Three Approaches to Predictive Analytics in Healthcare
When it comes to predictive analytics, the message is clear: it’s the intervention that matters. Crunching all the big data in the world and using fancy machine-learning math will never improve patient care unless the data is tied directly to driving appropriate and timely decisions.
To that point, for predictive analytics to be successful in healthcare, it must have three simple characteristics: timely, role-specific, and actionable.
And like a three-legged stool falling over when it’s missing one of the legs, the lack of any one of these three requirements significantly increases the chance of failure and waste during clinical implementation. Recent articles in the New York Times, Boston Globe, and Wall Street Journal highlight the inefficiencies in today’s healthcare systems.
At Health Catalyst, we are using three types of predictive analytics that directly support clinical decision-making and inform administrative priorities and action. These approaches include:
- Risk scores (stratification)
- What-if scenarios (simulation)
- Geo-spatial analysis (mapping)
Risk stratification scoring can assist in prioritizing clinical workflow, reducing system waste, and creating financially efficient population management. Well-established risk stratification scores of low-risk, high-risk, and rising-risk can play a key role in several healthcare scenarios. For example, a calculated risk score can help reduce system waste by setting workflow priorities for patient follow-up in heart failure patient populations. Based on this predictive risk score ranking, care managers are prompted to focus on those patients at highest risk and pre-emptively intervene with medication reconciliation, scheduling home visits, or follow-up appointments.
Other methods to calculate population risk include the CMS-HCC (Centers for Medicare and Medicaid Hierarchical Condition Categories) approach or population comorbidity algorithms, such as Elixhauser or Charlson-Deyo, which score to approximate the expected disease burden of a given geographic area. Patient-facing surveys such as SF – 12® and SF – 36® are another approach to approximating risk.
Another type of predictive approach we use with our client partners is simulation/what-if scenarios. These tools can be invaluable when decision-makers want to ask simple “what if” questions about a given clinical area or administrative function.
For example, our Key Process Analysis (KPA) application calculates the amount of opportunity dollars available to capture as variation is reduced in a specific clinical care process. The sum of these care process opportunity dollars across an entire care system can be quite substantial. The KPA tool also helps to prioritize which clinical areas to initially target using the Pareto 80/20 rule.
A more clinically focused example is found in our advanced analytics application for the appendectomy population module. In this example, we used a length-of-stay simulation to understand the financial impact (dollars saved) and improved patient satisfaction when patients have a shorter stay in the hospital. Predictive analytics used in a simulation environment allows clinicians and administrators a safe glimpse into “what if” and the likely outcomes of a given combination of events.
Geographic information systems (GIS) and geo-spatial analysis is a well-developed industry, having recently passed its 50th anniversary mark. Despite those five decades, the obvious overlap to leveraging GIS tools with healthcare and geomedicine is just now coming into focus. Only recently have large sets of national claims and payment data been made public. Additionally, private and non-profit institutions have large amounts of operational and patient data that could be mapped.
Mapping layers and predictive analytics are routinely used to forecast weather, optimize supply chains, and support military deployment. A natural extension of these established approaches is to leverage GIS mapping of health care facilities, patient disease burden, and accountable care populations. This is a very visual and effective approach to analytics and decision-making.
What Really Matters for Healthcare Predictive Analytics
Our message is simple: prediction is the easy part, it’s the intervention that matters.
Return on investment does not reside in data itself, but in timely interpretation of that data followed by appropriate intervention. When a calculated metric is timely, role-specific, and actionable, the return on dollars and quality follow. All the predictive bells and whistles in the world will not improve healthcare one iota without thoughtful and meaningful intervention and clinical leaders willing to base that intervention on proven predictive data.
Would you like to use or share these concepts? Download this Predictive Analytics in Healthcare presentation highlighting the key main points.