1. United States
  2. Health Care
  3. Health Information Systems
  4. Resources
  5. Library
  6. Health Analytics: More promise than performance?
Health Information Systems
Doctors looking at data on tablets

Data analytics: More promise than performance?

We live in an era of unprecedented data access.

But how can you leverage data to benefit your patients and populations?

Health care leaders have never had more access to data than they do today. This data can potentially provide a better understanding of their organizations and empower well-informed decisions in this new environment. But health care data is complex. When you begin to map out all the potential interconnections, it reveals a very intricate decision-making environment.

Where are the problems and opportunities?
What drives a winning strategy?

Data anaytics

Data and analytics are the #1 concern currently facing health care executives in the United States.

(According to the Healthcare Executive Group’s 2018 “Top Ten” study.)

Most health care analytics solutions strike out.

Why? Too many solutions look backward in an industry that needs to look forward.

A strong analytics approach is essential to draw needed insights from data. But for most organizations, today’s solutions offer more promise than performance. And since the data under analysis is retrospective, the solutions offer only a look backward, not forward.

But the promise of predictive, prescriptive data—data that tells you what actions to take before you need to take them—is a grand slam.

Most organizations are here...

  • Retrospective

    How well has your organization performed in the past?
    (Easy)

  • Real time

    How is your organization performing now at the point of care?
    (Challenging)

But they need to be here

  • Predictive

    Who is likely to develop Disease A or progress to Disease B? Can the progression be slowed?
    (Difficult)

  • Prescriptive

    What action plan can produce real results?
    (Most difficult and beneficial)

Relying on statistical models is not enough.

Analytic solutions must be clinically credible.

  • Statistical significance ≠ Clinical significance

    A strictly statistical approach to health care analytics risks identifying correlation without causation. Proven methodologies combined with clinical expertise are essential.

     

    Pure machine-learning models are “black boxes.” The machine cannot explain how it comes to its recommendations. When pure machine models deal with human lives, mistakes may be made without explanation or intervention. Consider the self-driving car that struck and killed a pedestrian. Or the drivers of those cars who stop watching the road because they have too much faith in the vehicle’s algorithms.

    A better model for health care begins with teaching the machine algorithms based on proven methodologies and then having subject matter experts review the machine output. With this approach, you can pursue opportunities consistent with your organization’s professional and ethical standards and knowledge.

    While new vendors and technologies promise tremendous value, clinical expertise and experience are must-haves for creating meaningful insights.

 
 
 
 
 
 
 
 
 
 
Data integration

Augment your understanding by expanding information sources. Consider how much more you can improve outcomes by integrating information from smart watches, gym participation tracking, healthy eating programs and more.

Risk-adjusted patient and performance benchmarks

Patients’ severity of illness and complications are not all alike, so account for differences when reporting benchmarks and performance rates. Opportunities lie in the gaps between the risk-adjusted, expected rate and the observed rate.

Clinically relevant performance measures

To improve outcomes and reduce the cost of health care, healthcare professionals must change the way they work. Example: instead of counting how many people got flu shots, determine if the flu shot program reduced ED visits and hospitalizations.

Clinically defensible risk adjustment

To be clinically credible, results must be risk-adjusted using proven and transparent methodologies.

Extensive baseline data

Machine learning and artificial intelligence (AI) require large data sets so that the machines can learn and then test algorithms. This requires the ability to combine very large data sets.

The right analytic solution reveals complex interdependencies.

  • For example:
     

    • Are hospital readmissions increased when the length-of stay is shortened for people with inadequate family support?
    • Are post-operative infection rates increased by using lower-cost sterilization techniques?
    • Do people with complex, chronic conditions experience fewer unnecessary hospitalizations when team-based ambulatory appointment scheduling is used?

It’s time for robust analytics to step up to the plate.

With a strong foundation, advanced technology can apply clinical methodologies at scale.

  • Icon of a brain with connections within it
    Natural language processing (NLP)

    Eighty percent of health information in medical records is unstructured notes, such as physician and nursing notes, and unusable. (Source) NLP is a technology that can read unstructured notes and turn them into structured information. NLP automates and augments the work of those spending time searching through medical records.

  •  Icon of a gear with addition signs below
    Automation

    People trying to understand data spend considerable time thinking about and acting on hunches to determine the best way to slice and dice that data. With the advancements of information technologies, much more of this work can be automated and the information even augmented. Cutting edge technology, properly programmed and trained with clinical models, can reduce the time spent in manual search and analysis.

Give your data analysts a winning strategy.

Example: Drilling down to the root cause of a length-of-stay (LOS) issue within a theoretical health system.

The example below clearly demonstrates how improvements in efficiency can be achieved through advanced technologies. In the past, a long LOS for a particular orthopedic procedure might mean working with all of the surgeons who perform that procedure. Instead, AI can provide more specificity more quickly and identify the areas to focus on. In this example, the analysis indicates there are only two surgeons whose performance needs review.

From broad issues (difficult to address) to specific, actionable issues

  • Icon of a hospital

    Hospital #1 accounts for 32% of total excess LOS

  • Icon of a knee joint

    Of that 32%, excess LOS as a result of orthopedic surgery accounts for 23%

  • Icon of a magnifying glass

    Within that 23%, potentially preventable complications are 46% higher than expected

  • Icon of a rectangle with a person inside it

    Of those potentially preventable complications, 68% occurred in patients with preexisting conditions

  • Icon of a physician

    Of those patients, 22% were treated by “Doctor A” and “Doctor B”

How do you address persistent industry challenges with analytics?

For each challenge, there is a potential solution.

More resources

Take the next step on your journey

Let's see how we can work together to optimize your organization today!
For more information, contact us or explore our other solutions and services.

CHAT WITH SALES

Connect with us

Discover the inspiration and information you’re looking for on the health care topics that matter most.

  • 3M is a trademark of 3M Company. Verily is a trademark of Verily Life Sciences LLC.
Legal
|
Privacy
|
HIPAA Privacy
|
DMCA
|
Accessibility Statement
© 3M 2021. All Rights Reserved.
Follow Us
The brands listed above are trademarks of 3M.