Successful PhD defense by Marit Dagny Kristine Jenssen

Congratulations to Marit Dagny Kristine Jenssen, who successfully defended her PhD thesis at UiT The Arctic University of Norway on April 10th

By Petter Bjørklund, Communications Officer at SFI Visual Intelligence

Jenssen is a PhD Research Fellow at Visual Intelligence and the UiT Machine Learning Group.

Her thesis, titled "Exploring patient generated health data in fibromyalgia: Statistical and machine learning approaches to chronic pain and physical activity",  combines machine learning, time series analysis, and clinical trial methodology to advance the understanding and treatment of fibromyalgia.

Jenssen's trial lecture was titled "Foundation Models for Time Series: Concepts, Methods, and Applications".

Summary of the thesis

In Norway, chronic musculoskeletal pain is a leading cause of disability benefits and sick leave. Fibromyalgia is a common chronic pain condition characterized by widespread pain, chronic fatigue, poor sleep, and depression. Despite extensive research, no efficient treatment has been developed. Several studies suggest that physical activity can help manage chronic pain, but interventions should be tailored to the individual.

This dissertation combines machine learning, time series analysis, and clinical trial methodology to advance the understanding and treatment of fibromyalgia. The work is organized around four research questions addressing the use of machine learning in chronic pain research, the predictability of pain from wearable data, the replicability of activity-based pain reduction, and the design of a factorial clinical trial.

The scoping review (Paper I) identifies opportunities for using machine learning to improve chronic pain management, finding that while diagnostic classification is well-studied, treatment and self-management applications remain underexplored. The case study (Paper II) applies SiZer, a statistical method for detecting significant changes, to analyze pain and activity patterns in seven fibromyalgia patients using Fitbit data. Results showed that three patients achieved pain reduction following physiotherapist-guided activity modifications.

Building on this foundation, a time series analysis investigates whether daily pain can be predicted from wearable-derived activity and sleep data. Both classical methods (ARIMA) and modern machine learning approaches (XGBoost, foundation models) were evaluated. The analysis reveals that pain exhibits strong temporal autocorrelation, and activity and sleep features provide only modest improvements for some patients. Predicting pain change direction showed more consistent gains over baseline than regression on exact levels, though overall prediction performance was modest.

Finally, the insights from Papers I and II informed the design of a randomized controlled trial (Paper III). The protocol applies a 2 × 2 factorial design to evaluate the effects of group exercise and somatic tracking—a psychological i ii intervention—on fibromyalgia outcomes.

Together, these contributions demonstrate the potential of combining wearable technology, advanced analytics, and clinical expertise to develop more personalized approaches to chronic pain management.

Supervisory Committee

• Professor Fred Godtliebsen, Institute for Mathematics and Statistics, UiT (main supervisor)
• Professor Johan Gustav Bellika, National Centre for E-Health Research
• Dr. Andrius Budrionis, National Centre for E-Health Research
• Senior researcher Phuong Ngo, National Centre for E-Health Research
• Researcher Miguel Tejedor, National Centre for E-Health Research

Evaluation Committee

• 1st Opponent: Associate Professor Lucia Sacchi, Universitetet i Pavia, Italia
• 2nd Opponent: Professor Kerstin Bach, NTNU
• Internal member and leader of the committee: Associate professor Kristoffer Wickstrøm, IFT, UiT