Insights Into Predictive Analytics and Population Level Health

Insights Into Predictive Analytics and Population Level Health

In the introduction to the November 2014 issue of HealthLeaders the editor states that “individual hospitals and health systems can analyze their own data sets to find opportunities to save money and provide more effective care for individuals or groups of patients.” I have found in my work with data that there are tremendous opportunities for healthcare providers to greatly improve the lives of patients while using fewer resources through the use of predictive analytics and population level management systems. By examining data from a global or population level and using predictive statistics to identify key performance indicators providers can improve the outcomes for many of their patients.

What are the key processes that can provide such success in providing high value outcomes? I would like to illustrate how a health system can leverage data to improve care. Much can be done with basic databases and advanced statistical analysis. A system does not need to invest in expensive IT solutions to achieve good results, although for larger systems such an investment would be worthwhile, I believe.

The first step is to set goals for the analysis and population level management. It is not enough just to collect data and present it to physicians. The goals should also include action steps. One goal, for example, could be to prevent patients of primary care physicians identified as being prediabetic from progressing to type 2 diabetes. This goal identifies a population-patients with prediabetes-and an action-keeping the patients from advancing to type 2 diabetes.

Goals should focus on preventive activities or strategies. These can be at the primary level, secondary level or tertiary level. Primary preventions strategies try to prevent the occurrence of disease or increasing resistance to disease. An example of this is having as many patients as possible have flu vaccinations. Another is counseling teens to avoid smoking and using alcohol.

Secondary prevention strategies seek to identify indicators or test results in patients that would predict the likelihood of developing a disease. For instance, a provider group could try to identify all patients with hypertension in order to provide services to prevent coronary disease or strokes.

Tertiary prevention strategies seek to prevent more serious outcomes for patients with serious conditions, such as type 2 diabetes. The goal here is help patients manage their condition so as to keep them from needing emergency medical help or from being readmitted to a hospital.

Once a goal(s) is set, a healthcare system needs to collect baseline data for future reference in data analytics. For instance, for systems with a goal of keeping patients with prediabetes from developing type 2 diabetes, data should be collected that indicates what percentage of their prediabetics developed diabetes. One could look at data over a one year period or a shorter time if the patient population of prediabetics is large enough. Another approach to this goal that would generate more useful data would be to track the fasting glucose level of patients with prediabetes. Such data would indicate how variable this measure is in patients, which would be an accurate indicator of how well patients and physicians are reducing the level of this indicator or at least keep it from getting worse.

As data is being collected and analyzed for variability and trends, clinicians along with support staff such as nurse coordinators should determine a variety of prevention strategies that can be employed to improve the outcomes. Administration should be involved in these decisions as their input on the costs of providing such services is important, especially in a value-based reimbursement environment. Strategies can be drawn from best practices research and from insights gained from clinician experience.

It is important that several strategies be employed concurrently in the designed prevention services. From the perspective of data analytics trying one strategy at a time provides much less predictive power than employing several at a time. Analyzing the interaction of several strategies through complex analysis provides much more useful information that can be used to provide better care.

For instance, strategies for the treatment of prediabetics could include having patients test their glucose level every three months, referring the patient to nutrition services if their payer will cover it or if there is a nutritionist on staff, have the patient join the YMCA as the Y has a nationally recognized diabetes prevention program that will work with physicians and providers, and have the patient keep a journal of their diet and exercise. Patients can even be asked to report selections from their journals through a patient portal, as directed by their physician.

After a sufficient data is collected as determined by a data analysis professional, it should be analyzed to determine what progress is being made in achieving the set goals. Using predictive analytics not only can progress be determined but also the most effective strategies or treatments can be identified that lead to the outcomes being measured. For instance, in the prediabetic example it may be determined that the best strategies are having the patient join the YMCA preventive program, report on his/her progress through a patient portal and interact with the nurse coordinator after each of the glucose tests.

Once the predictive analytic results are in the information should be shared with the clinical staff, including physicians. The results should be discussed in a group setting and ways to implement the new findings of the analysis should be discussed. Not all clinicians may be on board to modify the care that they provide but if several are then their progress in providing better care can be shared in future meetings. This will be very useful in motivating other physicians and staff to adopt the recommended changes, especially if the data from the clinicians adopting the recommended changes show success in their prevention work.

The work on providing improved care is not completed after the implementation of strategies shown to be effective by predictive analytics. Data should continue to be collected. Long-term data collection can provide refined strategies that deliver even better results. Plus, research and experience may identify new strategies that can produce even better care outcomes with improved savings for the providers. These new strategies can be implemented and after a period of time data analysis can indicate whether such strategies are effective for the providers. A word of caution, even though research of effective treatment by scientists and providers may show a strategy generally effective, it may not be in a given care setting. The strategy may not match the skill set of the providers, for instance.

In summary, a well-defined program of predictive data analytics and population health management can produce much better outcomes for the patient and the providers. The steps of collecting baseline data, identifying several strategies implemented concurrently, the continuation of data collection, the analysis of the data and the implementation of the best strategies as identified in the analysis should be carefully followed for optimum results. As payers such as Medicare base more of their reimbursement on the achievement of certain population level outcomes it is very important for healthcare systems to adopt the strategies that I have identified.