Where it makes sense
A predictive model is worth building when there is a precise clinical question and a dataset large and clean enough to answer it. Many of our telemonitoring projects and disease registries produce exactly that kind of data: repeated measurements over time, collected with the same questionnaire, on the same patients. That is where we work.
Predictive algorithms
We develop algorithms that estimate the course of disease activity indicators from data collected on a platform, such as PROs reported by patients. The aim is to give the clinician one more piece of information, not to replace clinical judgement. Models are therefore built and validated with the doctors who will use them, and the limits of each model are documented.
Chatbots for clinical studies and education
A well-trained chatbot can answer the questions study participants ask most often (visit schedules, how to fill in diaries, whom to contact at the site) and take some of the load off research staff. In education projects it can walk a doctor through a clinical case, answering from content reviewed by the project's scientific committee.
Data mining
On existing archives, such as registries, project databases and published literature, we use data mining to find correlations, patient subgroups and signals worth a closer look. The output is working material for the scientific committee and, when the data supports it, the basis for a publication.
Data and privacy
Before training, data is pseudonymised or anonymised as required by the study protocol and the impact assessment. Processing takes place within our information security management system, certified to ISO/IEC 27001.