|
Organizations must constantly adapt to rapidly changing business requirements
and increasing cost pressure. They need to optimize their business processes to
allocate scarce resources more efficiently. In today's businesses, many processes
are complex. Accurate data for optimization efforts is sometimes lacking, and
potential investments can be highly risky. Customers need our assistance to optimize
their processes and make quantitative simulations and predictions. Using our skills
in predictive modeling, stochastic optimization, portfolio theory, and simulation,
we provide our customers with highly customized solutions which address their
business needs.
|
|
|
When assessing the information needs of physicians in a clinical environment,
researchers have noticed that questions arise for two out of three patients, the
most frequent question being how to manage a patient who has a given disease.
This need for information is not surprising. The uncertainty in dealing with patients,
predicting how they will react to treatments, which side effects might occur,
etc., is not easy to manage using intuition alone, even for experienced physicians.
As a matter of fact, a study has shown that basic vocabulary to express uncertainty,
such as "likely", "highly probable", etc., has a wide range
of interpretation for individual physicians. The meaning of "likely",
for instance, can range from not probable (around 20% of occurrences) to highly
probable (around 95% of occurrences), depending on how an individual physician
defines it.
Medicine was one of the first fields to embrace computer science, artificial
intelligence and statistics to provide physicians with quantitative figures
to facilitate judicious decision-making. The first systems were aimed at diagnosis.
MYCIN, the Internist-1 and QMR are all well-known computer programs — also
called expert systems — that return a list of potential diseases from a
list of observed symptoms.
In this project, we focus on a different approach. Rather than taking only
the observations made at the time of diagnosis, we are designing systems that
draw on the patient's entire history. By taking into account the temporal nature
of the data and the sequential nature of treatments, an appropriate treatment
to improve the patient's state of health can be derived and returned to the
treating physicians. Applying their intuition to the computer output, physicians
can then delve deeper into the appropriate databases.
Treatment recommendation is only one aspect of the benefits of looking at temporal
trends of individual patients. Another aspect is the detection of abnormal behaviors.
Vital signs taken at an arbitrary point in time provide no context about what
is normal for that patient. For example, patients who formerly had high blood
pressure would exhibit abnormal readings as their blood pressure starts to normalize.
By looking at vital signs over a certain period of time, however, "normality"
becomes definable. It means stability, whereas abnormal behavior means unexpected
change.
|
|
