A fundamental algorithm for health analytics

August 19, 2016 / By Kristine Daynes

So much hope is invested in health data. As technology zooms into the 21st century (wearables, predictive modeling, precision medicine), health data organizations are scrambling to keep up. The promise is that the vast stores of data will reveal insights to detect disease, treat patients, predict outcomes and manage resources more effectively.

It’s a fair promise. Technology platforms like visualization dashboards and mobile apps make the data more accessible to decision-makers. But they don’t make the data better or more insightful if it isn’t already good.

Good data depends on advanced algorithms, fancy math that manipulates the data to produce answers. Wearables, predictive modeling and precision medicine are driven by advanced algorithms. The more precise and adaptable the algorithms, the better the insight or answers.

In health care, there are hundreds of algorithms that can do things like identify risk of infection, detect unreported diagnoses and recommend evidence-based treatment options. But before they can do that, they depend on another type of algorithm—risk adjustment.

Risk adjustment accounts for the clinical and financial risk associated with disease

Risk adjustment is a prerequisite to most advanced health analytics. It refers to any number of methods for determining how much a patient’s medical profile will affect use of health services. It can be done purely through statistics, although regression analysis alone produces a static, numbers-only equation.

3M’s approach to risk adjustment combines regression analysis with classification, a way of categorizing individuals with similar clinical characteristics into the same group. It allows us to make comparisons, measure changes and see patterns due to differences in behavior not inherent to a disease.

The accuracy of health analytics, particularly quality analytics, depends on risk adjustment. Without it, metrics such as readmission rates or preventable complications overstate the poor outcomes of providers with high percentages of patients with several chronic conditions, advanced age and disadvantaged socioeconomic status.

The importance of risk adjustment cannot be overstated. According to 3Mers Norbert Goldfield, MD, and Rich Fuller, risk adjustment is intrinsically linked to the measurement of value and health outcomes. They blog, “If we can’t accurately compare patients, then we can’t determine if we are paying too much for their care. We cannot be certain if their health outcomes deviate from what we should expect… Consumers certainly want and expect the best possible outcomes. To accomplish this fairly, it is important to stratify [risk adjust] individuals in a detailed clinical manner allowing fair comparisons among the institutions treating these patients.”

Risk adjustment is portable

There are innumerable ways organizations might use advanced algorithms: notifying clinicians, compiling dashboards, predicting readmissions, stratifying at-risk patients, streamlining transitions of care, assembling care teams, protecting patient safety, communicating with patients, reconciling medications, tracking operational costs, and so on. Although the advanced algorithms in each platform and application will be different based on the decisions they support, the risk adjustment model can be the same. Ideally, for consistent benchmarks and comparable results, the risk adjustment methodology should be portable, that is, applicable across different channels, vendor applications and platforms.

Advice for choosing a risk adjustment model

Organizations have a lot to prepare before deploying new data technology, especially in evaluating risk adjustment and other algorithms. Here are several questions that can guide the process to make sure you get trustworthy insights from the data:


  • Is the risk-adjustment methodology appropriate for the patient populations you serve?
  • Can the risk-adjustment model or its output be integrated into all relevant workflows (or it is tied to a fixed platform)?
  • Does it adequately predict or explain variation?
  • How was the algorithm developed? Was it validated by both actuaries and clinicians?
  • Is there proof of concept?
  • Are the risk-adjusted results granular enough to track measures at a patient level?


  • Can you easily extract and exchange data to support the risk adjustment model? Is your data set complete enough to produce reliable results?
  • How do you expect advanced algorithms, including risk adjustment, to decrease the time you invest in development, implementation, and maintenance of systems?
  • How will people use the risk-adjusted data to make decisions? What processes need to be in place for people to take action on the information?

See these other blog posts for more information:

Kristine Daynes is marketing manager for payer and regulatory markets at 3M Health Information Systems.

Want more insights like this delivered straight to your inbox? Sign up here.