Patient Record as Graph
ES, version 2022-05-02 /
- Addressed problems:
- Traditional narrative patient record would be very difficult for
decision support. Many traditional records contain many fact, but
relatively few up-to-date synthesis.
- Make patient records suitable for decision support.
- Make the patient record easier to share with colleagues.
- Associated nodes:
- At one side a node containing the actual value observed on
a patient at a given time, e.g a blood pressure measured
today. This may be measured any number of times.
- At the other side a unique node containing the permanent
knowledge about this type of observation as explained in the
knowledge base, e.g. expected normal values of blood
pressure and if out of range relations to potential
- The relations between the patient and all his/her instances of
concepts may be qualified. For example mild or severe
pain, occasional or frequent crisis.
- The most important concept. Any concern needing to pay
attention and to seek a solution. When more precision become
available a problem may become a diagnose.
- Problems relations:
- Symptoms leading to likely problems
- Problems to recommended actions.
- Typical scenario:
- Initially the patient is coming with relatively few
information, as the reason for encounter and the most important
- This lead to the identification of suspected problems.
- Suspected problems lead to the search of more information, to
more specific questions, maybe to lab tests, maybe to request
for radiologic images, etc... answer to questions lead to more
new related questions.
- Eventually the most likely problems will lead to
recommendations for treatments.
- Patient records are always incomplete:
- It would never be possible to get all the observable
parameters. Beside a set of minimal checkup, the content of the
patient record will depend on the particular situation.
- Evolution in time:
- All items have a date and time. Important for search and for
- Notion of "decay". For example an observation made 5 days ago
has more significance than an other made 5 years ago.
- When the understanding of problems evolve, new versions must
be created, but the previous versions must be archived and
remain available on request.
- Weights are important with multiple factors as severity,
- Care team collaborations:
- Multiple healthcare professional are often necessary, acting
from different specialized point of view.
- It is important that all the actors share a common view on the
central problem list. From here they may navigate ito details
related to their specific missions.
- Human interface:
- An appropriate human interface is essential for active
participation of healthcare people, doctors and nurses. A very
critical issue for the acceptance of this new system. Of course
motivation of this new way of working will be the quality of the
- Since the goal is to have patient information as a graph, the
challenge is here to let the users draw graph directly, reducing
the need of Natural Language Processing.
- Basic example:
- Starting from a patient having 4 symptoms, evaluation of
- In next versions attributes as likelihood, severity, etc ... will
have more visual styles like shape, size, color, thickness, etc...
- Up to now:
- Begin with a few typical patient records converted as graph
and discussion about recommendations. Discussion of the
necessary steering information in the knowledge graph.
- Improvement of the model.
- Working group documents:
- Contact person(s):