During Big Data Week 2014 I had the pleasure of talking to Michael Dulin of Carolina’s HealthCare System about their work using data to improve healthcare outcomes.
Q: I am sure that your healthcare system’s advanced analytics group, Dickson Advanced Analytics, did not happen overnight. What were the drivers for Carolinas HealthCare System to focus on predictive analytics?
During Big Data Week 2014 I had the pleasure of talking to Michael Dulin of Carolina’s HealthCare System about their work using data to improve healthcare outcomes.
Q: I am sure that your healthcare system’s advanced analytics group, Dickson Advanced Analytics, did not happen overnight. What were the drivers for Carolinas HealthCare System to focus on predictive analytics?
Carolinas HealthCare System’s Dickson Advanced Analytics (DA²) journey began over a decade ago with our desire to enhance the quality of care delivery in the hospital setting. At that time, our capabilities of integrating and processing data were identified as being foundational to driving changes in the clinical domain. Our healthcare system quickly recognized that it would be equally important to use the same data and analytics systems to drive operational and financial performance, as well as quality in ambulatory, or primary care, domains. We made significant investments building data and analytic capacity. This process really took off in 2010 when we made the decision to centralize our data and analytics teams from across the organization. We built one team. Our transformation is now central to driving our transition from a volume-based organization to a value-based organization. We are thinking about what health care delivery systems will look like in the future and how we can best enhance the lives and experiences of our patients – this vision is built upon the foundation of data and analytics.
Q: Today, everyone is talking Big Data, and we are in the midst of a data explosion. How has this affected the approach to population health management in the last five years?
In healthcare, a key catalyst for the exponential expansion of data availability was the move from paper-based records to electronic medical records (EMRs) along with a concomitant expansion of our medical knowledge. EMR implementation has been a difficult journey, but now that we have taken the first leap into the movement, we are able to move on to the fun and exciting stuff, which means extracting and integrating the data from the EMR, and using it to drive improvements in care. At Carolinas HealthCare System, a key step was the creation of a centralized team responsible for supporting all of the data and analytics work using a single enterprise data warehouse. Consequently, today, we are able to better leverage the clinical and operational data we receive from our hospitals and ambulatory care centers and use it to drive improvements in care delivery and efficiency. Having a centralized team to manage this data also allows us to work on predictive and prescriptive health management and enhancing decision support for our clinicians at the point of care.
In the past, part of a physician’s job was to take vast amounts of clinical data from medical records, think through the data, and find trends and patterns supported by current evidence to diagnose and treat patients. It is now beyond a single physician’s capacity to absorb all the data and compare it against all potential diagnoses and evidence-based treatment guidelines. Indeed, knowledge in the medical field is growing so quickly that providers would need to spend the majority of their time reading rather than seeing patients to stay up-to-date. This is particularly true in the primary care medical home, where physicians not only have to manage their patients and stay on top of the most recent evidence, but also coordinate their patients’ care across different specialties and facilities. This is my vision of population health – where data/analytics, medical evidence and care coordination come together to optimally provide care, information, and a personalized approach to a large group of patients. In the near future, population health will be comprised of a care team using data and technology to engage patients in the community.
Q: Can you provide one or two examples of how using bigger, better data has improved the way we can manage the health of a community?
Only a small percentage of the health outcomes are impacted by the medical system itself. The other 90 percent are related to genetics, social circumstances, behavior and environment. The explosion of big data allows us as a healthcare system and myself, as a primary care physician, to think beyond the 10 percent of a person’s health that is affected by elements outside of clinic walls. Access to data across the community allows me to consider the remaining key factors that affect a person’s health, including behavior, social conditions and environment. By combining these elements, we are able to better understand how these variables impact health and readjust how we approach our goal of improving a patient’s health.
For example, through these data capabilities, we can begin by looking at your neighborhood and identify environmental hazards and link these to certain genetic factors that predispose you to asthma in the future. The models will be much more complex than this as we start to leverage consumer information (like the data that you provide when you swipe your membership card at the gym and buy donuts at the store) along with data from your Fit Bit and Wi-Fi enabled scale, in addition to clinical and genetic data.
One example, in its early phases, is our team’s National Institutes of Health-funded work that explores hot-spotting across communities. We have brought in data from the healthcare system and from the community, combined it and created models to identify the people most at-risk for going to the emergency room for something that would be better treated in a primary care office, like a headache, sore throat or cough. These types of data models are able to very accurately identify the neighborhoods where this happens. Using these predictive models, we no longer have to wait for people to try and get into our clinic; we now can engage people living in these neighborhoods and guide their navigation through the healthcare system.
This project also has allowed us to identify environmental and social problems that have kept people living in these high-risk neighborhoods from being able to do the right thing – like exercise or eat healthy foods. What we have found in those communities are multiple barriers that prevent people from engaging in healthy behaviors. For example, there might be a park around the corner that community members can use for exercise, but because of high rates of violent crimes or homelessness in the neighborhood, they felt unsafe going out of their homes. By engaging with them as a healthcare system, we were able to help them overcome some of those barriers and find alternative ways to get exercise, like walking in groups and building relationship with neighbors.
Q: Have you learned anything new over the last three years while leading Dickson Advanced Analytics at Carolinas HealthCare System?
One of the key lessons learned is how important data analytics have been to evaluating and validating what we are doing as a healthcare system. In the past, prior to EMRs, we often had to go by gestalt to determine best practices in healthcare– this approach resulted in huge variations in how patients were treated. Now, I am excited that we have the opportunity to truly evaluate the work that we are doing and standardize around evidence-based practices. So, when we make changes in care delivery, when we redesign clinical systems and when we implement new decision support tools within the EMR, we are able to leverage our analytics team to evaluate the differences that those changes have made. One example, where we’ve put in significant work in terms of developing a decision support aid, is the electronic asthma action plan generator. This project was a lot of work, but we were able to first determine the need for the work and then evaluate its impact. This evaluation also kept our clinical teams engaged – when they saw their additional efforts pay off in patient outcomes, they remained motivated and were persevered.
The other lesson learned is that many people (including me) assume that data are always easy and that, when we deploy interventions that impact care delivery, the data will follow and be useful for evaluation. To fully understand the impact of our work, we need to spend time defining the metrics and the data elements that will drive those metrics over time. Often, the best approach is to use research methods to design a study in the real-world setting to truly understand the impact on patient outcomes. Much of my work is helping our teams to think ahead about defining and measuring our success.
Q: As you began your work with population health management, what challenges did you know you had to overcome? How have you managed to drive successes, and what role did data play?
Early on, one of the challenges and one of the lessons learned was one in the same: the importance of engaging end users. I mentioned before that as healthcare systems think about the interventions, they should also engage data and analytics teams to think about the metrics and evaluation. This engagement is critical because analytics teams cannot be effective if they are working in isolation of the rest of the system. Overall, we are delivering a model that engages all stakeholders. We need buy-in within the system, and we need the same from end users. When we produce a report or something like a predictive risk model, we must have someone on the other side ready to take that information and apply it to a pre-existing strategy that can make a difference within the system and in the community. Using data to change the traditional healthcare landscape, its delivery systems and results has been one of the major themes of the research that I’ve done over the last 10 years. I am committed to finding the best ways to engage end users to improve population health. The data alone might not always make sense or push us in the right direction unless we understand what that data mean to end users. To bring the insight of end users into what we do every day, we’ve done much qualitative work in the community, focused on interventions. We’ve gone into the community to ask the end users the questions that allow us to better understand the validity of our high-level data, what the data mean and how we use the data to drive transformations to overcome barriers. We are asking the same questions inside the clinical domains. Once we have produced the risk model and the variable, and understand a patient’s risk for a particular disease, it is very important that there is someone to take that information and use it to make a difference for the patient. That is what we are doing with our data models and analytics, but our System infrastructure is one of the most important pieces. Efforts are wasted if you produce the data without someone around to act on it.
Q: How have data contributed to population health management as it relates to common chronic diseases like asthma and diabetes? How do you see data changing the future for chronic diseases and population health?
In the terms of population health management, the first place we started was in identifying our patients that are at high risk for ending up in the hospital or the emergency department. These patients often have an advanced chronic disease and/or multiple medical problems requiring several different physicians, as well as social workers and care managers to keep them healthy. The real secret to managing population health is identifying patients who are healthy now but may have a hidden risk of developing a chronic condition in the future. This is the population we are looking to identify. As a primary care physician, I am also passionate about this – to engage people with the healthcare system in a way that keeps them healthy and productive members of society and away from the hospital.
For chronic disease management we look a simple formula that uses: (1) decision support for providers when they are seeing patients, (2) data sharing prior to the start of the day so that we can pre-plan the day and ensure that everyone gets the medical care needed; (3) a daily huddle where we look back over the work from the day before and identify any mistakes, so that we can correct the process that led to the mistake; (4) population management reports that allow the care teams to see what is happening with patients who are not in the clinic that day but may need some service; (5) patient access to their data so that they can better manage and self-govern their health; and (6) provision of data to patients about community resources that could be used to help them live a better healthier life. As you can tell, each of these steps requires accurate, up-to-date data.
Another component of population health is the use of predictive analytics. We have moved into this area with disease states, as well as with hospital readmissions and hospital length of stay. With asthma, we’ve created a database that includes all of our patients that have a diagnosis of asthma. We include elements that track if a patient has been in the hospital or emergency department because of their asthma and what factors might have led to that event. Then we add more information, such as the medications the patient is taking or whether the patient may have an underlying health issue like obesity. Next, we can start to look at environmental factors, like changes in weather, changes in pollen count, outbreaks of respiratory illnesses across communities and crime rates in the neighborhood. Layering these indicators over our data, we can use predictive models to drive care at the individual level. We can then tell a patient, “because of your condition and because of your prior experiences, we think you might be heading towards a worsening of your asthma if we do not make a change in your treatment regimen.” The healthcare system can can support that patient to do the right thing and keep them out of the emergency department and the hospital.
Q: Can you share more about your population health successes in the greater Charlotte area and what you hope to achieve in the next five years at Carolinas HealthCare System?
To be successful with the predictive modeling work, we also have to be efficacious at exchanging data with other key partners across the region. We are building the foundation for that now, but we have to understand what happens to patients when they are outside of the walls of the system. One important example is exchanging information with other healthcare systems. As a physician, when I see a patient that’s coming back after a recent hospitalization at a hospital outside of our system, I must know what happened, and why. In our current environment, I may have to call that hospital and receive a fax with the patient’s health information. However, the technology exists to make this information electronically available to me at the point of care. In the very near future hospitals will be exchanging data like this to advance patient care (with some already doing so).
Once this infrastructure is in place, systems must decide what exchange of data looks like and how we do that in a way that provides the best potential outcomes for the patient. As we begin to see that exchange of data, the patient will be able to gain greater control of their health information and data, and that is going to be a significant challenge in terms of redefining traditional physician –patient roles and exchanges. I believe this disruption will drive improved performance. So, in the future, when a patient comes in to see me, they have at their disposal all the information that’s important to their health. This will include data on their behaviors, including whether they’ve been exercising, if they’ve used the YMCA, their food shopping patterns, and utilization of healthcare services outside of my immediate periphery. The next step from this is to build even better predictive algorithms that are at the patient’s disposal and help them make informed decisions about their health.
The interviewee, Michael Dulin, MD, PhD, is Chief Clinical Officer for Analytics and Outcomes Research for Carolinas HealthCare System’s Dickson Advanced Analytics (DA2) group & Director of Research at the Carolinas Medical Center Department of Family Medicine.
