Optimal timing of anticoagulation in traumatic brain injury
A multidisciplinary research team worked to determine when to start anticoagulation in patients with traumatic brain injury, weighing bleeding progression against thromboembolic risk. The analysis found a non-linear relationship between prophylaxis timing and outcomes.
The challenge
Clinicians must balance the risk of intracranial haemorrhage progression against the risk of thromboembolic events, with both affecting patient mortality and outcomes. Current guidelines rest only on Level III evidence, and conflicting research adds uncertainty to practice.
What we set out to do
- Identify the best timing window for starting anticoagulation in TBI patients
- Assess how timing relates to patient outcomes
- Compare the effectiveness of different anticoagulant agents
- Account for patient-specific factors affecting coagulation
- Build a predictive model to support personalised decisions
Our approach
Data architecture analysis
Extracted patient data from the National Trauma Data Bank, cleaned data using protocols for handling missing values, used standardised mean difference analysis to find clinically meaningful imbalances, and categorised patients by when anticoagulation was started.
Technical implementation
Built propensity score models using Gradient Boosting Machines, fitted a Cox proportional hazards model with time-dependent splines, applied restricted cubic splines to model the non-linear effects of prophylaxis timing, and set up a cross-validation framework covering multiple machine learning algorithms.
Solution delivery
Built an interactive Shiny web application for clinicians, developed personalised risk predictions from patient characteristics, and added visualisation of optimised prophylaxis timing.
Results and impact
The analysis found a non-linear relationship between prophylaxis timing and adverse outcomes. Very early prophylaxis (under 12 hours) was associated with hazard ratios below 1, with risk rising in the 12-24 hour window and again after several days. Unfractionated heparin carried a hazard ratio 34% higher than the reference group. The work identified a timing "sweet spot" and patient-specific risk factors, providing an evidence-based framework to inform guideline development.
“This analysis gives us new insight into when to start anticoagulation in TBI, and makes clear that a one-size-fits-all approach is unlikely to be optimal.”
Timeline
- Initial data extraction and analysis: 1 week
- Model development and validation: 1 week
- Manuscript preparation: 1 week
Looking forward
Adding dynamic, patient-level markers of thrombotic and bleeding risk to tailor timing more precisely, running prospective studies with repeated imaging and standardised thresholds, and examining subgroups such as mild versus severe TBI or existing coagulopathy.
