Network analysis of multimorbidity and health outcomes among persons with spinal cord injury in Canada

Nader Fallah, Heather A. Hong, Di Wang, Suzanne Humphreys, Jessica Parsons, Kristen Walden, John Street, Raphaële Charest-Morin, Christiana L. Cheng, Candice J. Cheung, Vanessa K. NoonanFrontiers in Neurology 2024 · DOI

TL;DR

Researchers utilized network analysis to map the complex interactions between 30 secondary health conditions in Canadians living with spinal cord injuries (SCI). The study identified Autonomic Dysreflexia as a critical "hub" in the health network and successfully streamlined a clinical screening tool from 30 items down to 25 (the MMI-25) without losing predictive power for quality of life or healthcare utilization.

What Problem Does This Paper Address?

Living with a spinal cord injury (SCI) is rarely just about the injury itself; it involves "multimorbidity," or the coexistence of multiple chronic secondary health conditions (SHCs) like pressure ulcers, chronic pain, and depression. Traditional medical research often examines these conditions in isolation using standard regression. However, health is a complex system where one condition can trigger or exacerbate others.

The authors sought to move beyond one-to-one correlations by using network analysis to visualize the "web" of these conditions. Their goal was twofold: to identify which conditions are the most influential drivers of poor health outcomes and to refine the existing 30-item Multimorbidity Index (MMI-30) into a more efficient version for busy clinicians.

Methods

The researchers employed three distinct types of network models to ensure their findings were robust across different data types:

  • Gaussian Graphical Models (GGM): Used for continuous, normally distributed data.
  • Ising Models: Specifically designed for binary (yes/no) variables.
  • Mixed Graphical Models (MGM): A hybrid approach capable of handling both categorical and continuous variables simultaneously.

To determine which conditions were most "important," they calculated centrality measures:

  • Strength: The total weight of a node's connections.
  • Betweenness: How often a node acts as a "bridge" between other clusters of conditions.
  • Expected Influence: A measure of a node’s role in activating or deactivating other parts of the network.

Finally, they used logistic and linear regression to validate whether their new, shorter index (MMI-25) predicted health outcomes as accurately as the original MMI-30.

Datasets

The study analyzed data from the 2011–2012 Canadian SCI Community Survey.

  • Sample Size: 1,549 participants.
  • Subgroups: 1,137 individuals with Traumatic SCI (TSCI) and 412 with Non-traumatic SCI (NTSCI).
  • Measures: The dataset included the MMI-30, healthcare utilization (HCU) metrics, the SF-12 (for physical and mental health status), and Life Satisfaction (LiSAT-11) scores.

Key Insights and Ideas

  • Autonomic Dysreflexia (AD) as a "Hub": AD—a potentially life-threatening spike in blood pressure—showed high centrality. This suggests that AD isn't just a standalone symptom; it is a structural bridge in the SCI health network. Effectively managing AD could theoretically have a "cascading" positive effect on multiple other conditions like spasticity and lightheadedness.
  • The Upper Limb Cluster: The models revealed a very strong, consistent connection between elbow/wrist problems and shoulder problems. This highlights the "overuse syndrome" common in SCI patients who rely entirely on their arms for mobility and transfers.
  • The "Isolated" Conditions: Surprisingly, conditions like cancer, diabetes, and high blood pressure were relatively isolated in this specific network. While serious, they did not strongly interact with the secondary complications unique to SCI (like pressure ulcers or bowel issues) in driving overall quality of life in this population.
  • Mental-Physical Feedback Loops: The network showed tight clusters linking depression, mental health scores (MCS-12), and overall life satisfaction. This reinforces the idea that mental health is a central pillar that interacts with almost every physical complication.

Key Results

  • Creation of the MMI-25: By removing five "isolated" nodes (cancer, DVT, diabetes, high blood pressure, and liver disease) that didn't contribute to the network structure, the authors created a streamlined 25-item index.
  • Statistical Parity: Regression analysis showed that the MMI-25 was just as effective as the MMI-30 at predicting healthcare utilization, physical health status, and quality of life (p < 0.001).
  • Network Stability: The TSCI network was dense and stable, whereas the NTSCI network was sparser, suggesting that the drivers of health outcomes may differ slightly depending on the cause of the injury.

Limitations and Future Work

  • Directionality: Because the data is cross-sectional (a snapshot in time), the network shows associations but cannot definitively prove causality (e.g., does depression cause pain, or does pain cause depression?).
  • Self-Reporting: The data relies on participant recall, which can introduce bias compared to clinical records.
  • Sample Size: The NTSCI group was significantly smaller than the TSCI group, leading to less stable network estimations for that sub-population.
  • Future Directions: The authors suggest longitudinal studies to see how these health networks evolve over years of living with an injury, which could help identify "early warning" conditions that lead to a downward spiral in health.

Summary updated

Abstract

The network-derived MMI-25 was comparable to the MMI-30 and was associated with inadequate HCU, lower health status, and poor QoL. The MMI-25 shows promise as a follow-up screening tool to identify persons living with SCI at risk of having poor health outcomes.
BibTeX
@inproceedings{fallah2024network,
  author = {Nader Fallah and Heather A. Hong and Di Wang and Suzanne Humphreys and Jessica Parsons and Kristen Walden and John Street and Raphaële Charest-Morin and Christiana L. Cheng and Candice J. Cheung and Vanessa K. Noonan},
  title = {{Network analysis of multimorbidity and health outcomes among persons with spinal cord injury in Canada}},
  booktitle = {Frontiers in Neurology},
  year = {2024},
  doi = {10.3389/fneur.2023.1286143},
}
Figure