Network Modeling in Knee Osteoarthritis

Living with chronic musculoskeletal pain (cMSKP) is a complex, uniquely individual, day-to-day experience(1). Although the biopsychosocial nature of chronic pain is well documented, current management of cMSKP is focused on treating specific diagnoses and/or regional symptoms. There is no guidance on interpreting the relative contributions of many different biopsychosocial factors to the burden of pain for each individual. Thus, people living with cMSKP, many with co-occurring conditions, often receive fragmented healthcare, endure inappropriate diagnoses, and undergo expensive and invasive interventions, without long-term improvement.

Our goal in this proposal is to understand modifiable factors that determine inter-individual differences in the pain and functional trajectory of cMSKP. The investigators hypothesize that inter-individual differences emerge from a dynamic interplay of symptoms across multiple domains, including psychosocial, biomechanical and physiological(2). Conventional approaches study these domains separately. The investigators propose to study them as an interconnected system using time-varying network models. A network model(3) consists of nodes (variables representing system components) and edges (pairwise conditional relationships). The investigators will study whether clinically-relevant transitions in pain and functional states are determined by patterns of network connectivity, and analyze time-varying networks to identify critical pathways and personalized pain management strategies.

The investigators will use knee osteoarthritis (KOA) as a case study to test our hypothesis. KOA is a leading cause of pain-related disability(4). There is strong evidence that pain and pain-related functional limitations are not directly related to the severity of knee pathology(5-11) and psychosocial factors are an important contributor(7,10). To understand how psychosocial, biomechanical and physiological factors contribute to inter-individual differences in the lived experience of knee pain, the investigators will collect data from a prospective longitudinal observational study of adults with KOA recruited from community-based physical therapy (PT) clinics. Repeated measures performed every 3 months for a year will include patient narratives, questionnaires, clinical and functional exams, allostatic load measures, and quantitative sensory testing. Ecological momentary assessments will describe pain level, catastrophizing, function, and brief narrative reports for each day (in 5 waves), and pain and biomechanical function for each week. The investigators will develop patient-specific models to assess how pain management (PT for knee pain, or PT after TKA), and co-occuring conditions (knee pain alone or knee pain co-occuring with hip and/or low back pain) impacts individualized experiences of pain. Our specific aims will investigate:

Aim 1: Network Representations of Lived Experience: The investigators will utilize large language models (LLMs) to analyze patient narratives and create patient-specific networks utilizing their own knowledge and insight about their pain experience and overall health. The investigators will (1) Refine a protocol for conducting narrative interviews to solicit standardized, comprehensive narratives; (2) Create an annotation schema to map patient narratives to a structured, semantic representation informed by the International Classification of Function (ICF); (3) Develop and evaluate LLM-based models for automatically generating interpretable, fine-grained representations of patient pain narratives; (4) Validate these representations with clinicians and patients and (5) Connect patient narratives over time to identify change with changing clinical states.

Aim 2: Biomechanical Networks of Physical Function: The investigators will analyze the sit-to-stand (STS) task, a key predictor of function, to develop whole-body biomechanical networks and probabilistic models to determine the relationships along the whole-body kinetic chain. For our primary analysis, participants will record videos of themselves performing the STS task at home following a standardized protocol, which will be analyzed using open-source methods. Our working hypothesis is that clinically relevant transitions in pain and functional state can be predicted from week-to-week changes in task performance. As a secondary analysis to identify promising PT interventions, the investigators will probabilistically generate virtual cohorts of patients performing STS tasks based on the movement patterns from the patient-collected videos and then perturb model parameters to mimic clinically identified deficits/treatment targets (e.g., muscle strength of the gluteus medius). This secondary analysis will generate testable hypotheses for patient-specific PT interventions.