Anyone working in the healthcare field has heard the buzzwords "big data" and "analytics" with increasing frequency over the last five years. Interest in the topic is fueled by many factors, including accountable care, the Triple Aim and value based purchasing.
In the current era of accountable care, a healthcare organization must be data-driven to remain competitive. This realization has prompted the same data bonanza in medicine that several other industries experienced years ago.
New technology companies are offering data and analytics products that run the gamut from displaying your data in an interactive platform to a complete solution that:
1) Pulls data from your clinical and financial information systems
2) Standardizes it
3) Creates your data warehouse
4) Serves that data up in the form you like the best
Most commercial analytics systems offer canned reports that can be customized (though this requires some training, of course). Some vendors also base their business on proprietary algorithms that estimate risk and identify cohorts of patients on whom to focus resources (e.g., people at risk for 30-day readmission). Such systems are intended to improve quality and reduce cost.
As a result of this trend, data scientists are in high demand, enjoying bidding wars for their talents that rival those happening in the residential real estate market.
While this sounds exciting, the reality is that most hospitals don't have an integrated data warehouse that is inclusive of all of their clinical and financial information systems. Depending on the questions you ask, this could lead to an interfacing nightmare.
So how can organizations harness the power of big data without creating an information quagmire? A good way to start down this path is to focus on an initial set of metrics that are aligned with your organizational strategy and goals. Having a concrete game plan makes it easier to choose an appropriate platform and analyses for your needs.
Data analysis takes place on three levels. The lowest hanging fruit is what is referred to as descriptive analytics. This yields concrete information like how many patients a provider sees per hour, the average cost of a hip replacement or the average HgA1C of a defined patient population.
Given the right aggregation of the necessary data, metrics like these are relatively straightforward to calculate. And simple metrics can lead to more insightful ones.
For example, my medical center was involved in a pilot that aggregated data from a behavioral health database and a physical health database. The intent was to inform providers who shared patients (but used different EHRs) about all of their common patients' medications, allergies, problem lists and encounters.
The chief scientist from the data aggregation company was exploring the merged patient dataset and discovered that five patients in the 200-patient cohort had potential drug-drug interactions. So I received a call and verified that four of the five required medication changes. The next morning, the patients' providers were notified and the necessary changes made, diverting potential complications.
The next level of analytics is predictive. For example, we can use data to predict how many patients will be seen in the emergency department (ED) on a given day or time of day, which patients in the ICU are likely to deteriorate in the next 24 hours and which patients are most likely to be readmitted within 30 days.
Right now, a manual form of this type of analysis is done in many EDs. For example, ED directors collect patient arrival times over the course of a year or more, and these are segmented by day of the week. Directors use this data to predict future patient flow and staffing needs. Provider staffing is then adjusted on a daily basis, and shift length is configured per day of the week to accommodate patient flow.
Predictive analytics allows for more powerful analysis of this data. Most commercial analytics systems are not sold with predictive capabilities. However, these can be achieved by building upon existing systems or packages.
Finally, prescriptive analytics combines both of the previously described analyses and provides direction on what actions to take. Potential uses include defining the best way to treat diabetics of a specific demographic, dictating how many beds should be added to the ED to accommodate future demand and specifying the best treatment to improve clinical quality and lower costs for a specific demographic and disease state.
Becoming a data-driven organization is not easy — particularly in a field that has entrenched practice patterns and variable reimbursement strategies. Implementation requires a significant commitment to governance and transparency to achieve successful culture change. But when done well and tied closely to organizational goals, big data analysis can lead to a win-win situation for patients and providers alike.