The traditional use of cancer registries is to provide data for medical research, for administrative or political decisions, or as insight into the quality of the health services provided. Such data could also improve decision making in clinical practice if a patient´s profile can be compared to similar patients in the registry, exploring outcomes for different treatment options.
In this
project we examine whether the Norwegian Clinical Colorectal Cancer Registry
(NCCCR) could be used to give clinicians real-time clinical decision support by
providing statistics based on an individual patient’s characteristics and the
matching sub-population in the registry.
We
developed a prototype IT-system which matches a patient in real-time to a
similar patient cohort in NCCCR. Here, R is used for the analytics and RShiny
for visualisations. The system capabilities are:
- Capturing the following relevant clinical
parameters from patient medical notes; diagnostics, potential treatment
options, and related outcomes measure,
- relay
parameters to NCCCR,
- select relevant sub-population, and
predict outcomes, and
- return results to electronic patient
journal. By using all patients registered in NCCCR from 2004 to 2017, observed
survival was calculated by the Kaplan-Meier method.
To ensure clinical value, a small
group of clinicians and hospital leaders chose the relevant clinical parameters
and guided the visual presentation of the results.
The first
version of the current IT system was able to capture the relevant
characteristics, communicate with the cancer register, perform the necessary
analysis and return statistics for overall survival for a sub-population that
was similar to an actual patient. Tested on different patients with varying
profiles, the system returned relevant statistics that varied depending on the
patient. Predicted survival for the relevant patient profile was presented by
visualizing outcomes by Kaplan-Meier plots for different treatments. Clinical
specialists and hospital leaders were presented the visualization and have
subjectively verified the potential value of the information.
The next
iteration of the system will extend the number of features for clinical
parameters, treatment options and related outcomes. This is a novel use of
register information, and implementation of the described system in the clinic
will need clinical testing in order to prove the value for patient outcome and
patient experience.