framework for expanding the utility of 12-lead #ECG data to quantitative applications beyond its current uses #TelemedNow #ArtificialIntelligence @CircOutcomes Andrew Watson (@arwmd) With all of these different data types of channels, do you think we need a master table? Much like semantic interoperability? But focused on #Telemedicine? #TelemedNow Aditi Joshi (@draditijoshi) exactly depending on the specific triggers. it's more algorithm than ML. however if you get data sets from your patient population can start with inputting. perhaps we find out there is more local variance in dx and health due to environmental factors than we thought. #telemednow Ami Bhatt (@AmiBhattMD) Yes! #telemedicine is not a single point in time but a compilation of data that we use to help manage patients better @jpkca @jstclair1

Ritu Thamman (@iamritu) And #MachineLearning approaches is best if they openly share the source code used during the analysis with readers #TelemedNow #ArtificialIntelligence @rashmeeushah Shereesepubhlth (@ShereesePubHlth) I feel like this question has a lot to unpack. What problem are we looking to solve, vs. what will make the process most seamless for the patient. What workflows are currently in place? Tricky question, IMHO #TelemedNow Andrew Watson (@arwmd) I haven’t thought of this before, but can #AI somehow filter bad data and only input good data? Is there some algorithm that does this? #TelemedNow Obviously there is a learning phase, but once past it , can only good data be the sole input? Matt Sakumoto (@MattSakumoto) All clinical input. I am personally wary of any closed loop AI. Early stage should have eyes on for validation (treat almost like a clinical trial) #TelemedNow Salim Saiyed (@SalimCMIO) YES we can filter data #telemednow. but most #physicians #clinicians have a trust issue of missing something important Jim St. Clair (@jstclair) #AI can't really filter "bad" data. "bad" data in terms of..what? #telemednow Kamala Tamirisa (@Ktamirisa) #Telemednow ECG + AI: For DNN to work, need effective - Data storage platform - Abnormal vs abnormal recognition - list of inputs/diagnoses for recognition. Judd Hollander (@JuddHollander) I think most providers dont want to filter more data. They want help rather than more work. concerned more data may be more work without better quality. #telemednow Jennifer Co-Vu (@DrJenniferCo-Vu) In my complex #CHD patient population, #AI can identify red flag parameters to trigger a visit - O2 sats < 75% - No weight gain for 2 days - Weight loss - HR > 170 bpm And directly offer #TelemedNow appointments next day.

Q2 In the ideal #telemedicine model for your practice, which aspects would be driven by AI? A2 Notable Responses: Janice Tufte (@Hassanah2017) From the Patient Perspective: #DiscernablePatientHistory #ReasonforVisit #PatientPreferences #Values & #Goals #SDoHDataCollection #telemednow Shereesepubhlth (@ShereesePubHlth) Preregistration, H&P collection, patient engagement (chatbots & appointments) can all be aided greatly by #AI. #TelemedNow Helen Burstin (@HelenBurstin) #AI would logically support #telediagnosis, but we need more evidence that it works. I’m a big fan of this @mdhowell @google quote: “Have a little skepticism that AI will tell you the right way to treat every patient. It is math, not magic.” Ami Bhatt (@AmiBhattMD) I will start. Vital signs should be a natural input to every telemedicine visit with not only algorithms but #AI to help guide us to the patients who are Sick (rather than not sick, my favorite dichotomy). #TelemedNow Judd Hollander (@JuddHollander) May I politely disagree. If you stub your toe, you dont need vitals. Vitals are important for some patients some of the time, but not every time. #telemednow Judd Hollander (@JuddHollander) Should it be "driven by" AI or supplemented with AI? Maybe AI should be an adjunct rather than the driver. #telemednow Andrew Watson (@arwmd) Ami 2 main areas #AI #TelemedNow can help me: 1. Virtual office / clinic - which patients may be sicker or need more attention, if I don’t have a physical exam? 2. Virtual inpatient rounding - prioritize the sick / at risk to be seen first, versus following a routine plan. Ritu Thamman (@iamritu) #ArtificialIntelligence may help with simulation-based research, that will enable higher-quality, lower-cost, & more timely evidence #TelemedNow #DigitalTransformation Andrew Watson (@arwmd) That is a great one, the vital signs. The only challenge is that the respiratory rate is either always 16 or 20. Maybe a way to automatically measure that? #TelemedNow. Ha ! Helen Burstin (@HelenBurstin) #AI should help make sense of reams of #RPM data and provide needed information at point of care. What time of day is BP highest? When do sugars bottom out? #TelemedNow Salim Saiyed (@SalimCMIO) #primarycare #telemednow usually has the highest adoption, so to start there √ #AI to get patient to right appointment - urgent care, chronic vs new problem, physical, etc. √ base on above data collection -vitals, qns, meds, etc. √ diagnose help with #ehr workflow & fu Andrew Watson (@arwmd) Jennifer - what a cool idea to impact scheduling. That is very thought-provoking. #TelemedNow I wonder if it could also assess driving time, and the expense and risk of travel and correlated with no-shows. Aditi Joshi (@draditijoshi) and potentially avoid the ER! #telemednow Paneez Khoury (@KhouryMD) seems a good place for #telemednow for my world would be using AI for complex omics data and integration into clinical immunologic care. Ami Bhatt (@AmiBhattMD) Love this idea. Do you think your colleagues would take the time to help lead the clinical review of artificial intelligence to ensure that it is providing the correct answers? #TelemedNow Paneez Khoury (@KhouryMD) Completely. It’s something we think is important to ensure we have the right approach out of the gate!

Q3 Where have you tried AI or algorithms in your practice? How did it go? A3 Notable Responses: Jim St. Clair (@jstclair) How did you specifically know you were using AI or an algo? #telemednow Salim Saiyed (@SalimCMIO) usually in #sepsis #stroke protocols and models #telemednow Janice Tufte (@Hassanah2017) Needs for #AI seen by one #PatientPerspective: #ScreeningTestsNeeded #TreatmentSuccess #PatientSatisfaction #MedicationManagement #MedicalErrorAvoidance #Deprescribing #Overuse & Undercared #SDoH #TelemedNow Ritu Thamman (@iamritu) #TelemedNow Have you seen this : @binah_ai is one of several companies that are using #ArtificialIntelligence to gauge vital signals over video Salim Saiyed (@SalimCMIO) very intriguing. but more research is needed on what does "you are stressed" on a #telemednow video mean Jim St. Clair (@jstclair) Good example of where biometric patterns can be detected and measured beyond the tele encounter #telemednow Salim Saiyed (@SalimCMIO) is this something #psychtwitter #psychology colleagues appreciate in #telemednow ? I am still not sure of use of #hr patterns on an Apple Watch on a young healthy adult that goes to #crossfit daily Salim Saiyed (@SalimCMIO) another example diagnosing afib using Pulsatile Photoplethysmographic Signals via video I think the cardiologist on #telemednow would like Ritu Thamman (@iamritu) Photoplethysmography is what @Apple watch uses to detect AFIb & in the #AppleHeartstudy the detection yield was extremely low ~.053% Helen Burstin (@HelenBurstin) The use of #AI for #Covid-19 detection is an important use case. Great health system example. Jennifer Co-Vu (@DrJenniferCo-Vu) We currently use @locus_health #RPM for our Single Ventricle Home monitoring. We have set “Red Flag” parameters- flagged on pt end & physician dashboard alerted. Our “virtual rounding” time: from one hour went 10 mins for ~15 pts! Next stop: offer #TelemedNow

Q4 Which AI driven aspects of your non-medical life do you think could translate to #TelemedNow healthcare as well? A4 Notable Responses: Andrew Watson (@arwmd) I apologize about a “silly” answer, but honestly OR turnover / bedfellow. I think our ORs run about $150/min. With case stacking, complex scheduling, emergencies, re-do surgeries at a quaternary hospital -a lot on the table here. Esp with so many surgeons… #TelemedNow Salim Saiyed (@SalimCMIO) #automate it all #ai #Telemednow Is this same price to run an #plane? Aditi Joshi (@draditijoshi) using our home devices/phones to gather health data, tracking location and a reminder when we may need care/pick up meds/get a vaccine. using face id as ID, (could it recognize changes in voice/facial structure that is not age but rather health status related?) #telemednow Aditi Joshi (@draditijoshi) cont'd: figuring out the right type of exercise/food/rest based on climate, where i'm traveling, my vitals etc. recommending relaxation methods based on my history. essentially personalizing every aspect of health to my daily habits and hx. #telemednow Ritu Thamman (@iamritu) #chatbots Although they need to earn consugmer trust because they do a poor job of handling complex conditions but it’s a powerful #ArtificialInteligence tool that can check symptoms thru video chat, gain medical advice, & prescribe #TelemedNow Andrew Watson (@arwmd) Good answer Ritu. @MattSakumoto was Discussing this recently @TheHealthSummit webinar as well. I know @HCAhealthcare used them for their front line Covid work. Helen Burstin (@HelenBurstin) #AI already enables my (excessive) online shopping: If you like A, consider B. Could #AI support virtual shared decisionmaking by sorting through patient data to ID key decision points & options and offer targeted videos for digital waiting rooms? Matt Sakumoto (@MattSakumoto) The future is now for remote robotic surgery! #TelemedNow #teleSurgNow #5G Ryan Louie (@ryanlouie) With daily life becoming so connected with tech, #AI can maybe help with #cybersecurity #infosec. #TelemedNow Judd Hollander (@JuddHollander) We also used a COVID-19 chatbot. Tied into CDC recs, allowed one to escalate to video visit and schedule an appointment for COVID testing. Used several thousand times during peak of our cases. #telemednow Matt Sakumoto (@MattSakumoto) Similar setup at @UCSFHospitals @UCSFCDHI with a MyChart ChatBot to triage and allow patient self-scheduling (HUGE burden off our call center) #telemednow Ami Bhatt (@AmiBhattMD) Chatbot relieving call center is great. Did patient experience stay stable or perhaps improve?? Salim Saiyed (@SalimCMIO) Need more research but #ai #chatbot #telemednow in #mentalhealth #psychology #psychtwitter has a lot of promise btw are #chatbots claimed as #telemednow ? @iamritu @arwmd Gord Fogg (@GordFogg) #TelemedNow being in rehab we talk a lot about activity trackers. As an out there hypothesis(pending reliable trackers) use AI to learn individual PA patterns and abnormal deviation as surrogate for fatigue/HF exacerbation. Jennifer Co-Vu (@DrJenniferCo-Vu) This is absolutely true! And I like this idea! For example: If you have diabetes.. #AI could suggest “Patients who booked a clinic appointment with cardiology also booked clinic appts with.. Ophtho, Endocrinology, Nutrition, etc..” Salim Saiyed (@SalimCMIO) and #healthy food & #gymmotivation ! #telemednow Kamala Tamirisa (@Ktamirisa) #telemednow - Affordable choices: cheaper formulation - Post op recovery Q5 Which aspects of clinical care lend themselves to algorithms in your practice? A5 Notable Responses: bloomrhealth (@bloomrhealth) Patient Sign Up/In #TelemedNow #ArtificialIntelligence Andrew Watson (arwmd) Monitoring in near real time #TelemedNow satisfaction and patient comfort with #virtualcare. Also, helping to teach patients how to use #Telemedicine. We are seeing a lot of provider time wasted on teaching ptx how to set it up. #AI can help, not sexy but real $$ /time. Aditi Joshi (@draditijoshi) similar to my A1 of getting right level of care, information on what to do at home, understanding disease process. much of this can be done digitally. and eventually personalizing care. there's still so much to learn about the human body and health. #telemednow Ritu Thamman (@iamritu) Automating common administrative duties to ˅ burden borne by clinical admins & doctors, allowing focus to shift from paperwork to patients #TelemedNow Also #AI allows for 3D rendering of #whyCMR & the rotation necessary to overlay that image upon actual heart for procedures Helen Burstin (@HelenBurstin) It would be great if #algorithms and #AI could be used to identify patients at high risk for bad outcomes, detect disease at an earlier stage, and share results with clinicians AND patients! Judd Hollander (@JuddHollander) Many more than we can imagine. We need to recognize that individual variation in styles does not improve quality. We need much more standardization for most things. Then we can spend our time thinking about the things that dont fit neatly into a box. Everyone wins. #telemednow Q6 Do we think that #ArtificialIntelligence in #TelemedNow will help increase access to care in at-risk communities? A6 Notable Responses: Ritu Thamman (@iamritu) Depends who is writing the codes/algorithms for the #AI ( historically not a diverse group hence the bias) & will only happen if there is a concerted effort to include such groups #TelemedNow Matt Sakumoto (@MattSakumoto) Yes yes and YES. There's a reason it's called Diversity, Equity and *Inclusion* Let's have diverse voices included at the table #DEI #diversity #TelemedNow Aditi Joshi (@draditijoshi) wish i had a better answer than: god, I hope so. Otherwise what is the point if we can't change things for the broader population as a whole and ensure everyone can get that access. what is all this tech for then? Helen Burstin (@HelenBurstin) Double edged sword. #Algorithms based on faulty data could exacerbate disparities. But #AI built with #SDOH to ID pts at high risk could be game changer. @KBibbinsDomingo shared this cartoon on importance of "place" in recent #COVID19 disparities webinar. Matt Sakumoto (@MattSakumoto) Odds are not in our favor...unless there is a concerted effort and financial/institutional incentives. Some success stories on the upside of the #DigitalDivide but it's a long uphill road John (@johnhanacek) We think the AI needs a place to "live" that gets it out of screens and hands on - @Avatar_MEDIC

8 views0 comments

Recent Posts

See All