What's New in IHD?

Robyn: All right. Welcome, everyone. Thank you so much for joining us today for our What’s New in IHD webinar. I have two people with me presenting today: Matt Guthrie, director of product for IHD, and Becky Hollenberg, solutions manager for commercial strategy and innovation. We’re really excited to share what the team’s been building. We’ll start with Matt talking about our product vision. Then we’ll demo IHD’s newest features and capabilities live, and we’ll finish with a look at what’s on the horizon for IHD. If you’d like further demos or training, please reach out to your customer success manager or the study support team after this call. So let’s jump in. I’ll hand it over to Matt Guthrie.

Matt: Thanks, Robyn. Before we get into the individual products and features in the demos, I want to talk about our higher-level vision. It continues what we’ve always worked toward: one end-to-end platform where you can do everything you need to generate your real-world evidence. Now it’s about integrating AI on top of it in a way that feels natural and helpful. Our AI assistant, Ella, allows for that integrated flow from beginning to end.

We didn’t want to just slap an LLM on top and integrate a Claude or a ChatGPT. We wanted a layered approach that integrates our 10-plus years of domain knowledge and all the software we’ve built, so you get a much more nuanced, domain-specific application of AI. So it’s taking the IHD tool you know and love, expanding it in all directions so it’s truly end to end with all the tools you need, and then adding an AI assistant to help you do what you can already do in IHD even faster and more efficiently. That’s the overarching idea behind everything you’ll see today. Now we can jump into some specifics.

Becky: To start, we’ll talk about one of our brand-new modules, which lets you run raw SQL queries directly inside IHD. It’s called Data Sandbox. Matt, can you tell us why the team decided to build it?

Matt: Definitely. The pain point we explored is that IHD is very powerful, with a lot of tools and a lot of flexibility. But sometimes you want to do something simple, like some quick exploratory feasibility work, and then IHD is almost like using a race car to go down the street and get your groceries. It’s a hefty, powerful tool that may be more than you need in certain circumstances. So we wanted to create a new product in the ecosystem that lets you explore raw data with SQL queries for that simple stuff: quick feasibility checks, fill rates, and so on. Get in and get out. And, as I mentioned, we layered on the Ella assistant to help you do it much more quickly. Instead of the full IHD project-building toolkit, this is a lightweight exploratory environment, and it’s integrated into the IHD ecosystem, just one click away.

Becky: And what kind of data can you use in the Data Sandbox module?

Matt: It sits on top of the raw data, before it’s been mapped into the data model. That makes any data source compatible with Data Sandbox.

Becky: Great. Building on the theme of using AI in our platforms to speed your time to evidence, we also created an AI assistant within Data Sandbox, so you can spend more time on the answers to your questions and the analysis of your data instead of writing code. You can use Ella within Data Sandbox to write SQL, edit it, and run it yourself. You can pivot from Ella back to the SQL editor directly, or write your own SQL from scratch. You also get charts and visuals alongside your code, artifacts that show you what the data looks like, so you don’t just get a results table. Those visuals can then be added downstream to dashboards and other modules within IHD. It gives you exploratory flexibility and visual feedback without leaving IHD.

I’m sure everyone is eager to see this, so let’s look at it live. I’ll share the IHD homepage, which you should all be familiar with. To get to Data Sandbox, I go into the Data tab and click Data Sandbox. Here I can see my data sources and the tables available within each one. Let’s say I want to see what variables are in my patient table and how they’re named. I can expand it and see a patient ID, date of death, gender, race, and segmentation, all available to explore. This matches what I’d see in a raw data table set in whatever tool I normally use to write SQL.

I’ve set up a small exploratory question: how many patients are 18 years of age with New York as their state. I run the query, it runs live, and I see the results in real time: 266,000 patients. Now say I want to see how many patients are in Massachusetts. I just edit the SQL query directly, and I can see there are 70,000 patients in Massachusetts. So it’s iterative.

Now let’s say I want more direct answers with analytics built out, and I don’t want to go through data wrangling and coding. I can switch to the Ask Ella tab, where I can ask questions in plain language and get SQL queries and artifacts as output. Say I want to understand the line of therapy distribution in my cohort before building it, so I know the breakdown of therapies patients are on to inform my subgroup analysis and expected sample size. I’ll ask Ella, “What is the distribution of second-line therapies for metastatic breast cancer patients?”

The first thing Ella does is prompt me to pick a data source, suggesting options based on the sources I have, and ask for a date range. I’ll go into the data source panel and select a 5-million-patient sample data set. Next come the key parameters for the analysis: the date range and any definitions I want to use. I’ll tell Ella, “Use the past three years and a standard line of therapy definition.” If I had specific line of therapy rules, I could give those to Ella and it would incorporate them into the SQL. For now, I want Ella to suggest what it would use as the standard, because I want to explore what’s in the data first and then iterate.

Ella outputs a SQL query plus charts and artifacts to help me visualize the data. From there I can explore the data schema, meaning which tables and variables are available, so I can check the SQL it wrote against the underlying data, make sure it’s right, and ask it to make any edits directly. I can also jump from the SQL it outputs back into the SQL editor to work manually. It’s truly iterative, and you can move seamlessly between the SQL editor and the Ask Ella panel.

Robyn: It sounds like a huge time saver, especially if you’re not sure you’re heading in the right direction for what you want to build.

Becky: Yes. It’s a great time saver for fast exploratory feasibility work that previously would have required building an entire IHD project with inclusion and exclusion criteria. If I just want quick answers about what’s in my data and what the breakdown is, that lives here now. It saves time because you have more direction when you build your projects and cohorts. You know you’re building them right, and you’re not guessing about what’s in your data before you go through the work of building out your protocol and a full project.

As this ran, you can see I have the SQL, which I can copy or open in the editor, and here are my results. Since this is sample data, we’re only looking at 36 patients, but we can see the breakdown of each therapy. Ella also lists any assumptions and caveats for the analysis, plus next steps, and suggests follow-up questions so I can keep iterating: for example, the first-line distribution, what’s actually included in “other oral therapies,” or adding a time-to-second-line analysis.

Robyn: That is truly amazing. And all the charts just populate within the module? You don’t have to ask for a specific chart?

Becky: Right, they’re all instant and customizable within the module. So we saw that Data Sandbox with Ella really cuts down on manual work and gives you more time to focus on insights. Once you’ve explored the data and feel confident you’re ready to build, you’d move on to building a project in IHD, and now Ella can help with that manual work too. Matt, could you take us into this part of the workflow and show us Ella?

Matt: Definitely. As you said, maybe you used Data Sandbox for some exploration first, or maybe you’re starting somewhere else in your workflow. The Ella AI chat is where you can pick up the rest of the story and do a lot of different things. When we started, Ella had one skill. Now it has six and growing. The big thing it does is help you build an IHD project from scratch, measure by measure, just by outlining what you want your study to be. It walks you through the study design, picking the data source, and the assumptions and caveats, and gets you all the way to a full dashboard. It does a few other things, which I’ll show in the demo.

This is the Ella chat screen, which is easy to reach from the homepage. You can click here or over here. The new chat screen is very similar to other AI chats you may have used. You can ask anything you want, and Ella will tell you what it can and can’t do. One of its skills is being an IHD expert, so I’ll start there with a question I might have about IHD. Maybe I’ve been building a project, I’ve seen the list of measure types, and I’m not sure of the difference between two of them. So I’ll ask, “What’s the difference between the repeated encounter measure and the continuous encounter measure?”

Ella interprets the question and decides which of its skills is relevant. In this case, it decides this calls for the IHD expert skill, puts that hat on, looks through the documentation, and delivers a comparison of the two. It gave us a small table comparing them: one asks how many times adjacent encounters occurred within a specified gap, the other asks whether encounters covered a certain period. That’s correct. Another thing I like is that, depending on what you ask, it sometimes points out common pitfalls. Here it says that if you’re looking for something like at least two encounters within 30 days, you don’t want to use the repeated encounter measure, which is a common source of confusion. So Ella takes all our documentation and puts it together for you right in the chat. It also offers suggested follow-ups, so I might compare two other measures or ask about different measure types. That’s the IHD expert skill.

The big skill I’d love to show off is custom project building. I don’t need to start a new chat; Ella is smart enough to understand what’s going on. I’ll just say, “Let’s look at the six-month before and after index values of HbA1c for new metformin starters in 2021.” Ella thinks for a moment and decides which skill applies. It puts on its custom project-building hat, which takes a bit more thinking. You can see it says, “Build mode set to dynamic project builder.” It understood the task, and now it’s going to help me choose a data source. From the data source, it will understand which data components are available, such as whether the source has lab data, since I asked an HbA1c question.

It lists a few data sources it recommends. I can type which one I like, or select it from the dropdown up here, which is a little easier. I’ll pick our random sample of Norstella linked data. Now it combines my study design with what it knows about the data source. It’s asking about a washout period, noting that’s the standard approach, and suggesting six months. I’ll say yes, a six-month washout. The process goes back and forth like this. It asks questions and may suggest additional components to add to the study design. Next, I think it will ask how I want to define metformin and the HbA1c value.

To give it that definition, I’ll provide a code list, and we have a few ways of doing that, including one new one that’s pretty exciting. I may have already created code lists myself in the code list section, or a colleague may have built them in a central repository. If they don’t exist yet, Ella can help me build them on the fly, which is great.

Becky: That’s a great opportunity to plug another integration we currently have: Navidence code lists, which are clinically validated. Those are available within IHD and within the Ask Ella module.

Matt: Exactly, and I can show you that right here. Say I want to define the metformin code list. I go to medication, and I can see Ella is asking exactly that: how do I want to define it? I look up metformin, and you can see a Navidence code list for it. We’ve integrated that library of definitions, so I can just select it without creating it myself. I’ll pick that one and tell Ella that’s what I want. That’s how the back-and-forth goes. Next it will probably ask for the LOINC codes, and yes, it does. After I provide those, it puts together the study design.

For the sake of time, I’ll show you where the flow goes next using one I completed previously. This is a different study, on severe COPD. We went back and forth in the same way and defined COPD, and you can see I gave it a particular code list for that. What you end up with is a table of all the measures it will build for you, which you can also see in more detail in the sidebar. From there, you can go back and forth and ask it to add more of this or that. One thing I think is great is being able to say, “Can you add a set of baseline demographics?” Ella knows what’s available in the data source and can add that layer in, so you don’t need to know what’s available yourself.

Once everything was ready and it gave me the final build, I said, “Yes, submit.” It ran, told me I have 16,000 patients, and gave me three things: the IHD project, a dashboard, and a link to download a document with all the calculation details.

Becky: And that speaks to the auditability we built into the platform. You can download the calculation details that walk through everything Ella did, and you’re also left with an underlying IHD project. Since you’re IHD users, you can go back into that project, see how everything was built, and edit what Ella created. You’re not working in a black box where you don’t know what the AI actually executed. You can see it and edit it easily.

Matt: Exactly. The project it creates is just a normal IHD project. I can add more measures, open the ones it made, and make changes. For example, I can go into lines of therapy and change this from three to two. It’s now my project to continue however I want.

The other great thing it gives you is the dashboard. I click this, it opens a new tab, and this dashboard was created entirely by the AI. It gives me things it thinks I’ll find useful: the total number of patients in the cohort, their average age, a male-female breakdown, and some of that in table format. We’re looking to expand this so it gets even smarter about which charts and metrics you might want to see. It also set up some filters for me, which are ready to go.

What’s cool is that this dashboard tool isn’t just for Ella. It’s a new module we’ve added to modernize data visualization in the IHD workflow. Just like the project, the dashboard is entirely yours to edit. You can add more tabs to stay organized, add more widgets, and add more filters. I’d love to show how dynamic and easy we’ve made the new dashboard tool and how much we’ve upgraded the user experience, which is best seen in the new widget builder. If I want to add a new table, you can see we’ve completely redesigned the look and feel. I add some measures. It shows me the measures in my project, and I’ll pick a simple pair: age and sex. I check them off, add them, run it, and it shows me a preview. I can also add row groups and column groups. If you’ve used IHD for a long time, you’ll notice we’ve brought over many features from the reports section, but with a new and enhanced user experience.

It’s not just a reskin, though. There are a bunch of new things you can do. One is the filter widget shown here. I like being able to filter, then exit out and return to the default very quickly. One of the big things we think you’ll be excited about is the custom code widget. Previously we had integrated R into our dashboards, but now we’ve integrated Python and SQL as well, so whatever your language of choice, there’s an option for you. Anything you can build with the point-and-click widgets, such as pie, donut, area, or line charts, you can also build with custom code in Python, R, or SQL, and it integrates natively with those widget types. You can have an entire dashboard of Python and R widgets, customized however you like. We’ve also installed a bunch of new packages, which you can see here. I know people have been asking for that for a while. So this is the new and improved way of doing data visualization in IHD.

Robyn: That is awesome. Let me recap for the group. First, the new and improved dashboard module is included with your IHD subscription. You can access it through standard IHD projects, or, if you subscribe to the Ella module, you can get output of all kinds of charts and graphs and create your own widgets. It’s customizable, as you saw. I also want to go back to Ella for a second and reiterate that we are rolling out the IHD expert skill to all IHD users in a phased rollout, so some of you will already see the Ella icon. As for the custom project building and the other premium skills in Ella, we’re happy to show you more in a one-on-one demo. Just get in touch with your customer success manager so you can see even more of what Ella can do. It can really help build your projects so you spend your time on insights instead of ingestion and building.

Matt: And to build on that, some of you may remember Ella as the templated experience we launched to put guardrails in place while we learned how Ella would work as an AI tool. What you’re seeing now is a significant expansion. You don’t need to work within those three templates. You can build anything in your study protocol by starting out and writing all the measures you might need, and Ella will guide you through the workflow as a thought partner. The three templates are still there if you just want quick answers, such as simple counts, cohort builds, and treatment utilization. But we’ve really expanded what you can build with Ella.

Robyn: Thank you, Matt, for that wonderful demo. We’ll spend a few minutes on one of our modules called Model Studio, and I’ll hand that over to Becky.

Becky: Once your project and cohorts are built in IHD, you may need more advanced analytical methods and machine learning. Model Studio fits naturally at the end of the workflow, once your IHD project and cohort are set. For cases that require additional predictive modeling or machine learning, Model Studio is built right into the platform. You can compare predictive models across different methodologies and understand which variables are driving effects in each. You can then feed those predictions back into your IHD projects and use them for evidence generation. We won’t deep dive into Model Studio today, but we wanted to keep it on everyone’s radar as part of the full IHD platform experience.

Robyn: To add to that, I’ll pass it back to Matt to talk about some of the core feature updates we’ve made to IHD. As you can see from today’s presentation, we’ve put a lot of work into enhancements throughout the year to make your experience better.

Matt: Thanks. We’ve built a lot of new things, but we definitely don’t want to forget the core IHD experience you know and love. We’ve been adding small enhancements throughout the year, and here’s a sampling of the highlights. We thought about them thematically. In the first quarter, we wanted to respond to a feeling of slower project run times, so we put a lot of work into how we store data and into infrastructure enhancements to make things more efficient. You should see quicker run times. We also added a few other pieces in Q1 for extra customization, such as a processing mode for the one-inpatient, two-outpatient logic and changes to various event specs. These are things you might have run into before and not been able to do, and we want to make sure every customization you’d like is possible in the tool.

In Q2, the theme was addressing frustrations in the UI. One big focus was the dataset viewer and patient viewer, improving not just their speed but the information they provide. We also added more information to on-screen errors, so if something didn’t work, you get a better sense of why. On top of that, there were smaller improvements to various measures.

Q3 is a bit more spread out, but one big theme for this quarter and next is QC, meaning checking your projects. There are a few enhancements in that area, such as additional filters in the patient viewer and the ability to download a code list comparison, which help when you’re checking your own project or a colleague’s to make sure everything is right. Stay tuned for more features on that theme as we continue through the year.

Becky: On that topic, I’ll give you a sneak peek of what’s coming soon and where we’re going next with IHD. We’re actively shaping what comes next, and we want to make Ella available throughout IHD, with the AI expert feature rolling out to different teams. We’ll continue optimizing workflows with AI and additional assistance to help HEOR and RWE teams get the answers they need faster. We’re also bringing Ella into dashboards, so you’ll be able to ask Ella to create, edit, or refine widgets in the dashboards Matt showed today, reducing manual configuration time. We’re also adding an AI-driven document exporter that generates a structured summary of everything you built in your IHD study, building on the current export functionality for more transparency and visibility into what’s built in IHD. There’s more to share as we continue to make IHD a purpose-built, all-in-one platform with end-to-end capabilities.

Matt: And to add to that, this is where we’re pointing the ship right now, but as always, we’d love you to try these new features and tell us what you think: “I like this but not that,” or “I wish I had more of this or less of that.” That’s always what we use to decide what comes next. So that’s the plan for now, but as always, it’s up to you.

Robyn: We’re actively taking your input, and we’d love to hear from anyone listening who would like to join a focus group or have an informal one-on-one conversation about what you’d like to see in IHD, which modules you use most, and how we can help you get the most out of IHD and Ella. If you’re interested, please reach out to your customer success manager, and we’ll add you to the next round. Thank you so much for your time today. We hope this gave you a clearer picture of what’s new and how it fits into your workflow. If you’d like to see more demos of these features, or if you’re ready to add any of them to your IHD license, email us. We’re ready to talk with you and your team about everything we shared today and more. Thank you again, and if you have any additional questions, we’d love to hear from you.