Transcript#

This transcript was generated automatically and may contain errors.

Hi everyone, I am Samuel, I work at the Life Sciences team at Posit as a Solution Architect. Today we are going to discuss a real-world evidence generation workflow within Posit's team ecosystem.

We'll showcase how epidemiologists, data scientists, and RWD programmers can work collaboratively within the Posit team platform to generate high-quality evidence that can be shared with internal and external stakeholders.

Two quick definitions before we get started. RWD or Real-World Data is patient data collected outside a clinical trial. This includes electronic health records, insurance claims, patient registries, and variables. Real-World Evidence or RWE is what we get when we apply rigorous methods to real-world data for answering clinical or economic questions. We can think of RWD as the raw material and RWE as the finished insight.

Today we'll showcase how Posit can take us from fragmented analysis to a reproducible submission-aligned evidence package in one governed workflow.

Why RWE matters now

Real-World Evidence is reshaping how drugs are approved, priced, and used globally. A few forces are making this urgent. Regulators now actively accept RWE for level expansions and HTA decisions. Clinical trials cannot answer every question. Rare diseases, long-term safety, and pediatric dosing all require observational evidence, and payers own grand formulary access without it. RWE is no longer optional.

RWE is no longer optional.

Traditional use cases of real-world data includes Health Economic Outcomes Research or HEOR, market access, post-marketing commitments, safety monitoring. These are all well-established use cases. Teams are increasingly using RWD for clinical trial optimization, trial feasibility, external control arms, adverse event prediction, and inclusive recruitments. All of this requires the same thing, reproducible and auditable analysis.

The reproducibility problem

Most teams who are dealing with real-world data are stuck with fragmented data access, manual ad hoc scripts, results pasted into PowerPoint, and no audit trail. Every data refresh means rebuilding from scratch. Regulators and reviewers cannot trace findings to source. These are reproducibility failures.

Posit removes these frictions as an integrated data science platform built around three key components. Workbench, which is a governed compute environment for R and Python, Package Manager that locks package versions for long-term reproducibility, and Posit Connect, where users can publish Quarto documents, Shiny apps, and dashboards with audit trails and role-based access.

Posit team can be deployed in all major cloud provider. It can also be deployed on-prem within an air-gapped environment. It can also be deployed as Snowflake native application. In our demo, the environment that we use is deployed within Snowflake as a Snowflake native app. And the real-world data that we are using also sits within a Snowflake database.

So you may ask, why would you build your real-world workflow within Posit? The primary reasons are you can own your real-world evidence end-to-end. You prototype, validate, and deploy in one environment where governance is already built in. This is what will lead you to a faster innovation.

Demo study design

In our demo, we focused on a hypothetical study built on synthetic data set where we are looking at post-approval comparative effectiveness in type 2 diabetes. We want to identify new users of ASU and compare them with DPP4 inhibitors users within a synthetic Medicare claims population. And the data is already mapped to OMOP common data model. We'll use control for measured confounding using propensity score matching techniques. We'd like to estimate time-to-first inpatient cardiovascular hospitalization using survival analysis. We also want to quantify the hazard ratio within 95% confidence interval via Cox proportional hazards model.

At the end, we want to publish a pair evidence package to Posit Connect where internal and external stakeholders can access both the results and the underlying codes. Snowflake holds the data behind secure role-based access. Workbench runs the analysis and evidence is served through Posit Connect in forms of dashboard, HTML or Quarto documents, and web apps. Package managers ensures governed and curated R and Python packages are served to Workbench and Connect.

Business users like an epidemiologist will usually interact with dashboards and web apps that are hosted in Connect and published by data scientists and RWD programmer. Let's go into the live demo to get a better understanding of this type of interaction.

Live demo: Posit team on Snowflake

I just launched my Posit team from Snowpark container services. Here you can see the three components of Posit team, Posit Workbench, Posit Connect and Posit Package Manager.

Now let's see how an epidemiologist or a business stakeholder would interact with an existing application that has been published by their data science team. This application is built by the data scientist and exposed to business users through Posit Connect. The Shiny app is built using QueryChat. QueryChat brings in chatbot component within Shiny applications that allows end users to ask natural language questions to underlying data. QueryChat can connect to remote data sources, including Snowflake, Databricks, Redshift and Oracle.

As an epidemiologist, I can come here and I start asking different questions about my data. For example, how many patients have type 2 diabetes? So what's happening now, QueryChat is leveraging LLM to generate the SQL query. Then QueryChat is executing the query against the connected data source.

It's important to understand the separation here. Data is not exposed to LLM directly, only the data schema is. And once the LLM returns the SQL query, the application itself executes that SQL against the data source. So here you can see it identified 8,150 patients have type 2 diabetes in the data sets.

Now I may want to also do more complex jobs, right? For example, how many patients with type 2 diabetes are also taking metformin? This query will require joining of two different data tables, one for the patient condition, another for the patient drugs. And you can see it identifies properly the patients with type 2 diabetes that are also taking metformin.

So this application becomes the primary entry point for an epidemiologist when they are designing the observational study for real-world evidence generation. Before a data scientist or RWD programmer start coding, epidemiologists can first have a good idea of the underlying data source with the help from an LLM.

At the same time, we can ask API Assistant to update the dashboard. For example, if I tell it, show me type 2 diabetes patients who are not on metformin. You can see it's updating the dashboard and showing 1,455 patients with type 2 diabetes who are not on metformin.

In the same dashboard, we have Condition Explorer, where we can see top 15 conditions by patient count, top 15 conditions by error count, and then conditions summary. There is another tool within this web app, which is called Cohort Builder. This is where an epidemiologist can come in and apply different exclusion and inclusion criteria to ensure they are getting enough patients to conduct the observational study. So they can come here and without writing a single SQL query, they can start playing around.

So if I, as an epidemiologist, looking for patients cohort with type 2 diabetes, but at the same time, I want this patient to have a prior condition, such as hypertension, I can click on this build cohort, and what it will do is query the Snowflake database and bring back the summary results on the front end. So I can see here, if I apply all these conditions, eventually I'll have a patient cohort of size about 3,000.

So all these interactive tools, including the chatbot, which is built with QueryChat, allows the EPI to design the observational study. Once they are satisfied with all the exclusion and inclusion criteria and the data quality, they can provide a green signal to the data scientist to carry on the rest of the analysis.

Analysis in Posit Workbench

So that analysis will take place within Posit Workbench. Posit Workbench is where the data scientists usually live. You can see we have multiple projects here. And I can create a new session. And here I have my credentials to Snowflake attached to my session. So I'll launch this session.

And here I have a Quarto document open. So this Quarto document was built with the help of Positron Assistant, and we'll just look at the final results. But Positron Assistant has the full context of the data source that sits within Snowflake. So here you can see OMOP Common Data Model data source from where we are pulling all the data for our analysis. And to get to this data, I don't need to remember any token, username, or password. Everything is already managed through Workbench.

So I can click on Preview, which will render this Quarto document. And this Quarto document will act as our integrated evidence package. So from the guidance of epidemiologists, we build our cohort. We run all the different statistical analysis. And then we get a publishable document.

Once this run is complete, you can see the document on the right-hand side. Here you can see we have a diagram of the overall workflow, how the analysis was done. We have a study overview. Then we describe how the data was connected. If I click on Show Code, I can see the underlying code that pulled the data.

We can look at the different concept definition, like what is the concept for type 2 diabetes? What is the concept for a specific drug that we are interested in? This is the outcome, which is inpatient visit, hypertension, heart failure, and chronic kidney disease. So all these codes that are required to build a cohort. And these concepts are part of OMOP Common Data Model.

Now we can look at eligibility criteria. And this is where epidemiologists can confirm that the programmers followed the design they have created. And again, you can see we'll have code here, which we can hide or expand as needed. So this provides a nice summary of code and result living together.

We have the baseline characteristics with propensity score matching, how the standard difference between different confounding factors were before the propensity score matching, and how it looks after propensity score matching, and can compare and see. Here you can see the survival analysis with the Kaplan-Meier curves and Cox proportional hazard model.

We are not going to dive deep into the results because this is a synthetic data. What I want to highlight is the way this document is generated and what message it conveys. We can attach all our methodological notes and also determine where the outputs are saved.

Publishing to Posit Connect

So once the data scientist or the RWE programmer is satisfied with this document, they can decide to publish it to Connect. Positron IDE provides this one-button publishing option, where you can come here and just select the documents you want to publish and deploy your project. You can see your publishing logs here, and once it's published, it will be accessible through Project Connect. So now this document can be shared within my organization and also to external stakeholders.

So what we have covered in this demo is first, as an epidemiologist, you can interact through a web app with interactive chatbot and other tools to design your observational study. Once the design phase is complete, a data scientist or an RWD programmer can work within the workbench to generate a Quarto document that is also again published in Project Connect with necessary results, figures, and code.

Q&A and Posit Academy

Thank you so much, Samuel, and thank you all for joining us here today. We are always learning from you all from this wonderful community about what formats work best, and we're trying something a little bit different for Q&A this time. So instead of a large webinar Q&A format, we're hosting a few different small group sessions where you can connect directly with the Posit team, but also with a few other community members to ask questions, share your own experiences, and have a real conversation.

So if that sounds interesting to you, you can let me know using the form on the screen. I'm also going to drop that link into the YouTube description below for the recording, and I'll personally be reaching out to schedule those depending on everyone's interest here. We'll likely run a few sessions across different time zones, but I hope to see some of you there soon.

While Ryan Johnson and I started this monthly series about three years ago, we have something new to share with you all, too. So before I let you go, I want to let you know about this. We have a new Posit Academy learning space here at Posit, which is free and open to all. It's designed to help you use Posit tools more effectively, whether you are just getting started or looking to level up. You're going to find three types of offerings there. There's interactive courses, hands-on labs, and live virtual workshops, too.

I wanted Ryan to be able to give you a preview of that, so I'm going to play a short recording from him after this, just a few minutes. But I also want to personally invite you to go make a free account there on Posit Academy today. But thank you again for joining us today, and I hope to see you soon.

Welcome to the brand new Posit Academy. We are so excited to share our revamped learning platform with you. If you're looking to leverage Posit's open source and professional tools to advance your data science skills and your career, you're in exactly the right place.

Getting started is incredibly simple. Signing up is completely free. And because we use Posit Cloud's authentication, you can likely log in with one click and begin learning immediately.

Here is the Posit Academy homepage. The header contains convenient navigation links, and your profile icon provides a quick way to view your enrolled classes and resume your studies exactly where you left off. You can also jump out to learn more about Academy Apprenticeships or dive directly into our full catalog of on-demand and live training. Plus, we've created these handy quick start buttons at the bottom to guide you. With a single click, you can instantly filter courses based on your preferred learning format or the specific Posit product you want to learn more about.

Within the course catalog, we offer two main types of learning. On-demand courses you can take at your own pace and live instructor-led sessions as well. You will also see an option for learning paths. These are a collection of curated courses designed to guide you through a structured educational journey. Again, these sessions are 100% free to access.

Now you might be wondering about our premium mentor-led training, the program previously known as Posit Academy. That incredible experience has a new name, Academy Apprenticeships. Posit Academy is now the proud home for this paid cohort-based program. If you or your organization are looking for intensive mentor-led training with weekly group sessions and guided data projects to accelerate your team's capabilities, the Academy Apprenticeship program is exactly what you're looking for.

Now that you know what Posit Academy is, you can log in and get started right away. Happy learning!