Rheumatology Data Science

Horizon-EU SPIDERR

Stratification of Patients using advanced Integrative modelling of Data Routinely acquired for diagnosing Rheumatic complaints (SPIDERR)

SPIDERR is a project committed to targeted healthcare and disease prevention for patients with rheumatic diseases. Over 100 million people in the EU suffer from musculoskeletal symptoms leading to disability. However, the journey to effective treatment choices is long and inefficient.
SPIDeRR's approach goes beyond the state-of-the-art practice to change diagnosis, therapy prediction, and eventually even prevention.

SPIDeRR is funded by the European Union under grant agreement 101080711. For more information and news, please visit our website. 

Mining Electronic Health Record data

We created a machine learning pipeline that equips centers to build their own classifier e.g. to extract RA-diagnosis from format-free physician notes in the EHR. Due to the probabilistic nature of ML, users can tailor the case selection in favor of precision or sensitivity. Enabling centers to generate large retrospective cohorts for downstream observational studies within mere seconds.

Pattern recognition on Electronic Health Record data

Clustering on longitudinal data  - PheSpecs

Rheumatoid arthritis patients vary greatly in their prognosis, comorbidities and treatment response. We believe that these differences can, at least in part, be attributed to underlying differences in disease aetiology. By investigating the medical history of a large number of patients we mean to identify meaningful clusters of patients who all exhibit similar patterns leading up to their RA diagnosis. By creating more homogenous patient selections we will enable research into group-specific disease triggers and treatment options, thus taking important steps towards personalised medicine.

Clustering on baseline disease characteristics using integrating multi-modal data

We capitalize on the wealth of information from the EHR to disentangle the clinical heterogeneity of RA by identifying subgroups at baseline. Currently, there is a lack of studies that take full advantage of the EHR, mainly due to its multi-modal nature. Hence, we adopt powerful unsupervised techniques to uncover hidden structures in the patient dossiers, which may inform patient specific therapeutic choices in the future.

Identification of disease trajectories through clustering of Rheumatoid arthritis patients:

With the use of clinical data, we hope to identify different trajectories in the early stages of RA.

To achieve this goal we use multiple methods such as community detection to find disease stages and use the clinical data to identify the trajectories.

Next, we hope to link the communities and trajectories to a clinical and biological meaning.

Optimizing health care with eHealth tools for patients

Rheumatic?

Within the JPAST project we developed, together with Elsa Science, Karolinska Instituted and Erlangen university the eHealth tool Rheumatic?. This tool has attracted the attention of ReumaNederland, which led to a close collaboration where we now recruited 17,000 participants for a Dutch validation study.

Algorithm development

In collaboration with the systemic sclerosis (SSc) group of our department, we have developed a model capable of identifying patients with a low risk of disease progression. Our model will aid both patients in reducing excessive care, as well as the SSc carepath in determining where to best spend their resources. This project showcases the full trajectory of employing AI/ML to solve clinical questions; from project conception, through model development all the way to a clinical validation study currently underway.

G-Prob

To optimize knowledge utilization, we made the G-PROB tool publicly available, so everybody can freely use it. I built the tool such that it can be used for any clinical situation where patients present with similar symptoms.

Identification of RA patients in format-free text fields with Machine Learning 

In this study, we developed a pipeline that equips centers with a classifier that extracts the RA-diagnosis from format-free physician notes in the EHR. Due to the probablistic nature of ML, users can tailor the case selection in favor of precision or sensitivity. Enabling centers to generate large retrospective cohorts for downstream observational studies within mere seconds.

Genetic association studies

 Most of our genetic work is integrated in the previous projects, where we use genetics to explore whether the results of our data-mining could relate to any biological functions.

In addition, we contribute to/run genome wide association studies and sometimes run a candidate study upon request.

GWAS to remission – RTCure

In collaboration with a large number of European rheumatology groups we are looking into the genetic differences that play a role in who will and who won’t be able to reach early disease remission in the rheumatoid arthritis population. With this information we hope to uncover new pathways for treatment to improve disease prognosis for those lacking the favourable genetic background.

Squeeze

The SQUEEZE project will help with the optimization of existing treatments for Rheumatoid Arthritis (RA) by exploring the latest developments in the field of biomarkers, focusing on the effect and safety of medicines and analysing the best implementation science to enhance the therapeutic relationship.