Powering Medical Research With Data: The Research Analytics Adoption Model (Webinar)
Powering Medical Research with Data:
The Research Analytics Adoption Model
July 22, 2015
Thank you, Tyler. Before I get started with the presentation, I just wanted to provide a very brief background. I’ve spent the better part of my career connecting researchers to data. I spent about 10 years at an Academic Medical Center first on a genomics project, doing genome analysis and then disseminating that information to researchers through a public portal. And then I transitioned to a data warehouse team where I helped build the data warehouse and then ended up just kind of managing the whole research side of the data warehouse and helping researchers connect to that data warehouse to get enterprise data that powers that research. So, I’ve seen the process of adopting analytics to power research. And more recently, I’ve been in conversations with organizations across the country who are looking to augment their research with enterprise data. And what you’ll see presented today is kind of a compilation of both that experience in the past, as well as the conversations that they have been having nationally. And I hope it’s useful. I’m looking forward to the presentation and absolutely receiving your feedback. I will present my email address at the end. So if you would like to contact me and provide me with feedback, that will be very gratifying to me.
Why are we here? [01:16]
So why are we here today? First and foremost, we want to understand how we can make research better by providing data. And it’s really like everything we do in healthcare, we have to think about how we tie it back to the patients. And there are millions of both sick and healthy patients alike who want better care. These patients are volunteering their time and in some cases, risking, putting themselves at risk to participate in a research. And research really helps determine what we mean by better care. So it’s a critical part of creating a better healthcare system and providing better care in the future. So we owe it to those patients to make that whole process more efficient.
Also, the 27 percent, if not more, of folks on this call who are involved in research understand there’s a lot of waste in research, there’s a lot of duplication, there’s a lot of manual work, and we can greatly reduce the waste with data and analytics. If you’ve been to Health Catalyst webinars in the past, you are well aware that we’ve done – many, if not, most of our webinars are focused on how we reduce waste in the healthcare delivery system with data and analytics. And the same concepts apply to research.
And where we’re moving as a country in the future, precision medicine aims to deliver the right care to the right patient at the right time.
To get to Precision Medicine [02:51]
So, to get to precision medicine, we really need to work on improving research. So number one, identifying the underlying molecular causes of diseases and refining treatments based on those discoveries and that’s happening now absolutely, but again with data – and we’ll present a vision for how this works later in the presentation – with data, we can make this process of improving research a reality.
We also need to improve care delivery. So how do we deliver to care guidelines and adapt to new guidelines moving forward. And the quicker and more efficiently we can do that, the more value we get out of the research discovery platform. And as I mentioned before, we have been presenting a lot of webinars on really how to improve care delivery and monitor care according to guidelines and providing tools and technology to help healthcare systems adapt to those new guidelines.
Now, currently what we seen in the country is a lot of healthcare systems that are having trouble implementing care guidelines that are 10 years old and we really need to kind of shore up that whole ability to deliver on those existing guidelines before we can really speed up on the adoption of newer care guidelines.
Another thing that needs to happen in addition to improving both research and care delivery is increasing the coordination between the care delivery and the research enterprises. So currently the amount of time that it takes to go from a new clinical or operational best practice to the point where it becomes a habit of all frontline clinicians that’s typically measured in years and our goal is to help reduce that time to weeks or months so that as new discoveries come out, we can efficiently transition those discoveries really from a research project to the way they change in a way that care is delivered. And that’s really kind of setting up why we’re here, what we think the future of medicine is. Obviously there’s a lot more to the research than precision medicine and what you’ll see today applies to all aspects of research but we want to just specifically call that precision medicine because we do think it provides a nice vision for the future of how research and the care delivery systems can work together more coordinated.
So the agenda for the rest of the presentation is, first, we’ll present a review of the research process. We’ll look at the various steps in research and understanding that all research is pretty unique and we put together kind of a generic research process. Then we’ll be reviewing some roadblocks that prevent an investigator from conducting efficient research. And after that, we set up with the problems in the space. We’ll then present the research analytics adoption model and the vision for how analytics can be adopted to help remove some of these roadblocks. And finally we’ll end with a brief conclusion.
So let’s get started by looking at the research process.
Research Process [05:59]
So the first step is Hypotheses Generation, and an investigator can have hypotheses from many different sources, from their previous research, from researching and being current on literature, and also exploratory analysis. If researchers have good tools that enable them to look for trends and identify trends and formulate hypothesis, that actually creates a much more efficient hypothesis generation pipeline.
After a hypothesis is generated, a researcher then decides if they are going to study this further. And they engage in this process, called Cohort Exploration. And Cohort Exploration says, okay, if I want to study this hypothesis that I have in more detail, do I have access to patients that match my criteria? Or do I have enough patients to provide the statistical power that I need to generate conclusions for my study. And this actually can lead to a project not happening. If a researcher doesn’t feel they have enough patients that they can study, then they either have to expend where they’re recruiting patients, which greatly adds cost to the project, or maybe they just move on to another hypothesis. So it’s an important step and it also feeds in to the next step.
So once a researcher decides that they have a good hypothesis, they feel like they have enough patients to power the study, then they typically apply for assistance, financial assistance, in the form of a grant to make that research project happen. The data from that cohort exploration is critical in the grant application. Funders love to see applications that are backed up by data.
And furthermore, there’s other ways that grants can be strengthened with analytics and data. So if an organization is particularly adept at recruiting patients, you know, they have a good success rate for recruiting patients, that strengthens the grant application, as does a description of an enterprise data warehouse. If an organization provides access to researchers to that data warehouse, funders really like to see that that level of investment have been made in providing data to power research.
Another step in the research process is putting together an IRB Application. And if you’re not involved in the research, that IRB may be a new term for you. It stands for Institutional Review Board. And this was regulated by HIPAA and said that each organization who conducts research needs to review research to make sure that it fits ethical standards. So the IRB is typically making sure that the research that’s being conducted doesn’t put patients in harm’s way and it is unnecessary but often can be a time-consuming step in this process. And when the researchers put in together the IRB application, they also need to put in information about the cohort they are going to be studying, size of that cohort, the protocol that will be followed, as well as specifying what data needs they are going to need further on through the process.
The next step once the IRB and the grant comes through is Patient Recruitment. So we know there’s types of research where there’s a waiver of recruitment, but for the most part, investigators need to find patients who fit the criteria for their study and approach those patients and ask them if they want to be in the study. And then at that point, they are also presenting patients with information about the protocol and the potential risks of the study and asking the patients, do you want to be in this study? It’s a very important part of the process. As a matter of fact, if you look at clinicaltrials.gov at the reason why many many research projects fail is their inability to accrue enough patients in a study. So for patient recruitment, it’s not enough to just know who the patients are, although it’s a very important part of it. So oftentimes we see further refinement to that cohort. But the researcher needs to know who the patients are and where they are going to be. So information like scheduling data comes in handy. Information about patients who are admitted who fit the criteria comes in handy and that gives the researcher time to go approach that patient. And then how do I collect their consent. Their consent is when they sign the piece of paper agreeing to be on the study. And consents can be delivered through paper or electronically nowadays. But this patient recruitment set is time consuming and resource-intensive but at the same time a very very critical part of this whole process.
Once the patients are accrued in the study, then Data Collection starts. And there can be prospective data collection. So I need to collect data from a study that doesn’t exist anywhere yet. It might be patient outcome data, typically administered through a questionnaire. We call that prospective data collection, if you’re collecting new data for the research study. Alternatively, or also in addition to, data collection can also involve a data pull from an enterprise data warehouse or similar data resource. Where we have clinical data that exists in our systems and we need to put a data specification together and make a request and pull that data that is available through those systems on those patients. This data collection, the data pull needs to be approved by the healthcare organization. So even if there’s an IRB approval, a healthcare organization may wish to further review the data request to make sure that it is consistent with what was approved in the IRB. And so, data collection can actually be another bottleneck, and we’ll talk about that in a little bit.
Data Analysis, once we’ve got our data set involved doing this statistical analysis, using advanced tools for genomics, if it’s a genomics project, and this data analysis step needs to be secure. You know, I’ve heard stories about researchers doing this analysis in spreadsheets on their laptops that may not be encrypted. So there needs to be a good infrastructure to do this analysis in a secure way and there needs to be the right tools and the right skill sets to do the data analysis.
Next comes the publication step where the conclusions are compiled and manuscripts are submitted.
And finally, hopefully translation to clinical practice. Once a research conclusion is made, it doesn’t always lead to immediate translation to clinical practice. It might need to enter later stage trials. But eventually we hope that most research can find its way in the clinical practice. And the questions they could ask, how can this discovery now be used to treat patients? And how do we work more closely with the healthcare delivery systems to implement these latest best practices?
So this is, again, a very generic research workflow. We realize that there’s variations to this and all research is different but we want to kind of present a very generic research process.
Research Process – Roadblocks (Waste) [13:26]
Now we’re going to go through that process again and we’re going to talk specifically about where roadblocks and waste exist along the way. And this is where I would love to get your feedback at the end of the webinar over email, if you could. Just sharing some of these stories. We’ve collected lots of stories from investigators on where they run into roadblocks and it’s helpful for us as we seek to gain a deeper understanding of this research market.
So with hypothesis generation, cohort exploration, grant applications and IRB applications, the existence of exploratory tools is really important. And if those tools don’t exist, then this can slow down the process. So if you think about, you know, if I’m going to do a study on COPD patients and I’m in my cohort exploration tool. If I don’t have to go to a data analyst and ask them how many patients I have, if there is a tool that allows me, the researcher, to go and have that question myself, it’s going to eliminate a huge bottleneck. And when the groups that I ran at the Academic Medical Center first got started, I would say that 60 to 70 percent of our questions were simple counts of investigators putting together requests because they were going to put any grant application on IRB applications. So having tools that allow for the de-identified exploration of data is absolutely critical for all four of these steps.
When it comes to the IRB, I’m sure there’s lots of stories about how things often get held up in the IRB. Slow IRBs are unknown organizational issue and it’s not necessarily because the IRB is bad but it’s because they have a lot of things in their pipeline and most of them have other jobs and they get around to reviewing these things. It’s not part of their everyday job. So organizationally and they might (15:27) appropriately, they may not have enough committees to do the review process. On the technical side, in addition to having insufficient exploratory tools being a roadblock, insufficient tools to support the IRB process, technology tools. And organizations that adopt electronic IRB workflows tend to be more efficient than paper-based IRBs. So there’s some technical issues that can be resolved there as well.
For actually recruiting patients, one roadblock can be organizational restriction. So as I mentioned before, in addition to the IRB step, there’s oftentimes an organizational step where the organization may put additional restrictions on how researchers can approach the patients. So one very common restriction is that researchers need to work through the patient’s physician to understand if the physician approves of the patient being on the study or agrees that the patient will be good for that study, and this can slow things down in a couple ways – number one, if the physician doesn’t provide a timely response to the inquiry, that can slow down patient recruitment. Additionally, just identifying who the right physician is can often be a roadblock. Many patients, especially patients who have chronic conditions, see multiple physicians in this whole question of attribution – who is the physician that we should ask for it becomes a bottleneck in this process. On the technical side, I mentioned that you need to know not only who the patient is but where they are going to be and having insufficient data and tools that present that information to the person doing the recruitment can be a big bottleneck as well.
For data collection, organizationally, things work way better when there’s a process for the release of data. And organizations that invest in putting together a process, even a process that has a lot of steps do better than organizations with no process. When there’s no process, we see healthcare systems taking an overprotective stance on the data and not really releasing that data as efficiently as they could, as if they had a process. On the technical side, if there’s no data warehouse, there could be a lack of a single source for data. So investigators don’t even know where to go and they spend a lot of time cobbling together data from different systems. Insufficient self-service tools can slow the process down. So if investigators have access to tools that allow them to generate their own queries, those can make things way more efficient. And inefficient tools to support the data release process. So going back to that organizational issue, we’ve seen the process take place over email where an analyst will send a note to somebody at the healthcare system, asking for approval to release a particular data set and things just tend to get lost over email, and workflow tools are incredibly helpful in this situation where you have a good process and you can track each step. So insufficient tools to support that release process, not to another technical barrier.
For data analysis, having insufficient analysis tools and platform create some inefficiencies and roadblocks. What happens here is investigators invest in their own tools and it creates a much bigger expense. And organizationally looking and hiring people with the right skill sets is an important part of this. The data gets bigger and bigger. We hear all about big data and the ability to do large scale analysis. As that data becomes more and more available to researchers, there really needs to be an investment and the right skill set to do this analysis from an organizational perspective. And for publications, these bottlenecks are a little bit less than the organizational and technical sides but we did put in one bottleneck. Some organizations we’ve seen have support for manuscript preparation. Especially those organizations with a CPSI would be able to provide support for creation of the manuscript.
And finally from moving into clinical practice from an organizational perspective, the lack of a deployment system is one of the big bottleneck. And by deployment system, this is a term that we use for clinical quality improvement and it’s the way, it’s the people on processes in which we can provide insight into how an organization is doing, delivering to their own care guidelines and helping to improve them along the way. And organizations that are efficient at delivering the care guidelines and adding things into those care guidelines, they will be the most efficient organizations at bringing new practice and when it’s discovered. Another organizational roadblock is when research is not aligned with care improvement initiatives. So, doing a research with enterprise data from the healthcare system is really a strategic decision, and having support at the executive level and aligning research initiatives with care improvement initiatives will also help create that more seamless transition from research into clinical practice.
So that kind of summarizes the research process and roadblocks.
Research Process – Other Considerations [20:56]
We do have some other considerations that spend across the research process. So things that go across all research, things like data integration, you know, do we have siloed data, experimental data that’s siloed with limited ability to combine this data with clinical data? That can be a big roadblock that spend across all research studies. Research organizations are often required to do operational reporting. So how many patients in our healthcare system are involved in research, how many patients have denied studies, how many active studies do we have. And generating these operational reports, if there’s not good data systems, can actually take time away from people who could be adding value to the research process. So that’s another roadblock – is people getting pulled in to do more operational work. And finally, sharing of data. We see more and more multisite trials going on and more and more collaboratives across organization and not having a good infrastructure for sharing data can often lead to inefficiencies and organizations opting to write their own software when there’s other software out there that exists and could help them.
Agenda – Present Analytics Adoption Model [22:15]
So now we’re going to walk through the analytics adoption model and we’re going to go through it step by step and then we’ll end with a brief conclusion.
Level 0 – Manual dataset generation [22:23]
So level zero, the very bottom of our analytics adoption model, is manual dataset generation. And this is what typically occurs at organizations that haven’t invested or put together a strategy on the research used at enterprise data. And what happens in the manual dataset generation step is that data is delivered to researchers through operational analysts. These are people whose job is to generate operational reports of their existing systems and these data requests from research just typically go into report queues. And the problem is that research data requests are typically prioritized very low when put next to operational and quality improvement data requests. And oftentimes there’s no set research process or infrastructure and furthermore, those operational analysts may not know all the appropriate regulations and all the questions that they should really be asking of the researcher on the data. And the result of level zero is typically frustrated analysts and frustrated researchers, and we’ve seen this time and time again. This is not unique if you are experiencing this at your own healthcare system but it is addressable.
Enabler for all Levels > 0: Data Warehouse [23:38]
So as I walk through the later steps, we have eight steps in all, the Enabler for all of the levels above level zero is an integrated data repository or data warehouse. And I will not go into great detail. This kind of shows a high-level architecture of a data warehouse. And I think the most important take-home from the data warehouse initiative and using a data warehouse for research is that the data warehouse needs to very quickly and easily be able to ingest new sources. Research is a very fragmented enterprise typically and there’s lots of new data systems that pop up that will provide value when integrated with enterprise data. And if there’s a long process for incorporating those new data sources into a data warehouse, it can lead to frustration and in fact, it ends up being not scalable. So having an architecture that allows for very quickly and efficiently bringing in new data sources is a very important part of the data warehouse strategy that needs to be put in place that will enable the additional levels of the adoption model.
Level 1: De-identified tools and data marts [24:49]
So level one is about providing de-identified tools and data marts. And for each level, I’ll present a brief description of what we mean by that level. And then at the bottom, I’ll show all of the roadblocks that that level addresses from our roadblock slide. And so, from de-identified tools and data marts in the bottom, you can see how many problems to solve and that’s why it’s level one. By providing access to de-identified data, so data that doesn’t include medical record numbers or patient names or Social Security Numbers or even dates, this allows investigators to get access to that data without having to go through a lot of steps with the IRB, and it minimizes the data access roadblocks. And when investigators have access to de-identified tools, they can use them to power all these steps here, the hypothesis generation, cohort exploration, grant application, IRB applications, and even some of the data collection steps. So obviously this is a very high value activity and the quality of the tools dictate how far you can go in removing each of these roadblocks. But it’s a very very important step and a very foundational step to move to other levels of the adoption model.
Level 2: Delivery of customized data sets [26:11]
Level two is the delivery of customized data sets. So we can have all the great self-service tools in the world. Researchers are often dealing with the deep dark corners of the medical record. They are looking in clinical notes to pull out data fields that they think might be of value. And having tool sets is good but what we find is almost always there’s a group of researchers who need to get data that wouldn’t be exposed by those tools or the tools don’t expose yet, they might be on the roadmap. So having the ability to generate customized data sets is a very key competency. And at the bottom you can see that it solves the data collection issue in multiple ways organizationally and technically. And what we mean by level two is that there is clear guidelines for the data release process. Data stewards are involved and data stewards are folks who have a specific knowledge about data in the healthcare system and can provide information to people who are putting together data requests about what data is going to be valid for them and what data is not going to be valid. They also serve as the ones who can really decide if the data release is appropriate. Also important at this level is a dedicated research analyst. These dedicated research analysts understand research, understand the regulatory rules, the institutional rules, and how investigators want to look at data. It’s very different, although (27:45) very different. It’s different than the way an institutional analyst would look at things. And being able to speak the language of research is very important. So dedicating some resources to serve in this role.
The data sets should be delivered in an agile consultative manner. Zero percent of the time, I have seen a data set resulting from a researcher request exactly matched the request that was put in the first time. So a good analyst can walk the researcher through and ask the researcher what they are trying to accomplish and help assemble the data that is going to best answer their question.
And lastly, with level two, something that helps really well, I kind of mentioned it before, is a workflow tool that supports the data request workflow. So an investigator can put in a data request and in the workflow tool they can list their IRB number and we can track the status of the assembly and the approval of that data request so the investigator is not knocking on the door of the research analyst every day. So that’s providing insight into the process that needs to happen to get data out of the warehouse. It can be facilitated greatly by a workflow tool.
Level 3: Study Recruitment Facilitated by EDW [29:03]
Level three is about powering study recruitment with an Enterprise Data Warehouse. So providing tools that allow an investigator to define a population of eligible patients, patients who have not said “please don’t contact me. I don’t want to be involved in any research” and patients that match very specific recruitment criteria.
One thing, and we’ve heard the story over and over again, about death status and how important this is for study recruitment efforts. If you talk to any research assistant who has been involved in recruitment, they’ve made that unfortunate and uncomfortable phone call to the families of the patient who may have passed away, that the death status is not typically updated very well in the medical record system and it’s a very uncomfortable situation for the family who’s called. Sometimes it brings back some bad memories, and in fact, it’s uncomfortable for the recruiters. Well they don’t want to make that call. And if we can get better death status, that will provide great value add to this process of study recruitment.
Providing up-to-date scheduling information is very important. So I mentioned before how the investigator needs to know where the patient is going to be so that they can approach those patients. Mobile devices are going to be key here. And having a schedule that’s up-to-date. So a lot of organizations are doing up-to-the-date scheduling of patients. So people are scheduling their appointments first thing in the morning. And if we have a feed that baits on yesterday’s data, that’s not going to work for those patients.
Having an option for electronic consent and allowing the coordinator to get the consent from the patient electronically is another great time-saver. So there’s a lot of paper involved in study recruitment right now, that schedules are delivered by paper, the consents are collected by a paper, and that leaves a research coordinator lots of time in their schedule to actually print out the schedules, to go and enter all of their data they have been collecting through their recruitment process back into a system. So whether the patient consented or said “don’t contact me again”, that will have to get captured so that we can use it later. We don’t want to contact patients repeatedly who don’t have an interest at being in research.
Another important part of study recruitment is some kind of centralized coordination, so that the same patient doesn’t get approached five times in one hospital visit. This creates what well call study fatigue and patients who may be open to research just get so sick of being approached multiple times but they just end up saying no. In the long term, we see pretty good analytics being helpful here. So providing a score of how likely a patient is going to consent to be in a study based on historical data. That’s another thing that’s facilitated by the data warehouse and having a robust analytics engine that may not have been available before. And as you can see, this solves our patient recruitment issues of institutional restrictions. We can provide the data on who their providers are, as well as the insufficient tools to find patients.
Level 4: Research-Specific Data Collection is Centralized [32:26]
Level four is about how do we collect research-specific data collection. And typically data, when it’s collected for a specific study, if there’s not a good infrastructure, is collected in Microsoft Excel or Access. And these tools, while they are very powerful tools, they’re not great data collection tools, and the reason is, number one, security. So, Excel is typically sitting on somebody’s laptop. Number two, they don’t handle multiple people entering data very well. So, how many times have you got into a conflict where somebody enters the row in Excel and you’ve got a copy in your desktop and you enter a row and how do you merge all that together. So that whole process of putting data into Excel and Access is wasteful, so in order to facilitate this provider infrastructure to support self-service data collection form creation. So, a tool that allows me to specify the field that I want to collect and a way to collect that centrally is key here.
And another key point that should be made about these tools, it’s important that they both pull and push data from the data warehouse. So pulling in master data is important and by that, I mean – let me share actually an example here where I was working with a researcher at one time who was collecting information on breast cancer patients, and she was collecting tumor markers for these cancer patients that were kind of buried in the pathology report. It’s very difficult to abstract electronically. So she was abstracting them by hand in an Excel spreadsheet. And the time came where she wanted to do an analysis of the outcomes based on these tumor markers, and she sent that spreadsheet to my team and asked us to upload it to the data warehouse so that she could look at outcomes for these patients based on their various tumor marker combinations. And when we brought into the data warehouse, what we found was that 60 percent of the medical record numbers in the spreadsheet didn’t match data in the data warehouse. Some of them were mistyped, some of them had been out of date. And a good data collection tool will alert somebody when they’re putting in invalid data. So if I’m putting in a medical record number in a centralized system and it’s invalid, it shouldn’t allow me to put that data in there.
The next piece is that the data collection tool needs to be able to push the data back into the data warehouse. So going back to the research, for example, once we did get all the data issues sorted out, we uploaded her spreadsheet into the data warehouse and provided her with an outcomes data set that she could then slice and dice based on those tumor markers and a tool that automatically does this would have saved my team hours of work getting that into the data warehouse. And of course, through all of this, we need to be applying appropriate security to both the data collection tool, as well as the data that ends up in the data warehouse.
Level 5: Automated Reporting of Research Operations [35:31]
Level five addresses the problem of creating research operational reports, and a lot of time is spent doing this by individuals whose day job is to do research. And sometimes there’s questions like how many active studies do we have? How many of our patients are enrolled in trials? How many of our patients don’t want to be involved in research? These are tremendously valuable data points but the work that has to be done to assemble those is a lot and it has to be done through emails, how many trials are going on, and kind of assembled by somebody, and if we are loading the right data sources into the data warehouse – so, clinical trials, management system, any kind of patient recruiting system, any kind of electronic IRB system. We can provide a much faster and automated approach to providing this data. It’s up to the date. So we don’t have to worry about taking people away from their jobs to deliver these reports.
Level 6: Biobank/Genomic Data Integration [36:34]
Level six is about making the most value out of biobank and genomic data. So, there is an incredible richness of the data in a data warehouse clinically. And if we can combine that with biological inventories in data and we can provide a tool that easily answers the question “how many bio samples do I have for female COPD patients over the age of 65 across all repositories?” That is something that provides tremendous value. So, number one, it saves time. So I no longer have to go access all of my bio-repository separately and compile the data, but it also makes more efficient use of the samples by providing larger insight across the organization of where our samples are, will reduce the amount of things that have to be collected redundantly. So that’s an incredible richness that can be provided by loading biobank data.
And once genomic data is loaded, we can provide a platform for genomic discovery. One of the things that exist in a data warehouse with a rich clinical data set is the ease of defining a large number of populations. So, when you implement a data warehouse, sometimes you get starter set content that allows you to look across 500 to 800 populations out of the box. And if we can combine those population definitions with information that we have from the genomics world, we could easily answer questions, like what populations are enriched for a given set of gene variant that may be identified through an experiment. So, I’m looking at these gene variants and do I notice a particular population has more of the gene variants than other.
Another easily answered question is what gene variants are enriched for a given population? So if I’m looking at my COPD population, are there variants that show up for this population that are overrepresented in this population that might provide me some clue (38:36) genetic cause for this disease. But the thing that everybody needs to be aware of is the regulations on genomic data and providing the appropriate security miles. There are some local laws as well, some state laws that prevent the usage of genomic data and people just need to be aware of those regulations before proceeding on this level. But this is really about providing a lot more bang for the buck for our experimental data.
Level 7: Multi-site Data Sharing [39:09]
Level seven is about multi-site data sharing. So, coordinating centers for multisite studies need a quick and easy way to pull in data so that they can do combined queries for also potentially a platform to help them do federated queries if they don’t want to copy the data into a single location. And having a good infrastructure for this allows an organization to become a coordinating center and pull in data from multiple sites. The coordinated centers also need to provide good tools to help investigators to analyze the combined data set, again, either through copying it to a single repository or through a federated query approach. But for any multi-site study, it’s going to become more and more common and (39:57) that organizations can participate in these and that certain organizations can coordinate them efficiently and easily.
Level 8: Translational Analytics [40:07]
And level eight is what we call translational analytics. And this is about how we use care delivery analytics to power research. This is an important part of taking discoveries that have been made at the molecular level and helping to incorporate them into clinical practice. And I’ve got a few additional slides on level eight here. So rather than spending a lot of time on this slide, we’re running a little low on time, I’ll just walk through a scenario.
Translational Research Registries: Discovery [40:36]
So right here, this part of the slide here, we’ve got a heart failure data mart. And this heart failure data mart may have been developed by a care improvement team. And it’s a database within the data warehouse and it very carefully and accurately identifies a cohort of heart failure patients. And it was created with clinicians at the table. Those clinicians also define what outcomes they wanted to look at for heart failure patients. It might include things like length of stay. It might include things like readmission rate or observation stays and these were also defined from a clinical perspective – how do we want to measure what our outcomes for heart failure patient look like. And also there’s often process metrics in here. So for example, how many of these patients had their medication reconciled when they came into the hospital and how many had them reconciled when they left. These are processes that are known to positively affect readmission rates.
But the focus of a heart failure data mart is for care improvement. And an organization that’s regularly creating these data marts to look at populations for the purposes of care improvements knows what these look like and they are typically very sophisticated analytic tools and they really do help improve the quality of care. Now, if a researcher wanted to do a heart failure study, there’s no reason why this researcher should have to define another cohort, additionally define outcomes. So, the idea of de-identifying one of these heart failure data marts and making it available to researchers, it takes a lot off the researcher’s plate. Now, they can already use that clinically-defined cohort, they can already re-use the outcomes metrics that have been specified, and they may want to do something new in addition and look at a particular piece of genetic or genomic data and that data can also be de-identified but linkable to the patients in the underlying data mart and that allows the investigator to do a little bit of discovery. So I have a theory that this genetic marker might actually affect the outcomes of my patients. Why don’t I bring that into my de-identified version and I can test the hypothesis before I put together my grant application. So this is how an investigator might use a care improvement data mart to help them get started on the discovery process.
Translational Research Registries: Back to Clinical Practice [43:12]
So once that discovery process is done and the investigator decides to recruit patients, this is what a scenario might look like. Again, we go back to our clinically defined heart failure data mart. Now, the investigator put together grant and got IRB approval to recruit patients from this heart failure data mart. They use some set of tools to find these patients when they are going to be in the clinic or when they are admitted and enrol them into a study and then the patients who are enrolled in the study can then go into a data mart that’s dedicated to that heart failure study. Again, we can use the same outcomes criteria. We can leverage any data that was used in the care improvement data mart but we can add experimental data, including prospectively collected data. So questionnaires that are associated with my study or genomic data that may have been prospectively collected for the study. And this allows me to do a lot more in-depth research and asks some questions about the genomic underlying molecular causes for this disease.
And then if a discovery is made that has clinical value, that discovery and that data point can then be put into the heart failure data mart. So we work with the care delivery team to improve their collection of that data point and maybe a genetic test that needs to be done, but that new discovery can very quickly and easily be fed back into the same tool and same data structure that folks were involved in care delivery are using every day to monitor their heart failure patients. And this is just one way that we think a good data infrastructure can facilitate translational analytics and making new discoveries and bringing them back into care improvement quicker.
Research Analytics Adoption Model [45:04]
So we walk through the research analytics adoption model. This is the model stacked up together and with brief descriptions here. We will make sure that this model is on our website. Additionally, you will be able to download slides from this presentation, if you like. And we hope people use the adoption model. The feedback that we’ve gotten so far is this is a great way for me to help explain to my leadership where we are with analytics and I’m going to steal something from Dale Sanders who created the Healthcare Analytics Adoption Model, which looks a lot like this but it’s more for healthcare delivery system.
Research Analytics Strategy
(a starting point) [45:43]
And he said, if you take this model and turn it on its side, it becomes a good starting point for a strategy. So if you are involved in strategically assembling a strategy on research analytics and how you want to bring analytics into your research organization. This is a great starting point for that strategy and you can modify it and use it from there. Again, I would love some feedback on this. So if you’ve got feedback or specific experiences you’d like to share, that will be very helpful in refining this model and making it better. This is actually version number one and I do plan on making a couple of revisions over the summer based on feedback from presentations that we do in a live setting and also from this webinar.
Data Warehouse Considerations for Research [46:29]
So I’m just going to circle back, give some data warehouse considerations for research, and then end with my conclusion.
It’s critical for the purposes of research to provide both de-identified and identified access paths to data. Only de-identified will solve a lot of problems but it will not solve all the problems. Researchers often need to identify the patients, especially if they are recruiting patients live and we need to be able to pull appropriate data on those patients.
The warehouse must provide a security model that allows this kind of access. And organization, the mindset that an organization needs to adopt are policies that enable access for appropriate users. We often hear very restrictive organizations and sometimes it seems like the goal of those organizations is to limit it to as few users as possible and that is definitely one way to reduce risk, but it doesn’t really enable the mission of improving healthcare. So, adopting a policy that enables access for appropriate users does not break any laws with the appropriate users, as underlined, but it really is a different mindset and that’s the mindset that needs to happen in order to make this a reality.
The ability for a data warehouse to ingest a number of sources quickly and easily. Again, I talked about that before. It’s very important and because of the number of systems that are involved in the research and how quickly they can crop up and they all provide analytic value, so let’s get them into the system as quickly as we can when there’s a good use case for it.
So the conclusion here is that there is a very strategic initiative about aligning research and clinical enterprises, and making sure that that clinical data is available for researchers under the right circumstances. And in our experience, it requires executive engagement at its core. The roadblocks are too many to get result without very high level support in the organization, but once that support is there, there is a path to success and it’s played out at many different organizations across the country. But having that executive engagement is absolutely key.
Use our analytic adoption model to evaluate your current capabilities and put together your roadmap. We are happy to use it for a starter set strategy and please email me with feedback. I said I would put my email address up here. Here it is. I also am available to connect over LinkedIn as well. Contact me any way you see fit, but I would love to get your feedback and stories.
Healthcare Analytics Summit 15 [49:07]
Thank you very much for your time today. And I hope we do have time for questions. So before we do that, I just wanted to talk about our Analytics Summit coming up in September. We are excited about the agenda. This will be our second Analytics Summit. It’s a unique industry experience. It’s not an opportunity for us to sell our products. It’s an opportunity for us to bring in great speakers from around the country to talk about analytics in healthcare. We had a great success last year and we also do some interesting things like incorporating analytics into all of our presentations. So we’ve got a team of analysts who are analyzing data that’s collected real-time throughout the summit. It’s great, fun, and a great chance to network with other people who are interested in this field.
Tyler, would you want to follow up on this Analytics Summit?
I do. Thank you so much, Eric, for that. I will talk about the summit in a second.
How interested are you in someone from Health Catalyst contacting you to schedule a demonstration? [50:01]
But before I do, I did want to mention, now our webinars are meant to be educational about various aspects affecting our industry, particularly from a data warehousing and analytics perspective. We have had many requests, however, for more information about what Health Catalyst does and what our products are. So if you are interested in having someone from Health Catalyst reach out to you to talk to you more about Health Catalyst, answer this poll question.
And while this poll question is up, and afterwards I will put up a couple more poll questions to do the summit registration giveaways, but just to remind you all that if you’ve got any questions or comments, please make sure you put those, type those into the questions pane in your control panel. And thank you for mentioning, Eric, about sending out the slides. We’ve got a few folks asking about the slides. They will be available. We will be sending out an email to everyone with the link to the on-demand webinar, as well as the slides.
So I’m going to go ahead and close this poll now. And let’s see. We had a question. Someone just asked us where and when is the summit. The summit is in Salt Lake City and it will be September 8th through the 10th.
Alright. So let’s go ahead and get into those giveaways here.
Are you interested in attending the Healthcare Analytics Summit in Salt Lake City? (Single Registration) [51:26]
Are you interested in attending the Healthcare Analytics Summit? This is for the single registration. I’ll leave this up for a moment to give folks a chance to respond to this. If you schedule as such and you feel like you can meet us there, that will be great. So I’ll leave this up for a few moments to give everyone a chance to respond. Again, this will be September 8th through the 10th in Salt Lake City at the Grand America Hotel.
Alright. I’m going to close that poll. And let’s go ahead and start the one for those who are interested in a team.
Are you interested in attending the Healthcare Analytics Summit in Salt Lake City as a team? (team of three registration) [52:10]
So this is for a team of three registration. So while you’re filling that out, let’s go ahead and start with our first question. We can jump right into it, Eric.
QUESTIONS AND ANSWERS:
|For a researcher interested more in management and operational processes than clinical aspects of care, do you have recommendations for overcoming the classic barriers to implementing new processes and resistance to change, you know, like standardized reviews and other uses of EDWs?
Absolutely. So I think one of the keys there is that executive level support. Getting the executives bought into what you’re doing is extremely important. And then picking something small enough to where you can assemble a small team to accomplish that with the right stakeholders. The small team approach is very important because things tend to move a lot faster, and in our experience, if you’ve got a good use case and you can work that use case through the process, even if it’s a manual process and it’s a little bit messy, you can come back with specific recommendations when you’re done to what needs to be put in place. It’s very difficult to design a process for something that’s never been done before. So getting a small team together, working out on important process with executive support and then coming back and making recommendations on what the process should be and working on it on an agile iterative way seems to be the way that works best for most organizations.
Thank you, Eric. I would also like to add that we do have a webinar on our website about physician engagement and building physician engagement around process change. We have a couple of them. One is by Dr. Bryan Oshiro. That one deals very specifically about physician engagement. And also we have a very recent webinar, called Leading Adaptive Change, that also talks about strategies to be able to help to overcome some of these roadblocks from an organizational perspective. So I do want to mention we’ve got someone of those.
That is a great webinar and that would help with that question absolutely.
|How amenable is your tool to academic medicine where part of the research process is housed at the medical school and part of the process is at an affiliated hospital?
That is a great question and there is no question that that adds complexity to the process. I think most issues that arise from that are organizational and that is actually where my advice about involving executives come from because historically, at many academic medical centers, there’s not a great relationship between the academic organization and the healthcare delivery system, and it’s those processes that need to be worked out to make sure that the health system is bought in and supportive of the research mission, but also helping them to ensure that their rules are abided by. And I think a lot of the tools that I mentioned for workflow and tracking the whole process actually really helped build that trust. And getting a data steward from the health system to review, getting them engaged in the process, I think a lot of the complexities are more organizational than technical but they can be worked through. And some of the tools that we had actually helped by providing insight into the process.
|Do your Health Catalyst tools actually address all these issues all the way through level 8?
That’s a great question. We specifically didn’t talk about products because the webinar is more about education than a product but we have a product roadmap that addresses the different levels and we are in the early stages of building our product line around research and we would be happy to share that with anybody who is interested through a separate meeting. So please contact me if you are interested in learning what our product roadmap looks like.
Alright. Well it looks like we’re at the top of the hour. Thank you so much, Eric. Thanks everyone for joining us. Shortly after this webinar, you will receive an email with links to the recording of this webinar, the presentation slides, and the winners of the summit registration giveaways. Also, please look forward to the transcript notification we will send you once that is ready.
On behalf of Eric Just, as well as the rest of us here in Health Catalyst, thank you for joining us today. This webinar is now concluded.
Thank you everyone.
[END OF TRANSCRIPT]