A Guide to Successful Outcomes Using Population Health Analytics

a simulation based on similar patients with similar conditions and demographics and see how to move from a high-risk category down to a lower-risk category. The physician can also show the patient the impact improving (or not improving) his condition will have on his out-of-pocket costs. This is an actionable visualization that can help promote positive activities by both clinicians and patients.

Patient Flight Path for patient with diabetesFigure 10: An Example of an Actionable Visualization: The Patient Flight Path for a Patient with Diabetes


Analytics give the ability to measure how the organization and the individuals providing care within that organization are performing. Good analytics will consider compliance with best practices and whether care is being provided in the most effective and safest way. If an organization is unable to measure these actions, it proves really hard to actually improve.

An organization also needs to be able to find patterns in the data to show correlation and causation. This means integrating clinical data, financial data, and patient experience data. Eventually, clinicians should be able to use this integrated data to predict outcomes and prescribe actions that would be most appropriate for an individual patient.

The biggest issue organizational leaders must address is the amount of time it takes to get information. Strong analytics will limit the amount of time data analysts and data architects spend hunting down and compiling data. The majority of their time should be spent interpreting the data and answering questions.

Analytics add valueFigure 11: A Graph Showing How Strong Analytics Can Add Value Through More Productive Data Analysts and Architects

How to Start Building Analytics: Establish an Infrastructure to Bring Data Together

The first step to building a strong analytics is establishing an infrastructure that is capable of securely bringing all the data from EMRs, billing systems, etc. together. Every hospital has multiple source systems and data is typically siloed in different areas. An enterprise data warehouse can combine that data into a single source of truth.

There are three data modeling approaches often seen in healthcare.

The Enterprise Data Model

Technology vendors porting products from other industries and healthcare systems who have hired data modelers from other industries often adopt the enterprise data modeling technique. The model works well when data rarely changes, which is why it has worked in other industries like manufacturing and retail. Using this approach, data is organized and defined, and once it’s in the model, applications can use the data easily.

But enterprise data modeling has real challenges and limitations in healthcare. All data must be bound early and tightly to the predefined model. Changes in requirements, such as a regulatory definition modification, require a complete redesign or remapping of the model. Plus, developing the data model in the first place is time consuming. Using this method, it is difficult to model complex healthcare concepts.

Dimensional Data Modeling

Another common approach is the dimensional data model. EMR vendors and healthcare point solutions tend to favor this approach. It is quick to get started and to add value using this model. Only data to be used for the given analysis is extracted. Many traditional visualization tools prefer this type of model.

However, it carries some flaws. As additional data is needed, many redundant data feeds must be created. Changes to the underlying source systems cause a maintenance nightmare. Additionally, comprehensive atomic-level detail is not stored in the model meaning deeper questions can’t be answered without pulling more data. Dimensional data modeling is difficult to use in the healthcare environment.

Adaptive Data Modeling

One approach that has gained real traction at many successful health systems is the adaptive data model. The organization does not need
to predetermine how all the data will be used because data is bound to definitions late in the process. Data is pulled from transactional systems into source marts with very little transformation (thus, it is a quick and simple process). This model can grow incrementally and adapts easily to changing healthcare requirements. In addition, adaptive data modeling contains both atomic- and summary-level detail and can successfully support complex healthcare quality improvement initiatives.

However, the model is unfamiliar to many IT professionals because it’s been most commonly used in healthcare settings and military logistics; a lot of IT professionals come from data warehousing in retail, manufacturing, or finance environments. IT professionals evaluating data models can’t review the data model in this approach because it is always changing. Using an adaptive data model requires a mindset shift.

Information Management: Reoccurring Data Tasks

In addition to specific use cases, there are common tasks that hospitals will repeatedly do with data. For example, defining patient cohorts and registries, attributing patients to providers, and understanding the severity of the illnesses and comorbidities. To create the most effective system, the organization will need to have three distinct components working with its Late- Binding Data Warehouse.

The first is data capture. Here, application administrators acquire key data elements, assure that data’s quality, and ensure that data capture is integrated into the operational workflow. The second task involves data provisioning. Data architects and analysts move data from the transactional systems into the data warehouse, then build visualizations and generate external reports. The third component is data analysis. Data analysts and subject matter experts interpret the data, discover new information in the data, and evaluate the data’s quality (relaying that final point to the application administrators).

Information Management CycleFigure 12: The Information Management Cycle

It is a cyclical process, as seen in figure 12, with capture leading to provisioning, leading to analysis, which naturally circles back to capture.

What to Do with the Data

There are several ways an organization can decide what to do with the data—that is, which area to focus on for improvement and how to go about bringing that improvement.

One method involves setting a minimum standard of quality and then focusing the improvement effort on those not meeting that minimum standard. This might be called the “rank and spank” or “punish the outliers” approach. While those physicians who were below the minimum standard…

Page 3 of 5
1 2 3 4 5
Loading next article...