About Analytics

Until recently the ability to rapidly analyze and extract results from large quantities of data was unavailable to many organizations. Rapid advances in computer hardware and software have brought theses sophisticated techniques well within the range of smaller enterprises. These capabilities have allowed these organizations to “compete on analytics”, meaning that they make their superior ability to rapidly analyze data and act on the results a key competitive strategy.

“Analytics” includes the following building blocks:

  • Extracting and joining data from large, diverse data sets, frequently having different “key” values. For example, medical claims generally identify patients by a patient identifier, and physicians by a physician ID. However drug claims often utilize a medical records number as a patient identifier, and a BNDD number for physicians. “Lookups” must be created to allow the combined analyses of these two data sets; for example to evaluate the prescribing patterns of different drugs (from the drug data) for specific diagnoses (from the claims data). The process may also involve using one data set to specify conditions for another data set. For example, abstracting data may be used to identify Emergency Department patients, whose services will be summarized from claims data.
  • Developing data-based models, utilizing the data sets developed above as inputs. For example, models are developed to identify diabetic patients from a combination of medical claims (to identify certain diagnoses) and drug claims (to identify patients taking drugs used to treat diabetes). Other models determine the hierarchal condition categories (HCCs) of patients based on multiple diagnoses contained in claims data. Still others will be used in the formation of Accountable Care Organizations to associate patients with primary care physicians. Third-party payments can be modeled from combining claims or billing data with payment methods and rates.
  • Developing metrics and key performance indicators (KPIs) to allow well-defined and consistent tracking of achievement of the organization’s goals, based on the data models. To be useful, metrics must be well-defined, consistently measurable from accurate information, and linked to the goals and strategies of the organization. Key performance indicators create context for metrics by identifying the expected ranges of values of the metrics that indicate acceptable or unacceptable performance. They also define the type of graphic (stoplight, gauge, arrows, etc.) that will best provide an indicator for the metric. Careful attention to the development of these elements will be key to the organization’s effectiveness.
  • Developing a centralized data warehouse for data, models, metrics and KPIs that will be the central, consistent source of information for the enterprise.
  • Using data mining techniques to unearth hidden relationships. These techniques used to be the sole property of academics and researchers, but are now available to end users through Microsoft Excel windows into properly-designed data warehouses. They let users analyze key influencers if various metrics; for example the major drivers behind PMPM costs. They perform cluster analyses that can, for example, identify clinical services that are commonly ordered together by a particular physician. They allow sophisticated time series forecasting to identify cycles and trends in utilization. And the results can be provided directly into the user’s spreadsheet.
  • Designing reports and data delivery methods. The best analytics are useless if not delivered to the user in a meaningful way. Some data should be delivered as reports that contain detailed data. Some should appear as graphics on a dashboard that allow a quick overview of many different metrics. And some should be accessible in spreadsheets that allow drill-downs and stream-of-consciousness analysis. Tailoring the delivery system to the user’s needs is key to its effectiveness.

Many analytics projects don’t utilize all of these components – they involve development of reports and dashboards from existing data sets and models. Or they may involve creating a summary data model from several large, diverse data sources, which the user can then analyze using spreadsheets or simple reports. In the end, though, they all involve the use of data to help users form successful conclusions.