From Isolated Health Events to Disease Trajectories: Using Network Models to Understand the Path Toward Multiple Sclerosis
A diagnosis may appear as a single event in a medical record, but the path leading to it is rarely a single event. It is often shaped by years of symptoms, healthcare visits, diagnoses, investigations, and treatments. When these events are analysed separately, much of their temporal and clinical context is lost. Network modelling offers another perspective: it connects these events and represents disease as an evolving trajectory rather than a collection of isolated observations.
This is particularly important for complex diseases such as chronic inflammatory diseases such as multiple sclerosis (MS), where early symptoms including fatigue, pain, anxiety, or sensory disturbances are common, non-specific and rarely informative on their own. Conventional analyses typically examine whether each condition on its own occurs more frequently among future MS patients. Network modelling asks broader questions: Which conditions occur together? In what order? Which diagnoses form disease communities (known as clusters), become central hubs and essential for MS development, or connect otherwise separate clinical domains? A diagnosis may become important not because it occurs more frequently, but because its relationships with other conditions change over time. By capturing these connections, communities, and patterns of network rewiring, we can examine disease development as a changing system rather than as a list of independent risk factors.
In our recently published nationwide cohort study, at the first step, we reconstructed longitudinal disease networks from Swedish registry data across the 15 years preceding MS onset (the first time MS manifested itself clinically). The analysis showed that at the time before MS, there is a gradual reorganisation of common diagnoses into distinct multisystem patterns spanning psychiatric, metabolic, immune-related, neurological, and general symptom domains (Figure 1). Building on these findings, the second step was to develop a heterogeneous, multilayer temporal clinical graph connecting patients, visits, diagnoses, and treatments in chronological order. Covering 56,548 individuals, the graph contained more than 1.19 million nodes and 4.97 million connections, with additional layers translating fragmented medical codes into clinical concepts and identifying recurrent patterns associated with MS trajectories (Figure 2).
Figure 1. The 50 most connected diagnoses in the MS network (left) and control network (right). Nodes represent ICD-10 diagnoses, while directed edges capture their temporal sequence. Source: Ebrahimi et al. (2026), Multiple Sclerosis Journal.
Figure 2. Example of the patient-level layer of the heterogeneous temporal clinical graph. The graph connects the patient to chronologically ordered healthcare visits and links each visit to recorded diagnoses (ICD) and treatments (ATC).
This project demonstrate the value of network modelling at two complementary levels: revealing how disease patterns reorganise across a population when the onset of MS approaches, while preserving the structure of individual patient journeys. Because graph-based patterns can be traced back to the underlying visits, diagnoses, treatments, and temporal paths, they also offer a more interpretable foundation for AI than purely black-box predictions. As part of WISDOM, this work provides a basis for developing and validating trajectory-aware methods that may eventually support earlier recognition and a better understanding of complex diseases such as MS.
Ali Ebrahimi, Ali Manouchehrinia, Narsis Kiani