Predicting spinal cord injury severity and mortality
Routine blood tests could help predict spinal cord injury severity and mortality risk when tracked over time.
Using data from electronic health records, a research team from University of Waterloo (Canada) has developed machine learning models to analyze standard blood tests as predictive tools for spinal cord injury (SCI) outcomes, including mortality risk, the occurrence of SCI in spine trauma patients and the severity of the injury.
SCIs affected over 20 million people worldwide in 2019, with approximately 930,000 new cases annually, according to World Health Organization data. These traumatic injuries typically require intensive care and present varied clinical patterns and recovery paths, making accurate diagnosis and prognosis difficult, particularly in emergency and intensive care settings.
To overcome these challenges, the research team sought to find a way to utilize routinely collected data, such as blood laboratory values, as predictive markers in SCI.
Abel Torres Espín, professor in Waterloo’s School of Public Health Sciences, explained: “Routine blood tests could offer doctors important and affordable information to help predict risk of death, the presence of an injury and how severe it might be.”
You may also be interested in:
- Impact load and cartilage metabolic markers may predict injury risk in female athletes
- Scaling biomarker discovery: more data, deeper insights with Sapient Bioanalytics
- Implementing AI-driven digital transformation in bioanalysis
To build their predictive tools, the research team began by analyzing hospital data from more than 2,600 US patients, taken during the first 3 weeks after a spinal cord injury, to determine which specific blood markers showed the strongest relationships with patient outcomes. These markers included electrolytes, metabolic indicators like glucose, and complete blood components, such as red blood cells, white blood cells and platelets. The team then used these selected markers to train machine learning algorithms to recognize complex patterns and make predictions based on the changes in these markers over time.
“While a single biomarker measured at a single time point can have predictive power, the broader story lies in multiple biomarkers and the changes they show over time,” noted Marzieh Mussavi Rizi, postdoctoral scholar in Torres Espín’s lab.
The research team successfully developed three machine learning models with remarkable accuracy in three areas. The first model predicts whether a patient will survive their hospital stay, while the second model identifies if a spine trauma patient has sustained actual spinal cord damage. The third model distinguishes between severe spinal cord injuries with complete loss of motor function and less severe injuries where some motor function remains intact.
Notably, these models outperformed the standard Simplified Acute Physiology Score II (SAPS II), a widely used severity assessment tool in intensive care.
The accuracy of all three predictive models also improved over time as more blood test results became available. While other diagnostic tools like MRI and fluid biomarkers can provide objective data, they aren’t universally accessible across medical facilities. Routine blood tests, however, are economical, easily obtainable and available in every hospital.
Espín commented on the significance of the work, stating, “This foundational work can open new possibilities in clinical practice, allowing for better-informed decisions about treatment priorities and resource allocation in critical care settings for many physical injuries.”
