The report is no longer the center of your practice. Your patient is.

[01:41]

Hello, everyone. I may have just not had my audio on, so forgive me. We’ll start here about 12:02, 12:03, we’ll allow everyone to grab coffee, a sandwich, and we’ll get right into it.

[03:25]

Alright, let’s begin. I still see a few people rolling in. We’ll make sure that we send a recording afterwards, as well as any follow-up information. If anyone has questions, feel free to put them in the chat throughout, and we’ll be sure to respond to them. We do have some questions that we’ll be answering at the end as well. Alright, so I’m Dr. Barnes, and I’m the founder of PsychAssist. I’m a licensed clinical psych, and I’ve been doing assessment for about 15 years. I still see patients, and I’m still signing reports. About 2 years ago, technology finally caught up to what a purpose-built platform for assessment psychology could do. And so our team started building one. And we’ve been in the market for a year. In that time, we’ve gone over thousands and thousands of reports that have been generated in our system. And throughout that, we’ve learned a lot. And that’s really what I want to talk about today.

[04:15]

So, PsychAssist is an end-to-end intake-through-signed report in every document that comes after it. It’s not a report builder, and it’s a firm belief of our team that the document is really, subsequent to a well-crafted workflow, and the workflow is where we really put our time. So here’s what to expect over the next hour. We’re gonna spend about 10 or 15 minutes going through some underpinnings, some philosophy, some learning that we’ve made. some ways that we really think about solving this problem, and we’re gonna get into the platform. It’s also a feature release webinar, so I’m going to show you some new things that we’re doing in the platform, but it’s all built on this underlying idea that we’ve been really developing over the last year.

So, many of us in psych, assessment psych, we all have a backlog. And I think it’s really important to understand that the expertise is never really our bottleneck. It’s the keyboard. There’s so much documentation that has to happen, and everyone wants faster, quicker, and every practice has got this huge backlog, and it’s really dependent on documentation. But we really think at PsychAssist That a backlog is really a function of how quickly we can close cases, not the speed at which we can open them. It’s always easy to say yes to new things, but yet we are not done until we’re done. And because we need a document, this assessment report, to close that, it is the thing that we tend to avoid the most. And it oftentimes outlasts the time we’re with our patient.

So, we’ll do our interviews, we’ll collect our information, we’ll provide our feedback even, but that case isn’t closed until this write-up happens. But that write-up is done by the most expensive person in the practice, whether it’s a group practice, even if there are admin

[06:07]

Even if there are psychiatrists that help, the most expensive person in the office oftentimes is the one that controls that bottleneck. Many solutions in the area are really focusing on speed, and I think speed is, like, the easiest part of this, solving this problem. What happens… where it matters the most is where we… where and when we sign our name. And what we’re going to talk about in the next few minutes is really having a lot of confidence and the ability to get to that report in an efficient way, but feel really, really great about putting our name on that. And so… I want to talk about this idea that has really come up. The team has really started to put their head around conceptualizing this idea about how do we understand our patients in a full, holistic, contextual way. And we started really thinking about how do we do this in a way that technology can handle?

That really gives great feedback to a clinician, helps them along the way with their administrative and conceptualization processes, and we think about this in terms of a clinical reasoning engine, and this is something that is built into our platform, but I think even outside of our platform, it’s a way in which we can understand how to contextualize and understand data from a very holistic level. It’s not just… putting prose onto paper. It’s not just transcribing our inner thoughts on how we understand patients into a document. It’s creating a document that’s auditable, meaning we know where our information’s coming from. It’s defensible, which means we know that we can feel really great about this, and heaven forbid it get into an opposing counsel. We can feel really good about the considerations that were made along the way.

But most importantly, it’s true to the patient in front of us. At the end of the day, that’s the name of the game here. We want to serve our patients in the most effective, but furthermore, from a workflow perspective, efficient way.

[08:06]

when I think about the different tools that are available in our field, the team and I were really brainstorming about how do we just understand what’s out there and what are people using? we come up with this idea that there are about 3 main buckets, or 3 main solution categories. We think about what most people have used or even played around with. These are the general-purpose AI tools, ChatGPT, Claude, any of the HIPAA-compliant ones that we’re using. These are really good, and they make things sound fabulous. But they’re also potentially the most dangerous. And the reason being is it can smooth things together, but perhaps it isn’t looking at the full data set. I think a single source of truth of data is where we come from internally as clinicians. But when we’re looking to use technology, oftentimes that single source of truth doesn’t translate, and we’re having it do things that, again, makes beautifully sounding prose.

But it might be missing some signals in the noise. It may not be surfacing some of the nuanced information that we need as a clinician. The next area are the more refined report generators. Now, these are… built for assessment psychologists, school psychologists, those who have a particular interest in generating documentation. They solve a real problem, and oftentimes. they will also make things sound very, very good. But if we’re not stemming from a single source of truth, or this full contextual understanding of a patient, all of the information available to us. It may gloss over, or it may seam together some divergent information. And when there’s divergence, and it’s hidden in the record… It’s unfortunate, because sometimes that divergence, oftentimes that divergence, is the signal we need to be paying attention to. If there’s a teacher report that doesn’t match a parent report, they’re seeing different things. Well.

Potentially, these two categories may smooth that over, may make that look a little less important, but as clinicians, we know that that divergence is very, very valuable information.

[10:11]

Where PsychAssist has always been, and is continuing to go, and the team really feels this very firmly, is that we need a full understanding and an engine behind that. So a clinical reasoning engine is something that’s much more holistic. It works with divergence, it works with clinical reasoning. Now, I should note that PsychAssist or any other platform should not be making any decisions for anyone. As a clinician, we went to school for a long time. We understand the signals and the noise. And at the end of the day, we’re putting our license at the end of this document, and our name on that. And so, when we use a clinical reasoning engine, the work should not be to do the clinical side of things. It’s to get to a place where we can have a great document that feels great, is defensible, you can… it’s auditable, meaning you know where all the information comes from, it ties it to its source.

And it makes us much more efficient as a clinician. The goal of these tools should be to include… increase efficiency, excuse me, rather than do the heavy lifting of the thinking, which is what we all went to school for. When we think about, maybe even at that top level, the general purpose AI, I think of those as really fast typewriters. Great, great prose, perhaps not full context. Report generators, again, they’re great, and they are also very fast typewriters, but they went to graduate school, they have some more expertise. When we think about a clinical reasoning engine, it is this… This method in which the data we feed it is the gas, and the engine is the synthesis. I think personally about my own clinical work, and I think about when I’m gonna put my name on something.

I think about how I need to feel when I do that. And I have these two ideas. I need to be able to know where this evidence comes from. Not only the evidence that is supportive of my diagnostic outcome, including the rule-outs and considerations, I need to know what I’m missing, anything that might contradict it. Because again, those are signals and noise that are very important to us. And I need to know where the information came from, I need to tie it back to its source.

[12:19]

And that’s also why PsychAssist doesn’t compete in a 2-minute report. What we want to do is we want to make sure that we’re not sacrificing any quality for time or speed. But rather, we’re looking at the full record We’re making sure that it’s quality over speed. And more importantly, anything that we’re putting in our documents, such as claims, it’s going to take longer to write, but that’s fine, because it’s going to carry some provenance with it. Throughout… analyzing millions and millions of data points over the last two years, particularly very heavily in the last year. The team and I were sitting down, and we were talking about what needs to happen, what kind of standards are there for this industry, and really, there aren’t many.

What there are, however, is a lot of tangential things. There’s some great people writing about this stuff, but there still doesn’t seem to be an industry standard. So we brainstormed, we had a wonderful whiteboard session, we boiled this down to about 6 different things. We want to have provenance. We want to be able to identify and surface disconfirmation. We want to have clinician approval. There shouldn’t be machine-generated information going into a document without any approval process. There needs to be an approval gate. We want to make sure that it carries the voice of a clinician. We all hold that very deeply and important. And anything that’s written by any sort of technology, and this is the question I get all the time, is it going to sound like me? Well, of course, it’s an important part of your work.

In addition, we want to make sure that any claims that are made are auditable to the degree that we can go back and see when a clinician was doing what it was doing. Are we able to surface the interactions with the patient file? If we’re using a chatbot, are we able to get those chats? Are we able to look at that? More on this later, and that’s getting a little scary, but I want to talk about it, because I think we’re using technology these days in a way that we don’t fully understand.

[14:07]

And furthermore, are we going to disclose this information to our patients? And there’s a lot of consideration about that, and there’s a lot of anxiety around that as well. Nevertheless, you could look at the website that I have, psychassist.ai, you can go to the bottom and look at the standard. It’ll be sent to you after this as well, and it’s a pretty full text around all of the considerations that we want to think about as we’re using some of these tools. So we’re gonna move into a case here in just a moment. We’re gonna get into the platform, we’re gonna walk around, talk about how PsychAssist can really meet some of these needs that I’ve developed so far. One of the things that we’re going to do is we’re going to talk about how we can use a configured system, a specialized system, to help us argue against our case, to surface some divergent information, to help us find some holes in where we may not be communicating as clearly as we could. Not because we want to sound smart.

It’s because we want our patients to have a great outcome. We want to give them as much information as we can. And if we can find the weaknesses in how we’re writing, even though we don’t think we have them, other people think we do. And so if we’re able to preliminarily have a case conference, a peer review session, some of our work. We’ll be able to show that in the platform. as it relates to provenance and where the claims are coming from, we’ll be able to show you how it’s an important variable in our consideration. If we’re going to make a claim, we want to have the source. We want to be able to tie it back to a document, a data point, etc. And lastly, we want any generated document, at least from my perspective with a clinical reasoning engine as the forefront, to be able to read across the whole battery of information. I would rather not say, here are the scores, please write it like this.

that’s well and good, and it’s a fast typewriter, and if that’s your need, then that’s great. But when we think about a holistic view of our patients and how we’re communicating that through our documentation.

[16:00]

My belief, the team’s belief, is that we want to read across all the information. And you’ll see when we get into the platform that the goal of the platform is not just to produce documentation, although that’s what we all want. The goal of the platform is to create a single source of truth for our patients, so we don’t have 14 tabs open. We’re not looking at our Google Drive, and a paper folder, and any of the other platforms that you may be using. When we can create a single source of truth that does two things. The first of which is any clinical reasoning engine has full access to all of the nuance of very structured data. It’s not a PDF it’s trying to read, it is data points that have been extracted and approved by a clinician.

And secondly, we want to think about how do we speak holistically about our patients. You know, we all have our reports, and in certain sections, it is a bit more holistic, and in other areas, it’s much more acutely understood. But when we can read across the whole battery of information, the outcome is much more robust. So I want you to think… we’re gonna get into the platform in a moment. I want you to think about a place we’ve all been. We’re going to use some synthetic information here, but it’s not a synthetic run. We’re using the platform on synthetic information, but a configured, real workflow. We’re going to use Maya, we’re actually going to look at a few patients here. And Maya’s coming in for a child ADHD assessment. We’re just about ready to write this report. Our workflow is such that we’ve done all the information gathering, we’ve done our interview, we’ve done our assessments, we’ve collected our information, we’ve even scored them.

And we’re going to get into the platform and chat just a bit about how we can get to this document in a really structured way, with a wonderful infrastructure behind it, and a clinical reasoning engine on top of it. Before we get in there, though. I want to think about some very important things. Auditability is something that is, I think, very important. We want to know who does what and when, particularly for larger teams. When we have multiple people in the file, whether they’re psychometrists, students, clinicians, other clinicians, clinic directors, department heads.

[18:00]

If someone’s interacting and making some sort of decision about a data, accepting it, closing a record, looking at a transcript, we want to be able to trace that back. We also want to make sure that we’re classifying documents and information. The single source of truth is a structured set of data, and when we’re putting information into a system, we want to make sure that a clinician or someone with their name on the line is… has the authority for which we are configuring that. We want to look at extraction, how we’re pulling information. A system shouldn’t necessarily just be doing it for you and saying it’s good. A system should be extracting the key components, and even some of the signals and the noise, and a clinician, someone with the authority and the experience, should be saying this is true, or editing it otherwise.

And as I’ve said many times, the report generation should be a synthesis of the whole file at once. Of course, background section, we know that that’s factual information. There’s not a lot of interpretation that goes there, but when we start looking at data, when we start looking at the synthesis, when we start looking at how we put this all together. Having a full data set, having the clinical reasoning engine run all over that all at once, not just section by section, is where we start to see some really magical things, because we can surface some patterns. Alright, give me… a moment. So, here we are inside PsychAssist. When we get in, again, this is all synthetic information, but a real organizational configuration here.

When we get inside PsychAssist, we know that we have multiple patients working through multiple components of multiple different workflows. You may have adult ADHD, you may have child ADHD, you may have multiple different questions you are answering through your practice. And what we want to do is not only just make it easy to put in information, we want to make it easy to find information. So, I mentioned we’re going to chat about Maya. We’re going to look at how this system here can really understand a lot of information, and we’re just up to the point in this workflow where I want you to orient yourself is just before this document is generated.

[20:06]

When we look at a patient. We have workflows configured for organizations. In my practice, we have four main. We have adult and child ADHD, we have adult and child ASD. Each one of those has a different configuration. We know that we’re using different measures for the age groups. We know that when we’re answering different questions, the battery is different. We also know all of the known knowns about this workflow that will fit 95% of the people that come in for your particular stuff. We know… so for this instance, we have Maya did an intake interview. We know that we have these different measures that we’re using, so we have some asynchronous measures, we also have some things that we’re going to be doing inside, with that patient, some data collection.

If you do a school observation or any sort of ad hoc or collateral information, we can configure that as well. And for this workflow, there is also a feedback session. The reason I show this for you is because, again, the single source of truth, this full contextual data set, needs to have a way for a clinician to put in information. And when we can configure a workflow based around a particular assessment workflow. We’re able to give a clinician, as they walk through that workflow with the patient, the opportunity to contribute to that structured For instance, if we’re doing an intake interview, what we want to do is think about this in terms of a clinician’s perspective. I’m with a patient, I’m taking notes, and I can take notes, and I can put them in an EHR, I can leave them in a Google Doc, but they… they rest in a read-only state.

When we’re using a system with a clinical reasoning engine behind it, not only can we take our notes, not only can we put information in, whether it’s during or after. we wanted to do double duty. We want to have downstream effects in any of the clinical information that we’re writing about later, particularly as it relates to assessment reporting. When we’re doing an intake interview, PsychAssist allows you to do recording, of course with consent. So we were able to have an ambient scribe playing and listening to this conversation, using contextual information to understand who’s in the room.

[22:12]

To surface some of the additional signals, signals and the noise, and some of the things that may be otherwise missed because we’re busy interacting with the patient. Similarly, we can do this in multiple other interactions. But what I want to think about is, remember, we’re just about ready to write this report. We’ve already done the assessment. And the way that this works to put more structured information into the platform is through upload. Of course, you’re still going to score the way you’re going to score. You still need to use PAR, you still need to use Q-global, you still need to give them the opportunity to understand your data and give you some sort of structured output. Following that, however, when we have some standardized scores, statistics have been run, and comparisons have been made, that’s where we start to put information into this system to once again create that single source of truth of structured information.

So I left one of these. To be reviewed. And I did this for a very important reason. A clinician has to affirm all of this information. The system will never make any claims for you, the system will never make any assumptions for you. What it will do, however, is it’ll extract the necessary information. It will make that available to you as a clinician to approve, because nothing enters the patient’s dataset until a clinician has approved it. You can see here that there’s many things that we take from a document. Of course, synthetic information. We’re gonna pull the exact facts. We’re gonna pull the exact scores that we see in this document. The document’s always rendered on the left side, so you can go back and review. Should there be any edits that are necessary, it’s very easy to do within this.

Another thing, when we’re looking at a whole document. We’re gonna look at high level here, not only at the acute level of the data we have, the scored, structured data. We’re gonna look at, are there any insights that we might be able to pull?

[24:02]

Are there any high-level understanding within this dataset that might be valuable to a clinician? So not only does this serve the clinician as they’re reviewing a document and ensuring its quality and fidelity, it’s also giving a clinician an opportunity to find some of the signals and the noise. We’re all busy, we all have our blinders on, whether we believe it or not. I mean, we can be jolted every once in a while to find some of the signals and the noise. It allows us to get to a more robust data set for which we can then further communicate and even benefit the patient downstream. Again, accept or reject here, full clinical control. Once a document has been uploaded and approved, it then goes to the patient’s data file.

So now that we have everything approved, it’s time to put the final polishing on this document. Remember, the clinical reasoning engine’s running behind it the whole time. Because the clinician is the one rendering any diagnosis. I have this pre-filled for the sake of time. The clinician needs to then instruct. How are we going to then write all of this information? How are we understanding this? So the clinician then is able to add any diagnostic information, any rule-outs that they’re rendering, but that’s not where it stops, and this is where the clinical reasoning engine becomes quite prolific. For every diagnosis or rule out a clinician puts in. The clinical reasoning engine is going to then go back to that single source of truth of structured data, the full data set, and it’s going to provide that clinician with some supporting evidence within the document. It’s going to find some converging patterns or evidence across all the data set. It’s going to provide that to the clinician to find some signals in the noise.

But, as you recall earlier in the webinar. I discussed the ability to find some of the divergent evidence that exists. That doesn’t mean the system is saying, hey, you’re wrong. It’s there to help you feel right. We want to make sure that we’re finding the divergence. We want to make sure that we’re looking at all the corners, and when we can do that in a really efficient way, because we have this structured information, a system with a clinical reasoning engine is able to deliver that to you in a way that you can find it quickly and leverage it again.

[26:13]

To take that one step further, once we’ve identified the clinical constellation, whether it’s one or three diagnoses, including any rule-outs. The system will then go back to the data set. It’ll take that constellation of diagnoses, rule-outs, etc. And with that lens in mind, the clinical reasoning engine is going to find some patterns that exist. But once again, it’s going to find information that supports that claim, and it’s going to find information that might be divergent with that claim, and furthermore, cite its source. So we’re finding that ability, that auditability, how can we get back to the information? How can we cite its source? Again, the clinician retains full clinical control, and so they are able to accept or reject this.

The reason I’m showing this is to highlight the degree to which a reasoning engine can be very valuable. We’re not looking at data in isolation. We’re looking at data holistically. And for a clinician to be able to see this, which then downstream can prevent any uh-ohs, any… I wish I would have thought of those, I didn’t even think about this. And where some of the general-purpose AIs might just gloss over, we make it a point to prove it, I’m sorry, show it to the clinician so they can make the decision about acceptance or rejecting. So what we see here, I skipped a step, of course, we also want to retain voice. Now, for all of the organizations that sign up for PsychAssist, we take lots of samples, and from those samples, we extract how does this organization, or how does this clinician write? And this is always pre-filled for every, every patient before you get to generate the report.

Now, we know that sometimes a different writing style might benefit a client or a stakeholder, and so if we’re predominantly writing strength-based, clinician-oriented documents, well, perhaps this client may benefit from something different.

[28:03]

And we’re able to do one-offs. We’re able to make a different style of report that still retains your voice, because this clinical reasoning engine is… has a robust data set, and then can map it however you’d like afterwards. Another important thing here is the use of disclosure. That was one of those six that I mentioned previously. And so, the clinician or the organization has the ability to include or not include that, and it’s auditable. And so, if we ever have to show a chain of custody of data, or thinking, or interaction with some sort of chat, we have this all built into the platform. So, you don’t have to think about doing it yourself. The very last thing I’ll talk about before we get into the actual generation of the report is every account has a recommendations bank. Actually, it’s a language library that includes much more than that. But here is the very final polishing of this document. You’re able to indicate any of your own individual recommendations.

Or any banks of recommendations. So, for this individual, we had ADHD inattentive. Well, if you have a bank of those individual recommendations, you could also just include them all. what would typically happen is we would hit Generate Report. That’s gonna take, depending on the size, 10, 15, 20 minutes, but that is because we’re looking at that full, robust data set. In the interest of time, I have one already. Completed. And this is where the clinical reasoning engine becomes even more powerful. Now, before anyone has a “that doesn’t look like mine” moment, we want to make editing very, very easy in the platform. And so we deliver every preliminary first draft in a text editor. This is just simply a text editor. Every report should be edited.

And so, we’ll look at this. And you’ll see that it gives a great, well, in my opinion, a very wonderful first draft. It follows the template that I gave it, and it’s using client information to do this. But remember, we don’t just want to start editing. We want to start using a new feature that’s available to PsychAssist users. It’s rolling out today, we’re doing it case by case. And what we want to do is to utilize the technology that exists now in such a way that we can then reason or have a case conference with this whole data set.

[30:15]

We have several pre-built things, particularly based on the information and the thinking we had around the standard that we wrote, and it’s certainly available on our site. We want to be able to find some defensibility. We want to be able to find any holes in the defensibility. And so when we think about defensibility, it’s really as though you are handing your document signed, all ready to go, to opposing counsel. It’s likely that individual is going to start finding holes and weaknesses to come at you with. And if we can start thinking this through, not only because we’re afraid of going to court or afraid of being called for something, because If we think about it in those terms, even at the extreme level, it forces us to really move through how we can communicate better within these documents, or even find where some of our claims might not have as much,

Power behind them as we could. You can have a peer review with your document. a peer review is if you were to hand your document to someone else. Hey, would you read this through for me? Tell me what you think. It’s going to surface, potentially, some weaknesses, some, I wonder, moments. And the reason that this is important is, again, to tie it back to the full data set in that reasoning engine. It’s going to have the whole set available to it, and it’s going to compare all of the data that you obtained for this patient, and it’s going to then start saying, hey, why didn’t you think about this? Or you’re making a claim that might not have full support, not just in this document, maybe it’s not supported in the document because of the writing, but there may not be data to make the claim as well.

If you’re running, like I mentioned, we have… the diagnosis is rendered before we’re generating the document. Maybe you want to start thinking about it a little bit more. You want to run a second opinion. In fact, I’ll hit that now. So, for this document. Remember, this is a different case.

[32:04]

We have ADHD inattentive, and so let me just move this over to the side. It’s gonna go back to the data set. It’s going to act as though it is a peer reviewer. It’s also going to review the written text you have in this particular version, and it’s going to start asking some questions. Maybe the current diagnosis doesn’t fully account for something, maybe there’s some data in the data set that might not be, considered. Perhaps there’s some alternative formulations. Maybe there’s some data available that you can’t really separate the two possibilities. This is your case conference that doesn’t have a schedule. If you have other clinicians in your office, they’re busy, we’re busy, and so if we can do some thinking through without having to work around someone else’s schedule, or feel perhaps as even though we’re burdening them, then we can not only create a better document, which then subsequently serves the patient, we can do this on our time.

What I should also note is auditability. There’s been some scary things in the literature that I’ve come across. In fact, there was a legal case where an attorney was using, I believe it was ChatGPT, don’t quote me, and they were writing some documents. And we’ve all heard about the hallucinating fake cases, and this is not where I’m going with it. This individual was… because they asked for the chat history, and they found that the individual was Asking a chatbot. To write in a certain way, to bypass and gloss over information that would otherwise contradict their statements. The reason I bring that up is not that we all have to have panic attacks about that every day. But if we want to feel good about keeping all of our contexts together, again, ChatGPT and Google Docs are not going to talk.

When we can do it all together, it stores the audit log, the chain of custody, the patient data, and all of the documents together. To highlight that, we have this.

[34:00]

So this is going to keep a log of your interactions. Should you ever have to export them, then you have it available to you. But furthermore, you’re not sifting through the tree of chats and any of the other tools that you may be using to try to find where you were. It’s all available to you right when you need it. I mentioned earlier, this doesn’t look like me, Chris. Show me what it’s going to look like. And so we’re able to provide stylized views of reports, and this is a working document. When you finalize a report, it adds your signature, and also it has the styles to it. I’ll briefly go through this for you. So, we can make these look however you’d like to meet your organizational needs, any templated tables you have at this point. But it also is a great way to start seeing how is your document flowing? Is it looking like I want it to? Is my brand really represented here?

Regarding audits as well, or audit log, rather, I showed you what the chat can do. So we have an audit trail, a chain of custody for our interactions, our case conference with the patient. We also are auditing all of the interactions that we have with this file. Every time a clinician or a human does something, it’s going to be audited, or it’s going to have a timestamp, rather, and you can export this at any time. Heaven forbid you need to show your receipts, it’s quite easy to do so. But it will also allow any directors, it’ll also allow any clinic and organizational leaders to understand who’s doing what. If you’re sharing a patient, and many clinicians are working simultaneously, sometimes you just want to know what’s going on and who’s doing what. So we provide that opportunity to surface that information.

Now, we’ve gotten to the report. Maybe the report is done. I mentioned at the beginning of our talk this afternoon. that not only are we trying to get to a document, and I’ll underscore, I think the document is subsequent or secondary to an incredibly well-crafted workflow, there are additional documents that we need along the way. Maybe we want to send a summary to the primary care. Perhaps we’d like to send something to the school, but they don’t need mom and dad’s dirty laundry in it.

[36:10]

The system is able to take that report. Anything secondary to it. The report becomes that single source of truth. It is the final outcome of a clinician’s thinking. And what we can then do is to generate some secondary documents. We have plenty of them built into the platform. a full library. But where it becomes quite interesting is if you have your own document that you’d like to prepare, whether it is secondary or even through the workflow process, we can build that in for you as well. And then, therefore, that clinical reasoning engine and that single source of truth is able to populate this information in such a way that is Producing the information you need in the way that you need it. I’ll show you.

So if you wanted to create a document, let’s say you wanted to send something to the school, maybe it’s a… Maybe… actually, let’s think about this. Let’s have talking points. for, feedback. session. And here I’m gonna say… Give me the high level. understanding… That I’d want to… communicate. to a parent. So you can put in here whatever you need. You can just speak to it, tell it what your goal is. that reasoning engine is going to also work at an administrative level. It’s gonna say, hey, I know this is what you want, I think this is really gonna then drive it home. And so I put in, give me the high-level understanding of what I’d want to communicate to a parent on a feedback session. It said, great, I think I get it, and it’s gonna make it even better.

So therefore, this is the first step. The second step is, how do we need this thing to look? Who is the audience? What kind of tone do we need? How is the length? Do we have the right structure for you? And it’ll therefore then generate a sample.

[38:10]

I mentioned you can have these automatically triggered after a document is done, a well-crafted document, but at any time, if you need something throughout the pipeline, whether it’s an insurance authorization letter, whether it’s something that you just typically do, maybe you’re pulling from a template, you’re able to generate these in the platform. Again, this is just a text editor, but this gives you just a high-level understanding of what’s going on. And then not only can you save it to this patient, you can save it to the entire library, even just the workflow or a population of patients. So we have our chat, we have the additional documents, all of which is being driven underneath by the clinical reasoning engine, and that clinical reasoning engine is really to support our quest towards

Meeting the needs of those 6 different standards. We want auditability, we want to have it really surface that divergent information. And the chatbot, because it has access to that information. Well, we’re satisfying that as well, in addition to the audit log. Now, not only are you able to use the chat within the patient, you’re able to use it organization-wide. And I won’t show you now, because I want to move towards some questions. we have questions about our organization. How many patients do I have in this stage? Are there any unmet documents that I haven’t approved yet? What is pending? Do I have any tasks? From an account owner’s perspective, who is seeing the most patients? How long is it taking to write reports? What is the timeline between intake and feedback? What is the timeline between feedback and report delivery? We’re able to get these metrics available to us, and as an account owner, we’re able to surface that quite quickly.

So it’s a multi-layered hierarchical approach, and underneath it all is data, not only for the patient.

[40:00]

Not only for a service, or a question, or a workflow, but furthermore, at the organizational level, how can we just become more efficient? So therefore, we can reallocate that cognitive bandwidth to serving patients, to doing high-quality work, and to not having to shuffle through multi-tabs and multi-folders. So, what I hoped to have shown you, and I think I did, was being able to argue against a case, being able to do that because we have all of the information from that case inside that case. The evidence that, if any evidence diverges, what was the conclusion that was made, was it surfaced with or without being asked? Another thing is the provenance. You saw that it was citing actual data from actual sources. A clinician doesn’t have to wonder where all that information’s coming from. In every claim, you’re able to open that source.

Nothing enters the patient data set without some sort of human approval. So we have the ability to make sure that we’re putting information in, but it’s not true until someone says it’s true. No one’s going to make a mistake, necessarily, unless you, of course, make that. But we put in the speed bumps to make sure that you’re having the opportunity to review information, because garbage in is going to be garbage out, and if we have information that’s going in that might have some deleterious downstream effects, we want to catch that before we get to that, because heaven forbid it get into the patient report, and then downstream to them. We also talked about auditability, so every interaction with the patient file, it’s timestamped. You can keep track of your receipts, and should you ever have to produce them, well, then they’ll be there ready for you.

I’m wondering if anyone has any questions, and if you do, please put them in the chat. I’ll be sending a response to most of them after. We did have some pre-submitted questions. Let me get those going for you. My team put those together for us.

[42:00]

One of the questions that was submitted previously, by someone they read the standard, and they were just, what does this defensible stuff mean? What does writing a defensible report mean? Well, defensibility is making sure that you’re dotting your I’s, crossing your T’s, not from a prose perspective. But really, can you support the claims you’re making? Did you review all of the information? In the case of that legal situation I mentioned just previously, are you trying to… are you trying, or is a system perhaps glossing over very important Although divergent information. And so when it comes to defensibility, we really think about how do we put high-quality report out there that looks like yours, sounds like yours, has some of the weaknesses exposed, because we all have them, but we do this in a way before it gets to a patient.

It increases their ability to take what we have and move forward with a high probability of good prognosis. But furthermore. If we even just crank up the anxiety, we think about if this document was in opposing counsel, where are they going to go? Let’s try to close some of those gaps before we get there. And if we can do that in the system that has access to all the data, not just some, we can have more confidence in that robustness. Another question was about chat and discovery. I alluded to this earlier when we were thinking through. Now we can really look at the interactions. If I’m asking questions, I have a history of those. This case that I read was quite alarming, and, you know, when we’re thinking about those general AI tools that I mentioned very early on in our talk.

They’re great, and a lot of people are embarrassed to admit they use them, and some people even feel a little slimy for using them. When we get into a situation, or when this individual got to a situation. their chat history was discovered, it was produced, and it therefore was used against them, because it was really an active attempt to, like, produce something in a way that glossed over divergent information.

[44:03]

But when we think about clinical work, we want that divergent information. And should you ever have to then produce any of that documentation or the chat history, it’s available to you through the platform via export, but we have those pre-built tools. The tooling, then, can also do this for you in a way that surfaces divergent information on purpose, so you can then reason with it, have an interaction, even chat with that data file and have a case conference. So again, we’re not putting the juju out there that anyone’s going to have a report in opposing counsel, but if we can increase that anxiety ever so slightly about the interactions we’re having with our data, whether it be in a general-purpose AI, whether it be with a more robust engine.

That we can have a lot of confidence that we’re doing it in a way to surface that information and understand it, rather than smooth around it. The last question, I think there was a lot of anxiety in some of the questions that came in, and rightfully so. We’re in a bit of a wild time. I do believe that the Wild West is starting to end, and I think that the way that PsychAssist has been built by the team is a way to move from the Wild West and into the era — the next era, Era II. And… Someone asked, okay, so you have a report, maybe something got through, maybe you were in a rush. Who’s liable? Is it the system, or is it the clinician? And quite boldly, and almost arrogantly, it’s the clinician. Which is why we at PsychAssist have built it in such a way to surface information, to bring it to your attention, before it ever becomes a question.

If we can look at the divergence, if we can look at the information from a holistic perspective, then we’re able to surface it, we’re able to chat with it and reason with it, and most importantly, we then put ourselves at less risk. Because at the end of the day, we’re putting our licenses on the line here. And we have to have confidence in the tools that we’re using. So, I hope that through discussing the standard, through discussing the conclusions we’ve made over just looking through millions of data points, that

[46:10]

we have built a system that really meets those needs of those standards. The auditability, the citing the sources. Everyone wants a faster report, and I’m with ya. I would love a faster report. But where PsychAssist, I think, thrives is a full understanding of a data set. It’s understanding the holistic view, not just making us happy, which we’ve all perhaps seen Claude, and it says you’re absolutely right. You give it one more review, and you’re like, actually, I don’t think I’m right, how about this? And then they say you’re absolutely right again. PsychAssist isn’t a people pleaser, it’s a data surfacer. And when we can have that full context, when we can access it from various aspects, when we have human approval gates, what we’re doing is we’re increasing confidence. We’re also getting to a faster report, but not at the cost of quality. Quality is of utmost.

So to leave you today, again, please keep asking questions. I’ll be sure to respond to them in an email that follows this. I want to underscore that a report just can’t be assembled. It has to be derived. And when we can think about it from those terms, it helps us really think about intentionally looking in the corners, even if there’s some data that may not jibe with how we’re seeing this patient, that we can move through that in a way that is clinically relevant, helps our patients. But the goal of the system is not just to get to report, it’s to help the whole organization. It’s supposed to… it’s intended to clear up the cognitive and administrative burden. Of course you’ll get a faster report, but not at the cost of quality.

So, thank you all for attending today. There’ll be a follow-up with a video for anyone who couldn’t attend. Any questions that have come in that I didn’t get an opportunity to respond to, I’ll be sure to put those into the email response as well. And my email address is at the bottom, chris@psychassist.ai. Please message me directly if any of this seems interesting. If you’d like to potentially look at the platform and see how it may meet your needs, I’d be happy to have some of those conversations.

[48:10]

But the goal of today was not just to have a show and tell of the cool things that I think PsychAssist does. The team has developed a really robust platform. The goal is to communicate how I think we should all be thinking about this. And if we can all attack this from a much more higher level. If we can start to develop some standards, if we can start to think about this, and how does this then play out moving forward? There’s a lot of fear about how this could play out in moving forward. Then I think all tools in this field, all tools in this space, are going to become much more robust. Thank you all again for attending.