Solving Problems in Industry - Java Data Mining: Strategy, Standard, and Practice - page 42

To realize these benefits requires a certain infrastructure, one that
is able to collect electronically meaningful data about patients, their
treatments, and healthcare providers such as labs, doctors, and
nurses. Hospitals and other care providers are slowly moving into
the digital age. For those who have, they can use this data to identify
the important factors that determine a patient's likelihood of
responding to a particular treatment. Using the association tech-
nique, providers can determine probable causes of a patient's death
that may be a result of external factors such as staffing shortages,
prescription mix-ups, or process flaws.
As for financial institutions, fraud is a major concern for health-
care insurance companies, specifically in the area of billing fraud
[SPSS 2003a]. Being able to identify which providers are likely billing
incorrectly and by how much is an area where data mining can be
readily applied. A fairly common technique is to cluster healthcare
providers to identify those with unusual billing patterns. There may
be a few clusters of providers that warrant investigation since they
are outside the norm of other providers. There may be individual
providers that do not fit any particular cluster and warrant further
investigation. Association can be used to identify uncommon
relationships expressed as rules that appear with low support, that is,
few cases exhibit that behavior. Using regression techniques, an
expected number of claims or monetary value of claims for each
provider can be predicted. By comparing the actual number of claims
or monetary value to the predicted, those providers who signifi-
cantly differ from the predicted value are identified as candidates for
investigation.
Still other areas for healthcare fraud involve doctors prescribing
unneeded medications, perhaps due to receiving kickbacks from
drug companies. Developing predictive models on patients who
legitimately use certain drugs can yield profiles of such patients.
When a drug is prescribed, the profile of the patient can be matched
against known profiles for legitimate users of the drug. Mismatches
are candidates for investigation.

2.2.3 Higher Education

Higher education institutions such as colleges and universities are
faced with a growing number of concerns, including reducing opera-
tions costs and the resultant tuition increases, attracting students
who will flourish in their respective environments and complete a
degree program, and increasing alumni donations.