We use cookies and other similar technologies (Cookies) to enhance your experience and to provide you with relevant content and ads. By using our website, you are agreeing to the use of Cookies. You can change your settings at any time. Cookie Policy.

Predictive Analytics in Clinical Healthcare

Originally published March 22, 2005

Predictive analytics is the hot topic in business intelligence right now, and for good reason. Since the first commercial transaction took place thousands of years ago, sellers have wanted to predict who will buy, what they will buy, when and how much. Significant amounts of time, energy and money are devoted annually to making these types of predictions across virtually every industry.

The patient encounter is the universal nugget in clinical healthcare, driving every decision and action by this group. What if one could predict the specifics of your patient encounters, and use this information to improve decision-making across the entire organization? The benefits of answering this question to patients, to the organization and to the stakeholders are potentially tremendous.

This article details the specific dimensions of a patient encounter prediction, and more importantly how such predictions can be used to support operational and strategic decisions for greater effectiveness, efficiency, safety and cost-effectiveness in clinical healthcare organizations.

Predictive Analytics Examples

Commercial enterprises of all types expend great effort and resources attempting to predict the future of their businesses and the environmental factors affecting their business. Here are a few examples:

  • Insurance. An insurance company could not exist without an actuarial function, which is essentially prediction of claims in terms of when, why, how much, etc.
  • Retail. Marketers seek to predict consumer behavior in order to present the products with the best likelihood to be bought, in order to make intelligent decisions regarding product, price, promotions, distribution, etc. Some retailers even use models of weather patterns that may affect people’s willingness to go out and shop.
  • Manufacturing. Manufacturing also needs to be able to anticipate consumer behavior in order to predict sales. In addition, this industry often needs to predict the costs of its raw materials in order to succeed. I once worked with a candy bar company who spent 12 years perfecting a model to predict the cost of cocoa for the following year.
  • Telecommunications. The phone company spends a lot of time and money on churn prediction, or the ability to predict when its customers are likely to leave their service.

For the clinical healthcare organization, these examples provide ideas for potential predictive analytics subject areas, but what if the core event and the details surrounding that event could be predicted?

Elements of a Patient Encounter Prediction

A patient encounter isn’t just a simple event like buying a pack of gum. Rather, it entails many details spanning a number of dimensions such as who, what, when, where and why. Prediction of this encounter, therefore, is a complex undertaking. But it is also rich in possibilities for improving both operational and strategic decisions within the organization. Here are just a few of the elements of a patient encounter.

  • Who the patient is as well as his or her health history.
  • What healthcare conditions or changes to those conditions caused the encounter.
  • Where the patient encounter took place.
  • When the encounter took place. This includes the time of day and, more importantly, at what point in time during the patient’s health history.
  • Why the patient had the encounter (emergency, routine visit, hospital admission, etc.)
  • How the encounter took place. Increasingly, remote care such as phone or e-mail is used in place of an in-person encounter.
  • Who provided the care and;
  • How much revenue the encounter produced, who paid for it, what it cost, namely, the inescapable financial questions.

Predicting even one of these elements with enough precision to make intelligent decisions can be tricky. In addition, each of these factors can affect several others, which compounds this complexity.

Driving Operational Decisions with Patient-Encounter Prediction

Two situations drive operational managers crazy. These are tasks without resources and resources without tasks. The first is obviously bad. You have a job to get done but not enough people, information, systems, tools, etc. to get it done. The second is less obvious, but equally annoying. That is having a person standing at your desk fully loaded with information, knowledge, tools, etc. and no tasks, except maybe some low-value filler work.

Predicting your patient encounters can create a single payoff for the operational manager—readiness. By knowing what your tasks are most likely to be (i.e., how many patients, what conditions they have, what brought them in, etc.), you can be ready with the right resources, in the right place, at the right time and in the right quantities.

The benefits of having this level of readiness include:

  • More intelligent staffing and scheduling;
  • Equipment and supplies in the right quantities, available at the right time and in the right place;
  • Information about patients, groups of patients, etc. ready for the care team’s use; and
  • Training and development for the right providers, in the right place, at the right time.

The quality of service to the organization and the patients, as well as the efficiency and cost-effectiveness of that service can be improved simultaneously with patient encounter prediction.

Supporting Strategic Decisions with Patient Encounter Prediction

It is estimated that up to 30 percent or $510 billion of the $1.7 trillion U.S. healthcare industry is wasted effort and cost. Industrial firms have found that the best way to avoid waste is to focus the provider’s resources on the right activities. Rolling up the provider’s patient encounter predictions helps find this focus. Once found, this information can be used to support the following key strategic decisions, among others:

  • Revenue prediction. Providing visibility to the sources and timing of revenue from patients, health plans, employer plans, pay-for-performance programs, etc.
  • Cost prediction. Providing the ability to place the right resources in the right place at the right time, and uncover where wasteful expenditures are being made.
  • Investment decision support. Providing evidence on which to base decisions such as type and location of facilities, type and amount of equipment, etc.
  • Knowledge and skills management. Understanding the characteristics of the anticipated patient population supports recruiting decisions, staff development decisions, training program content, etc.
  • Marketing approach determination. How to “go to market” is determined by what the “market” expects. For example, in one community where I have worked, one large integrated healthcare organization focused on providing the best access to healthcare services, while another focused on having the latest medical technology and techniques.

Driving performance throughout the organization in a consistent manner requires a common base of evidence regarding where the provider organization is headed. Predicting patient makeup helps align this information from top to bottom.

Requirements of Patient Encounter Prediction

Predictive analysis capabilities are not easy or cheap to develop to obtain decent results to use in making the kinds of decisions I have described. The something-for-nothing myth is just that, a myth. Three key requirements are necessary to deliver these capabilities:

  • Differentiating Predictive Analytics from Proactive Analytics. The difference between these two is the degree of control one has over the outcome of the decision. Predictive analytics refers to situations that are for the most part often external and therefore not within the organization’s control, such as revenue trends, propensity of payers to pay, propensity of patients to follow their doctor’s advice, etc. Proactive analytics refers to situations where there is a high degree of control over the outcome. An internal decision such as purchasing supplies falls into this category. One example that illustrates this difference is the reason for a patient visit. If it is the first encounter with this patient, then the visit could be predicted using statistical evidence. If it is a follow-up visit, then it was the result of proactive action by the providers and patient. Before investing a dime in predictive analytics technology, please be sure the organization and its vendors can differentiate between these two situations.
  • Sufficiently Deep Evidence Base. The organization needs a repository of evidence about patient encounters, as well as potentially from external sources in order to make reliable predictions about patient behavior. A deep repository provides the data points to make deep, meaningful predictions. Shallow evidence means shallow predictions. Expecting a decent prediction from a handful of transactions and applying some pixie dust is a recipe for wasting a lot of money both now and in the future. Plan to base predictions on a sufficiently deep base of statistical evidence, just as would be required for a medical study.
  • Statistical Tools. The most visible components of predictive analytics are the statistical tools and algorithms used to generate the prediction. This is the sexiest, most magical part. Many of the truly serious business intelligence tool vendors offer predictive analytics functions as well as the ability to assemble highly sophisticated predictive models. But before applying these functions, algorithms and models, the base of evidence described above is needed.

Next Steps

The promise of predictive analytics to clinical healthcare is great, but so are the potential pitfalls. The next step in pursuing these capabilities is to begin to organize the data already owned in order to make predictive analytics possible. If done wisely, everyone in the organization can benefit greatly at both an operational as well as a strategic level. One way to focus efforts is to concentrate on a core question such as predicting patient encounters.

Thanks for reading. I look forward to your comments.

  • Scott WanlessScott Wanless

    Scott is the Healthcare Analytics Director for Cipe Consulting Group. He has more than 30 years of experience in business intelligence strategic planning, analytics application development and business analysis across numerous industries including hospitals, physician groups, healthcare payers, laboratory research, insurance, lending, manufacturing, retail and state government. Scott can be reached at scott.wanless@cipeconsulting.com.

    Editor's note: More healthcare articles, resources, news and events are available in the BeyeNETWORK's Healthcare Channel featuring Scott Wanless and Laura Madsen.

Recent articles by Scott Wanless



Want to post a comment? Login or become a member today!

Be the first to comment!