Reducing EMR Burden with AI

[00:00:00] Karl Pister: Welcome again to the Healthcare Leadership Excellence Podcast. We are always delighted to have you with us. We know you could be spending your time in many other places, and we’re happy you’re with us today. And you will be happier with us today because you’ll walk away much better educated on something that has most of us a little terrified, because we don’t know a lot about artificial intelligence and how it applies to healthcare.
We hear all sorts of rumors. We hear about ChatGPT having errors. We wonder about how accurate that’s going to be when we’re doing a diagnosis. We wonder about how it’s going to affect electronic medical records, if it can replace it, but how it’s going to happen.
Well, we have a person today who is going to answer a lot of those questions. He is CEO of Technology River, founder and CEO , one of the premier East Coast houses for development of software and use and integration of AI into that. I would like to welcome Ghazenfer Mansoor with us today. Ghazenfer, thank you for taking time to educate us today.
[00:01:20] Ghazenfer Mansoor: No problem. Thanks, Karl. Thanks for having me.
[00:01:23] Karl Pister: Give us a little background on yourself, on Technology River, and let’s get started here.
[00:01:35] Ghazenfer Mansoor: Okay. My name is Ghazenfer Mansoor. I’m the CEO of Technology River. We’re a healthcare software development company. We started as a regular software development company, but our second customer was a health tech company.
Then we started building different applications, and gradually we started building a lot of HIPAA-compliant applications. So we work a lot with health tech companies, helping them create HIPAA-compliant, innovative software products, as well as helping healthcare services businesses improve their operations through AI and technology. AI is a big part of what we do. HIPAA compliance is a big part of what we do. We’re based in Virginia.
[00:02:23] Karl Pister: Excellent. What background brought you to being CEO of a development company on such a cutting-edge subject?
[00:02:33] Ghazenfer Mansoor: I have a computer science background , I’ve done a master’s in computer science, done development for various different types of companies. I worked with two startups as an online engineer. So the tech background is one part of it. Many times, as tech people, we think that if you’re a tech person, you could run the company , but that turned out not to be true. There was a lot of learning along the way. I created a recruitment SaaS startup, so there was a lot of learning there, but in 2015 I started this business.
It came along as part of a gradual, natural process of helping businesses. As I was moving up in my career, I was helping companies with product development. Along the way, I realized we could set up our own thing , we weren’t just helping one or two companies with these products.: That’s when I decided to start this.
[00:03:39] Karl Pister: I know a lot of people hear “expert” and think, “Yeah, that’s fine, but they don’t understand healthcare.” So share this , you’ve done dozens of healthcare platforms and applications. I think you mentioned around 45?
[00:03:57] Ghazenfer Mansoor: That’s correct. You’re right on 45.
[00:04:00] Karl Pister: So as you’re listening to Ghazenfer Mansoor today, please realize he’s been waist-deep, neck-deep in a lot of the questions you have today, and we’re going to answer some of those. I don’t think anyone can take a breath in healthcare without wondering if they’re stepping on something with HIPAA, if they’re going to get sued over something.
So when you’re talking about HIPAA-compliant software, walk us through what that actually means. You talked about how AI can help with that. You also mentioned device recording for SOAP notes, which is familiar to most healthcare people. Walk us through some of that thinking.
[00:04:49] Ghazenfer Mansoor: Okay. When it comes to HIPAA, a lot of the time people think about a checklist of things you have to do to comply. Yes, there’s a checklist , that’s the compliance part , but I think the bigger part is the trust part.
As we’re working on these products, the users, whether patients or providers, need to have confidence that this application is HIPAA compliant, because there’s very sensitive information involved. So trust is a bigger part, and how you build that is also part of the process you have to bring in. As we’ve worked on many of these applications , as you mentioned, 45 healthcare applications, and out of those, 24 are HIPAA compliant , we’ve seen a lot. In many cases we’ve started working on a project that was previously done by somebody else; sometimes we’re assessing existing software for compliance.
A bigger misconception people have is that if you just encrypt it, it’s HIPAA compliant. That’s just one small piece of HIPAA compliance. Think of it this way: in a traditional doctor’s office, you see those big filing cabinets , they’re locked. The doctor has access, the nurses have access, the front-desk person may have limited access, but the only way traditionally was to put locks on things. It’s hard to know when somebody opened the file , who actually looked at it? Was it a receptionist? Was that being recorded? Those are administrative processes, but you may or may not have had them.
When it came to the web, that data became available in many different places. Now you have to make sure that if you have a computer screen open and you leave for the bathroom, the data doesn’t leak. The next part was mobile apps. Your app is HIPAA compliant, but if you drop your phone somewhere or lose it, can you make sure the data doesn’t leak if someone gets access? And then the next step is AI. Everything becomes another layer on top, because each of these brings different complications. With web and mobile, you can achieve a lot of things you couldn’t in a traditional way.
Now you can look at the data , for example, if I’m a provider looking at it, we can track exactly how long somebody looked at a certain page, what permission they had, who the person was , was it a nurse, staff, a billing person? They all need data for different purposes.
Does a specific person need to see a specific prescription? Then you can restrict that kind of access. If you’re on a list, you might see multiple patients with just a name and address, but if you go into detail you may see a lot more, like prescription notes. So these are more refined controls you can put in place , access control, audit logging , in the web application as well as on mobile. On mobile, and even now on the web, you can use fingerprints or Face ID, which connects you to your user so you can handle authorization more easily.
In the past you had to remember all your passwords. Now it’s making things easier as well. The bigger part now is AI. As soon as AI came in, now you have all that data, and if you start using AI to get answers from data that contains PHI, there’s a big risk that your data is being used to train models. So when it comes to compliance, we have to make sure we anonymize or remove any PHI-specific data before sharing it.

You do need that data to create predictions , say an 80-year-old patient has X, Y, Z; if you’re seeing a pattern, you can create predictions without providing specific identifiable data. So these are things you have to be aware of when it comes to AI.

[00:09:52] Karl Pister: Walk me through what is probably the nightmare for any clinician. I’ll give you a quick story. I was talking with a cardiologist years back, probably right before the pandemic, and he was working a pretty tough workload. He said, “Carl, I go from nine to six every day with my clinic.”
I thought, “Okay, that doesn’t sound too bad.” I think he caught my look , “what’s the problem?” He said, “Karl, what you don’t realize is that’s probably 30 to 35 patients a day. For each one, I have to spend a minimum of 10 minutes on notes.”
Suddenly the math hit me. This guy was doing 350 minutes, nearly five hours a night, just on Epic or Cerner , it was Epic in his case. 

So put on your crystal ball hat here for me, Ghazenfer. How do you see either the work you do with Technology River, or AI in general, helping that poor cardiologist reduce his clinical load of EMR work? Any thoughts on how that can take the weight off?  It’s the number one reason for physician burnout and people leaving the profession right now , the electronic medical record. Thoughts?
[00:11:37] Ghazenfer Mansoor: Yeah, absolutely. It’s very normal. When I go to the doctor, right after the meeting, I often see them recording , “I talked to this patient, he has this, I issued this prescription.”  That voice note gets sent to another company for transcription, someone logs the notes, and it comes back ready the next day. Now you can do a lot of that through AI right away.
As you mentioned, the SOAP notes app , one of the apps we created lets you have an app right in front of you, put the phone down, record the conversation with the patient, and it generates the SOAP notes. Whatever was happening traditionally through an outside company, someone listening and transcribing, now AI does it.
Not only does it do the actual transcription, it generates notes in the format required by the EMR, including all the dates and medicine references. It can also push that to the EMR with the specific codes. So you still have to communicate some of it, but you no longer need to spend hours pushing that data. Using AI, you can create AI agents for a lot of this. Automation is a big thing , we look at the workflows, the bottlenecks, where doctors or businesses are spending time, and see if we can automate some of it.
This is one example , there could be many others, whether it’s remote patient monitoring, generating codes, pushing into EMRs, or pulling data out of EMRs. You can do a lot more.

[00:14:27] Karl Pister: Are there individual things a clinician or office manager can do to start implementing AI, or are they pretty well tied to an overall platform and specialized software? What do they have to do to get that into their clinic?

[00:14:51] Ghazenfer Mansoor: One of the bigger challenges right now is the sheer availability of so many different tools. People get confused , “what do I use?” For normal content, you have ChatGPT, but there are ten other tools that do similar things, like Gemini, Copilot, or Claude. That’s just for content , if you look at creating a presentation, another document, or any type of work, you’ll see a lot of options. That’s confusing for most users: which one do I use, do I use Perplexity for this specific thing?
Some tools are good at one thing, some are good at others , there’s no right or wrong, you just have to pick. That’s where an expert comes in , they look at the specific things you’re doing and figure out which tools are best, and how to integrate them. If you have five different tools doing five different things, you may need to integrate them, push data into your EMR, or pull from the EMR to get more insight , whether it’s improving scheduling or sending notifications.
One piece of advice I’d give anybody starting out: don’t start big, don’t start with a big project. Start with one basic thing , simple administrative stuff. Once you feel comfortable, you get a feel for what it can do and what solutions are available.
A simple example: you might load some internal documents into ChatGPT or another tool and do internal searches that stay local to your organization. There are ways to do that. By default, a lot of the data becomes part of AI training. But in a different version, you can turn that off and make sure the data isn’t used for public training. So depending on the specific data, you can start connecting and integrating with these tools and get some training on them. Once you’re comfortable, you can start producing more and more.
[00:16:59] Karl Pister: Let’s go into a bit of business talk here. Not only is AI, software, and HIPAA compliance daunting for many people, there are a lot of people out there saying they can help who maybe aren’t as helpful as they need to be. Let’s say I’m a clinic manager , I’ll refer to one clinic I’m working with. They have over 100 providers, 300 total employees, and a large orthopedic practice. Let’s say they’re looking for a really good platform that can do their software and is also on the cutting edge of AI.

Ghazenfer, can you walk me through, as that clinic manager, what should I be looking for? What questions should I be asking as I look into this for my clinic, so that I don’t drain a lot of money into something I don’t need? As CEO of Technology River, teach us what questions need to be asked to vet a good vendor for my clinic.

[00:18:28] Ghazenfer Mansoor: I think we need another podcast for this. There’s a whole lot we could talk about. It’s not an easy thing, because a lot of it is very specific.But I’d start simply , this even applies to non-AI software evaluation. We don’t go with what the marketing says , “we do X, Y, Z” , because if you start comparing things they’re telling you, you have to look at your specific use cases.

You have to list your use cases. Remember, AI is like a mirror , it can automate what you have. If you don’t have the right data, if you don’t have the right processes, you can’t automate it. Even for a normal CRM, if you don’t put the right data in, it’s not going to help you. Your EMR isn’t going to help you if you’re not putting everything into it. It’s the same for any AI medical software you get , you have to have the right data. A big part is defining your data, cleaning it up, doing the mapping , but more importantly, the use cases.
Once you define your use cases , say, every time I see a patient, I want to do X, Y, Z, or whenever we submit something, we need to be notified by a certain date , whatever the rules are, you list those and compare the software against them.

Second, the bigger part is HIPAA compliance. How do you make sure the software they’re offering is actually HIPAA compliant? You want to look at their processes. For example, in our business, we have a big checklist that makes sure all the software our team builds complies with HIPAA regulations, and we share that with our customers. 

So on the buying side, you want to check , what does HIPAA compliance mean to them? Is it just encryption? How is the audit log handled? I gave an example earlier about how different records are created , can you show me an example of how I’d visualize that?
Another big part is authentication and authorization , who is the person, and what permission do they have? Do they have access to everything? As you start using it, you’ll start to realize what the right things are. You want to make sure you understand their process , how are they vetting and verifying all of this. That’ll give you some confidence.
As I said earlier, HIPAA isn’t just about the checklist , it’s the confidence and trust you’re building by creating HIPAA-compliant applications. For AI applications specifically, there’s not a whole lot of difference when it comes to buying , the additional pieces are: how is the data trained, what tool are they using on the backend, and is our data being used to train the larger LLM? Is the data anonymized? Is it my responsibility to load anonymized data, or is it the tool’s job? And if the tool does it, how accurate is that?
[00:22:27] Karl Pister: What other questions should I be asking? What should I be looking out for as a real red flag that the organization marketing to me may not know what they’re doing? That’s where I hear so many people don’t know what they don’t know, and suddenly they’re being sold a development option they really don’t need. Any obvious red flags for basic entry-level users?
[00:23:03] Ghazenfer Mansoor: Who are the people building it , are they tech generalists, or are they aware of healthcare specifically?
Going back to the checklist , that’s important. Once you talk to them and validate a few of the steps… I think as a buyer, you also need your own checklist. Many people just ask “is it HIPAA compliant?” but that’s not enough. What does “yes” actually mean? Because that covers 20 different things , how is authentication handled, how is authorization handled, how is data deletion handled? That’s a big question , what if my business shuts down, what do you do with the data?
All of these are important. In HIPAA, there’s no true deletion , if I’m a doctor and I look at somebody’s record and say “I want to delete it,” no , I looked at it, so that record can’t be deleted. It can go away from view, but it’s stored somewhere so an audit can pull the data and show who made a query at the database level and looked at it. These are compliance steps you have to validate to make sure something is HIPAA compliant. When you start asking about deletion and data retention , what if somebody loses the device, do we get notified, is there an MDM (mobile device management) so you can remotely erase the phone?
I think you’ll get your answer there. If you’re just getting an app installed on your phone or computer without ways to delete your data or remove access, you know it’s not right.
[00:25:02] Karl Pister: I want to jump from actual development to larger AI perspectives we can pick your brain on. From what you’re seeing, what should the average provider, nurse, or clinic manager do to best prepare for AI coming into the field? What do you project will happen with AI, and how can we best be prepared? How do we open our minds to the new perspectives this technology brings, and some of the risks? Talk to us a bit about that, remembering that many of us , myself included , know enough about AI to be extremely dangerous, and influenced by myths and untruths. Teach us a bit here.

[00:25:58] Ghazenfer Mansoor: I’d start with what you said , most users don’t know, so training is a big part of it. For any organization adopting AI, the key parts are data infrastructure and people. If your people don’t know, that’s going to be a challenge, so you want to start with training.
You want to pick the right tools, and data is the important part. As I said earlier, AI is like a mirror , it’ll show you what you have. If your data isn’t right, if you’re not tracking all the information, that’s a problem. You want to make sure the tool you’re picking lets you extract all the data you could use AI to predict with. 

If a tool is proprietary and keeps everything for its own training without giving you access , say you have a large organization, 3,000 people, collecting millions of data points every month , that’s a huge dataset, but where is it going? Is it staying in Epic? Are they giving you just summarized data, or raw data? Once you have raw data, you can do a lot more , that’s the goldmine you’re sitting on. So pay close attention to the data you’re creating. 

You have to have a data-cleanup process, you have to know your use cases , how your users are going to use the application, what you want out of it. If you know the steps, you can automate them. The bigger myth on the AI side is that it’s plug-and-play , you bring in a tool, plug it in, and AI is ready. 

In some small use cases, maybe, but in most cases, it needs a huge infrastructure, a lot of training, on your data specifically. You don’t want to rely only on public LLMs , ChatGPT, Vertex, Gemini, all available for general use. What you want is training on your own data, and that requires significant infrastructure. You need to be ready for that investment as well.
[00:28:38] Karl Pister: How can one best self-educate, or where would you send me if I said, “Ghazenfer Mansoor, I want to become more versed with AI and its implications for my practice,” whether I’m a physician or a clinic manager?
[00:29:01] Karl Pister: Where would you send the average person like me to get more versed, so we’re better consumers and better adapters for this wave of change?
[00:29:17] Ghazenfer Mansoor: There are many different ways. One thing I tell everybody, including my kids and anyone who asks, is: the question you’re asking me , if you put the same question into ChatGPT, you’ll get a lot more resources than you’d find anywhere else.
I use ChatGPT even to ask what I should be asking, because you can use it in an innovative way. You can have AI agents that ask you questions. For example, for any project we do, we have our own AI agent for requirements , it asks us questions. Traditionally we had forms that people filled in. 

Now the AI agent asks questions and changes them based on your responses, so you get a lot more clarity.
You can also find organizations that specifically train people on AI for the healthcare space. But you have to start somewhere. I usually start with ChatGPT to get basic insights on what to look for, and those questions help me get to the next step.
[00:30:46] Karl Pister: Excellent.
[00:30:46] Ghazenfer Mansoor: Because when you go through training, you also see a practical way of doing it , going to the use cases. Many times these tools will give you an answer, but people are reluctant and shy about trying things.
Content is one thing, asking one question is one thing, but if it gives you a recommendation involving complicated work , even ten simple steps , surprisingly, most people won’t follow through. That’s where a workshop or training would definitely help.
[00:31:29] Karl Pister: Excellent. Talk to us a bit about how, if people have been listening and thinking, “I need to get a hold of Technology River and see how they can help me” , who are your ideal clients, who should be contacting you, and how do we contact you?
[00:31:58] Ghazenfer Mansoor: Multiple health tech organizations, any healthcare organization, service businesses with over 70 people , those are our ideal clients, because that’s when you’re outgrowing your technology.
You may have ten different tools, and now you need consolidation, or your specific workflow needs internal tools to optimize processes and get more value.
[00:32:37] Ghazenfer Mansoor: Those are our ideal customers. How do you contact us? The website is technologyrivers.com , R-I-V-E-R-S. You can also find me on LinkedIn , just type in my name. I’m the only one with this name spelling, so I’m easy to find. Website, contact, LinkedIn , and for this podcast, my domain is ghazenfermansoor.com, so you can find me there as well. In many different ways , I’m easy to find if you type my name correctly.
[00:33:16] Karl Pister: I imagine a lot of our listeners are West Coast. You go nationally, correct? Not just the East Coast?
[00:33:26] Ghazenfer Mansoor: Yes, our customers are all over the US.
[00:33:29] Karl Pister: Excellent. So again, you have a superb resource in someone with dozens of healthcare applications behind him, CEO of Technology River.
[00:33:46] Karl Pister: I’d really encourage you, the listener, to think about what you’ve heard today and use this resource if you think it’s appropriate to help you go to the next level. Because one thing I’m hearing as you talk today, Ghazenfer, is we’re going in this direction whether we like it or not, and we’d better be prepared as much as possible to be educated consumers.
[00:34:12] Ghazenfer Mansoor: Absolutely. I’ll also add a couple more things.
[00:34:16] Karl Pister: Please.
[00:34:16] Ghazenfer Mansoor: We talk everywhere, including in our company, that everybody needs to be 10x. You said it right , whether we like it or not, it’s already here. We need to be on that wagon, or we’ll be left behind. It’s going to impact our jobs, our businesses, everything. But the good thing is you can catch up. See how you can be 10x in whatever you’re doing , whether you’re a doctor, a COO, or a developer. There are many different ways to do it, and AI can help you get there. Technology is here , if you’re not taking advantage of it, you’re going to stay behind.
[00:35:15] Karl Pister: I couldn’t say it better. We can have all sorts of strategies, but if we’re not well connected to this wave of change, we’re going to be playing catch-up, and perhaps eventually we’ll never be able to catch up if we don’t adapt.
Before we close , there’s the first user, the early adopter, and the mid-level adopter. Where do you recommend people come in? Should they be on the cutting edge, or wait and see what others are doing? Any thoughts as we close up?
[00:35:53] Ghazenfer Mansoor: I’d say it’s very personal-specific , that’s just the personality people have. I think you have to really figure it out , you have to look ahead and see how the world is moving. There are some problems that aren’t solved yet, but we know some of those will improve and get solved. Do you compete on those, or pick something different? You really have to figure out how the business is changing too, because healthcare is going to keep changing. For example, post-COVID, remote patient monitoring increased a lot , it wasn’t there before. With AI, there will be a lot more changes coming. So just be ready, get yourself educated, and the sooner you get in, the better.
[00:36:55] Karl Pister: Excellent. Well, thank you for taking time out of a busy CEO schedule to talk to us for as long as you did today. I really appreciate it, and so do our listeners. Thanks for being with us today.

[00:37:08] Ghazenfer Mansoor: Oh, thanks for having me.