S1E3 Texting Patients with GenAI (ft. Neda Khan, Mount Sinai)
Healthy Uptime Podcast Neda Khan (Mount Sinai) _ Jordan Cooper (Rackspace)-20260909_113357-Meeting Recording
September 9, 2026, 3:33PM
20m 45s
Jordan Cooper started transcription
Jordan Cooper 0:03
Khan, the Director of Digital Experience at Mount Sinai Health System. Neda, thank you for joining us today.
Khan, Neda 0:09
Thank you for having me. It's great to be here.
Jordan Cooper 0:12
So for our listeners, Mount Sinai Health System is headquartered in New York, New York, with 3,200 beds across 7 hospitals supported by 9,000 physicians. Now, Neda, as a director of digital experience, I think you have a lot of, well, interaction with or a mandate over patient access and patient engagement. So I think I'd like to talk about those topics today.
Let's start with patient access. I know that you've been running through a variety of change management initiatives. Would you like to provide some context on what you've been working on regarding patient access and change management initiatives?
Khan, Neda 0:49
Yeah, absolutely. So, you know, when we think about change management and patient access, we want to think about like how we can reduce patient friction and how we can reduce the complication to patients. And sometimes when we think about technology and implementing new technology, you know, we're always thinking about all the new bells
and whistles, how we can add to the workflows, add technology. And a lot of times we're not thinking about how we can reduce, you know, the workflow, reduce the amount of clicks for patients. And so my role in thinking about access to patients is
First, you know, what do patients, you know, have access to? Not every patient is going to have access to a smartphone. Not every patient has access to reliable signal. And so when we have, you know, patients, you know, in the worst case scenario, what kind of
you know, programming, what kind of technology can we provide to them that's going to be accessible to them 24-7. And so oftentimes we think about like text messaging. That's something that everybody has access to. Patients are always going to be able to access texting. They're doing it constantly.
And so how can we use texting as a way to change behavior, to get information to them about their care, about, you know, their appointments? That's just a really easy way to get information to patients. When we're changing, you know, their care, texting is usually the easiest way to do it. Instead of trying to add an app that they have to download,
Jordan Cooper 2:28
Mhm.
Khan, Neda 2:34
That always adds friction. You have to provide instructions to that workflow, things like that. Adding friction always just reduces the amount of adoption that you're going to have.
Jordan Cooper 2:45
And so what are some of the business and clinical drivers that are that that you're trying seeking to address with this text messaging?
Khan, Neda 2:53
Yeah, so some of the business drivers there is one cost. That's the number one. When we're thinking about new technology, text messaging costs almost nothing. It's like.005 cents to deliver a text message to patients. And then on top of that, when we think about, you know, the clinical side,
You can actually do a lot with text messaging when you're thinking about remote monitoring. You can do, you know, collect patient reported outcomes through text messaging, conduct, you know, symptom reports, symptom surveys. Tons of information can be collected and sent through texting. And so we can actually
you know, move a lot of the needle through text messaging to help improve patient behavior and patient health outcomes downstream.
Jordan Cooper 3:50
I know that you've been working on prioritizing digital and AI solutions, as has everyone across the country, to improve patient access and utilization. How are, I guess, how is text messaging supporting this, or what other initiatives are you pursuing to improve access and utilization? And I guess,
I'd love if you could ground it. For example, you said remote monitoring or patient reported outcomes, kind of like, you know, what's very specifically, what's one application where you've been seeking to leverage in order to improve access and utilization?
Khan, Neda 4:26
Yeah, so one thing that we've been prioritizing is, and it's actually this functionality through Epic because we are an Epic shop here at Mount Sinai, which is called Cheers Campaigns. And it's these text messaging campaigns that you can actually build based off of
demographic information of the patient based off of clinical information. And it basically builds these like text messaging campaigns to promote either, you know, patient behaviors, promote
you know, appointment reminders. So we have a few different ones that are coming up. We have ones around no show, no show recommendations. We have ones around breast cancer screening, other different types of cancer screenings. We also have ones around
Jordan Cooper 5:22
Yeah.
Khan, Neda 5:27
Uh...
Jordan Cooper 5:29
So I think, Neda, you work with the nudge unit to implement these kinds of nudges to improve patient engagement. Does the nudge unit kind of work with breast cancer screenings and no-show recommendations? Can you explain a little bit about how the nudge unit works?
Khan, Neda 5:46
Yeah, so the nudge unit also worked with those types of screenings. And essentially, the goal for that was to really implement small changes in the patient workflow to really nudge them towards the right direction. So whether that be sending a small
you know, message to remind them to do something or, you know, sending them an educational reminder or changing a default in the workflow to nudge them towards the right decision and making a different change.
Jordan Cooper 6:23
So how are you supporting these kind of programs in the back end? Are these, what sort of infrastructure is supporting this text message or nudge unit programming?
Khan, Neda 6:37
There are a couple different infrastructures that we have in place. So, you know, as I mentioned, the EHR is the big one where we're pulling all this patient data, but then we're connected to our text messaging vendors. We also have an analytics team that we work with to
help drive a lot of the, you know, a lot of the decision-based like logic to, you know, send these messages to the right patient populations based off of the different criteria that we have for the different campaigns.
Jordan Cooper 7:14
So you're using, you're pulling clinical data from the EHR, you have text message vendors that that data is getting sent to in order to push to the right patients, and you have an analytics team identifying which patients you want to be reaching to.
Has there been have there been any technical challenges or challenges in staffing to manage all these applications and the servers and infrastructure uses supported, or is it hosted in a hyperscaler or on-prem? Kind of what are some of the technical challenges of maintaining?
kind of a viable program.
Khan, Neda 7:54
Yeah, so a lot of this is hosted on the cloud, but I think the biggest challenge that we face is with the scalability sometimes, you know, our, actually it's the texting vendor. Typically, if we're pushing out a lot of text messages at once,
Our texting, or actually our cell carriers can sometimes read mass messages as spam. So sometimes messages can get filtered out. And so that is a risk for when we're doing bulk messaging to patients through these texting programs.
Jordan Cooper 8:31
Mm.
Khan, Neda 8:34
Other issues are, you know, cell service. You know, if a patient's not near cell service, they won't get the message. It's not that they, it gets delivered later. They just don't get it. So that's also an issue to be concerned about too, when you're relying solely on text.
Jordan Cooper 8:51
Right. And so how are you leveraging AI to drive these outreach campaigns?
Khan, Neda 8:56
We are, yes, because AI, a lot of this is natural language processing. So when we're having the conversations with the patients, when they respond back in, we're using that AI to help drive what, you know, what the response is and what the, you know, response back should be based on that patient incoming response.
Jordan Cooper 9:19
So you're saying that a message gets sent out to the patient, the patient replies, and then you leverage some Gen. AI capabilities to automatically respond to the patient's reply and kind of go back and forth in a conversation?
Khan, Neda 9:36
Right. And it's still contained to a certain number of responses, so we can follow the right tree of our programming to get to the end result. But it's still, it's smart enough to know that like a yes is the same as yeah, or the same as okay.
Jordan Cooper 9:48
Mhm.
Khan, Neda 9:58
so that it doesn't get stuck along the conversation.
Jordan Cooper 10:02
At what point does a person read the messages or respond to the patient instead of the Gen. AI?
Khan, Neda 10:09
There would be some trigger points along the way. And sometimes it's when, you know, the patient has responded with a certain keyword like, you know, stuck or, you know, stop. Those are certain keywords built into the system that would trigger
Jordan Cooper 10:30
Mhm.
Khan, Neda 10:31
a human to step in. We also have some failsafes in there too, where, you know, if at any time a patient needs to speak to a human, they can indicate it by texting humans so that it can trigger that failsafe.
Jordan Cooper 10:48
And what about instead of just a human text message response, a human call or a, yeah, I guess a call, is there ever a time when somebody would actually call the patient?
Khan, Neda 10:59
That is mostly for our like remote monitoring programs. And it's typically for, you know, if a patient's indicating, you know, they're feeling worse or, you know, if they've triggered a response that requires that human call. So, for example, we, you know, at the nudge unit, we had implemented a few
Jordan Cooper 11:13
The.
Khan, Neda 11:20
blood pressure programs. And if a patient's blood pressure seemed to be very, very low or a very concerning blood pressure, that would trigger an alert in the Epic system to a provider that was on call to call the patient within the hour.
Jordan Cooper 11:27
Mhm.
Got it. And that's all from data that patient input into the text messaging program. Do patients need to opt into this program?
Khan, Neda 11:44
Yes.
So it depended on the program and what, you know, what we were able to get approval from our privacy team on, because it really, you know, some programming, you can have a default for an opt out, opt out enrollment for some low risk programs, but then for higher risk programs, you did require
Jordan Cooper 12:04
Mhm.
Khan, Neda 12:12
opt-in enrollment.
Jordan Cooper 12:14
And then what have been patient reactions to this text messaging engagement program?
Khan, Neda 12:21
We've actually had really high feedback, you know, NPS is in the high 90s on a lot of these programs because it is, as I mentioned, it's just part of the regular workflow for a patient to like be responding to text messages versus having to log in to their MyChart app and then
respond to a MyChart message from a provider, you know, you're having to constantly log in, remember your login. That workflow is just really annoying to a patient. And so being able to use a text messaging system to do the same thing is much more easier.
Jordan Cooper 12:59
Did you have any controlled group of patients who did, who are compared to the text messaging patients but did not receive text messages to evaluate the extent to which this intervention with the text messages has had any positive impact on patient adherence and patient outcomes?
Khan, Neda 13:17
Yes, at the nudge unit, we had done several RCTs to test, you know, how the text messages compared to a survey or even sending messages in the MyChart system and texting always prevailed over other usage systems.
Jordan Cooper 13:36
So, texts prevailed and meaning that there was more responsiveness.
Khan, Neda 13:41
Yes, adherence, satisfaction was always higher.
Jordan Cooper 13:46
Okay, and okay, so you've been finding a success with this program and success with leveraging AI. What kind of, have there been any AI governance challenges that you've run into or what has the process been like to bring on vendors or to roll this out from a pilot to enterprise-wide?
How have you gone about those processes?
Khan, Neda 14:11
Yeah, so I would say at Mount Sinai, it's been a different process than at the Nudge unit. At Mount Sinai, we actually have a really robust AI governance process where we run an assurance lab. So that is essentially a lab that like pilots out
these use cases at a really small scale prior to us launching with patients to do some testing with the vendors, with the LLMs, in, you know, a controlled test to ensure that it's going to be, you know, viable and successful when we launch this with patients.
Jordan Cooper 14:51
How large of the team, like what's the, I guess, is this a program that pays for itself or did you need to have leadership allocate FTEs to manage the texting vendor and to manage the public cloud instances in which this program is hosted or to manage the APIs with?
Epic, kind of what, how has that worked?
Khan, Neda 15:16
Yeah, so we have a digital team already and they manage our other digital applications. And so it didn't take additional FTE to manage these programs. I would say where it does take some additional FTE is on the operational side. So when
you know, we have alerts coming in for that low BP or, you know, a concerning message from a patient. We need that, you know, provider on the other end or some operational partner to respond to those messages. And so we did have to ensure that there were FTEs on the other side that would be able to
You know, staff these programs.
Jordan Cooper 16:01
Was there an executive champion of this program? I have to imagine there needed to be clinician buy-in here. Was there the CMIO involved or anyone of that nature in order to ensure that if physicians or nurses or NPs or PAs were required to respond in the instance of, for example, low blood pressure,
Khan, Neda 16:02
And.
Jordan Cooper 16:20
that they actually had availability and interest in doing so.
Khan, Neda 16:24
Yeah, at Sinai, we have our Chief Digital Transformation Officer, who's our key executive sponsor. And then we also work really closely with our Chief Population Officer on a lot of these remote monitoring programs.
Jordan Cooper 16:41
And I suppose that a lot of these patient reported outcome measures kind of feed into HEDIS measures or other public reported quality measures to CMS and effects reimbursement rates. Have you had any challenges kind of tying your program success to improved revenues in order to secure funding? Or were there any other hoops you had to jump through to secure initial funding
to try out this program when you are getting started.
Khan, Neda 17:09
Yes, I would say that there definitely have been some challenges initially, especially with, you know, some of the CMS funded programs. You know, for the blood pressure programming, some of them where we were actually providing cuffs to patients required initial funding from
the health system to, you know, be able to provide cuffs for every patients because we knew they didn't have cuffs at home to be able to take their blood pressure. So that took initial, you know, business case providing, you know, showing that the outcomes on the end would have that change and then be able to provide
ROI at the end, preventing patients from, you know, requiring extra hospitalization, you know, reducing readmissions, reducing visits that would eventually lead to, you know, saved costs.
Jordan Cooper 18:05
So how long, I don't know, did you do a retrospective analysis to see how long it took to prove your ROI that you originally projected to the executive sponsors so that they would pay for those blood cuffs, blood pressure cuffs?
Khan, Neda 18:18
Yeah, so we were still in that process because these programs are still very early on. I would say they, we've been live with some of them for six months. I would say that we had more data at the nudge unit to prove our ROI for some of those programs.
Jordan Cooper 18:36
And how did that go with the nudge unit?
Khan, Neda 18:38
It went well. We were able to prove we had, I think, saved, I think it was like a two to one or a three to one savings for every patient.
Jordan Cooper 18:49
That's $3, for every dollar spent you save, you generate, save $3 or generated $3 and okay, saved, not generated. Okay, thank you. So again, Neda, we're coming up to the end of this podcast episode. And just to remind our listeners, we've been discussing patient access and patient engagement at Mount Sinai Health System.
Khan, Neda 18:53
Okay.
Saved.
Jordan Cooper 19:12
talking about text messaging and talking about prioritizing AI solutions to improve patient access and utilization. Any advice that you'd give to yourself a year ago, maybe when you began some of these programs that you'd like to share with people listening right now who may be interested in replicating your experiences at their own health system?
Khan, Neda 19:35
Yeah, I mean, I think a couple things that I've already mentioned are like, you know, thinking about how you can reduce friction instead of adding friction to the patient workflow, thinking about, you know, as much as possible, how you can build in defaults instead of thinking about how you add in reminders or education. I think that's
typically what people think about when they're thinking about how to change patient behavior or increase engagement. And then I think one thing that is a new thing from, you know, the behavioral economics perspective is how you add in gamification into the way that you build in, you know, patient engagement in some of these programs.
How can you add incentives to the patients? You know, add in like streaks, games to like get them, you know, thinking and promoting positive behavior.
Jordan Cooper 20:29
Fun. Gamification, streaks and games. That's a great way to end it. So for our listeners, this has been Neda Khan, a director of Digital Experience at Mount Sinai Health System. Neda, thank you so much for joining us today.
Khan, Neda 20:42
Thank you so much. It's been great.
Jordan Cooper stopped transcription
