Healthcare Support beyond Record Keeping
Dieter Gawlick, Oracle Corporation

The use of electronic healthcare records should be pervasive by
now, but it is not. While there is plenty of healthcare software,
the focus is not on the patient's health but rather on administering
and charging for procedures that have been performed. Most likely,
electronic healthcare systems will not be widely deployed until
they become effective in providing critical information in a timely
fashion to doctors.

A traditional database stores information to be accessed by (transient)
queries. In an ICU (intensive Care Unit) environment this is not
sufficient. As patient data is gathered, critical conditions will
arise; however, using the traditional data access model nobody will
notice until someone looks at the data in the right way. So we need
the ability to register (lots of) queries, execute them continuously,
and have the database notify doctors if any of the query results
indicate a critical situation. Writing these queries (think about
them as rules) manually is sometimes easy, sometimes difficult and
often impossible. If writing queries manually is impossible, data
mining and machine learning can be used to develop and improve
(predictive) models, which represent diseases or critical situations.
Once developed, these models must be permanently scored and those
with scores high enough to indicate a critical situation must be
brought to the attention of the doctors. Developing the models and
scoring large numbers of these models (tens of thousands) for
hundreds of new data elements per second is a very challenging task.

We - Oracle in cooperation with the University of Utah (USA) and
the University of Coimbra (Portugal) - have developed a prototype
as a proof of concept. The prototype is mainly the effort of Diogo
Guerra (Coimbra) and Ute Gawlick (Salt Lake City), supported by a
team of experts from Oracle. Diogo and Ute had to overcome many
obstacles with ingenuity and hard work.

The prototype consists of a single data repository which combines
a highly configurable rule-based system; (push-based) alerts, data
mining models and enhanced user interface. Everything is highly
customizable to the preferences of doctors and the specific
circumstances of patients. The prototype is able to predict life
threatening events (e.g., cardiac arrest) before they happen. This
approach has the potential to significantly increase the value of
healthcare information as the database can store and analyze health
care records continuously, find critical situations, explain why
they are critical, allow for detailed investigation and provide
recommendations -- even for diseases or critical situations that a
specific doctor has never encountered or does not even know about.
A side effect of this approach is that the number of alerts and
additionally the number of false positive and negative alerts is
significantly reduced; with it the alert fatigue significantly
reduced. We contend that this approach will save lives and improve
health care not only in ICU settings by also for any inpatient and
outpatient services. The lessons learned should be applicable to
other domains as well.

A wide array of database technologies has been used in an integrated
way: OTLP, (transaction) temporal, OLAP, data mining, continuous
queries, and CEP (Complex Event Processing) technology.

This prototype illustrates a new style of rule driven applications that
provides timely access to critical information and that can handle enormous
complexity while remaining highly customizable and extensible.  Classical
streams/CEP technology is neither able to provide the required functionality
nor has the necessary operational characteristics.