Enabling precision medicine in multiple sclerosis
Multiple sclerosis (MS) is a chronic neurological disease that affects the brain and spinal cord. It is an autoimmune disease where the immune system attacks the body's tissues. In Sweden, MS is the most common cause of mobility impairment in young people, and each year approximately one thousand people become ill with MS. Today, there are a total of around 20,000 people with MS in Sweden. Most people who fall sick are between 10–60 years old (usually 20–40 years), and women are twice as affected as men. Untreated, the disease leads to severe disability and premature death.
Project description
The cause of MS is largely unknown, but epidemiological and genetic studies indicate that MS is triggered in genetically susceptible individuals following exposure to lifestyle and environmental factors. In most cases, the disease begins in so-called relapsing-remitting MS. The disease comes in periods of symptoms called relapses; the relapses can then go back in whole or in part. After about 10 to 15 years with the disease, the disease becomes more progressive, which is referred to as secondary progressive MS (SPMS). It is not possible to cure MS, but today, several drugs can slow down the course and relieve the symptoms, especially for patients with RRMS. In some cases, so-called hematopoietic stem cell transplantation may be relevant as the medicines do not have an effect.
Today, there are no measurable biomarkers that can be diagnose SPMS early or that can predict the development of the disease. We want to change that. We develop highly accurate predictive models using machine learning and AI to distinguish between RRMS and SPMS, trained on data from electronic health records (EHR) collected at routine hospital visits and complemented with novel biomarkers. Proactive recognition of patients with progressive disease could limit exposure to ineffective medications and their side effects.
Valid measures of uncertainty at each patient level
To be helpful within a clinical setting, we apply conformal prediction (CP) to deliver valid measures of uncertainty in predictions at the level of the individual patient. Early identification of patients who eventually fulfill the criteria of SPMS would be a valuable addition to the armamentarium of clinical practitioners, enabling meaningful intervention.
We also propose how these predictors could be used to monitor a patient’s disability trajectories in the spectrum between RRMS and SPMS and for early prediction of SPMS. Also, adding molecular biomarkers to an existing EHR-based AI model with CP could open up the door for producing forecasting in MS and earlier meaningful interventions.
Significant general interest
The initial stages of this project have aldready generated significant interest withn the research community following Kim Kultima's presentation at MSMilan 2023. Our ongoing work was selected as one of the highlights of the conference. To showcase our approach, we have also launched an anonymized and accessible version of the model MSP-Tracker.
Funding
The project is funded by group leader Kim Kultima, receiving a grant from the Swedish Research Council (VR) Medicine and Health, NEURO Sweden, and Åke Wiberg Foundation.