Editor’s note: Nancy V. Paddison is a marketing communications specialist at Customer Potential Management Corporation, East Peoria, Ill.

Getting to know customers’ preferences, needs and likelihood to use a company’s products is the starting goal for customer relationship management (CRM). On the basis of customer data analysis, companies continue or modify existing products and develop new ones. To solidify one-to-one relationships with their customers, companies are relying on data analysis for relevant customer intelligence that can lead to a competitive edge, increased loyalty and lifetime value.

In the health care market, which struggles consistently with financial strain and little product differentiation, CRM solutions can help providers to be more targeted and efficient in reaching the right patients with the right information to improve wellness, approach disease more proactively, or to lessen the physical and financial burden of chronic illness.

The success of such a solution hinges on information - obtaining it, analyzing it, acting on it and measuring it to achieve stated goals and objectives. But there has to be a top-down commitment to the power and correct use of such data. In a survey conducted by the META Group, only 29 percent of 800 business and information technology executives polled said their company was adequately using customer data, while 67 percent disagreed. In addition, while 92 percent of respondents said achieving customer intimacy is a priority, four out of five individuals said “No” when asked, “Does your company know who its customers are?”

Despite the fact that health care traditionally lags behind other industries in adopting technology, there are some encouraging signs that data analysis, segmentation and prediction are making inroads and a difference.

Cluster codes and beyond

Traditionally health care used mass direct marketing techniques to reach people in their service areas, but as the costs continued to rise and responses fell, they looked for a better solution. The first improvement in target marketing, or the beginning of customer relationship management (CRM), was segmentation by geography. Planners looked at where customers were buying and focused on those areas, omitting unproductive regions.

Next, cluster codes enabled planners to take a large data set and divide it into smaller groups or “clusters” based on similarity. Geographical areas were thus defined by smaller groups containing similar people or demographic attributes such as age, marital and economic status.

“The idea behind all demographic cluster systems is the same. Each system divides neighborhoods into groups based on similarities in income, education and household type, as well as attributes and product preferences.”1 In the absence of any individual information, geodemographic clusters are a reasonable solution. Cluster coding thus became a standard by which many health care organizations sought to reach households and even individuals. The rationale was “we know who you are because we know where you live.”2 However, information to form demographic clusters comes from data collected by the Census Bureau only once every 10 years about averages of age, income and other variables.

All of these facts are collected and summarized at the census tract level. Unfortunately, any individual uniqueness is lost during geographic aggregation and summarization. Cluster systems are based on geographic similarities and averages, rather than individual attributes and differences.

The fatal flaw of cluster systems in health care is that the variables used to design the clusters have nothing to do with health care. Health care strategists could not use such cluster codes alone to develop long-range and strategic plans about specific health care services and products needed by individuals. Heart disease doesn’t discriminate by cluster group or neighborhood. Neither does breast cancer, diabetes or even birth defects.

To move beyond the generalization of clusters and information such as sex, zip code, income and geography, more predictive models that can segment individuals on the basis of health care variables and status have been developed.

Technology has come a long way in the 20 years since clusters were conceived. Today we have sophisticated databases, data mining, neural networks and statistical capabilities that allow health care organizations to seek smaller and more precise bits of information to forecast health care needs. Data mining techniques have been used to build predictive models for health care that can identify who will most likely need certain services and who will most likely fall ill. Such capabilities actually expand CRM beyond marketing and information provision to better disease prediction, service utilization planning and cost efficiencies.

The ultimate goal is wellness and improved health by drawing a more complete picture of patient needs and then offering programs and services necessary to help change their lifestyle and health care management. Providers hope that this ongoing responsive and more pinpointed approach will help establish better patient relationships and increase patient loyalty. The chart below shows the critical differences between predictive segmentation and cluster codes.

Chart 1

Predicting who needs care

One program, the Consumer Healthcare Utilization Index (CHUI), provides a health risk profile similar to the way a credit rating calculates an individual’s credit risk. The profile is based on one of three levels: major diagnostic categories (MDC), such as pregnancy and childbirth and the circulatory system; the top 100 identified medical service areas of the 21st century such as weight management, diabetes and cardiovascular disease; and diagnosis related groups (DRG) such as cardiovascular medicine and urology.

The index provides a number between 0-999 that indicates an individual’s propensity to use specific health care services. The higher the score, the greater the likelihood the individual would need that service in the category selected. The scores are gathered through an algorithm based on the results of data mining research. Using that empirical model, scores for patients and prospects are calculated from enhancement data that is overlaid onto the health care organization’s database for data analysis and use with communication management tools.

This index can forecast the need for certain health care services throughout the lifecycle. What one person requires today will change over time with age, and probably differ from those of a spouse, friend, or co-worker.

A post-campaign analysis shows how predictive modeling might have changed the approach and results of one CRM effort. A Midwestern medical center, part of a larger health system, did a direct mailing of 6,887 postcards to females ages 41 to 55 with incomes greater than $25,000 in a selected geographic region. The postcard was part of a monthly women’s educational series designed to provide health information and increase use of the hospital’s cardiology services.

In terms of revenue received and new patients, the campaign had good results. In the first six months, 14 new cardiology patients resulted from the mailing, with 16 services required by those 14 patients. The ratio of received to spent dollars was $126 to $1. But did it reach the right individuals? Could it have been more effective?

When MDC 05 Cardiology (diseases and disorders of the circulatory system category) was appended to records used for the initial mailing, a better picture of which women are most likely to require cardiology services based on their health risk emerged.

The numbers showed that:

  • All women who responded to the promotion had scores of 450 or higher, which indicates a moderate risk for cardiac disease and therefore, need for services.
  • The higher the predictive score, the greater the use of services and revenue generated.

The medical center then scored all women in its entire database to see how many women with scores of 450 and higher compared with the initial list segmented by age and income. Here’s what they found:

  • The number of women in the overall database with MDC 05 scores of 450 and higher was 29,859, but only 5,146 females with these scores received the initial mailing. Therefore, more than 24,000 appropriate individuals - those who could benefit most from this information, as identified by the predictive model - did not receive the mailing.3

Beyond better targeting for health information and education campaigns, predictive modeling is making it into the disease management market. With more than half of health care costs attributable to approximately one-fifth of the population with a chronic medical condition such as asthma, heart disease or diabetes, effective targeting and prediction about this minority could save billions in health care costs each year. More importantly, it could motivate and help people with chronic diseases avoid health care crises.

Al Lewis, president of the Disease Management Association of America, and executive director of the Disease Management Purchasing Consortium, says the sales of disease management software and services approached $390 million in 2000. He predicts the number will grow by about 30 percent annually for the next two to three years. While still a small percentage for health care, Lewis notes that it’s a huge increase over just a few years earlier.4

For example in California, the parent company of a large insurer is partnering with a health care company to offer a disease management program to plan members who have cardiovascular disease. The goal is to stratify people based on medical risk and their ability to change.

Even small compliance rates, for example with a medication plan, can be impressive. In Texas, the cost of hospitalization for congestive heart failure can be about $10,000 a year, but the cost of treatment with a specific enzyme inhibitor runs between $300 to $400.5

In the Midwest, a Medicare managed-care organization used predictive modeling in combination with the identity of a wide range of contributing factors — such as stress emotions, function level, treatment compliance and personal beliefs — to predict near-term health care consumption of 4,632 of its risk members.

After 12 months, the predictive algorithm was able to identify with 92.5 percent accuracy members who incurred more than $25,000 in health care costs. In combination with stratification from the identifying factors, the organization realized a 50 percent reduction in costs per member, per month, and a 38 percent savings over the cost of traditional managed care.6

Predictive segmentation, which allows organizations to focus more time and resources on the most relevant issues of its market, is a prudent departure from cluster code methods designed to reach groups solely on the basis of geography or other attributes unrelated to health. As part of a total customer relationship management solution, this enhanced customer intelligence can result in better patient health and loyalty, retention and lifetime value.

References

1 Mitchell, Susan, “Birds of a Feather,” American Demographics, 1995.

2 Harris, Rich, Adding a Qualitative Depth to Quantitative Models? An Introduction to Geodemographics and Lifestyles, School of Geographical Sciences, University of Bristol, England.

3 CPM Corporation, Predictive Segmentation, Increasing the Effectiveness of Healthcare Market Segmentation and Communications, 2000.

4 Baldwin, Fred, “Disease Management: Technology That Helps Patients Manage Their Chronic Disease is Proving its Worth,” Healthcare Informatics, February 2001.

5 Ibid.

6Meek, Julie, “Rein in Costs: Predictive Modeling Pinpoints Populations Most Likely to Require High-Cost Care,” Health Management Technology, February 2001.