Healthcare organizations are constantly bombarded with data and sifting through all of that information may seem like a daunting task. However, that data holds valuable insights and the potential to provide a better understanding of the health system as a whole, making the tedious work worth it.
To achieve the most accurate and useful insights that can help organizations develop best practices and improvement strategies, strong analytics techniques that can comb through and explain such complex healthcare data are a must, according to 3M Health Information Systems Senior Medical Director Gordon Moore, MD, who spoke at Becker's Hospital Review Health IT + Clinical Leadership 2018 conference May 11 in Chicago.
According to Dr. Moore, it's time healthcare pull analysts out of the data maze and empower them to go beyond data silos to inspire real change. "When we have vast amounts of data, we [need to be] using risk-adjustment that makes sense to clinicians," Dr. Moore said.
Many solutions don't live up to the hype, Dr. Moore said, adding popular approaches look backwards by using retrospective data to explain how well organizations performed in the past. However, solely relying on such statistical models poses the risk of identifying correlation without causation.
Models that could predict certain outcomes or put plans into action are the most useful in healthcare, Dr. Moore explained. Coupled with clinical expertise and experience, analytics solutions can uncover hidden interdependencies and inform best practices.
Here are Dr. Moore's five steps for building an intelligent, useful analytics approach.
1. Extensive baseline data. To effectively use a big data approach, a large test dataset is crucial for teaching the machine learning algorithms how to understand the best path to the goal.
2. Clinically defensible risk adjustment. Instead of using complex statistical models that can help identify what's wrong, it is important to consider the whole patient and not just their problem. That way, when optimizing a certain outcome, the risk of compromising other aspects of the patient's health is decreased.
3. Clinically relevant performance metrics. When building an approach, it is important to be very specific about what is being optimized and to keep those performance metrics rational and clinically relevant. Possible outcome measures include hospitalization rate for people with ambulatory care sensitive conditions or risk-adjusted cost of care.
4. Risk-adjusted patient and performance benchmarks. Asking the questions such as "Compared to what?" and "Compared to whom?" can be very useful when targeting a specific metric and establishing benchmarks.
5. Data integration. It's important to take into consideration data from various sources, such as historical claims data, which could reveal information about the patient that the organization may not have readily available.
More articles on data analytics & precision medicine:
Geisinger to integrate DNA sequencing into routine clinical care
4 things to know about the national 'All of Us' biobank at Mayo Clinic
A direct-to-consumer DNA testing service fails to distinguish dog DNA from human DNA