Best Solution to Aggregate Healthcare Data Including Clinical, Financial, Research, Population Health, and More
Being in the healthcare industry means generating and collecting enormous amounts of healthcare data. Systems collecting clinical data, financial data, staffing data, human resources data, supply chain information, research data, among other data sources, are all part of the mix. In addition to capturing so much healthcare data, health systems also need to analyze the data for many different needs, such as quality improvement, operations, research, and financial analytics.
Is there a one-size-fits-all solution that can aggregate all of a health system’s data from the many disparate source systems, so analysts don’t need to build or purchase a different analytics platform for each use case? The answer is yes: an adaptive, scalable, flexible enterprise data warehouse (EDW).
But here’s the catch: the EDW must be designed to flexibly support the various analytic needs each health system will have. And even though there are several models of healthcare data warehouses (enterprise data models, independent data marts, late-binding) to choose from, there’s only one architecture that’s proven to be the most flexible for the healthcare industry: the late-binding architecture. Here are five important qualities that make the late-binding model the best solution to aggregate data for healthcare.
5 Ways a Late-Binding Healthcare Data Warehouse Supports the Many Needs of a Healthcare System
1. New source data feeds can be developed quickly
The late-binding architecture is successful because the data does not undergo significant transformation or scrubbing as it’s loaded from the source systems into a source mart within the data warehouse. Once in the source mart, the data resides largely in its native format until it’s needed, eliminating the traditional method of assigning business rules or definitions up-front. This allows new data feeds to be developed quickly and more complex data operations are performed only when a bona fide use case exists and business users can help to fine-tune the logic. With so many data sources needed to support the myriad of data analysis needs, you need to be able to incorporate those data sources quickly and efficiently.
In comparison, data in an early binding architecture is transformed early in the process — an inefficient, difficult, and inappropriate task for healthcare-related data. Any architecture that requires substantial modeling of data up-front misses the most important aspect of healthcare data: it’s difficult to define a data element until actually working with data consumers. Even a calculation as simple as length of stay can have different meanings to different data consumers: a surgeon often views length of stay from incision time to discharge, while an administrator tends to look at admit time to discharge. Both are valid calculations and need to be implemented appropriately to support the analytic questions of each user.
2. The architecture is flexible.
Unlike a traditional, early-binding enterprise model, a late-binding architecture provides analysts with the flexibility they need to use the data however they see fit when they’re creating analytic applications and data marts. For example, an analytic application for quality improvement might contain clinical data, patient satisfaction data, and costing data, but an application for operational purposes might contain staffing levels and clinical data. A research analytic application, however, may need to combine clinical outcomes data with a research registry. With all of these various data needs from throughout the system, a flexible architecture provides just-in-time data for any use case.
A flexible architecture also allows you to combine different data sources in different ways. For example, when you need to link data across different source systems, a data bus in a late-binding architecture serves to link elements together using common linkable identifiers. The ability to mix and match data sources and data elements from different data sources for different purposes is absolutely necessary and should be the design goal of any good healthcare data warehouse.
3. Data definitions match their context.
With a late-binding architecture, different data marts can contain definitions that match their context: a surgery data mart aimed at surgeons can use the surgeon definition of length of stay while another data mart aimed at operational administrators might use the “admit time to discharge time” definition for length of stay. The underlying data for the calculations is represented in the different source systems loaded into the warehouse, but that data is used differently for different targeted users. Instead of creating a whole new data warehouse to support the different use cases, it’s only necessary to create different calculations that feed from the same source mart layer.
4. A single source of truth supports all use cases.
When organizations create separate data warehouses to support individual departments and use cases, there can be a lack of consistency about which data source holds the source of truth. Usually, this is because each different data warehouse transforms and scrubs data in different ways and does not have good tools to track important details like lineage and transformation logic.
A Late-Binding™ data warehouse supports all analytic use cases because the underlying healthcare data is the same for all departments. By bringing the data into the data warehouse in a raw format, there’s now a foundation of “raw ingredients” through which any department can craft their own data mart. In fact, the more data that goes into the warehouse means more options for how analysts can use the data. Also, since the data warehouse architecture does not heavily transform the data, lineage can also be easily tracked and catalogued through a single enterprise metadata dictionary.
The following video highlights the problems with a traditional EDW model by showing how difficult it would be to try and make different recipes with repeated trips to the store for each ingredient. Think of the ingredients like source data and recipes like data marts. I hope you like it! (Don’t worry — I won’t be leaving for Hollywood any time soon. J)
Click Below To Watch Brief Video Demonstration
5. Data access can be customized.
In order to support a diverse and thriving user-base, a healthcare data warehouse must have a solid security framework. For example, HIPAA has very specific rules about who can use medical data for research, and health systems need to ensure that their security model allows granting access to appropriate users. Being able to grant access to authorized users is just as important as controlling access to non-authorized users. I’ve seen too many healthcare systems focus on who should not have access without focusing on who should have access to data. If data access is over-restricted, nobody will be able to use to improve patient care through quality improvement and research analytics. This is the major part of the value proposition of the huge effort of collecting good data.
The Late-Binding™ architecture also supports granular security. For example, a researcher could be granted access only to data marts that have been de-identified. Additionally, researchers who have been approved to access patient data can be granted access to only the specific data about patients in their study. This allows researchers to use the same infrastructure as other departments like quality and financial. Granular security also provides a single point-of-access that when implemented with good data stewardship, will serve many different needs across the organization.
The Best Data Warehousing Approach for Healthcare: A Late-Binding™ Data Warehouse
Traditional approaches for data warehousing were developed for certain industries like retail and banking. For those industries, an enterprise data warehouse is easy and efficient to implement because the underlying data and use cases are relatively static. But the enterprise data model simply can’t provide the scale and flexibility the healthcare industry needs. As the demand for analytics continues to grow, and healthcare knowledge continually expands and changes, the data model needs to be flexible and adjustable. That’s where the adaptive, Late-Binding™ model fits in: it’s customizable but has the right amount of structure to efficiently work with the many sources of healthcare data, giving health systems access to all of the data they need for any use case.
Did the Grocery List video strike a chord with you? Have you experienced first-hand the frustration of trying to collect data from the many data collection tools in your healthcare organization? If so, what has your organization done to streamline data collection?